Research On Tree Species Recognition Method Based On Deep Learning

In China,forestry production and management,tree species identification and classification is the basis of many forestry work.Therefore,improving the efficiency of tree species identification and classification is of great significance and can bring great value to the development of forestry engineering.The automatic and efficient identification and classification of tree species has always been an important subject in forestry research.In recent years,deep learning technology has made breakthroughs in image classification tasks.People have gradually used the theoretical methods of deep learning to make up for the shortcomings of traditional image classification methods and improve the efficiency of image classification.The purpose of this paper is to study the techniques of image classification for complex tree species.This paper proposes the use of deep learning theory to complete the automatic identification and classification of tree species based on the image of tree leaves,which solves the difficulty of identifying and classifying complex tree species in many categories.This article begins with the development of tree species recognition classification and the deep learning theory of image classification methods.It also describes the shortcomings and improvement directions of tree species recognition methods at the current stage.In order to make up for the lack of sensitivity and generality of neural networks to input data,the current problem of difficult classification of complex tree species is actually solved.The main research contents of the paper are as follows.(1)For the classification task of multi-class complex tree species,in order to improve the feature extraction ability of the neural network and reduce the impact of classification performance caused by data sensitivity,fusion based on Spatial Transform Network(STN)and Dense Neural Network(DenseNet),and then proposed a new network model ST-DenseNet.The ST-DenseNet network uses the spatial mapping network to normalize the key point detection invariance of the input picture,and combines the subsequent dense convolutional network layer to classify and recognize tree species based on the tree leaf image.(2)In order to better complete the classification and classification task of multi-class complex tree species,this paper proposes an improved Convolutional Neural Networks(CNN)model based on the attention mechanism-Attention Branch based Convolutional Neural Networks(ABCNN).The ABCNN model can obtain the region of interest(ROI)of the image through the added attention mechanism,and enlarge the important information through the designed reconstruction process,so as to reduce the overall performance of the network caused by data sensitivity and improve the network versatility.(3)In this paper,the leaf images of 185 trees provided by Leafsna are selected as experimental data,and the two models proposed are verified respectively and compared with the results of other mainstream CNN network models.The method proposed in this paper achieves more than 90%test accuracy on the Leafsna dataset,which is significantly better than other network models.At the same time,in order to prove the generality of the network model proposed in this paper,the proposed method was experimentally verified on the general public data set SVHN.The results prove that the proposed method also has strong classification performance and generalization in the general public data set ability.It proves that the proposed network model can solve the problem of complex tree species identification and classification,and can also be used in practical engineering.