Application of Deep Learning in Wood Classification

This paper presents a deep learning model for the classification of 25 precious wood species. 25 precious wood species were sampled, observed and photographed under a 20-fold magnifying glass after polishing, and then classified by professors to mark the pictures. We removed the blurred part of the original images and got the effective region of images. The 25 kinds of wood pictures were flipped, and 192 pictures were obtained for each type of wood. A total of 4800 pictures were collected, so the experimental data sets were obtained. In this paper, data sets are divided according to the ratio of 60%:20%:20%. Among them, the 60% of the dataset is used as the training set, 20% as validation set and 20% as testing set. The purpose is that using training sets to train the models as much as possible, and retain a certain number of images to validate and test the model. The corresponding structure and weight of the model were directly used which the pre-trained model of the VGG19 network trained on the ImageNet dataset, and applied to the classification of 25 precious woods. The deep learning network achieves a good fitting effect by learning from a large number of labeled wood sample data and extracting effective features. It also has a good generalization performance on the test set.