International Journal of Science, Engineering and TechnologySmart Bio Waste Management System: A Review
Authors: Aditya Chanodiya, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Professor Karad V. A.
Abstract: Handling bio-waste becomes harder as cities expand, and a number of efforts now try to make the process smarter through sensors, optimization, better treatment planning and policy design. This review looks at five selected scholarly works that cover different parts of the problem: an IoT-based intelligent waste management system for smart cities, a location-allocation model for treating hotel bio-waste on site, a smart-bin study carried out in ten hospitals, a review of thirteen biowaste treatment technologies, and an edited volume on composting through nexus thinking. The literature suggests that sensing and optimization can cut energy use, data gaps and transport effort, but that these gains depend on what comes after collection: which treatment technology is chosen, where facilities are sited, whether the products can be sold, and whether policy backs the whole chain. We compare the methods and limits of the works, point out gaps, and propose directions for a smart bio-waste management system. Since the review rests on five selected works and not on a database search, it is a focused narrative review and not a systematic one.
Ethical Hacking for Cloud Security: A Comprehensive Review of Methodologies, Tools, Legal Frameworks, Bug-Bounty Ecosystems, and AI-Assisted Penetration Testing
Authors: Ananya Sunil Mengade, Professor Bangar Abhishek Popat, Professor Bhosale S.B.
Abstract: Organisations now run critical services on networks, web applications and cloud platforms that attackers probe continuously. Ethical hacking, the authorised use of attacker techniques to find weaknesses before criminals do, has become a standard part of cyber security practice. This paper presents a focused narrative review of verified sources on ethical hacking for cyber security. It examines four connected themes: the foundations and legal basis of ethical hacking, the testing methodologies and frameworks that structure an engagement, the tools and attack surfaces involved, and newer developments in bug-bounty programmes and large-language-model (LLM) assisted penetration testing. The literature shows that ethical hacking is distinguished from malicious hacking mainly by authorisation and scope, that established guidance such as NIST SP 800-115 organises testing into planning, discovery, attack and reporting, and that reward programmes can be an effective way to obtain vulnerability reports. Early studies suggest LLMs can assist some penetration-testing tasks, but they remain limited and need human supervision. Most sources, however, are descriptive or guidance documents; few report controlled comparisons of methods, and evidence on testing for Indian organisations is sparse. The review contributes a comparative analysis of representative sources, a proposed conceptual seven-layer governance-first engagement framework, and a prioritised list of research gaps.
Smart Umbrella Reminder Using AI and IoT: A Comprehensive Review of Enabling Technologies, Rain-Prediction Models, Existing Designs, and Deployment Challenges
Authors: Aniket Digambar Fartade, Professor Bangar Abhishek Popat, Professor S. B. Bhosale, Dr. Khatri A.A.
Abstract: Forgetting an umbrella is a small problem, but it happens to almost everyone, and ordinary weather apps do little about it because they cannot tell whether you are actually carrying one. Smart umbrellas try to close that gap with sensors, a microcontroller and a phone connection. This paper presents a focused narrative review that synthesises twelve sources: five foundational works on IoT architectures and protocols, Random Forests, LSTM networks and TinyML [1]–[5], four research prototypes of IoT umbrellas [6]–[9] and three commercial or hobbyist designs [10]–[12]. The review examines three connected themes: the technologies and design elements on which a rain-aware umbrella rests, the strength of the evidence for existing designs, and the conditions that matter for deployment. Existing designs mostly detect rain, track location or guide navigation. In the sources we could access, none documents a system that joins a learned rain-probability estimate, on-device inference and carry detection to decide when a reminder is really worth sending, and none reports a head-to-head accuracy test of prediction models on umbrella hardware. The review contributes two comparative tables, a conceptual ESP32-based hybrid edge–cloud architecture and a prioritised agenda of research gaps and future directions. The architecture is a proposal only; it has not been built or tested, and no accuracy figures are claimed for it.
Explainable Artificial Intelligence for Loan Credit Approval: A Literature Review of Transparency, Interpretability, and Fairness
Authors: Ashish Uttam Kakade, Professor Bhosale Sachin Bajirao, Dr. Khatri A. A, Professor Bangar Abhishek Popat
Abstract: Artificial Intelligence (AI) is widely used in loan credit approval to assess applicants' creditworthiness. However, complex AI models often lack transparency. Explainable Artificial Intelligence (XAI) helps explain loan approval and rejection decisions using techniques such as SHAP and LIME. This paper reviews existing research on XAI in credit assessment, focusing on transparency, interpretability, and fairness. It also identifies key challenges and future research directions for developing reliable and understandable AI-based loan approval systems.
A Review on Real Time Security Risk Assessment on Cloud Computing
Authors: Bhor Shravani R., Professor Bangar Abhishek P., Professor Karad Vandana A., Professor Bhosale Sachin B.
Abstract: Cloud computing gives organizations flexible storage, computing power, and access to services, but it also creates security risks that can change quickly. A one-time secu-rity check may miss a new attack, a dangerous configuration change, or a threat that moves from one connected system to another. This paper reviews six supplied re-search papers and develops a practical model for real-time security risk assessment in cloud computing. The reviewed work covers holistic risk scoring across cloud and edge layers, searchable encryption and secure data deletion, artificial neural networks for threat detection, secure network design, AI-based anomaly detection with Zero Trust, and a systematic mapping of cloud risk assessment methods. Based on these sources, the paper proposes a continuous cycle that identifies important assets, col-lects security signals, detects unusual activity, combines technical findings with busi-ness impact, assigns a risk level, and supports an appropriate response. The model uses simple and explainable risk scoring alongside optional machine-learning alerts, identity checks, network segmentation, encryption, and audit records. The paper is a literature-based design and does not claim to have run a new experiment. It also iden-tifies practical challenges, including false alarms, limited visibility, data privacy, model drift, and the cost of continuous monitoring. The main conclusion is that real-time assessment works best when monitoring, vulnerability information, access con-trol, and response actions are connected in one process rather than operated as sepa-rate tools.
AI-Enabled LoRa Mesh Wearable for Personal Emergency Alerts Using Edge AI: A Review and Conceptual Framework
Authors: Bombale Sukanya Vikas, Dr.Anand.A.Khatri, Professor Bangar Abhishek Popat, Professor S.B.Bhosale
Abstract: A personal alarm is useful only if it can send a message at the moment help is needed. A person who has fallen may be unable to reach a button, mobile coverage may be missing in remote or enclosed places, and a cloud service may be out of reach. Progress in embedded machine learning, inertial sensing and low-power long-range radio makes it realistic to build a wearable that screens body movement locally and relays a small alert through neighbouring radio nodes. This paper presents a focused narrative review of eight studies on LoRa mesh routing, wearable fall detection with edge inference, LoRaWAN-based wearable networks, radio localisation and multi-sensor emergency alerting. The studies are discussed under three themes: sensing and on-device classification, multi-hop alert transport, and system-level evaluation. Reported fall-detection accuracies of about 95.6–95.9% and an inference time below 37.84 ms show that local classification is feasible, yet these values come from individual datasets and prototypes and cannot be transferred to a different wearer, sensor or outdoor setting. None of the reviewed work evaluates the complete chain from body-worn motion sensing, through mesh relaying, to notification of a trusted contact. To address this gap, the paper outlines a conceptual seven-layer architecture that combines an IMU-based event score, a manual SOS path that bypasses classification, compact authenticated packets, duplicate-controlled relaying and explicit delivery-state reporting, together with evaluation metrics and a four-stage test plan. The main risks (false alarms, missed events, radio congestion, battery drain, location error, privacy loss and spoofed alerts) are examined. The proposed system is an assistive alert aid, not a guaranteed rescue service or a diagnostic device.
