Ocean Small Target Detection in SAR Image Based on YOLO-v5

Qilong Dou ； Mingzhong Yan

《International Core Journal of Engineering》 7卷3期 (2021/03) Pp. 167-173

https://doi.org/10.6919/ICJE.202103_7(3).0024

摘要

Aiming at the problem of high missing rate and weak generalization ability of small targets detection in SAR image by traditional detection methods, this paper proposes a small target algorithm of sea surface based on improved YOLO-v5 SAR image. First production of a wide range of ports and sea small target data set, and then the characteristics of the data set to improve optimization of deep learning network, including through adaptive algorithm convergence speed and improve model anchor box detection effect, increase the residual network to stem the spread of deep neural network gradient disappeared and excessive fitting, add different neural network model. The experimental results show that the miss detection rate of yolov5 is 1.78% lower than that of yolov3, and it has a lighter model than yolov4 model.

關鍵字

YOLO-v5 ； Ship Inspection ； Residual Network ； K-Means ； Synthetic Aperture Radar Image

參考文獻

Girshick R.Fast R-CNN[J].Computer Science,2015.

Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(6).

Feng Hong, Changhua Lu, Chun Liu, et al. PGNet: Pipeline Guidance for Human Key-Point Detection. 2020, 22(3).

AI Jiaqiu, Cao Zhenxiang, Mao Yuxiang, Wang Zhanghuai, Wang Fanfan, Jin Jing. An improved ship detection algorithm for SAR image bilateral CFAR in complex environment [J/OL]. Acta radar Sinica: 1-17 [2020-12-23] http://kns.cnki.net/kcms/detail/10.1030.TN.20201214.1001.002.html.

Chen ququ. Research on SAR image target recognition method based on deep learning [D]. Hefei University of technology, 2020China National Standardization Management Committee. Specifications of Crane Design (China Standardization Press, China 2008), p. 16-19.