基于被动水声信号的加州鲈摄食强度识别  

Identification of Feeding Intensity of Micropterus salmoides Based on Passive Underwater Acoustic Signals

作　　者：李路[1,2] 周玉凡 孙超奇 周铖钰 朱明 谭鹤群[1,2] 万鹏 LI Lu;ZHOU Yufan;SUN Chaoqi;ZHOU Chengyu;ZHU Ming;TAN Hequn;WAN Peng(College of Engineering,Huazhong Agricultural University,Wuhan 430070,China;Key Laboratory of Agricultural Facilities Engineering,Ministry of Agriculture and Rural Affairs,Wuhan 430070,China)

机构地区：[1]华中农业大学工学院,武汉430070 [2]农业农村部水产养殖设施工程重点实验室,武汉430070

出　　处：《西南大学学报（自然科学版）》2025年第2期1-16,共16页Journal of Southwest University(Natural Science Edition)

基　　金：国家重点研发计划项目(2022YFD2001700);湖北省科技重大项目(2023BBA001);中央高校基本科研业务费专项资金资助项目(2662023GXPY006)。

摘　　要：为解决高密度养殖加州鲈摄食强度识别难的问题,实现加州鲈的精准投喂,提出了一种基于被动水声信号的加州鲈鱼群摄食强度识别方法。该方法以高密度养殖条件下的加州鲈为研究对象,采用摄食声信号采集装置获取加州鲈鱼群的摄食声信号,经预处理后提取摄食声信号的多种特征,通过随机森林、皮尔逊(Pearson)相关性分析及主成分分析筛选出较为重要的特征,基于粒子群算法(particle swarm optimization,PSO)和多层感知机(multi-layer perceptron,MLP)建立PSO-MLP鱼群摄食强度识别模型。结果表明:基于PCA特征选择的PSO-MLP识别模型的分类识别效果更好,其准确率达到97.88%,识别时长为6.24 s,与基于随机森林和基于皮尔逊相关性分析的模型相比,识别准确率分别提高2.61%和1.02%,识别时长缩短1.04 s和1.09 s。说明基于被动水声信号的加州鲈鱼群摄食强度识别方法有效提高了鱼群摄食强度识别的准确率和效率,可为智能投喂系统的开发提供技术支持。To solve the problem of identifying the feeding intensity of high-density cultured Micropterus salmoides and achieve precise feeding of M.salmoides,this paper proposed a method for identifying the feeding intensity of school of M.salmoides based on feature of passive underwater acoustic signals.This method took the school of M.salmoides as the research object and used a feeding sound signal acquisition device to obtain the feeding sound signal of the school of M.salmoides.After preprocessing,multiple features of the underwater sound signal were extracted.The important features were selected through Random Forest,Pearson Correlation Analysis,and Principal Component Analysis,and a PSO-MLP fish school feeding intensity recognition model was established based on Particle Swarm Optimization and Multi-Layer Perceptron.The results showed that the PSO-MLP recognition model based on Principal Component Analysis feature selection had better recognition performance,with a classification recognition accuracy of 97.88%and a recognition time of 6.24 seconds.Compared with the RF and the Pearson correlation analysis based PSO-MLP recognition model,the recognition accuracy was increased by 2.61%and 1.02%,and the recognition time was shortened by 1.04 seconds and 1.09 seconds,respectively.This method effectively improved the accuracy and efficiency of fish feeding intensity recognition,and can provide technical support for the development of intelligent feeding systems.

关 键 词：水产养殖 鱼类摄食强度 被动水声信号 加州鲈 特征提取 

分 类 号：S969.3[农业科学—水产养殖]