Brain Computer Interface: A Review of EEG Decoding Adaptive Rehabilitation and Edge AI
Authors: Research Scholar Dhome Pranav Pradip, Professor Bangar Abhishek Popat, Professor Bhosale Sachin Bajirao
Abstract: Brain-computer interfaces establish a communication pathway between neural activity and external devices, offering opportunities for assistive interaction, motor rehabilitation, and cognitive-state monitoring. Despite advances in electroencephalography and artificial intelligence, practical deployment remains limited by noisy signals, variation across users, calibration burden, computational constraints, and uneven clinical evidence. This review examines five publications from 2023 to 2026 that address complementary aspects of these challenges: adaptive rehabilitation after stroke, tactile stimulation for cross-subject motor-imagery calibration, convolutional Transformer-based EEG decoding, embedded Edge AI systems, and clinical translation. A structured narrative synthesis is used to compare study objectives, methods, evaluation settings, findings, and limitations. The reviewed evidence suggests that adaptive feedback can support participation by users with poor initial BCI control, while tactile calibration may reduce dependence on demanding imagery tasks. Hierarchical temporal feature learning improves performance on selected cognitive classification benchmarks, but benchmark accuracy does not establish rehabilitation efficacy or reliable operation in everyday environments. Edge deployment introduces additional requirements for energy efficiency, latency, memory use, and data protection. Based on these findings, a conceptual seven-stage EEG architecture is proposed to connect signal-quality assessment, individualized calibration, efficient decoding, confidence-aware feedback, and monitored adaptation. The architecture is not experimentally validated. Future work should prioritize independent participant evaluation, prospective online studies, clinically meaningful outcomes, and transparent reporting of the trade-offs between accuracy, usability, and hardware efficiency.
Overcoming the Static Knowledge Bottleneck in AI Chatbots Using a Dual-Path Hybrid Architecture: A Review
Authors: Doke Sakshi Vitthal, Professor Bangar Abhishek P., Professor Karad Vandana A.
Abstract: AI chatbots built on large language models (LLMs) answer from parametric knowledge that is fixed when training ends, so they cannot reliably reflect updated, institution-specific or access-restricted information unless the model is retrained or connected to external sources. This paper presents a structured narrative review of 17 sources published between 2019 and 2026, centred on five peer-reviewed IEEE Access studies from 2024–2026 and 12 foundational works, organised into seven themes: the static knowledge bottleneck and knowledge injection; retrieval foundations and pipeline design; fine-tuning, retrieval-augmented generation (RAG) and hybrid comparisons; retrieval-bypassing and question-indexed alternatives; corrective and adaptive retrieval; permission-aware retrieval; and local deployment, evaluation and reliability. Reported evidence shows that retrieval substantially improves on closed-book generation, that fine-tuning and RAG have complementary cost and quality profiles, that a combined approach attained the highest semantic scores in the one study that tested it, that similarity-based answer validation and question-indexed retrieval improve grounding on domain corpora, and that per-user access control can be enforced at retrieval time with modest overhead. However, none of the five core studies evaluates a mechanism that routes queries between a parametric path and a retrieval path, none measures answer quality after the underlying knowledge changes, and most validation is limited to one corpus or benchmark. The paper contributes a thematic synthesis, a comparison of twelve representative studies, a proposed conceptual seven-layer dual-path architecture and a prioritised research agenda.
Smart AI-Driven Governance System: AI-Based Multilingual and Intelligent Citizen Grievance Management: A Review
Authors: Dongare Suhana Saibaba, Professor Bangar Abhishek P., Professor Karad Vandana A.
Abstract: Public administrative grievance portals across developing and developed nations remain severely strained by high filing volumes, multilingual heterogeneity, unstructured text submissions, and manual departmental routing. This paper presents a structured narrative review of 22 verified sources published between 2018 and 2026, including state-of-the-art public sector deployable models, organized into seven themes: administrative grievance dynamics and digital governance barriers; multilingual natural language processing (NLP) and dialect translation; automated department routing and multi-label text classification; sentiment analysis, priority ranking, and urgency detection; Optical Character Recognition (OCR) for hand-written citizen petitions; generative AI and Large Language Model (LLM) decision support; and privacy-preserving edge deployment and ethical AI governance. Large Language Models, Fine-tuned Transformer architectures (mBERT, XLM-RoBERTa, IndicBERT), and hybrid classifier ensembles achieve over 92% precision in routing accuracy and priority score allocation. However, literature evidence is dominated by single-jurisdiction pilots, synthetic datasets, and fair-weather operational conditions, while cross-lingual dialectical shifts, low-resource language processing, systemic bias in priority scoring, and data privacy remain under-reported. The paper contributes a comprehensive thematic synthesis, a comparative breakdown of fourteen representative studies, a proposed conceptual seven-layer governance architecture, and a prioritized research agenda for trustworthy public sector AI integration.
Real-Time Prompt Injection and Guardrails for Large Language Model Applications: A Review of Attacks, Defense Mechanisms, and Research Challenges
Authors: Durgude Siddhi Murlidhar, Professor Abhishek Bangar P., Professor Karad Vandana A.
Abstract: Large language models (LLMs) are now embedded in applications that read emails, documents, web pages and images and that can call external tools, which makes prompt injection one of the most important application-level security threats. Static rules and a warning in the system prompt cannot guarantee protection. This paper presents a focused narrative review of five recent primary studies (2024–2026) on prompt-injection detection, guardrails and security evaluation, supported by background works cited within them. The literature is organised into seven themes: the threat landscape; the attack taxonomy; reasoning-based detection; mutation-based universal detection; scannerbased detection; empirical attack simulation; and quantitative security assurance with guardrail requirements. The reviewed studies report that a chain-of-thought (CoT) detector reached 93.20 % accuracy and 90.13 % F1 with a 14B model on 2,176 test prompts; mutation-based detection reached 86.14 % (text) and 82.90 % (image) accuracy on an 11,000-sample dataset; an open-source scanner reached 89 % accuracy on 200 synthetic prompts; plaintext injection succeeded in 20 of 20 trials for three of four techniques on GPT-4o Mini while encoded variants failed; and a CVSSweighted assurance metric ranked four model deployments. However, the evidence is dominated by small, synthetic or single-model evaluations; detectors are compared on different datasets against different baselines; latency and query cost are rarely controlled; and none of the reviewed experiments evaluates detection together with action-level controls in a tool-calling agent. The paper contributes a thematic synthesis, a comparison of the five studies, a conceptual eight-stage layered framework and a prioritised research agenda.
