基于Faster R-CNN和时序统计的肉鸭行为节律分析  

Analysis on Rhythmic Behavior of Meat Ducks Based on Faster R-CNN and Time-series Statistics

作　　者：刘啸虎 肖德琴[1,2] 刘又夫 王春桃[1,2] 周敏[1,2] LIU Xiaohu;XIAO Deqin;LIU Youfu;WANG Chuntao;ZHOU Min(Key Laboratory of Smart Agricultural Technology in Tropical South China,Ministry of Agriculture and Rural Affairs,South China Agricultural University,Guangzhou,Guangdong 510642;College of Mathematics and Informatics,South China Agricultural University,Guangzhou,Guangdong 510642)

机构地区：[1]华南农业大学农业农村部华南热带智慧农业技术重点实验室,广东广州510642 [2]华南农业大学数学与信息学院,广东广州510642

出　　处：《中国家禽》2023年第11期95-104,共10页China Poultry

基　　金：现代农业产业技术体系建设专项资金(CARS-42-13)。

摘　　要：为实现自动化笼养肉鸭的行为识别及行为节律分析,文章提出了一种基于Fast⁃er R-CNN和时序统计的笼养肉鸭行为节律分析方法(Temporal statistical behavior recognition algorithm for caged meat ducks,TSBR算法)。该算法利用Faster R-CNN检测出常态化监控视频关键帧中的鸭头和乳头式饮水器,形式化定义并识别笼养肉鸭的饮水、进食、站立和趴卧四种关键行为,采用时序统计方法分析各行为的发生时间与时长,得到肉鸭行为节律。对连续3 d采集肉鸭视频进行试验分析,结果显示:算法检测的平均精度为0.93,肉鸭站立、趴卧、进食和饮水行为占比分别为:30.52%、28.05%、1.16%和40.27%,肉鸭在2∶00~3∶00和19∶00~20∶00较为活跃,6∶00~10∶00和20∶00~21∶00较为安静。综上,在相同试验条件下,文章提出算法的精度能够支撑肉鸭的日常行为识别;为保证肉鸭正常休息,饲料投喂应当避免6∶00~10∶00以及20∶00~21∶00两个时间段。研究结果可为笼养肉鸭精准管理及健康状况分析提供技术支撑,具有较好的应用前景。In order to realize the behavior recognition and rhythmic behavior analysis of caged meat ducks,this paper pro-posed a temporal statistical behavior recognition algorithm(TSBR)for caged meat ducks based on Faster R-CNN and temporal statistics.The algorithm used Faster R-CNN to detect the duck head and nipple waterer in the key frames of the surveillance vid-eo,defined and recognized four key behaviors of caged meat ducks:drinking,eating,standing and lying down,and adopted time-series statistical methods were used to analyze the occurrence time and duration of each behavior to obtain the rhythmic behavior of meat ducks.The experimental analysis of the three-day continuous meat duck video showed that the average detection accura-cy of the algorithm was 0.93,and the proportions of meat duck standing,lying down,eating and drinking behaviors were 30.52%,28.05%,1.16%and 40.27%respectively,meat ducks were more active from 2∶00 to 3∶00 in the morning and 19∶00 to 20∶00 in the evening,and quieter from 6∶00 to 10∶00 in the morning and 20∶00 to 21∶00 in the evening.In summary,under the same ex-perimental conditions,the algorithm proposed in this paper can support the daily behavior recognition of meat ducks.In order to ensure the normal rest of meat ducks,feed should be avoided in the time periods of 6∶00 to 10∶00 and 20∶00 to 21∶00.The re-sults could provide technical support for precise management and health status analysis of caged meat ducks,and had a good ap-plication prospect.

关 键 词：肉鸭 行为识别 行为节律 FasterR-CNN 

分 类 号：S834[农业科学—畜牧学] TP391.4[农业科学—畜牧兽医]