Deep Learning-Based Optimized YOLOv5 for Enhanced Surface Defect Detection in Ceramic Tiles

This research delves into the ensuring the quality of ceramic tiles, is an important but difficult task in the manufacturing sector. For defect identification, many businesses still use human inspection, which is labour-intensive, unreliable, and time-consuming. With the use of cutting-edge deep learning models such as YOLOv5 and Convolutional Neural Networks (CNNs), this paper presents an automatic flaw identification method. Without the need for human feature engineering, these models can recognize and learn complex patterns in tile images, including textures, edges, and forms. The study compares the performance and accuracy of CNNs with different activation functions and YOLOv5 in order to establish the most effective method for detecting surface flaws. The results apparently show that the Mish activation function outperforms in detecting various defects across the two proposed architectures. This solution paves the path for quicker and more dependable quality control by increasing accuracy while lowering the need for human intervention.