Research On Tree Species Recognition Method Based On Feature Fusion And Deep Learning

Forest ecosystem is one of the main bodies of national nature reserves in Northeast China.It is of great significance to realize the automatic identification of tree species for the protection of biodiversity,which is conducive to increasing people's understanding of plant resources and making people better use of these resources.In addition,people's demand for wood products is increasing day by day,so it is particularly important to popularize species identification.In this paper,a tree species recognition method based on feature fusion and deep learning is proposed.The tree species recognition experiments were carried out from five aspects:traditional classification algorithm,deep learning,transfer learning,target detection and recognition,and feature fusion.The main contents of this paper are as follo ws:(1)Tree species recognition based on transfer learning.The premise of this method was that a large number of samples were needed to support the direct training of deep learning,and transfer learning was suitable for the recognition training of small-scale data sets.The first was the pre training model on the transfer ImageNet,and then the transfer learning training was carried out on the dataset.Finally,the feasibility and effecti veness of the transfer learning were further explained and analyzed by using the confusion matrix.(2)Tree species identification based on target detection.This method was based on migration learning.Firstly,the transfer convolution neural network model was used to extract image features,which were used as the input of target detection.Finally,images with complex background were detected and recognized.It could effectively solve the problem of image input,no longer limited to the unified size of the image,and could take any size as the input.In addition,due to the influence of environmental factors,the growth direction of trees would be different.Whether the growth direction would affect the recognition results was also discussed and explained.(3)Tree species recognition based on image fusion of leaves and bark.(1)And(2)were used to recognize the leaf image or the bark image separately.Through the confusion matrix,it was found that there were problems of the similarity between trees and the difference within trees.Based on this,an image fusion method of leaves and bark was proposed.Firstly,the leaf image and the bark image with the same number and species are selected for the fusion experiment,then different fusion strategies were selected to combine with the convolution neural network.Finally,the feasibility and effectiveness of the fusion method were discussed and analyzed by using the confusion matrix.12 kinds of trees in Northeast China were selected as the research objects,and the main research contents were based on transfer learning,target detection and feature fusion.Compared with the traditional classification methods and other recognition methods,the final recognition result was 97.80%.The method proposed in this paper not only improves the training speed,but also improves the recognition results.