Cross-Scale Attention and Multi-Layer Feature Fusion YOLOv8 for Skin Disease Target Detection in Medical Images

Abstract

With the continuous development of deep learning technology, skin disease target detection has been increasingly widely used in medical image analysis. This paper proposes a skin disease target detection method based on cross-scale attention multi-layer feature fusion YOLOV8. By introducing cross-scale feature fusion and attention mechanism, the performance of the model in processing skin disease images is enhanced. First, YOLOV8 is used as the basic framework, and its original structure is improved. The cross-scale feature fusion module is introduced to improve the detection ability of skin disease targets of different scales. Secondly, combined with the cross-scale attention mechanism, the key areas of skin disease targets are focused on by dynamically weighting feature maps of different scales, which significantly improves the robustness of the model in complex backgrounds. Experimental results show that the proposed model outperforms traditional mainstream detection algorithms such as YOLOV5, YOLOV8, and DETR in multiple performance indicators such as precision, recall rate, and mAP, especially when dealing with skin disease targets of different shapes and scales. Through further ablation experiments and comparative analysis, the positive impact of cross-scale attention mechanism and multi-layer feature fusion on detection performance is verified. This study provides a new solution for skin disease target detection, which can effectively improve the automated diagnosis capability of skin diseases and provide strong technical support for future medical image analysis.