The Research on Sheet Metal Part Recognition Technology Based on Improved Mask R-CNN

The visual recognition of sheet metal parts is a crucial component of industrial automation sorting, particularly under complex lighting and background interference conditions. To enhance the accuracy of sheet metal part recognition, this paper curates a dataset of sheet metal parts acquired by an intelligent light field image sensor and proposes a visual recognition model, PA-CBAM, based on an improved Mask R-CNN. By improving the original feature pyramid FPN and introducing the Convolutional Block Attention Module (CBAM), the model enhances multi-scale feature fusion and the focus on key features: (1) Bottom up paths and shortcut connections have been added between different layers of the traditional FPN feature pyramid, thereby improving multi-scale object detection capability. (2) CBAM further optimizes the model's channel and spatial attention mechanisms, effectively enhancing the capture of edge details and the suppression of background noise. Experimental results show that the proposed PA-CBAM model significantly improves detection and segmentation performance compared to traditional methods, particularly in terms of edge recognition accuracy for sheet metal parts under complex lighting conditions.