Smart Disaster Response Drone for Rapid Relief Operation
Authors: Gangule Krushna B, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Dr. A. A. Khatri
Abstract: Disasters such as floods, earthquakes, landslides and large-scale accidents can damage roads, power systems and communication infrastructure, delaying search, assessment and relief delivery. Unmanned aerial vehicles (UAVs) can provide rapid aerial observation, survivor-search support, delivery of lightweight emergency supplies and temporary communication coverage. This review synthesizes three supplied papers: a global review of UAVs for search and rescue, a YOLO-based radio-frequency (RF) spectrogram detector, and a lightweight authentication protocol for emergency UAVs. The search-and-rescue literature emphasizes sensor integration, multi-UAV coordination, communication recovery and the persistent constraints of battery endurance and payload. The RF detection study compares YOLOv8n and YOLOv9t, reporting mAP50–95 values around 0.80–0.85 and inference times of 1.6 ms and 3.0 ms, respectively, in its experimental setting. The authentication study addresses mutual authentication and session-key establishment among an operational drone, ground station and rescue drones, using formal protocol analysis. These contributions are relevant but do not individually demonstrate a complete end-to-end disaster-relief platform. This review proposes a conceptual architecture combining multisensor reconnaissance, edge-based AI, task allocation across multiple UAVs, secure communications and human-supervised relief decisions. Major gaps include limited field validation, uncertain performance under severe weather and radio interference, battery and payload trade-offs, and the need to test security and detection jointly in realistic missions.
Machine Learning and Gas-Sensor Arrays for Early Detection of Spoilage-Related Volatiles in Onion Storage: A Critical Review and Conceptual Early-Warning Architecture
Authors: Ganjave Snehal Laxman, Professor Bangar Abhishek P., Professor Bhosale Sachin B., Professor Karad Vandana A.
Abstract: Postharvest losses of stored onion bulbs (Allium cepa L.) arise from storage rots, sprouting, rooting and respiration-driven weight loss, and are often recognised only after visible or odorous decay has spread through a bin or shed. Because infected bulbs emit altered headspace volatiles before symptoms become obvious, continuous gas sensing combined with machine learning (ML) has been proposed as a basis for early-warning monitoring. A structured, PRISMA-informed literature search was carried out through publisher platforms, PubMed Central, Crossref metadata and institutional repositories, prioritising 2022–2026 publications and retaining foundational onion studies; every bibliographic record was verified against an authoritative source, and 29 sources were retained, including 13 onion-specific and 5 related food-spoilage primary studies. The evidence shows that onion storage pathogens, notably Fusarium oxysporum (basal plate rot), Botrytis allii (neck rot), Burkholderia cepacia (sour skin) and Penicillium spp., alter the abundance of organosulfur compounds such as methyl propyl disulfide, dipropyl disulfide and 2,2-bis(methylthio)propane, but many of these compounds are also emitted by healthy bulbs, so no single gas uniquely indicates spoilage. Metal-oxide sensor arrays, ion-mobility instruments and GC-MS-guided pattern recognition have separated inoculated from healthy bulbs with best reported accuracies of roughly 85–100 %, yet almost all onion studies used small numbers of artificially inoculated bulbs in sealed laboratory containers, random rather than grouped data splits, and no long-term drift or humidity compensation. Commercial-scale validation in onion sheds has not been reported in the peer-reviewed literature. The review contributes (i) a disease-to-volatile evidence table, (ii) a comparative analysis that separates onionspecific from analogous evidence, (iii) an eight-layer, literature-justified conceptual architecture for an edgecapable early-warning system with uncertainty handling and a gated validation loop, (iv) a leakage-resistant evaluation protocol, and (v) a prioritised agenda of research gaps that must be closed before such systems can be deployed with confidence.
Integrated Edge AI Framework for Non-Destructive Gold and Cultural Heritage Authentication: A Review
Authors: Ghuge Renuka Sanjay, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Dr. A. A. Khatri
Abstract: This review explores non-destructive testing (NDT) methods for authenticating and characterizing gold jewellery, bullion, and gilded cultural heritage artifacts without causing physical damage to the samples. Because gold is frequently targeted for counterfeiting using materials like copper or tungsten, traditional destructive assays are increasingly being replaced by rapid, non-invasive analytical techniques. This review synthesizes three key research studies by examining low-cost automated detection for bare and packaged gold items using acoustic impact sound frequency analysis and high-resolution image processing with convolutional neural networks and support vector machines, which achieve high classification accuracy. Additionally, it incorporates the application of portable X-ray fluorescence (p-XRF) spectrometry to analyze historic architectural gilded wall ornaments in-situ. By detecting negligible gold content alongside a characteristic zinc and copper proportion, these elemental analyses successfully rule out genuine gold and identify the use of brass-powder-based imitation paint over a lime-plaster substrate. Together, these works highlight the effectiveness, accuracy, and complementarity of NDT methods—ranging from computer vision and acoustics to elemental spectroscopy—for reliable authentication and conservation decision-making.
Intelligent Hybrid Smart Shopping Trolley for Shopping Malls: RFID-Based Automated Billing with Edge–Cloud Analytics and Autonomous Assistance
Authors: Gundal Kirti Sandip, Professor Bangar Abhishek
Abstract: Long checkout queues, manual barcode scanning, billing errors and the physical effort of pushing a trolley through a crowded mall remain persistent problems in organised retail. Existing smart-trolley research addresses these problems in isolation: RFID-based systems automate billing, AI-driven carts add tracking and demand forecasting, IoT and fog-computing architectures reduce latency, and sensor-driven carts add autonomous following. This paper reviews four representative studies and proposes an integrated, low-cost Hybrid Smart Shopping Trolley for shopping malls. The proposed system combines an RFID reader and microcontroller for real-time item identification and billing, an ultrasonic and infrared sensor layer for obstacle-aware assistance, a fog/edge layer for time-critical processing, and a cloud layer for inventory forecasting and customer analytics. A layered architecture, working flow, component design, security model and evaluation methodology are presented. Based on results reported in the reviewed literature, the approach is expected to cut checkout time by roughly 70%, reach about 98% product-identification accuracy, and keep system latency below 200 ms, while remaining cheaper than camera-based cashierless stores. Limitations and future directions are also discussed.
