Research on High-Precision Detection Method for PCB Surface-Mounted Components Based on Machine Vision

To address the issue of reduced detection accuracy caused by the loss of edge information in PCB patch inspection using machine vision, this paper proposes a high-accuracy detection algorithm based on a multi-scale deep neural network incorporating edge-preserving pooling. The model is specifically designed to mitigate the edge loss of small targets commonly observed in traditional max-pooling operations. First, the Edge Enhancement Module (EEM) retains edge details by introducing learnable parameters, weights, and a specialized computation method. Second, the Edge-Preserving Pooling (EPP) module compensates for edge loss by utilizing dilated convolution to extract features across multiple receptive fields. Additionally, the pixel distribution of patch elements is clustered and analyzed using the K-means++ algorithm to generate anchor boxes aligned with the spatial distribution of PCB patch components. Experimental results demonstrate that the proposed model achieves a recall of 89.8%, an accuracy of 87.1%, and a mean average precision (mAP) of 88.2% on the patch dataset, thereby satisfying the requirements for efficient and high-precision detection of PCB patch components.