Research on Wood Strip Classification Method Based on Deep Learning

Wood strip is one of the raw materials for manufacturing finger plate, and the classification quality of wood strip determines the surface quality of the finger plate. This paper aiming at the problem of the wood strip classification, this paper collects wood strip image data with the industrial camera on factory production line and takes a series of data augmented ways to simulate the production line vibration and other factors, making augmented wood data. This paper proposes a solution based on deep learning image classification, which adds channel attention mechanism to the deep residual network and constructs the SEResnet50 network model based on the Resnet50 network model. The experimental results show that the Accuracy, Precision, Recall, and F1 scores of SEResnet50 are 98.54%, 98.54%, 98.54%, 98.54%, and Classification accuracy for all categories exceeds 95 %. Compared with Resnet50, our model's Accuracy, Precision, Recall, and F1 scores are improved by 2.4%, 2.0%, 1.8%, and 1.8%, Compared with Resnet18, Vgg16, Densenet, Alexnet, SEResnet50 has the best performance.