Research on Classification of UAV Optical Image Tree Species Based on Res2Net

Abstract

The internal features of remote sensing images of plant communities are complex and the boundaries between classes are blurred. The traditional image processing methods based on pixel spectral information cannot make full use of the image feature information, making the extraction effect poor. Therefore, this paper proposes a deep convolutional neural network. Convolutional neural network (CNN) high-resolution remote sensing image plant community automatic classification method. Segment drone images to obtain regular images, and then use the CNN-based Res2Net model to abstract and learn the features of the image to automatically obtain deeper abstractions and more representative image deep features, realize the extraction of the distribution area of the plant community, and output the automatic classification result of the plant community in the form of the original image and the result image superimposed on each other. The number of samples with different gradients are used as training samples, and the method proposed in this paper is used to analyze the influence of the number of training samples with different gradients on the results of automatic classification. Experimental results show that the number of training samples has a significant impact on classification accuracy, the modeling accuracy of the ResNet50 model and the Res2Net model are increased from 82% and 83% to 90% and 92%, compared with the traditional supervised classification method, the deep convolutional network significantly improves the classification accuracy. The classification results show that when the number of training samples is not less than 200, the CNN-based Res2Net model shows the best classification results.