A Lightweight Wood Image Recognition Model Based On Convolutional Neural Networks

Timber species recognition plays an important role in the wood science and wood industry.However,the conventional classification methods are not robust to extract the wood features,and the classification accuracy cannot meet the requirements of practical applications.In recent years,deep learning technology has made major breakthroughs in the field of image recognition.It has been widely used in many fields and has achieved good results.Therefore,a new wood scanning electron microscope image classification method using deep convolutional neural networks is proposed in this paper.The proposed WSEMNet which is a new Convolutional Neural Networks is constructed for the classification of lightweight wood scanning electron microscope images.In the paper,wood microscopic image dataset were collected by the Japanese Forestry and Forest Products Research Institute,and a total of 1 210 images including 10 tree species were collected after image preprocessing.The WSEMNet structure used two Inception V1 modules to increase the efficiency of parameter utilization in the network,and optimized by using Batch Normalization which accelerating the training convergence and reducing the training time.The initial model is trained on WSEMDataset to obtain the final classification model.Finally,we use the test images as the input of the network to verify the classification effect of the WSEMNet model,which compare the classification result with the known label to obtain the accuracy of the test dataset.The result shows that(1)The average recognition rate of WSEMNet was 99.15%,and the average recognition rates of Goog Le Net and Alex Net models were 58.49% and 75.47%,respectively,by classifying ten different timber species on the WSEMDataset.(2)Comparing with the Google Net,the training time of WSEMNet was reduced by 25%.(3)Using the Batch Normalization layer after the Re LU function is beneficial to improving the overall effect of the network.(4)WSEMNet model uses a less training time to achieve higher classification accuracy and better robustness to classify lightweight WSEMDataset.