Unmanned aerial vehicle small target detection method and system and storable medium

Abstract

The invention discloses an unmanned aerial vehicle small target detection method and system and a storable medium, and relates to the technical field of computer vision. The method comprises the steps: constructing an unmanned aerial vehicle detection network based on a lightweight depth model; collecting unmanned aerial vehicle images and constructing an unmanned aerial vehicle image database to obtain a training data set; inputting the training data set into the unmanned aerial vehicle detection network for neural network training until the network converges; and inputting a to-be-detected unmanned aerial vehicle image into the trained unmanned aerial vehicle detection network to obtain an unmanned aerial vehicle small target detection result of the to-be-detected unmanned aerial vehicle image. According to the method, the problem that complex scenes are difficult to deal with due to the fact that semantic information extraction is incomplete when a traditional target detection method is used for unmanned aerial vehicle small target detection tasks is solved, the unmanned aerial vehicle detection precision can be effectively improved on the premise that the model light weight is guaranteed, and the method has practical engineering application value.

Description

technical field [0001] The present invention relates to the technical field of computer vision, and more specifically relates to a small target detection method, system and storage medium of an unmanned aerial vehicle. Background technique [0002] With the promotion and popularization of drones, the difficulty of maintenance and management at many levels such as public safety and airspace traffic has also greatly increased, which also poses challenges to the traditional civil aviation regulatory system. How to carry out the necessary supervision and management of the flight of civilian drones is an important and difficult point to be solved urgently. [0003] In order to improve the efficiency and strength of supervision, it is necessary to introduce automation technology to realize UAV monitoring. Traditional detection methods mainly use acoustic detection technology and radar detection technology to detect UAVs, but this method has many limitations: for acoustic detection...

AI Technical Summary

Problems solved by technology

[0003] In order to improve the efficiency and strength of supervision, it is necessary to introduce automation technology to realize UAV monitoring. Traditional detection methods mainly use acoustic detection technology and radar detection technology to detect UAVs, but this method has many limitations: for acoustic detection Technically speaking, it is difficult to filter the noise in the sound signal and extract effective features when encountering a complex environment. Effective detection of medium and long-distance UAVs; for radar detection technology, when facing low-altitude UAVs and other low-slow and small targets, due to not only the interference of ground clutter, but also the radar scattering interception generated by UAVs The area is small (the surface is mostly composed of non-metallic composite materials), and it is usually difficult to achieve the ideal detection effect. In addition, radar equipment usually has a working blind area and is not sensitive to headspace, low altitude and short-range targets, which also limits its application scenarios

[0005] Most target detection methods based on convolutional neural networks (such as classic Faster, RCNN, YOLOv3, SSD, etc.) can show excellent performance on general benchmark datasets for target detection such as PASCAL VOC, MS COCO, etc. The performance on such small target detection tasks is slightly insufficient, and there are two main problems: On the one hand, these classic algorithm models mainly solve general target detection tasks, including large, medium and small targets of different scales. , so the network cannot fully extract the semantic information contained in the low-level feature map, which is extremely important for the detection of small targets such as drones; on the other hand, the classic target detection method relies on a large number of network parameters, which will eventually generate Larger weight files (usually larger than 10MB) are used to save these parameters, and for low-altitude UAV detection tasks, it does not have a large number of complex scenes in general target detection tasks, and the background information is relatively simple. In practical application scenarios There is no need for a large number of parameters, and while reducing parameters, it can further reduce the requirements for video memory and deployment overhead

Images