An Improved Mobile Phone Screen Defect Detection Based on Multi-Level Feature Fusion

With the growing demand for cell phone products, the quality of cell phone screens has become increasingly crucial. Accurate detection of mobile phone screen glass (MPSG) defects is of utmost importance. However, traditional manual and machine vision methods of detection suffer from low accuracy and slow speed. In this paper, a novel MPSG detection method that addresses the limitations of existing detection methods. The proposed method utilizes a YOLOv8 model with multilevel feature fusion necks. We develop a cutting-edge solution to address the issue of defect scale imbalance. Our High-level Screening-feature Fusion Pyramid-Path Aggregation Network (HS-FPN-PAN) enables seamless integration of features at multiple levels, resulting in enhanced performance. The model utilizes high-level features as weights, effectively filters low-level feature information using the channel attention module, and integrates the filtered information with high-level features to enhance the model's feature expression capability. Additionally, we implement the variability convolution module to enhance the neck's capability in extracting irregular shapes. This module is designed to be lightweight, ensuring optimal performance. Our experiments demonstrate that the enhanced YOLOv8 model exhibits superior detection accuracy and inference speed, providing strong evidence for the effectiveness of the proposed model in this paper.