Accelerated identification method of industrial instrument based on yolov5
Shimao Yu, Haiyang Zheng, Chunhe Song, Pengpei Gao · 2022 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) · 2022
There are many problems in industrial instrument detection, such as complex background and visual Angle, etc. Traditional machine learning methods are difficult to achieve high-precision object detection and real-time monitoring applications. This paper proposes an accelerated identification method of industrial instruments based on YOLOv5. First, sample data are collected on site for image annotation, and YOLOv5 network structure is designed for transfer learning training, so as to realize image recognition of four types of industrial instruments. After that, the model is accelerated based on TensorRT on Jetson TX2 platform. The speed of instrument online detection reached 30 fps and mAP0.5 is 0.989.