A Small-scale Object Detection Method for LCD Defects Based on Improved YOLOv8

In recent years, the integration of image processing and computer vision techniques has demonstrated significant potential in the realm of automatic liquid crystal display (LCD) defect detection. Small-scale and blurry features characterize defects in LCD images, with the demand of deployment of defect detection methods on edge devices, necessitating considerable demands on the model's flexibility. To address these challenges, this paper presents a lightweight LCD defect detection algorithm for small-scale defects based on the enhanced YOLOv8 architecture. The feature extraction capability of the network for small-scale defects is bolstered by the addition of an extra small target detection head and modifications to the loss function. Furthermore, we reduce the number of network parameters and model size through a lightweight design of the network structure. The proposed algorithm achieves commendable detection results for various types of defects on mobile phone LCD panels, with a particular emphasis on small-scale defects. Experimental results on our proprietary LCD dataset reveal a notable 8.7% improvement in mean average precision (mAP) compared to the baseline, concurrently reducing the model size by 52.9%.