Building vector extraction model based on deep learning and extraction method thereof

Abstract

The invention discloses a building vector extraction model based on deep learning and an extraction method thereof, and the method comprises the steps: S1, receiving a remote sensing image, carrying out the feature extraction of a plurality of stages of the remote sensing image, and obtaining the extraction features of a plurality of scales; s2, performing feature optimization on the extracted features of each scale to obtain optimized features of a plurality of scales; s3, performing feature fusion on the optimized features of the multiple scales to obtain fused features; s4, performing feature recovery and category judgment on the fusion features to obtain a building preliminary extraction result; and S5, post-processing the preliminary building extraction result to obtain a final building vector result. According to the scheme, the feature extraction advantages under multiple scales are fused, the feature extraction capability is effectively improved, the information extraction capability of the network on the original image is enhanced, the extraction result is extracted according to the fused features, post-processing is performed on the extraction result, a relatively regular building vector result can be directly obtained, and subsequent editing and application are greatly facilitated.

Description

technical field [0001] The invention belongs to the field of building extraction, in particular to a deep learning-based building vector extraction model and an extraction method thereof. Background technique [0002] Building extraction based on remote sensing images is a technology that analyzes the spectral and texture features of remote sensing images and determines the pixel-level semantic categories of the images. Commonly used traditional classification methods mainly include methods based on the overall characteristics of buildings, methods based on auxiliary information and object-oriented building extraction methods. The methods based on the overall characteristics of the building include corner detection, line grouping, building index and other methods based on the structural characteristics of the building itself, combined with empirical knowledge. Auxiliary information-based methods utilize auxiliary information such as Digital Surface Model (DSM) and building ...

AI Technical Summary

Problems solved by technology

[0003] There are many subjective factors in the above traditional methods, the segmentation scale and feature selection need to be manually adjusted according to different images, the image data features are not fully exploited, and the classification robustness is poor

Compared with traditional algorithms, the deep learning model uses multi-layer convolution and activation functions to simulate the neural structure of the human brain, and automatically learns the geometric structure, spectrum and texture characteristics of ground objects from a large amount of data, which can reduce the influence of subjective factors on the results. Therefore, the use of deep learning technology in the process of building extraction has become a new trend. Although remote sensing image classification based on deep learning has been proposed, the current methods are not very ideal in terms of shallow feature extraction and restoration, and still need to be improved.

The convolutional neural network for building extraction can only get raster extraction results, and cannot be edited later, which affects the subsequent application of the extraction results

Images