Deepfake Detection Using Hybrid Deep Learning with Spatial-Frequency-Temporal Feature Analysis
Authors: Indore Tanuja Sunil, Bhosale Sachin Bajirao, Bangar Abhishek Popat
Abstract: The rapid advancement of artificial intelligence and deep learning has made it in-creasingly easy to create realistic manipulated images and videos, commonly known as deepfakes. Such manipulated content can create serious risks to digital security, privacy, identity protection, and the reliability of online information. Existing deep-fake detection methods mainly focus on spatial features, frequency artifacts, or temporal inconsistencies, but their performance may decrease when dealing with compressed videos, unseen manipulation techniques, and cross-dataset conditions. This paper proposes a Hybrid Spatial-Frequency-Temporal Siamese Deep Learning Framework for robust deepfake detection. The proposed approach combines spatial features extracted from RGB facial frames using a convolutional neural network, frequency-based features obtained using Error Level Analysis (ELA) and frequency analysis, and temporal features learned from sequential video frames using a bidirec-tional recurrent model. A Siamese learning mechanism is incorporated to improve the separation between authentic and manipulated representations, inspired by recent forgery verification research. The extracted features are fused and classified into Real and DeepFake categories. Grad-CAM is also considered to provide visual expla-nations of the model's decisions. The proposed framework aims to improve detection robustness, generalization, and interpretability under different manipulation and compression conditions.
A Machine Learning, Internet of Things, and Techno-Economic Approaches for Solar-Powered Street Light
Authors: Jadhav Sanchita Chaitanyakumar, Professor Bangar Abhishek P., Professor S.B.Bhosale, Dr.A.A.Khatri
Abstract: Street lighting consumes substantial electricity, and photovoltaic (PV) street lamps are increasingly promoted as a way to cut energy cost and emissions. Their performance, however, depends on weather, the dimming strategy, battery state of charge, and lamp condition, which makes prediction and monitoring central to reliable operation. This paper reviews three recent IEEE Access studies that address complementary parts of the problem: a comparison of five machine learning algorithms for predicting the operating status of PV streetlights, a year-long field trial and artificial-intelligence-assisted techno-economic analysis of a PV-integrated pedestrian streetlamp, and a vehicle-mounted Internet of Things device for mapping light intensity. The studies are summarised, compared on data, methods, and outcomes, and examined for internal consistency and transferability. The review finds that deep models give the highest reported accuracy but at much higher computational cost and on very small datasets, that compact neural networks combined with dimming-aware energy management achieve large grid-energy savings in field conditions, and that mobile sensing offers a scalable route to compliance checking but has not yet been validated against regulatory illuminance standards. Open gaps and research directions are outlined.
Blockchain-Based Supply Chain Transparency for Agricultural Produce: A Comprehensive Review of Distributed Ledger Architectures, Smart Contracts, QR-Code Traceability, and IoT Integration
Authors: Jaydip Dilip Barse, Professor Bangar Abhishek P., Professor Bhosale Sachin B., Professor Karad V.A.
Abstract: Agricultural produce passes through the farmer, processor, distributor, retailer and consumer before it is eaten, and fragmented, manual or centrally held records make it difficult to verify the origin, quality and movement of a product. Blockchain, a distributed ledger that keeps time-stamped, tamper-resistant records, has therefore been widely proposed to improve transparency and traceability in agri-food supply chains. This paper presents a focused narrative review of verified peer-reviewed studies on blockchain-based traceability of agricultural produce. It examines four connected themes: blockchain and distributed ledger foundations; smart contracts for automated, batch-level recording of supply-chain events; consumer-facing verification through QR codes; and integration with Internet of Things (IoT) and radio-frequency identification (RFID) sensing for data capture. The literature shows that blockchain, smart contracts and IoT can be combined into working farm-to-consumer prototypes, and that review articles consistently point to benefits for traceability, trust and food safety. However, most studies are prototypes or conceptual frameworks, and the reviewed work offers limited evidence on large-scale deployment, scalability, cost, farmer adoption, regulatory fit and, above all, the trustworthiness of data entered at the source, since a ledger can preserve but not verify what is written to it. The review contributes a comparative analysis of representative studies, a proposed conceptual architecture linking batch-level ledger records, smart contracts, QR-code access and IoT sensing, and a prioritised agenda of research gaps and future directions.
Secure Data Wiping for Trustworthy IT Asset Recycling: A Critical Review and Conceptual Verifiable-Sanitization Architecture
Authors: Kahadane Shruti Sandip, Professor Bangar Abhishek P., Professor Bhosale Sachin B., Professor Karad Vandana A.
Abstract: Enterprises, schools and households retire large volumes of laptops, servers, phones and storage devices each year, and the recycling or resale of these assets is only trustworthy if the data they held can no longer be recovered. Failures are rarely cryptographic; they are procedural and technical: hidden storage areas are skipped, flash translation layers retain stale copies, wiping is not verified, and no auditable evidence is produced. A structured, PRISMA-informed narrative review was carried out covering standards (NIST SP 800-88 Rev. 1, IEEE 2883-2022), empirical studies of disk sanitization, flash and SSD erasure research, cryptographic erase, and the assured-deletion literature for cloud and IoT storage, including the recent Efficient and Verifiable Scheme for Secure Data Deletion (EVSD) [20] as a related primary study. The evidence shows that (i) single-file overwriting is unreliable on flash media, (ii) device-level sanitize commands and cryptographic erase are the preferred techniques for modern drives but depend on firmware correctness, (iii) verification is the weakest link in practice, and (iv) cryptographic assured-deletion schemes offer verifiable proofs but prove only limited properties and have not been evaluated for physical asset recycling. The review contributes (i) a media-to-method evidence table, (ii) a comparative analysis separating asset-recycling evidence from cloud-deletion evidence, (iii) an eight-layer conceptual architecture for verifiable, auditable data wiping with a gated validation loop, (iv) a leakage-resistant evaluation protocol, and (v) a prioritised research agenda. The architecture is conceptual and has not been implemented or experimentally validated.
AI Smart Allocation Engine For The PM Internship Scheme: A Comprehensive Review Of Scheme Design, Matching Theory, Recommendation Methods, Fairness And Reservation Constraints, And A Conceptual Engine Architecture
Authors: Khilari Atharv Sopan, Professor Bangar A.P., Professor Bhosale S.B., Professor Karad.V.A
Abstract: Large public internship programmes must match very large numbers of candidates to a limited number of company-posted slots while respecting preferences, eligibility rules and priority for under-represented groups. The Prime Minister's Internship Scheme (PMIS) of India, launched as a pilot in October 2024, faces exactly this allocation problem. This paper presents a focused narrative review of verified sources relevant to an AI smart allocation engine for PMIS. It examines four connected themes: the design of the scheme and its reported pilot outcomes, matching theory and allocation mechanisms, recommendation methods used in e-recruitment, and fairness and reservation constraints in Indian centralised allocation. The literature shows that official guidance describes a portal that shortlists candidates using preferences, company requirements and priority groups, that companies made 82,077 offers in Round I of the pilot against more than 1.27 lakh posted opportunities, that only about 33% of offers were accepted over the first two rounds, and that deferred acceptance based allocation with reservations has been used at scale in Indian engineering admissions. However, the exact PMIS shortlisting method is not public, reasons for non-acceptance are not evidenced, and no source evaluates an AI allocation engine for internships. The review contributes a comparative analysis of representative sources, a proposed conceptual seven-layer allocation engine architecture, and a prioritised list of research gaps.
