Method for extracting remote sensing image interesting area based on multi-scale feature fusion

Abstract

The invention discloses a method for extracting a remote sensing image interesting area based on multi-scale feature fusion, and belongs to the technical field of remote sensing image processing. The method comprises the following steps: (1) adopting a saliency analysis method based on a multi-scale spectrum residual error to generate a multi-scale low frequency saliency image of the remote sensing image, (2) obtaining multi-scale high frequency saliency images in the horizontal direction, the perpendicular direction and the diagonal direction through integer wavelet transformation, (3) obtaining a luminance saliency image and a direction saliency image through cross-scale weighting and cross-scale fusion, (4) combining the luminance saliency image and the direction saliency image to generate a main saliency image, and (5) carrying out threshold segmentation to extract the interesting area through the OTSU. Compared with a traditional method, the method has the advantages of being low in computing complexity and effectively improving area extraction efficiency and area extraction accuracy. Due to the fact that spectrums and color information of the remote sensing image are not required to be obtained, the method can be directly used for interesting area extraction of a high-resolution full-color remote sensing image. The method has good application value in the fields of land planning, environment monitoring, forestry investigation and the like.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and relates to a remote sensing image interest region extraction method based on multi-scale feature fusion. Background technique [0002] With the rapid increase of the spatial resolution of remote sensing images, the time for analyzing and processing massive remote sensing data has increased sharply, and research on more efficient and fast remote sensing image processing methods has become an urgent topic. Accurate extraction of regions, and then targeted analysis and processing” provides a good idea to effectively reduce the complexity of massive remote sensing data analysis and processing and reduce processing time, and is also a research hotspot in the field of remote sensing image processing technology. [0003] Visual attention methods have received more and more attention due to their advantages of fast and accurate extraction of regions of interest without prior k...

AI Technical Summary

Problems solved by technology

[0005] Although the above visual attention methods have many advantages, they have many disadvantages when directly applied to remote sensing images, especially in high spatial resolution remote sensing images.

For example, the resolution of the saliency map generated by the ITTI method and the GBVS method is too low to achieve an accurate description of the region of interest in the remote sensing image, and many important regions will be missed

The WT method produces a saliency map of the same size as the original image, but the saliency map contains many fragmented regions

The SR method reduces the original image to 64×64 pixels and then calculates the spectral residual, and generates a saliency map based on a single low-scale image. Although it is conducive to the rapid extraction of the region of interest, it will cause the edge of the region to be too blurred, making the region of interest The accuracy of region extraction is greatly reduced

In addition, remote sensing images contain a lot of information such as edges and textures. Generally, regions with richer edges and textures are more likely to become regions of interest. The above methods are not ideal for edge and texture information detection in remote sensing images. Exacerbated the missed detection and false detection of the region of interest

[0006] On the other hand, most of the currently widely studied visual attention methods use interpolation to a specific scale for the fusion of multi-scale saliency maps and then add them directly, so that the regions of interest with larger sizes are continuously strengthened in the saliency map, while The saliency of ROIs with smaller sizes is continuously weakened, which may lead to failure to detect

Images