Aerial image small target detection method, device and equipment and storage medium

Abstract

The invention provides an aerial image small target detection method, device and equipment and a storage medium, and relates to the technical field of digital image detection. The method comprises the steps: obtaining an aerial image; detecting the aerial image by using a small target detection network to obtain a detection result, wherein the small target detection network is obtained by training an improved Yolov3 network model, the improved Yolov3 network model comprises five lightweight residual modules and an improved output layer, and the five residual modules are input to the output layer through specific layer feature fusion. By adopting the above method, a problem that small target features are lost due to the fact that the network is too deep during aerial image small target detection is solved. According to the method, the problem that the model is too large when small target detection is applied to the mobile platform is solved, the detection speed is higher through the light-weight small target detection network provided based on cloud computing, the real-time detection effect is achieved, and the problem that real-time target detection cannot be carried out on the mobile platform is solved.

Description

technical field [0001] The present invention relates to the technical field of digital image detection, in particular to an aerial photography small target detection method, device, equipment and storage medium. Background technique [0002] With the rapid development of UAV technology, the automatic detection and tracking of UAV ground targets plays an important role in the fields of reconnaissance and early warning. Small target detection under aerial photography is the core technology to solve this kind of problem, and it is also one of the technologies that need to be overcome in the field of computer vision. In addition, with the rise of deep learning, the network is gradually deepened, and the model is gradually increasing. It often takes a long time to train on a high-performance GPU computer, and it is even more difficult to perform target detection on a mobile platform drone. , and cloud computing has powerful computing resources, which can realize rapid training a...

AI Technical Summary

Problems solved by technology

[0005] 1. The Darknet53 network is used in the Yolov3 target detection framework, which contains a total of 53 convolutional layers and 8 downsampling layers. Such excessive downsampling will cause the characteristics of small targets to be lost in the deep layer, resulting in the final deep layer output already Cannot detect small targets;

[0006] 2. The Yolov3 target detection framework has a deep network and complex floating-point operations, and the final output model size reaches 240.6M, which is not conducive to running on some platforms with small storage space and weak computing power

[0007] 3. The training of the YOLOv3 model requires a lot of computing resources, and the training takes a long time

When detecting on mobile platforms such as drones, it is often impossible to achieve real-time detection

Images