YOLACT in Micro-Assembly Robot System

摘要

With the development of micro-assembly systems, the requirements for micro-assembly vision recognition and detection are getting higher and higher. Combining with deep learning, which has made a great splash in image processing, this paper uses the YOLACT instance segmentation algorithm to recognize and segment the targets in microscopic vision. ResNet-101 backbone is used in the experiment to obtain higher detection accuracy and mask quality. The experiment finally achieves an accuracy of 89.43% for the recognition of targets with boundary mAP values higher than 0.5.