Leather Defect Detection System Using Machine Learning

Abstract

The quality of leather products is significantly influenced by the presence of surface defects such as scratches, wounds, insect bites, tanning irregularities, and texture variations. Traditionally, leather inspection is performed manually by trained workers, a process that is highly time-consuming, inconsistent, and subjective. With increasing demand for high-quality leather goods and the need for standardization in quality assurance, automated defect detection systems have become essential. This project proposes a deep learning–based approach for automated leather defect detection using the MobileNetV2 Convolutional Neural Network (CNN) architecture, specifically designed for efficient image classification tasks.

The system utilizes a comprehensive data set of leather images, categorized into defective and non-defective classes. The methodology includes image acquisition, preprocessing, augmentation, feature extraction, and model training. MobileNetV2 is chosen for its lightweight structure, fast computation, and strong performance on image recognition tasks. The model architecture is enhanced with additional dense layers, dropout regularization, and a soft max classifier to improve generalization and accuracy. A systematic training and validation process is conducted using augmented datasets to ensure robustness against variations in lighting, orientation, and texture.

Experimental results demonstrate that the proposed model achieves 98% accuracy on the MVTEC anomaly detection data set, effectively distinguishing between defective and non-defective leather surfaces. The system not only identifies defects but also ensures consistent evaluation standards across all inspected samples. A user-friendly interface is developed to enable real-time detection, allowing users to upload an image and receive instant defect classification results.

The automated approach significantly reduces manual labor, minimizes inspection time, and increases overall production efficiency. This research validates the effectiveness of deep learning in industrial quality inspection and highlights its potential for large-scale deployment in leather manufacturing units. Future work may include incorporating advanced AI models, expanding the defect categories, integrating segmentation for precise defect localization, and deploying the system on edge devices for real-time industrial automation.