A Real-Time Skin Disease Recognition System Based on YOLOv5 Algorithm

Skin diseases are common and frequently morbidities in medicine. This paper focuses on common skin diseases such as acne and dermatitis, and selects deep learning target detection algorithms such as You Only Look Once (YOLOv5), EfficientDet, M2det and single shot multi-box detector (SSD) using cloud and edge end. A skin disease recognition system for rapid identification of skin diseases has been developed for popular science dermatology. In practice, the system quickly recognizes twenty common skin diseases, and the system deployed based on the YOLOv5 algorithm achieves a mean average precision (mAP) of higher than mAP@.5:.95 compared with the algorithm EfficientDet, M2det and SSD. The average mAP value of YOLOv5 is about 8.3 and 3.1% points higher than that of SSD and M2det algorithms, respectively, and more in line with system requirement.