Research on Lightweight Wound Recognition Algorithm and System for UAV in Accident Emergency Treatment

Abstract

Aiming at the problems of slow manual rescue response for emergency hemostasis at accident scenes, single emergency function of emergency UAVs, insufficient lightweight degree of airborne wound recognition algorithms, and poor hemostatic effect of traditional hemostatic materials, this paper takes YOLOv5s as the basic model. Through channel pruning, INT8 quantization and knowledge distillation, a lightweight wound recognition algorithm suitable for edge deployment on UAVs is developed. Integrating temperature-sensitive self-expanding nanofiber hemostatic materials and an umbrella-type precise delivery mechanism, a fully automatic emergency UAV system is constructed. Experimental results show that the optimized algorithm model is only 89.2 KB in size, with a wound recognition accuracy of 93.6% in complex scenes and an inference speed of 34.2 fps, which meets the requirements of real-time airborne deployment. The hemostatic material expands completely within 28 s at 37 ℃ with an expansion ratio of 2.4 times, and the in vitro coagulation time is 76 s, improving hemostasis efficiency by 72% compared with traditional gauze. In a level 4 wind environment, the system delivery error is controlled at 2.1 cm, and the total hemostasis time is 110 s. The system realizes integrated operations of automatic wound recognition, precise delivery and rapid hemostasis, which can improve emergency efficiency at accident scenes and provide technical support for emergency medical rescue.