Building extraction method based on gating depth residual error optimization network

Abstract

The embodiment of the invention discloses a building extraction method based on a gating depth residual error optimization network. The method comprises the steps of obtaining an image feature combination of a high-resolution aerial image and airborne LiDAR point cloud data; the diversity of image samples is enhanced through the modes of random cutting, rotation, overturning and light and shade adjustment; using the improved deep residual convolutional neural network to automatically learn multi-level features of the image to obtain a rough building extraction result; a gating feature markingunit is adopted to screen and fuse effective features, and a high-quality building extraction result is obtained through successive up-sampling. By implementing the embodiment of the invention, a feature information gating transmission mechanism is combined with the deep residual convolutional neural network, and the method is used for building extraction of high-resolution aerial images and airborne LiDAR point cloud data.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a building extraction method based on a gated deep residual optimization network. Background technique [0002] Automatic acquisition of building information from remote sensing data plays an important role in topographic map updating, urban 3D modeling, urban sprawl analysis, population estimation, and environmental research. However, how to accurately and automatically obtain building information from remote sensing images has always been a major problem in the field of remote sensing and computer vision. The main reasons include: 1) Buildings in most scenes, especially in developed urban areas, They all have different shape characteristics and roof surface materials, and their spectral reflectance varies greatly, and they are easily blocked by the shadows of surrounding high-rise buildings and tall trees; 2) The high-resolution remote sensing images have large intra-cl...

AI Technical Summary

Problems solved by technology

However, this type of method has the following limitations: First, most research methods use lower-level features of the image to distinguish building and non-building pixels, which often need to be combined with certain threshold setting or rule determination, resulting in such The method is not universal; secondly, many algorithms pre-segment the image when extracting buildings, and the result is highly dependent on the setting of segmentation parameters, which is easily affected by environmental factors during imaging, such as solar radiation, shadows and even random noise

[0005] (1) FCNs use CNNs as its encoder for image feature extraction. Although such output contains high-level semantic features, it is too rough and easy to lose the edge details of the image. For example, in building extraction, building Rich low-level image features such as edges and right angles are easily overlooked;

[0006] (2) Although the segmentation results can be optimized by passing low-level features to the decoder of FCNs by "jumping" connections or using the maximum position of the maximum pooling layer, this method easily leads to the generation of redundant features and reduces the network efficiency. learning efficiency

In addition, the output features usually contain category uncertainty or non-boundary related information, which affects the optimization of classification results

Images