Tree Species Classification and Mapping Based on Deep Transfer Learning with Unmanned Aerial Vehicle High Resolution Images

A tree species classification and mapping method is proposed based on the deep transfer learning with unmanned aerial vehicle high resolution images. The image features of tree species arc extracted using a large convolution neural network trained on ImageNet. The features of tree species images arc compressed by the global average pooling. A simple linear iterative clustering method is used to generate the super-pixel, which arc used as the minimum classification unit to generate tree species maps. The experimental results show that the proposed method can accelerate the convergence of the training process. The overall accuracy and Kappa coefficient arc increased by 9.04% and 0.1547, respectively, compared with the small convolutional neural network method in the case of small inter-class gap and the large intra-class gap, and the boundary of the super-pixel tree mapping is more accurate.