DeepCrackNet: MobileNet-Driven Deep Learning Framework for Image-Based Crack Detection Using Transfer Learning

Authors- B.L.K.Vamsi, K.Sravya, K.V.V.Sai Karthik, L.Niha Jyothi, Y.P.S.Bhavaraja Praneeth, Mrs.P.Satyavathi

Abstract-Crack in infrastructure present significant challenges to public safety, necessitating timely detection for effective maintenance. This research introduces Deep Crack, a novel deep-learning approach for image-based crack prediction. Utilizing the power of Convolutional Neural Networks (CNNs), the study employs the Rfcn_b architecture as its backbone, leveraging transfer learning to enhance crack detection capabilities. The proposed methodology incorporates extensive data preprocessing, including image augmentation techniques to mitigate data scarcity. The model is trained and validated on a diverse dataset, effectively distinguishing images containing cracks from those without. A custom top layer is integrated into the classifier, featuring global average pooling and dense layers to optimize performance. To evaluate the model’s effectiveness, a comprehensive assessment framework is introduced, including confusion matrices and classification reports. The results demonstrate that Deep Crack achieves high accuracy, precision, and recall, validating its potential for real-world infrastructure monitoring and maintenance. This research contributes to the intersection of deep learning and image processing, offering an innovative solution for automated crack detection. The proposed methodology not only highlights the effectiveness of Rfcn_b and transfer learning but also underscores the broader applications of deep learning in civil engineering and infrastructure management.

DOI: /10.61463/ijset.vol.13.issue2.255