DCA-MVIT: Fused DSGated convolution and CA attention for fish feeding behavior recognition in recirculating aquaculture systems

文献类型： 外文期刊

第一作者： Hu, Weichen

作者： Hu, Weichen;Yang, Xinting;Ma, Pingchuan;Fu, Tingting;Zhou, Chao;Hu, Weichen;Hu, Weichen;Yang, Xinting;Ma, Pingchuan;Fu, Tingting;Zhou, Chao;Hu, Weichen;Yang, Xinting;Ma, Pingchuan;Fu, Tingting;Zhou, Chao

作者机构：

关键词： Recirculating aquaculture system; Fish feeding behavior; Feature extraction; DCA-MVIT

期刊名称：AQUACULTURE （ 影响因子：3.9； 五年影响因子：4.4 ）

ISSN： 0044-8486

年卷期： 2025 年 598 卷

页码：

收录情况： SCI

摘要： In recirculating aquaculture systems, real-time and accurate recognition of fish feeding behavior from video is the foundation for the development of intelligent feeding equipment and system. However, complex features and abnormal postures have brought great challenges to achieve the above objectives. Therefore, this paper proposes a fish feeding behavior recognition model, DCA-MVIT (DSGated convolution and Coordinate Attention for Multiscale Vision Transformers version 2), with enhanced feature extraction ability. Firstly, data enhancement such as horizontal flipping, random noise and random filtering are introduced to solve the dataset equilibration issues. Then, the Coordinate Attention enhances the feature capture ability of the feeding domain. Finally, the proposed DSGated convolution further improves the feature capture ability and reduces model parameters. The experimental results showed that the Precision and Top-1 Accuracy of DCA-MVIT in fish feeding behavior recognition reached 96.62 % and 83.13 %. These are 1.86 % and 5.52 % higher than the original model, respectively. The proposed model provides important technical support for developing scientific feeding strategies and intelligent feeding systems.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序