AI Powered Predictive Self-Healing Software Systems: A Review
Authors: Mansi Suryakant Kyatankeri, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Dr. A. A. Khatri
Abstract: Modern software, networks, operating systems and factories have grown too complex to be repaired by hand. Self-healing systems try to notice a problem, find its cause and fix it without waiting for a person. This paper reviews four recent studies that attack this goal at four different levels: finding faulty lines of code with a bee-colony and chaotic genetic search, choosing repair actions for software-defined networks with reinforcement learning, repairing Android, Linux, macOS and Windows machines from system logs, and a large survey of fault diagnosis and self-healing in smart manufacturing. We explain each study in plain language, compare their methods, data and results side by side, and point out where their numbers or claims need careful reading. All four follow the same basic loop of monitor, detect, diagnose and repair, and all report clear gains over simple baselines. However, most results come from simulations, emulators or benchmarks rather than live systems, multiple simultaneous faults and slowly growing faults are handled weakly, and almost nothing acts before a failure happens. We close with research gaps and directions for predictive, safe and explainable self-healing software.
BLE Beacon-Based Micro-Tracking Systems: A Review of Positioning Methods, Signal Processing, Applications and Research Challenges
Authors: Nalawade Sneha Sunil, Professor Bangar Abhishek Popat
Abstract: Finding people or objects inside a building is not always straightforward. GPS works well outdoors, but walls and roofs weaken its signals indoors. This creates a need for affordable systems that can identify the room or area where a person or item is located. Bluetooth Low Energy (BLE) beacons are small, battery-powered transmitters that offer one practical option. This review examines 17 sources published between 2012 and 2024 on BLE-based indoor positioning and micro-tracking. The studies are organised around positioning techniques, RSSI fingerprinting, signal filtering, machine learning, Bluetooth 5.1 direction finding, and applications in places such as museums and hospitals. Overall, fingerprinting tends to perform better than estimating position from signal strength alone. In one testbed, 95% of position estimates were within 2.6 metres when the beacons were placed relatively close together. Filtering and machine-learning methods can make estimates more stable, while Bluetooth 5.1 direction finding may support more precise positioning. However, signal strength is affected by people and the surrounding environment, and many studies test their systems at only one site for a limited period. Battery replacement, ongoing maintenance and privacy also receive less attention than accuracy. The review compares twelve representative studies, summarises recurring research gaps and suggests practical steps towards reliable systems that can operate beyond controlled experiments.
Low-Cost AI-Based Multimodal Sensor Fusion System for Pesticide Contamination Risk Screening in Fruits and Vegetables
Authors: Naykodi Sanika Dnyaneshwar, Bhosale Sachin Bajirao, Bangar Abhishek Popat, Karad Vandana A
Abstract: The extensive and uncontrolled use of agrochemicals in contemporary agriculture has sparked serious worries about consumer health and food safety. Conventional laboratory-based analytical methods, like high-performance liquid chromatography (HPLC) and gas chromatography-mass spectrometry (GC-MS), offer high precision but are limited by high costs, time-consuming sample preparation, produce destruction, and a lack of portability for regular users. This research examines the architecture, techniques, and integration of low-cost, AI-driven multimodal sensor fusion systems intended for real-time pesticide contamination risk screening in order to close this gap. Multimodal frameworks greatly improve classification accuracy by storing fruit or vegetable samples in a controlled chamber and combining heterogeneous inputs, namely optical-spectral signatures from an AS7341 multispectral sensor and volatile organic compound (VOC), temperature, and humidity telemetry from a BME680 sensor aggregated via an ESP32 microcontroller. We look at the underlying machine learning pipelines, discuss important technical issues including data heterogeneity and sensor cross-sensitivity, and suggest future paths for edge-computed, mobile-integrated food safety screening.
Smart CCTV Surveillance using Edge AI and Blockchain: A Review
Authors: Pawar Sakshi Ajinath, Bangar Abhishek Popat, Bhosale Sachin Bajirao, Khatri Anand A.
Abstract: The rapid expansion of smart cities and urban populations has created an urgent demand for real-time, highly secure, and privacy-preserving public surveillance systems. Traditional CCTV surveillance architectures heavily rely on centralized cloud servers, making them vulnerable to single points of failure, network latency, data tampering, and severe privacy breaches. To address these critical limitations, this paper proposes a secure, decentralized surveillance framework that integrates Edge AI with private blockchain technology and InterPlanetary File System (IPFS). The proposed system utilizes lightweight deep learning models—such as YOLOv8 for high-precision real-time object detection, DeepSORT for multi-object tracking, and ArcFace for robust facial recognition— operating locally on edge devices to achieve low-latency processing without relying on centralized cloud infrastructure. To ensure strict data confidentiality and integrity, a hybrid cryptography model combining AES-256 for video metadata encryption and RSA-2048 for secure key management within a Trusted Execution Environment (TEE) is implemented. Furthermore, a private Ethereum blockchain leveraging Quorum Byzantine Fault Tolerance (QBFT) consensus and smart contracts is deployed to enforce role-based access control (RBAC) and maintain tamper-proof, immutable audit trails. Experimental results demonstrate that the proposed edgeblockchain integration significantly improves detection precision, ensures real-time performance at 60 FPS, and provides robust resistance against data tampering, offering an efficient and secure solution for next-generation intelligent urban surveillance.
Early Stroke and Panic Detection using AI in Neurology
Authors: Sachin Phuldev Yadav, Professor Bhosale S.B, Professor Bangar A.P, Professor A.A. Khatri
Abstract: Neurological emergencies span a wide spectrum, from sudden vascular events such as ischemic stroke to acute psychiatric-neurological crises such as panic attacks. Both categories are time-sensitive and both currently rely heavily on manual clinical assessment, which limits the speed and consistency of detection. This paper presents a detailed review of six recent peer-reviewed sources (2024–2025) on artificial intelligence (AI) in neurology, and uses their combined findings to build a case for, and a design of, an integrated two-module early-detection framework: an imaging-based deep-learning pipeline for stroke triage, and a wearable-sensor-based sequence-modelling pipeline for panic/anxiety detection. The review covers foundational AI concepts (machine learning, deep learning, neural network architectures), the current clinical evidence base for stroke-imaging AI, the comparatively nascent evidence for physiological/behavioural AI in affective and psychiatric-neurological states, and the shared ethical, technical, and regulatory challenges — interpretability, bias, data privacy, and validation — that recur across every source reviewed. The paper then details a proposed system architecture, dataset and preprocessing requirements, candidate model architectures, an evaluation plan with specific metrics, expected outcomes, and a discussion of limitations and future work. The overall conclusion is that stroke-detection AI is a mature and clinically validated field that can serve as a design template for a still-emerging class of wearable, sensor-driven AI tools for detecting acute panic and anxiety states.
An AI-Based Early Detection, Risk Assessment, and Decentralized Security Architecture for Preventing Question Paper Leakage in High-Stakes Competitive Examinations
Authors: Samiksha Sanjay Jadhav, Professor Bangar Abhishek P, Dr. Anand Khatri A., Professor Bhosale Sachin B.
