Rotating target detection method based on multi-scale feature integration and attention mechanism

Abstract

The invention belongs to the field of computer vision, particularly relates to a rotating target detection method, system and device based on multi-scale feature integration and an attention mechanism, and aims at solving the problems that a rotating target detection method is low in detection precision and poor in robustness. The method of the system comprises the following steps: acquiring a to-be-detected image; extracting multi-scale features of the image by adopting a feature pyramid network, and processing, zooming and integrating the multi-scale features to obtain an integrated multi-scale feature map I; processing the I to obtain a foreground feature map and a background feature map, and performing matrix point-by-point multiplication operation on the foreground feature map and thebackground feature map and the I to obtain a foreground saliency feature map and a background saliency feature map; obtaining an attention feature map through a preset first method, and obtaining a candidate region of the to-be-detected target through a region suggestion network; and according to the positions of the candidate regions and in combination with the attention feature map, obtaining adetection result by rotating the target detection network. According to the invention, the detection precision and robustness are improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a rotating target detection method, system and device based on multi-scale feature integration and attention mechanism. Background technique [0002] Object detection is an important computer vision task widely used in remote sensing, industrial inspection, robotics and other fields. Among them, target detection under conditions such as overlooking angles has problems such as diverse target angles, dense and complex distribution, large scale differences, intra-class differences and inter-class similarities, and complex backgrounds. Conventional object detection is a horizontal frame. For natural images, the objects are mostly distributed vertically and horizontally. However, for images acquired from a top-down perspective, the objects will appear in any angle. Horizontal frames often cannot meet the precise detection of such objects, especially For the situation where ...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above-mentioned problems in the prior art, that is, in order to solve the problems of low detection accuracy and poor robustness of the existing rotating target detection methods in the case of various target angles, large scale differences, and complex backgrounds under the overlooking angle, this paper In the first aspect of the invention, a rotating target detection method based on multi-scale feature integration and attention mechanism is proposed, which includes:

Images