Robot Arm Grasping and Automotive Unidirectional Device Defect Detection Based on Dual Modal Mask R-CNN

Chunsheng LI, Jianlin CHEN

Abstract

This study proposes an improved mask model based on a bimodal convolutional neural network for addressing the limitations of the existing detection methods with regard to the processing of multi-source information. The purpose is to improve the model`s recognition accuracy for complex targets in the context of robot arm grasping tasks, as well as the accuracy of defect detection for automotive unidirectional devices. Firstly, based on the mask model including a convolutional neural network, a bimodal input model is constructed by combining red-green-blue images and depth images in order to enhance the model's ability to perceive the three-dimensional structure of objects. Secondly, the model’s feature extraction performance is enhanced by introducing a convolutional block attention module, and the final robot arm grasping system and the automobile unidirectional device defect detection system are constructed. The experimental results demonstrated that the proposed model achieved a detection accuracy of 0.97 for benchmark performance testing and a recognition time as low as 0.01 seconds for actual robot arm grasping tasks. In addition, the recognition accuracy of the proposed model in the context of five different robot arm grasping scenarios was as high as 98.84%, and the defect detection accuracy was as high as 98.25%. Therefore, this model can well optimize the efficiency of the current robotic arm grasping and defect detection tasks.

Keywords

Mask R-CNN, Defect detection, Mechanical arm, Deep learning, Unidirectional device, Target recognition.

View full article