Deep Learning-Based Framework for Leather Surface Defect Detection and Classification Using CNN Architectures

The most critical procedure in the leather processing industry is the detection of surface defects in leather, which helps determine the amount of treated leather that is usable. Thus, identifying and locating leather flaws has a significant influence on leather hide grading. A system for identifying and classifying leather surface flaws is presented in this research. Deep learning models are used by the suggested framework to identify and categorize leather flaws. Convolutional neural networks, or CNNs, are used to identify flaws and categorize their types using the Multi Class Model (MCM) and Binary Class Model (BCM), respectively. Various CNN designs, including LeNet-5, VGG-13, VGG-16, VGG-19, ResNet-50, MobileNet, and MobileNetV2, were employed for the BCM and MCM. Out of all the models, the BCM model's test accuracy was the highest at 93.65% for the VGG-16 architecture, while the MCM model's test accuracy was 81.24% for the MobileNet architecture. The best models have been incorporated into the suggested framework once their performances are compared.