Multi‐Scale Ship Target Detection Based on Improved Yolov5

Da Huang ； Wenhua Li

《World Scientific Research Journal》 8卷8期 (2022/08) Pp. 132-135

https://doi.org/10.6911/WSRJ.202208_8(8).0018

摘要

In the actual inland waterway scene, it is often affected by natural factors and physical conditions, resulting in the problem of inconsistent ship size. To solve the above problems, this paper proposes a ship detection algorithm based on improved yolov5. Firstly, the 3*3 convolution in the residual block of the backbone network CSPDarkNet is changed into deep separable convolution to reduce the network parameters; Secondly, by adding the improved SPP feature pyramid pooling, semantic information of different sizes can be extracted from the feature map, and a multi‐scale fused feature map can be obtained. Finally, the feature layer is strengthened and CBAM attention mechanism is added, so that the network can adaptively focus on the areas that need more attention. The improved detection algorithm based on yolov5 has significantly improved the detection effect of ships of different scales, and compared with the current mainstream SSD, Faster‐Rcnn and so on, it has significantly improved the accuracy and speed.

關鍵字

Attention Mechanism ； Yolov5 ； Multi‐Scale ； Ships

參考文獻

Song. Application[2]Song Binbin. Analysis of the application of computer vision technology in automation [J]. Electronic World, 2020 (20): 10-11. doi: 10.19353/j.cnki.dzsj.2020.20.004.

Zhang Hui, Wang Kunfeng, Wang Feiyue Application progress and Prospect of deep learning in target vision detection [j] Journal of automation, 2017,43 (08): 1289-1305 DOI:10.16383/j.aas. 2017.c160822.

Woo S, Park J, Lee J Y, et al. Cbam: Convolutional block attention module[C] //Proceedings of the European conference on computer vision (ECCV). 2018: 3-19.

Howard A G, Zhu M, Chen B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J]. arXiv preprint arXiv:1704.04861, 2017.

Ge Z, Liu S, Wang F, et al. Yolox: Exceeding yolo series in 2021[J]. arXiv preprint arXiv:2107.08430, 2021.