Abstract: Leakage of question papers before a high-stakes competitive examination damages fairness, wastes institutional resources, and erodes public trust. Most technical research on online-exam integrity, however, concentrates on what happens after the paper reaches the candidate. This review examines three recent IEEE Access studies that represent three complementary defensive ideas: blockchain-backed storage of exam answers and results, screenshot-based activity monitoring in bring-your-own-device (BYOD) exams, and a multi-stage agent that joins IP-address screening with a deep-learning behaviour classifier. Each study is summarised, compared on technique, evaluation scale, and reported outcomes, and critically assessed. The comparison shows that decentralized tamper-evidence, behavioural analytics, and risk scoring are individually mature enough to be useful, yet none of the reviewed systems protects the question paper itself during preparation, storage, and distribution. The review closes by identifying the resulting research gaps and outlining directions towards an integrated architecture that unites AI-based early detection, risk assessment, and decentralized security for preventing question paper leakage.
Artificial Intelligence and IoT-Enabled Chemical Pollution Monitoring and Remediation of Rivers: A Comprehensive Review of Sensing Technologies, Machine Learning Methods, and IoT Integration
Authors: Sarthak Sunil Khandagale, Professor Bangar Abhishek Popat, Professor S. B. Bhosale, Dr. A.A.Khatri
Abstract: Rivers receive chemical contamination from industrial effluent, domestic sewage and agricultural runoff, including nutrients such as phosphorus, organic load, and other pollutants. Conventional assessment relies on periodic sampling and laboratory analysis, which needs specialised staff, chemicals and expertise, and therefore cannot give continuous, early warning of pollution events. Artificial intelligence (AI), combined with low-cost sensing and the Internet of Things (IoT), is increasingly proposed for continuous river monitoring and for the data-driven control of chemical treatment such as coagulant dosing. This paper presents a focused narrative review of studies that use machine learning (ML) and deep learning (DL) with IoT sensing for river water quality monitoring and chemical-treatment control. It examines three connected themes: sensing technologies, learning methods, and IoT architectures. The reviewed literature reports water-quality classification accuracy of up to 98 % with an Arduino-based IoT device, XGBoost-based prediction in an industrially affected river basin, ML-assisted phosphorus sensing in river samples, and a deep-learning coagulant-dosing model that reduced dosage by about 22 % in a plant study. However, most studies are small, single-site, short-term and laboratory-oriented, and few report field validation, sensor fouling and drift, or closed-loop treatment control in rivers. The paper contributes a comparative table of representative studies, a conceptual river-to-decision architecture (not experimentally validated), and a prioritised list of research gaps.
AI PCs and On-Device Artificial Intelligence: The Future of Intelligent Computing
Authors: Sasane Sakshi Shirish, Professor Bhosale S.B, Professor Bangar A.P, Professor Khatri V.A
Abstract: AI PCs and on-device artificial intelligence are changing how everyday computers use AI. Instead of sending every request to a remote cloud server, a device can process suitable tasks locally using its CPU, GPU, and specialised neural processing unit (NPU). This can reduce response time, limit data transfer, support some functions without a stable internet connection, and give users more control over personal data. The practical value of these benefits depends on the task, the model, and the computer's hardware. Local processing also has to work within limits such as memory capacity, power consumption, heat, and model size. This review discusses three research works: a comprehensive survey of on-device AI models, a study proposing a collaborative consumer edge-AI ecosystem, and research on on-device AI for smart environments and Industry 4.0. The papers are compared by focus, approach, contribution, and limitation. The review identifies model optimisation, efficient hardware use, privacy-aware collaboration, and energy management as important areas. It also finds that the selected works do not establish a common benchmark specifically for AI PCs. A task-aware local/cloud architecture is proposed for future study; it has not been experimentally tested.
AI and IoT-Enabled Smart Helmets for Accident Detection, Health Monitoring, and Automatic Emergency Alert Systems
Authors: Saurav S Borase, Professor Abhishek Popat Bangar
Abstract: Motorcycle accidents remain a leading cause of death on the roads, particularly in developing countries where two-wheelers are the primary means of transportation. The delayed emergency medical response and lack of information about the victim's health status increases the risk of death during the crucial first hour following the incident. WHO estimates that Powered two-wheelers (PTWs) account for nearly 25% of global road deaths, with some Asian countries witnessing even higher rates due to intense traffic and insufficient protective infrastructure.. Researchers have proposed various Internet of Things (IoT) and sensor enabled smart helmets to address these issues. Previous works use a combination of motion sensor and communication modules coupled with micro-controllers to identify accidents and relay user location information. However, a deeper inspection of the recent literature suggests pertinent limitations with the proposed solutions. While most works demonstrate successful crash identification using accelerometers and force sensors, there was an apparent lack of integration of biometric monitoring (heart rate and SpOn), intelligent AI driven severity assessment tool, cloud assisted medical record retrieval and robust false alarm mitigation mechanisms. This paper presents a detailed analysis of five IEEE research papers published in 2023-2025 that explore smart helmet design. The papers examined various existing works related to smart helmet based accident detection systems, user health monitoring and emergency alert relaying mechanisms. The review highlights potential research opportunities based on the gaps identified in the surveyed works. Leveraging the lessons learned from the review and the opportunities highlighted, this work also proposes an original AI driven smart helmet system design. The proposed model uses a combination of sensors for accident detection, biometric health monitoring and alcohol content estimation. The model also implements an AI based impact assessment algorithm and intelligent dual mode emergency message relaying system. The observations made in this review can form a robust foundation for the development of next generation smart helmet systems that canreduce post-crash mortality rates due to timely emergency response
AI Diffusion Models for Image Generation: A Review
Authors: Shaikh Iqra Arif, Professor Bhosale Sachin Bajirao, Professor Bangar Abhishek Popat
Abstract: Diffusion models have become an important approach for generating images from text descriptions, random noise, and other conditions. They work by learning how to gradually remove noise from data until a meaningful image is formed. This review studies five publications from Springer and IEEE related to diffusion-based image generation: a broad survey of image-generation applications, a survey of lightweight diffusion models, the original Latent Diffusion Models paper, a survey of diffusion models in computer vision, and a review of text-to-image generative models. The selected works are compared in terms of their focus, main contribution, practical value, and limitations. Together, they show that image- generation research has progressed in three connected directions: improving visual quality and controllability, reducing computation, and better measuring how closely generated images follow user prompts. The review also identifies open issues involving evaluation, computational cost, dataset bias, privacy, copyright, and responsible use. This is a literature-based review; it does not report new experiments or claim that the five papers were tested under identical conditions.
