Small target detection algorithm based on improved YOLOv5

Abstract

The invention discloses a small target detection algorithm based on improved YOLOv5, and the method comprises the steps: 1, carrying out the enhancement of a collected to-be-detected face mask data set through Mosaic-8 data, i.e., carrying out the random cutting, random arrangement and random zooming of eight pictures, combining the eight pictures into one picture, and reasonably introducing some random noises; and a second step of adding a new-scale feature extraction layer in the YOLOv5 feature fusion network, adjusting a target frame regression formula of the YOLOv5 network, and improving a loss function. And step 3, sending the enhanced data into the network for iterative training, and adjusting the learning rate by using a cosine annealing algorithm. And step 4, after training is completed, sending a to-be-detected picture into the optimal model obtained after training, detecting the category and the position of a target, and finally obtaining a recognition result. The improved algorithm is applied to protective mask wearing detection in a dense crowd scene, and an experiment result shows that compared with an original YOLOv5 algorithm, the algorithm has higher feature extraction capability and higher detection precision in small target detection.

Description

technical field [0001] The invention relates to the technical field of small target detection, in particular to a small target detection algorithm based on improved YOLOv5. Background technique [0002] As one of the core problems in the field of computer vision, target detection is to use image processing, deep learning and other technologies to locate the object of interest from the image or video, judge whether the input image contains the target through target classification, and use target positioning to find the target. The position of the object and the target are framed. Its task is to lock the target in the image, locate the target position, and determine the target category. It is widely used in computer vision fields such as face recognition, automatic driving, pedestrian detection, and intelligent monitoring. The traditional target detection algorithm consists of three parts, namely region selection, feature extraction and classifier. However, due to the characte...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to propose a small target detection algorithm based on improved YOLOv5, by using the Mosaic-8 method for data enhancement, adding a shallow feature map, adjusting the loss function to enhance the network's perception of small targets, and modifying the target frame Regression formula, solve the problem of gradient disappearance in the training process, and improve the detection accuracy of small targets

Images