Research on image recognition based on the yolov3 model in an extreme environment

Abstract

In recent years, the human demand for the exploration and application of extreme environment is increasing. However, the light, climate, object motion and other factors in extreme environments have great uncertainty and complexity, which makes the image recognition technology face great challenges. This study aims to investigate the image recognition techniques based on the yolov3 (You Only Look Once Version 3) model in extreme environments. For the problem of image recognition in extreme environments, this study compared the identification gap between the initial data set in the yolov3 model, and optimized the yolov3 model (The main way is to prune and quantify, and fine-tune the algorithm) to improve its accuracy and stability in extreme environments. In this study, the feasibility of yolov3 model for image recognition in extreme environment was verified by comparing the performance of yolov3 model before and after optimization. The experimental results show that the image recognition technology based on yolov3 model proposed in this study has high accuracy and stability in extreme environments.