Super-resolution building fine identification method based on multi-scale feature deconvolution

Abstract

The invention discloses a super-resolution building fine identification method based on multi-scale feature deconvolution. The method comprises the following steps: making a training sample set; constructing a super-resolution semantic segmentation convolutional neural network comprising an encoder module, a decoder module, a multi-scale feature deconvolution module, a feature fusion module and a category judgment module; inputting the training sample set to train the super-resolution semantic segmentation convolutional neural network; and processing an image to be recognized by adopting the trained super-resolution semantic segmentation convolutional neural network to obtain a recognition result. The method has the advantages that the super-resolution semantic segmentation network MLSRSS-Net based on multi-scale feature deconvolution is provided, and the multi-scale primary feature deconvolution up-sampling of an encoder and advanced object semantic feature deconvolution up-sampling of a decoder are integrated; the remote sensing image with the low spatial resolution is input, the building semantic graph with the high spatial resolution is output, and the target extraction precision is remarkably improved.

Description

technical field [0001] The invention relates to the technical field of remote sensing image feature extraction and semantic segmentation, in particular to a super-resolution building fine recognition method based on multi-scale feature deconvolution. Background technique [0002] Buildings are one of the important types of ground features. Accurately identifying buildings from remote sensing images can provide important decision support for government departments such as land management and urban planning to carry out land use status investigations and macro-planning. It has important applications in many fields such as the detection of illegal building and occupation of cultivated land, and military reconnaissance. [0003] Recognize buildings from remote sensing images. Depending on the image resolution, the details of the recognition results are also different. For example, high-resolution images at the meter level or sub-meter level can reflect the fine outlines of build...

AI Technical Summary

Problems solved by technology

At present, there are many related technical methods for super-resolution image reconstruction (interpolation and reconstruction from low-resolution images to obtain high-resolution images), but there is a lack of super-resolution semantic segmentation (high-resolution semantic segmentation results obtained from low-resolution images) Technology

Images