A Semantic Segmentation Method of Remote Sensing Image Based on Downsampling

Abstract

The invention discloses a feature fusion remote sensing image semantic segmentation method based on downsampling, which includes the following steps: cutting the high-resolution remote sensing image used for training and its corresponding label map into small images in the same way as the original Input image, the model structure is divided into downsampling module, high-level semantic feature extraction module, feature fusion module and classifier module. After the downsampling module extracts high-resolution low-level semantic features from the input original image, they are divided into two All the way into the high-level semantic feature extraction module to extract high-level semantic features, together with the low-level semantic features directly extracted by the other down-sampling module, enter the feature fusion module for feature fusion, and finally, the fused feature map is processed by the classifier module Classification, update model parameters by means of stochastic gradient descent. The invention improves the segmentation accuracy while reducing parameter calculations.

Description

technical field [0001] The invention relates to the technical field of remote sensing image semantic segmentation, in particular to a feature fusion remote sensing image semantic segmentation method based on downsampling. Background technique [0002] Semantic segmentation of images is to perform pixel-by-pixel category classification of input images to achieve pixel-level segmentation of objects and scenes. In recent years, deep learning methods have made good progress in semantic segmentation of remote sensing images. MFPN (Multi-Feature Pyramid Network) is a pyramid network for multi-feature extraction of remote sensing image roads. The network provides a weighted balance loss function to solve the problem of classification imbalance caused by road sparseness; FCN (FullyConvolutional Networks) Trained in an end-to-end, pixel-to-pixel approach, this framework has the advantage of being able to use the raw semantic information produced by the trained network for image segm...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the problems of many parameters, large amount of computation and low segmentation efficiency in the prior art, the purpose of the present invention is to provide a semantic segmentation method of remote sensing images based on down-sampling feature fusion, which can improve the segmentation accuracy of remote sensing images and reduce network overhead at the same time. Complexity and save training time

Images