Research on the Identification of Particleboard Surface Defects Based on Improved Capsule Network Model

Aiming at the problems of low classification accuracy and overfitting caused by limited number particleboard image samples, a Capsule Network algorithm based on improved CBAM (Convolutional Block Attention Module) attention model is proposed. The utilizes GELU equation to improve incorporates it into convolutional layer Network. In this way, optimizes feature maps surface defects and, meanwhile, improves training efficiency stability model. alleviates problem adding dropout layer, which makes more suitable for small sample classification. effectiveness method proposed in paper verified experiments dataset defect images.