Greenbelt water system vector extraction method based on convolutional neural network and edge constraint energy optimization

Abstract

The invention provides an end-to-end greenbelt water system vector extraction method based on a convolutional neural network and edge constraint energy optimization, and designs a greenbelt water system extraction network architecture suitable for remote sensing images. The architecture comprises the steps of: adopting remote sensing image context feature extraction and fusion, and achieving basic feature extraction of an image of a to-be-processed region; on the basis of extracting rich features, combining a convolutional layer and an up-sampling layer, and adopting an end-to-end energy optimization iteration mode to obtain a fine and smooth green land water system edge; and finally, finely adjusting to a fine green land water system edge by using a full connection layer or a graph convolution layer. Besides, two loss cross entropies and Dice loss are used for semantic recognition of the green land water system, a recognition result is constrained at a full convolutional network end and an edge energy constraint optimization end, a multi-layer coordinate point matching loss function is provided to realize constraint of contour points, and the model can enable the predicted node to be better close to a true value contour point.

Description

technical field [0001] The invention relates to an end-to-end vectorized extraction and boundary optimization method of green space and water systems, which is used for automatic extraction of green space and water system contours from remote sensing images. Background technique [0002] In recent years, with the rapid development of deep learning and big data technology and the significant improvement of computer hardware technology, breakthroughs have been made in the extraction technology of typical features in remote sensing images. Among them, technologies such as image classification, target detection, and semantic segmentation have been widely concerned and deeply researched in remote sensing intelligent applications. Image classification and object detection techniques are difficult to apply to the extraction of object contours in remote sensing images. Semantic segmentation technology assigns a corresponding semantic label value to each pixel on the image to be int...

AI Technical Summary

Problems solved by technology

Therefore, accurate vector edge extraction based on remote sensing images of green space and water systems is facing a major challenge

Images