Cultivated land information extraction method of high-resolution remote sensing image based on edge enhancement

Abstract

The invention designs a cultivated land information extraction method of a high-resolution remote sensing image based on edge enhancement, designs and constructs a combined edge enhancement loss function BE-Loss, and correspondingly designs an efficient BEC-Net network framework to realize accurate extraction of cultivated land parcels and edge conditions thereof. A UNet network with an OfficientNet as a backbone network is used as a semantic segmentation module, and training precision is improved by integrating an scSE dual-channel attention mechanism, a Cot module, gating convolution and the like in a boundary enhancement module. And the learning rate is updated by combining a cosine annealing attenuation algorithm, so that the training result is more comprehensive. And finally, the classification result subjected to edge enhancement is obviously improved in precision compared with other models. According to the method, model optimization can be realized by adjusting a network framework and a learning means; the efficient shallow structure is utilized to focus on processing edge semantics, and the method has the same important significance for improving the extraction precision of cultivated land type plots.

Description

Technical field [0001] The present invention relates to the field of remote sensing image cultivated land information extraction technology, in particular to a high-resolution remote sensing image based on edge enhancement of the cultivated land information extraction method. Background [0002] For the extraction task of cultivated land categories in remote sensing images, while the traditional unsupervised classification, supervised classification, and object-oriented classification methods continue to develop, the use of deep learning algorithms for farmland information extraction shows application advantages and potential. Semantic segmentation networks such as U-Net, DeeplabV3+, etc., have a significant effect on farmland extraction, but these algorithms are prone to edge blur and boundary misdivation when extracting cultivated land. Marginal characteristics are of great significance for the division of cultivated land plots, and it is necessary to identify cultivated land a...

AI Technical Summary

Problems solved by technology

Semantic segmentation networks such as U-Net, DeeplabV3+, etc., are effective in extracting cultivated land, but these algorithms are prone to blurred edges and misclassified boundaries when extracting cultivated land

Images