Building target detection method based on compact quadrilateral representation

Abstract

The invention discloses a building target detection method based on compact quadrilateral representation. A feature extraction network obtains rich multi-size feature information through a feature pyramid structure; and recommends an area generation network to obtain an area of interest. A regular rectangular bounding box branches perform classification tasks and regression tasks of a regular rectangular bounding box on different sizes of feature maps of the feature pyramid structure. A compact quadrilateral boundary frame of a building target is generated by branches of the compact quadrilateral boundary frame, and the contour of the building target in a remote sensing image is precisely positioned. According to the building target detection method provided by the invention, the generation of an irregular shape can be avoided, and the shape constraint of a certain structure can be maintained. Experiments prove that the building target detection method provided by the invention not only can extract more nodes of the building target and more accurate edge feature information, but also can obtain more accurate detection results.

Description

technical field [0001] The invention relates to the technical field of target detection in remote sensing images, in particular to a building target detection method based on compact quadrilateral representation. Background technique [0002] With the continuous and rapid development of remote sensing satellite imaging technology, people can obtain more and more high-resolution remote sensing images through satellites, which also means that the spatial information and semantic information of remote sensing objects will be more abundant. This brings many benefits to human life, such as crop surveying, forest fire monitoring, and vehicle detection for traffic guidance systems. Therefore, extracting information from remote sensing images will play a pivotal role in the field of remote sensing and computer vision. [0003] Target detection, as one of the most challenging problems in image understanding, remote sensing image target detection algorithms that automatically extract...

AI Technical Summary

Problems solved by technology

One is a regular rectangular bounding box. The method of using this bounding box cannot well represent the location of the building, and cannot generate the outline of the building

The other is the polygonal bounding box (i.e., segmentation mask). The method using this bounding box is usually based on an instance segmentation detector, such as Mask R-CNN, which can predict the segmentation mask corresponding to each building. However, due to the uncertainty of the number of nodes and the irregularity of the shape, these polygonal outlines based on segmentation masks are often not very accurate, and it is easy to get irregular shapes, which cannot be very accurate. Good representation of building geometry

Images