Federated Learning For Fraud Detection: A Review
Authors: Research Scholar Shivam Babasaheb Rakh, Professor A. P. Banger, Professor S. B. Bhosale, Professor V. A. Karad
Abstract: Fraud detection in banking and payments depends on learning from transaction data that institutions are often unable or unwilling to pool, because of privacy, regulatory, competitive, and security constraints. Federated learning (FL) offers an alternative in which institutions train a shared model by exchanging model updates rather than raw records. This review synthesizes selected scholarly works on FL for fraud detection, covering credit-card and payment fraud, financial-crime detection with graph learning, aggregation methods, non-IID data, class imbalance, differential privacy, secure aggregation, poisoning attacks, communication overhead, concept drift, and real-world deployment. Fraud-specific federated studies are compared with foundational FL, privacy, security, and fraud-detection literature. The literature suggests that FL is a plausible route to collaborative fraud detection, but that its benefits in the fraud setting depend on how well it handles extreme class imbalance, heterogeneous institutional data, evolving fraud patterns, and layered privacy and security protections. Many enabling results were obtained in non-fraud settings and should not be assumed to transfer unchanged. Because the review is based on a selected set of publications rather than an exhaustive database search, it is presented as a focused narrative review and not as a systematic review.
AI and IoT-Enabled Safety Monitoring for Mines: A Comparative Review of Gas Detection, Wearable Sensing and Integrated Smart-Mining Frameworks
Authors: Shivani Nandu Shinde, Bangar Abhishek Popat, Bhosale Sachin Bajirao
Abstract: Underground mines are hard places to keep safe. The air can turn dangerous without any smell or warning sign, communication is patchy, and help is usually a long way off. Over the last few years many research groups have tried to close this gap by combining low-cost sensors, wireless links and machine learning. This paper reads four recent studies from 2026 side by side: a LoRa-based gas monitoring node with Random Forest analysis, a wearable wristband and belt hub with edge intelligence, a Random Forest gas classifier with a Python dashboard and cloud alerts, and a large integrated smart-mining case that joins electrification, safety analytics, biodiversity and accounting. We compare their sensing, communication, analytics and alerting layers, and we recompute several figures directly from the numbers the authors report. Our reading is that the hardware ideas are sound and increasingly affordable, but the evidence behind them is thin. Datasets are rarely described, labels are often derived from the same thresholds the model is meant to improve on, and underground field trials are scarce. We close with a layered reference framework and a short list of practical recommendations for future prototypes.
Smart Pothole Detection and Reporting System: A Comprehensive Review of Sensing Approaches, Deep-Learning Detectors, Empirical Evidence, and Reporting Challenges
Authors: Tambe Harshal Maruti, Professor A.P. Banger, Professor S. B. Bhosale, Professor V. A. Karad
Abstract: Potholes wear out vehicles, hold up traffic and cause crashes, yet manual road inspection is slow, expensive and cannot provide a continuous, geotagged record. This paper is a focused narrative review of five IEEE and IET studies on automated pothole detection: a comparison of YOLOv3, SSD, HOG with SVM and Faster R-CNN; a study that fine-tunes four deep detectors on road images from around the world and runs the chosen detector on a real vehicle that sends alerts to an Internet of Things platform; a convolutional-network prototype for autonomous vehicles; an assessment of R-CNN and SSD MobileNet through an object detection API; and a ResNet-50 system that detects potholes and estimates their depth. The review looks at three linked themes: the sensing approaches and detection models in use, how strong and how comparable the reported evidence is, and what matters when detection is stretched into a reporting pipeline. Only one of the studies could be read in full. It reports a mean average precision of 82.02% (IoU 0.4) for Faster R-CNN ResNet101 and 5–6 frames per second for a lighter model on an embedded automotive platform. The other four were judged from their published abstracts alone; among other figures, they give 99.02% accuracy for a prototype CNN and a maximum detection confidence of 93%. Because these numbers rest on different data and metrics, they cannot be ranked against one another. Just one study carries detections onward to a cloud platform. The review offers a comparative table, a conceptual architecture for the proposed pothole detection and reporting system (not experimentally validated) and a prioritised list of research gaps.
A Review of Reversible Logic-Based Architectures for Ultra-Low-Power CPUs
Authors: Tambe Pranjal Machhindra, Bangar Abhishek Popat
Abstract: Energy dissipation, rather than transistor density, now limits the performance of general-purpose processors, and conventional complementary metal-oxide-semiconductor (CMOS) logic is approaching practical efficiency limits because every erased bit and every abruptly charged node releases energy as heat. Reversible computing addresses this problem by preserving information throughout a computation so that, when combined with adiabatic switching, most of the signal energy can in principle be recovered rather than dissipated. This paper presents a focused review of reversible logic-based architectures for ultra-low-power central processing units (CPUs). Literature drawn from IEEE Xplore, SpringerLink, ScienceDirect, American Physical Society and Nature Portfolio journals, and related sources is organized into four themes: thermodynamic foundations and reversible gate libraries; reversible instruction-set and processor architectures; physical implementation technologies, including adiabatic CMOS, adiabatic superconductor logic, ballistic fluxon logic and quantum-dot cellular automata; and evaluation practice. The comparative analysis shows that logically reversible processors have been specified and simulated, and that adiabatic superconductor microprocessors have been demonstrated, but no reviewed study reports measured system-level energy savings for a fully reversible CPU. Key gaps include efficient resonant power-clock generation, leakage and cryogenic overheads, garbage management and inconsistent evaluation metrics. A conceptual layered architecture is proposed to frame these challenges and guide future research.
VisionMate: Multimodal AI Smart Glasses for Visually Impaired People: A Review
Authors: Tavhare Sakshi Hemant, Bangar Abhishek Popat, Bhosale Sachin Bajirao
Abstract: Visually impaired individuals face daily mobility and information-access challenges that traditional aids, such as white canes or guide dogs, cannot fully resolve, particularly regarding dynamic obstacles, environmental hazard detection, and reading printed text. This paper presents a structured overview of the VisionMate Multimode AI Smart Glasses ecosystem, a wearable assistive technology designed to bridge these gaps through real-time computer vision, optical character recognition, GPS mapping, and audio feedback. The system architecture distributes workloads between a lightweight, camera-enabled smart glasses frame and a companion mobile processing unit running offline-capable deep learning models. Key functional modes include real-time obstacle detection with alert throttling, instant text-to-speech conversion using OCR pipelines, spatial GPS navigation, and emergency SOS integration. However, assistive wearable deployments face ongoing design constraints, including battery limitations, processing latency, and the need for seamless, hands-free accessibility controls. The system demonstrates how multimodal AI integration can enhance autonomy, spatial awareness, and independence for users with visual impairments.
AI-Based Sign Language Recognition and Conversion to Text Using Computer Vision: A Review
Authors: Tejas Shantaram Jadhav, Professor Bangar A.P., Professor Bhosale S.B., Dr. A. A. Khatri
Abstract: This paper reviews recent computer-vision approaches that convert sign language into text. Sign language is the main means of communication for people with hearing and speech impairments, yet most hearing people cannot understand it. Four recent IEEE Access papers are studied: a comprehensive survey of traditional and deep-learning recognition methods, a two-stream graph convolution network with attention and residual connections, a Vision Transformer for static Indian signs, and a lightweight pose-based graph convolution pipeline called SignGraph. The review compares their input modalities, models, datasets, accuracy and computational cost. The study shows that skeleton (pose) based methods built on MediaPipe-style landmarks give accuracy close to or better than RGB-based models while needing about twenty times fewer parameters, which suits real-time webcam use. Open problems such as large vocabularies, signer independence, non-manual features and the scarcity of Indian Sign Language data are discussed, and directions for future work are suggested.
