Automated Asphalt Crack Detection Using Convolutional Neural Networks
Dr. Kanmani P1, Anushka S2, Anushka Agarwal3
1Department of Data Science And Business Systems, SRM Institute of Science and Technology, Chennai, India
2 Department of Data Science And Business Systems, SRM Institute of Science and Technology, Chennai, India
3 Department of Data Science And Business Systems, SRM Institute of Science and Technology, Chennai, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Road damage caused by asphalt cracks is a significant issue in the civil engineering industry as it poses a threat to road and highway safety. Detecting and classifying cracks is a difficult undertaking because of the intricate pavement conditions created by various factors such as shadows, oil stains, and water spots. These factors can create challenges in differentiating cracks from the surrounding pavement. The focus of our study was to put forward a architecture of a deep convolutional neural network (DCNN) that can automatically detect and categorize pavement cracks. To train DCNN, we utilized RGB images of pavement cracks that were captured manually with a resolution of 1024x768 pixels. These images were then segmented into patches measuring 32x32 pixels. During the training of the DCNN, we employed two filter sizes, which were 3x3 and 5x5. Our presented approach achieved a recall of 98%, precision of 99%, and accuracy of 99%, successfully detecting the presence of cracks in the images. The DCNN was also capable of classifying With fair classification accuracy for both filter sizes and no noticeable difference in accuracy between the two filter sizes, pavement cracks into no cracks, transverse, longitudinal, and alligator. In contrast to bigger filter sizes, smaller filter sizes required greater processing time during training. Overall, 94.5% accuracy was achieved while using our suggested method to classify different kinds of cracks.
Key Words: Crack detection, DCNN, Faster R-CNN, Unet, YOLO, ResNet-50