Vision Based Leather Surface Defect Detection & Classification using Convolutional Neural Networks

Leather surface defect detection is the most important operation in the leather processing sector which aids in measuring the usable section of processed leather. The detection and localization of the leather defects thus have much impact on the grading of leather hide. This paper presents a methodology for leather surface defects detection and classification. The proposed framework has Deep learning models to detect and classify the defects on the leather. Convolutional Neural Networks (CNN) are used for detecting the defects and to classify the type of defects with the Binary Class Model (BCM) and Multi Class Model (MCM) respectively. The different CNN architectures that were used for the BCM and MCM are LeNet-5, VGG-13, VGG-16, VGG-19, ResNet-50, MobileNet and MobileNetV2. Among all the models VGG-16 architecture achieved better test accuracy of 93.65% for the BCM model and MobileNet architecture achieved a test accuracy of 81.24% for the MCM model. The performances of these models are compared and the best models have been used in the proposed framework.