YOLOv5-Based Automated Skin Cancer Detection: Robust Lesion Localization

Abstract

This study proposed and verified an automatic skin cancer lesion detection algorithm based on YOLOv5. By comparing the performance of mainstream target detection models such as Faster R-CNN, SSD, RetinaNet and YOLOv3, the experimental results show that YOLOv5 performs well in key indicators such as mAP, recall rate and precision rate, and can more accurately identify skin cancer lesion areas, especially with good robustness in detecting small lesions. YOLOv5's fast detection capability and high-precision performance prove its potential for application in early screening of skin cancer, and provide technical support for the realization of efficient automatic skin cancer detection. Future research can further optimize the model performance by combining multimodal data and transfer learning methods to improve its generalization ability in clinical applications.