Multi-scale object detection method, model, electronic equipment and its application

Abstract

This application proposes a multi-scale target detection method, model, electronic equipment and its application. The multi-scale target detection method optimizes and improves the multi-scale target detection model. The convolution of the backbone network residual unit of the multi-scale target detection model It is improved to a dual multi-scale attention module, which can adaptively integrate local features and their global dependencies, and attach two dual methods to model the semantic interdependence in the spatial and channel dimensions respectively on the traditional full convolution. Attention module, the dual attention module enriches the context feature dependence to significantly improve the detection results, which helps to detect multi-scale targets more accurately, and can accurately detect targets of different scales and scenes, especially suitable for detecting complex Multi-scale targets in the scene, and then expand the use of target detection algorithms in various application scenarios.

Description

technical field [0001] This application relates to the field of target detection, in particular to multi-scale target detection methods, models, electronic equipment and applications thereof. Background technique [0002] Target detection has always been a hot spot in the field of deep learning. It refers to the technology that uses computer vision technology to accurately detect the required target from the image, and can replace the manual to quickly obtain the target that needs to be marked from the image or video. At present, the mainstream target detection algorithms are divided into two categories according to the algorithm stage: the first category is a two-stage target detection algorithm, the more famous one is the R-CNN series and its variants; the second category is a single-stage target detection algorithm, such as The YOLO series and its variants. [0003] Many researchers have made some progress in the field of target detection and recognition. However, consid...

AI Technical Summary

Problems solved by technology

[0004] Taking the non-motor vehicle in the elevator as the target to be detected as an example, the surveillance camera in the elevator is often set in a fixed corner, and in the actual scene, there are various parking situations of non-motor vehicles entering the elevator, some non-motor vehicles are far away from The surveillance camera is very close, and the target scale obtained at this time is large; some non-motor vehicles are far away from the surveillance camera, and the target scale obtained at this time is small; some non-motor vehicles are blocked by human bodies, and the target obtained at this time is blocked; The lighting conditions in the elevator are good, and the target images acquired at this time are clear and bright; some elevators have poor lighting conditions, and the target images acquired at this time are blurry and dim. As mentioned above, the current target detection algorithm cannot accurately detect various targets Very accurate and efficient detection, which makes it unable to meet the needs of actual application scenarios

Images