Attention Deformable Convolutional Networks for Wooden Panel Defect Recognition
To solve the problem of high labor cost and low efficiency in wood defect detection,this paper proposes an end-to-end neural architecture model based on a deformable convolutional network and an at-tention mechanism.Firstly,the deformable convolutional network(DCN)enables the model to focus on regions with more useful image information by converting a rectangular grid into a deformed grid.Using a deformable convolutional network can ignore the irrelevant coefficients in image features,addressing the limited ability of traditional convolution to learn more information in features.Then,the DCN output is fed to the gated recurrent unit(GRU)layer to learn high-level features of the defective image.Finally,by focusing on the most important features of the input image,the attention mechanism is applied to enhance the high luminance of the defective regions,thus improving the accuracy of the model recognition.Using the Matlab platform to compare and analyze this paper's method with other existing methods with four wood panel defect datasets,the proposed method improved the accuracy by 2.4%to 13.2%in dimen-sions,sensitivity by 3.3%to 16.6%in dimensions,and specificity by 4%to 21%in dimensions over the other three compared methods.The experimental results show that the method in this paper outperformed the existing methods in terms of detection accuracy and various other performance aspects,and the best accuracy was 99.2%,which proves the effectiveness of the proposed method.
NETL
NSTL
万方数据
