YOLOv7-DCN-SORT: An algorithm for detecting and counting targets on Acetes fishing vessel operation

文献类型： 外文期刊

第一作者： Sun, Yueying

作者： Sun, Yueying;Zhang, Shengmao;Shi, Yongchuang;Tang, Fenghua;Dai, Yang;Sun, Yueying;Chen, Junlin;Xiong, Ying;Li, Lin

作者机构：

关键词： Offshore fishing; Deep learning; YOLOv7; Hungarian matching method; Kalman filter

期刊名称：FISHERIES RESEARCH （ 影响因子：2.4； 五年影响因子：2.6 ）

ISSN： 0165-7836

年卷期： 2024 年 274 卷

页码：

收录情况： SCI

摘要： The quantification of fishing information on fishing vessels is a prerequisite for implementing refined management of quota-based fishing. In order to address the target detection and information quantification issues in the quota-based fishing of Acetes chinensis, this study installed an Electronic Monitoring (EM) system on Acetes chinensis fishing vessels. Using the EM system video as a data source. Based on YOLOv7, an improved object detection algorithm (YOLOv7-DCN) is proposed. Additionally, drawing on the main ideas of the SORT algorithm, a target counting algorithm is also proposed (YOLOv7-DCN-SORT). YOLOv7-DCN object detection algorithm uses DCNv2 as the backbone network to detect the main targets in fishing vessel operations, improving the network's ability to detect deformable targets. The YOLOv7-DCN-SORT target counting algorithm utilizes the YOLOv7-DCN obtained in the detection phase as the target detection model. It applies the Kalman filter and Hungarian algorithm from the SORT algorithm to track and predict the counted targets. By setting collision detection lines, timestamps, thresholds, and counters, this algorithm can accurately count the number of baskets filled with Acetes chinensis and the number of nets deployed during fishing operations. The results show that: 1) The improved YOLOv7-DCN achieved precision, recall, mAP, and F1-score of 98.21%, 98.43%, 99.19%, and 98.33%, respectively, for each target detection category on the test set. These values represent improvements of 2.06%, 0.64%, 0.08%, and 1.37% compared to the original YOLOv7 model. 2) The YOLOv7-DCN-SORT algorithm achieved counting accuracy rates of 82.00% for counting the number of Acetes chinensis baskets and 96.61% for the number of deployed nets. In summary, this study provides methods for automated recording and intelligent information processing in operations on offshore fishing vessels, serving as a reference for quota-based fishing management decisions.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序