Proceeding ICSEMT June 2026

12 May
Posted on Author

An Arduino-Based Integrated System For Environmental Control And Intrusion Detection In Smart Agriculture

Authors: Kulkarni Mrunmayee Sanjay, Prof. Dr. Shirish Kulkarni

Abstract: This research provides the design, development, and evaluation of an affordable smart agriculture system using an Arduino Uno microcontroller. The system includes real-time environmental monitoring—temperature, light intensity, and soil moisture—coupled with a local security aspect. A digital temperature sensor, e.g., the DHT11, and a Light Dependent Resistor (LDR) measure the ambient conditions, while a soil moisture sensor measures the moisture levels in the soil. An automatic cooling fan, regulated by a relay module, is powered to mitigate excessive temperature, thereby providing an ideal microclimate. In the security aspects, an Infrared (IR) sensor detects illegal access, which causes an alarm through a buzzer. Experimental verification verifies the system performance in providing favorable environmental conditions while issuing timely alerts for illegal access. This combined approach offers a realistic and general solution for small to medium-scale agribusinesses, thus improving crop management and farm security.

DOI: https://doi.org/10.5281/zenodo.20132972

Virtual Canvas: A Dual-Pipeline Benchmark Of MediaPipe And YOLOv11-Pose

Authors: Jatin Jain, Dr. Sakshi Indolia

Abstract: This paper presents Virtual Canvas, a real-time touchless drawing application that simultaneously executes two hand-pose estimation pipelines — Google MediaPipe Hands and a pre-trained YOLOv11-Pose model — on every captured webcam frame. The dual-pipeline architecture eliminates input variance between models, enabling a controlled side-by-side experimen- tal comparison under identical real-world conditions. Across 11 sessions spanning 10 days of live evaluation on CPU-only hardware, 1,391 performance samples were captured at 500 ms intervals via automated CSV logging, covering inference latency, frames per second, CPU utilisation, hand detection counts, and lighting conditions. Results demonstrate that MediaPipe achieves a 2.35× lower mean inference latency than YOLOv11-Pose (t = −120.68, p < 0.0001, Cohen’s d = 4.58). Under dim lighting, YOLOv11-Pose inference variance increased by 133% while MediaPipe remained stable, though MediaPipe latency itself rose by 18.2%. YOLOv11-Pose exhibited systematic over- detection, reporting two hands in 81.6% of single-hand frames. Exponential Moving Average (EMA) smoothing (α = 0.35) and 5-frame gesture debouncing enabled fluid drawing interaction despite sub-5 FPS dual-pipeline throughput. The system provides a practical, data-driven benchmarking framework for selecting between lightweight pre-trained detectors and heavier single- stage models in human-computer interaction applications.

DOI: https://doi.org/10.5281/zenodo.20135316

FairScan: A Dual-Stage Bias Detection And Mitigation Framework For Machine Learning Classification Models

Authors: Shubhi Bhardwaj, Dr. Yatu Rani

Abstract: Bias in machine learning models is one of the biggest concerns in today’s AI-driven world. When models are trained on data that reflects real-world inequalities, they end up making unfair predictions that can harm people based on their gender, race, or age. This paper introduces FairScan, a two-stage framework designed to first detect and then actively reduce bias in classification models. The detection stage uses a new metric called the Statistical Parity Divergence Score (SPDS), which measures bias not just across individual groups but also at the intersections of multiple sensitive attributes. The mitigation stage applies a custom training strategy called Reweighted Fair Gradient Descent (RFGD), which adjusts how much the model learns from different groups during training to push it toward fairer outcomes. We tested our approach on the UCI Adult Income dataset and found that FairScan reduced the Demographic Parity Difference by up to 79.4% while maintaining a classification accuracy of 86.7%. Our results show that it is genuinely possible to build models that are both accurate and fair, which is a step forward for responsible AI development.

DOI: http://doi.org/10.5281/zenodo.20226534

AegisIDS: An Adaptive Hybrid Intrusion Detection System For Intelligent Cyber Defense

Authors: Muskan, Dr. Yatu Rani

Abstract: The evolution of cyber threats requires security methods that are smarter, more adaptive, and tailored to the unique properties of web technology beyond the capabilities of traditional IDS. AegisIDS [10] – An adaptive hybrid intru-sion detection system combining signature based and machine learning-driven anomaly detection for greater accuracy and responsiveness. The new system has been proposed using several techniques such as dynamic data sampling technique, optimized feature selection, and ensemble learning to solve problems related to class imbalance, false positive rate and detection latency. AegisIDS is built to work well for today environments including Cloud, Internet of Things (IoT), and enterprise network. Ex-perimental insights from recent hybrid IDS studies demonstrate that combining adaptive learning with hybrid architectures significantly improves detection rates and reduces false alarms. This paper discusses the architecture, methodology, performance considerations, and future scope of AegisIDS.

DOI: https://doi.org/10.5281/zenodo.20227328