Small target detection algorithm based on YOLOv4
Quanfang Li, Liucun Zhu, Boquan Li, Junqi Luo, Daopeng Liu, Mingyou Chen · 2022
In recent years, deep learning has been rapidly developed in the field of target detection. How to accurately and efficiently locate small target objects is one of the main difficulties in target detection research. This paper proposes a small target detection method based on YOLOv4 algorithm. In the process of data preprocessing, the small target enhancement method is used instead of the Mosaic method to increase the number of small targets in the image and improve the small target detection ability of the network model. Using the K-means to get the best size of the anchor box to improve the accuracy and speed of small target detection. Experiments on the DOTA data set show that it meets the performance requirements and can effectively and accurately locate the target object in the image with the accuracy of 93.2%.