AI-Powered Smart Ticketing and Bus Management System for MSRTC: A Review
Authors: Unde Manasi Ramdas, Bhosale Sachin Bajirao, Banger Abhishek Popat, Khatri Anand A.
Abstract: The Maharashtra State Road Transport Corporation (MSRTC) runs one of India's largest state bus networks and has recently added digital ticketing machines, UPI payments, RFID smart cards and GPS-based vehicle tracking. This paper reviews selected work on electronic ticketing, real-time bus tracking, bus arrival time prediction, passenger demand forecasting and AI-based fleet management, and relates it to MSRTC's operating context. The reviewed studies use GPS traces, smart-card records and neural-network or ensemble models. Reported results are encouraging, but a direct comparison is difficult because the studies use different cities, datasets and metrics, and almost none are tested on Indian state-transport conditions. The review highlights gaps in data availability, evaluation practice, rural connectivity and integration with existing ticketing systems, and proposes a conceptual system workflow for future work.
Image-Based Breed Recognition System for Cattle and Buffaloes of India: A Review of Computer Vision and Deep Learning Approaches
Authors: Vaishnavi Dattaray Nalawade, Professor Bangar Abhishek Popat, Professor Bhosale S.B.
Abstract: Reliable breed identification underpins conservation, breeding, registration and traceability in India’s cattle and buffalo sectors, yet it still depends largely on expert visual judgement. This review examines how computer vision and deep learning have been applied to image-based breed recognition of Indian cattle and buffaloes. A structured narrative search was carried out in October 2026, and 17 records were examined: studies on Indian breeds, adjacent non-Indian breed-classification and identification work, a public dataset, and an earlier systematic review of cattle identification. Reported accuracies for Indian cat-tle range from 72.5% for the visually similar Tharparkar and Hariana pair to about 95% for a lightweight CNN on four zebu breeds, while a YOLOv12 sys-tem covering 15 breeds reported 69.8% mAP at IoU 0.5. These figures cannot be compared directly because datasets, class definitions, data splits and met-rics differ. No peer-reviewed, breed-level study of Indian buffaloes was identi-fied, and the buffalo evidence is limited to a muzzle-based individual-identification study from Türkiye. Recurrent weaknesses include small and imbalanced datasets, unclear animal-level splitting, absent external validation, internal inconsistencies in reporting, and little attention to crossbreds, explain-ability and low-cost deployment. We synthesise these gaps and outline a con-ceptual, not yet validated, framework that combines image-quality control, multi-view fusion, uncertainty-aware output and expert verification.
Remote Classrooms for Rural Colleges: A Comprehensive Review of Delivery Approaches, Low-Bandwidth Techniques, Empirical Evidence, and Deployment Challenges
Authors: Research Scholar Wable Sarthak Sandip, Professor A. P. Banger, Professor S. B. Bhosale, Professor V. A. Karad
Abstract: Rural colleges often have too few subject-expert teachers, limited laboratories and few specialised courses, and an unreliable, low-bandwidth internet connection makes standard video teaching hard to keep running. This paper is a focused narrative review that brings together five studies of remote and distance classrooms in rural and developing-world settings: a low-bandwidth virtual classroom assembled from Google Docs and Skype (IEEE, 2010); IIT Bombay's distance education through satellite, video conferencing and webcast (IEEE Frontiers in Education, 2008); live classroom sharing between urban and rural schools in Karnataka (IATED, 2013); the Digital StudyHall project on facilitated video instruction in rural India (ICTD, 2010); and video conferencing over analogue telephone lines for rural technology education in South Africa (ASEE). The review looks at three linked themes: how lessons are delivered and how bandwidth is managed, how strong and how comparable the reported evidence is, and what matters when these ideas are carried over into the proposed Remote Classroom system. None of the five studies could be read in full, so each was assessed from its abstract, catalogue record or partial text excerpt. The only quantitative results available come from a perception survey in the classroom-sharing study: 92% of respondents agreed that urban and rural schooling differ in quality, 92% of rural schools were willing to take part, and 82% believed that classroom sharing would help rural students if it were done well. These numbers capture attitudes, not learning outcomes. In the material accessed, none of the studies deals with adaptive bitrate streaming, H.265/AV1 compression, audio-first prioritisation, edge caching or cloud and IoT infrastructure. The review offers a comparative table, a conceptual architecture for the proposed system (not experimentally validated), and a ranked list of research gaps.
Secure Multi-Cloud Data Storage with Encryption: A Review
Authors: Yash Sukhdev Jadhav, Professor Bangar A.P., Professor Bhosale S.B., Dr. A. A. Khatri
Abstract: Storing data with a single cloud provider creates a single point of failure and ties an organisation to one vendor’s security posture. Multi-cloud storage spreads data over several providers, but it also multiplies the places where keys, ciphertexts and integrity evidence must be protected. This paper reviews four recent research contributions that approach the problem from complementary angles: hybrid authenticated encryption with sliced storage (CloudLock), publicly verifiable equality testing over encrypted replicas (PVPKEET), blockchain-coordinated, collusion-resistant consistency verification (MDCV), and password-protected key management that survives provider collusion (PECS). We compare their system models, threat assumptions, cryptographic tools, reported costs and limitations, and we outline open issues such as dynamic data support, trusted middleware, and post-quantum readiness.
AI-Based Smart Cargo Booking and Empty Return Load Matching System: A Review of Rule-Based, Optimization, and Hybrid Matching Approaches
Authors: Yevle Pruthviraj Govardhan, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Professor Karad V. A.
Abstract: Booking cargo takes far more than listing loads. Shippers, carriers, vehicles, routes and time windows all have to fit together under firm rules and cost goals, while many trucks still return empty after delivery. This paper looks at five publications relevant to the problem: a review of the causes of empty miles in freight trucking, a two-part survey of pickup and delivery problems (including vehicle routing with backhauls), a study of dynamic pickup and delivery problems, and a seaport truck-sharing model that matches loads to empty truck slots. Between them they draw on search-and-matching economics, vehicle routing with backhauls, dynamic re-planning and computer-based load matching. The trucking review reports an average empty-mile share of 29% across 54 collected estimates, with matching technology lowering empty-trip shares by 4 to 5 percentage points in some studies. Overall, the papers suggest that no single method suits every platform. A practical smart booking system should pair checked input data and strict hard-constraint enforcement with a feasibility-first, improvement-second matching strategy, optional AI-based prediction, open evaluation and human review. Key open problems are scalability, benchmark coverage, fair evaluation, real-time re-matching, and the risk of accepting a generated match without checking it.