Remote-sensing image rotating target detection method

Abstract

The invention discloses a remote sensing image rotating target detection method based on a global-local attention mechanism. The method comprises the following steps: S1, extracting depth features; s2, extracting a saliency feature map; s3, constructing a significance pyramid; s4, extracting features and coordinates of candidate regions which contain foreground potential targets and are ranked atthe top; s5, generating a global scene semantic constraint feature, and splicing the global scene semantic constraint feature with the feature from S4; s6, carrying out a RoIAlign pooling operation; s7, generating a final feature map; s8, category prediction and prediction of bounding box coordinates represented by five parameters are carried out; and S9, analog loss and coordinate loss are calculated, a calculation result of a non-maximum suppression algorithm is optimized by adopting an IoU, and a target detection result is displayed. The method can effectively reduce the interference of background noise, improves the precision of a detection result, can effectively eliminate an obvious false detection result, and finally obtains a high-precision and reasonable type and rotating boundingbox result.

Description

Technical field [0001] The invention relates to a remote sensing image rotating target detection method, belonging to the technical field of aerial image target detection. Background technique [0002] Simultaneous positioning and category recognition are the foundation and difficulty of aerial image target detection. As the number of aircraft and satellites increases, more and more aerial images are available. Target detection in aerial images has become one of the research hotspots in the field of computer vision, and has a wide range of applications in the fields of traffic control, airport monitoring, oil depot monitoring, near shore ship detection, and military target discovery. [0003] When the target scale changes drastically, the background is complex, the target direction is arbitrary, and the meteorological environment is complicated, it is an important and urgent problem to use the semantic information of the scene and the target and the difference between the target a...

AI Technical Summary

Problems solved by technology

[0014] The purpose of the present invention is to provide a method for detecting rotating targets in remote sensing images, to solve the problem that the detection of optical remote sensing targets is not robust enough when the background is complex, the coordinates of the rotating quadrilateral surrounding the target cannot be predicted, and the scene and target are not fully utilized. The problem of semantic information between

Images