Method for rapidly detecting small target under view angle of unmanned aerial vehicle based on yolov3

Abstract

The invention discloses a method for rapidly detecting a small target under a view angle of an unmanned aerial vehicle based on yolov3, and the method comprises the steps: obtaining an image of a to-be-detected scene, inputting the image into a trained small target detection model, and outputting a detection result of a small target, wherein the model building process comprises the following steps of: adding a prediction unit on a quadruple down-sampled feature map output by a second group of residual blocks in the Darknet53, wherein the prediction unit comprises two residual units which are connected in sequence; carrying out 2-time up-sampling operation on the feature map output by the 8-time down-sampling prediction branch output by the third group of residual blocks, and then carrying out feature fusion operation on the feature map and the feature map output by the prediction unit; adding two sequentially connected residual units behind the feature map output by the second group of residual blocks of the YOLOv3 network so as to establish a small target detection network; and training the small target detection network by using the preprocessed data set, and then carrying out network pruning on the trained small target detection network to obtain the small target detection model.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a yolov3-based rapid detection method for small targets under the perspective of an unmanned aerial vehicle, combined with model compression technology, to realize rapid detection of small-scale targets on mobile embedded devices. Background technique [0002] As an emerging technology, UAV target detection has a wide range of applications in aerial image analysis, intelligent monitoring, and route detection. Object detection has made great progress in recent years, especially with the development of large-scale visual datasets and the improvement of computing power, deep neural networks (DNNs), especially convolutional neural networks (CNNs), demonstrated record-breaking performance in computer vision tasks. However, it is still a challenging work due to the view-specific target scale issue. [0003] At present, some excellent target detection algorithms have achieved...

AI Technical Summary

Problems solved by technology

Since there are a large number of convolution kernels with a size of 1x1 in the currently designed neural network, it greatly limits the use of low-rank approximation methods to compress the parameters of the convolution kernel.

Images