Remote sensing forest tree species classification method for multi-scale convolution and multilayer fusion

The invention discloses a method for classifying forest tree species through multi-scale convolution and multi-layer fusion remote sensing, which comprises the following steps of: firstly, acquiring a remote sensing image through a remote sensing satellite, and carrying out image processing and data enhancement on the image; then, in the ResNet-50 network, a shallow multi-scale convolution attention combination module is adopted to replace 7 * 7 convolution, a 3 * 3 convolution layer is inserted between two 1 * 1 convolution layers of residual blocks to form a new convolution structure, and the residual blocks in the network are arranged as 3, 4, 6 and 3; feature information extraction is carried out on the last residual block of the last feature layer in ResNet-50 and the FC layer and the multi-layer selection feature fusion module, and then the last residual block and the FC layer and the multi-layer selection feature fusion module are added; and finally, carrying out classification by using a softmax class