Remote sensing image segmentation algorithm based on convolutional neural network

Abstract

The invention relates to a remote sensing image segmentation algorithm based on a convolutional neural network, and the algorithm comprises the steps: employing ResNet34 pre-training weight parametersas the depth guarantee of a network layer, so as to solve a problem of gradient disappearance when the network layer is deepened; a pyramid pooling module is adopted to aggregate context informationof different areas in the image so as to improve the capability of obtaining global information; designing a loss function by adopting a method of combining a cross entropy loss function and regularization constraints, wherein the cross entropy is used for judging the closeness degree of actual output and expected output in the multi-classification problem; the regularization term can reduce the complexity of the model so as to prevent overfitting and improve the generalization ability of the model. According to the method, the designed composite convolutional neural network algorithm is applied to semantic segmentation of the remote sensing image, so that the speed of information recognition in the remote sensing image and the capture rate of effective information are greatly improved, and the method is of great significance to information analysis of the remote sensing image.

Description

technical field [0001] The present invention proposes a semantic segmentation algorithm for satellite remote sensing images, utilizes the excellent characteristics of convolutional neural networks in image segmentation, applies it to the field of semantic segmentation of satellite remote sensing images, and realizes pixel-level segmentation of remote sensing images, and the segmentation results can be It is used in satellite map construction, military target detection and other fields. Background technique [0002] Remote sensing images refer to images of the earth that are aerially photographed in space by drones, remote sensing satellites, and other aircraft. Because the earth image obtained by aerial photography is a bird's-eye view, it is beneficial to clearly observe the surface information, which is objective and macroscopic, and has great value for the study of surface characteristics. [0003] There are many remote sensing satellites currently operating in Earth orb...

AI Technical Summary

Problems solved by technology

Traditional remote sensing image recognition basically relies on manual analysis. For remote sensing images that contain a lot of complex information, the speed of manual analysis is too slow, which is stretched for information processing in remote sensing images.

Images