基于改进U-Net的PCB涂胶区域识别研究

Research on PCB Glued Area Recognition Based on Improved U-Net

摘要: 针对PCB涂胶区域识别的问题，提出一种能够高效分割PCB涂胶区域的改进U-net模型，首先在U-net编码器部分使用自主设计的Block模块替换U-net原有的双层卷积模块，提取图像特征；然后插入添加ESE注意力机制的下采样分支，将两路特征图拼接完成U-net编码部分；最后将拼接后的特征图在解码器部分上采样，上采样的过程中和编码器相应尺度特征图拼接，完成特征图解码，提高模型的特征提取能力和特征表达能力。实验结果表明：改进U-net的Acc为90.25%，比原算法提高了4.12%，它的Dice为84.57%，比原算法提高了4.95%，能够较好地识别涂胶区域，具有实际使用价值。

Abstract: Aiming at the problem of identifying the glued area of PCB, an improved U-net model is proposed, which can effectively segment the glued area of PCB. Firstly, the self-designed Block module is used to replace the original double-layer convolution module of U-net in the encoder part of U-net to extract image features. Then insert a down sampling branch with ESE attention mechanism, and splice the two feature maps to complete the U-net coding part; Finally, the spliced feature map is sampled in the decoder part, and it is spliced with the corresponding scale feature map of the encoder during the up-sampling process to complete the feature map decoding, which improves the feature extraction ability and feature expression ability of the model. The experimental results show that the Acc of the improved U-net is 90.25%, which is 4.12% higher than that of the original algorithm, and its Dice is 84.57%, which is 4.95% higher than that of the original algorithm. The improved U-NET can better identify the gluing area and has practical value.