Comparative Analysis of AlexNet, ResNet18 and SqueezeNet with Diverse Modification and Arduous Implementation.

Road cracks are caused due to the abundant usage of the roads, heavy traffic, and increased prerequisite of transportation. So road maintenance is an important aspect of such a huge number of vehicles on the roads to have safety measures and continuity. Besides the traffic, bad weather is also contributing its part in creating road cracks. In the proposed research automatic road cracks have been detected, it looks simple apparently but the intensity, and complexity of the background make it a challenging task. In this challenging task, the contrast of the processed image, complexity of different kinds of crack recognition, assembly of proper database images, elapsed time, and approximate classification, etc., are processed to get the optimum results. Deep learning has multiple neural networks among them few networks are used like AlexNet, ResNet18, and SqueezeNet for the data recognition mainly for the huge database having the optimum results in minimum throughput. In the accomplished research 4333 images with eight diverse road cracks classes are used. The classified images are processed by utilizing three different networks through a supervised dataset. Dataset is built by mixing the university resources-collected images and some online datasets. The best expected result is gained after proper training and tested along with classification. In this experiment, the training and testing images were kept the same in epoch and iteration. But the ResNet18 was superlative with an accuracy of 85.20%. [ABSTRACT FROM AUTHOR]