| 摘要: |
| 为深入了解基于视觉智能感知的畜禽智慧养殖管理与疫病诊断的研究现状，本研究以“深度学习”、“个体检测”、“畜禽身份识别”、“体尺体重评估”、“体温检测”、“行为识别”、“疫病诊断”等为关键词，在Web of Science核心集合、Science Direct、CNKI等数据库就1990—2022年已发表的文献进行检索，从5个方面对研究畜禽智慧养殖管理与疫病诊断的方法与技术进行总结、归纳、分析。结果表明: 1)畜禽身份识别主要通过畜禽面部识别实现，针对单帧的畜禽面部数据设计无约束方法是未来研究方向。2)畜禽体尺体重智能评估研究中，基于三维点云的畜禽体尺体重高精度快速测量技术是研究的重点。3)由于畜禽疫病数据集的稀缺，基于小样本的畜禽疫病识别技术是突破疫病诊断的关键。4)畜禽体温检测关键是在复杂养殖环境下畜禽热窗的准确定位，通过检测分割算法对热红外模式下的图像进行精准检测。5)日常行为识别主要难点为长时间畜禽密集目标检测与跟踪，并计算其行为轨迹与特点；异常行为通过连续帧间的上下文关系进行识别，主要难点为畜禽异常行为数据稀少性和正负样本不均衡的问题。本综述对基于视觉智能感知的畜禽体温检测、体尺体重评估、行为识别与疫病诊断技术方法进行了研究现状阐述、难点分析和未来趋势展望，为视觉感知技术在畜禽养殖的技术演进和应用发展提供了参考方向。 |
| 关键词: 智慧养殖 视觉感知 身份识别 疫病诊断 行为识别 体尺体重评估 |
| DOI：10.11841/j.issn.1007-4333.2023.10.13 |
| 投稿时间：2022-12-09 |
| 基金项目:国家自然科学基金项目(61976219, 31971792)；中国农业科学院院级基本科研业务费专项(Y2021LM02,Y2021XK10,Y2022XK24)；中央级公益性科研院所所级基本科研业务费专项(Y2022QC17, JBYW-AII-2022-14,JBYW-AII-2023-06)；中国农业科学院创新工程(CAAS-ASTIP-2023-AII,ZDXM23011,CAAS-ASTIP-2016-AII)；三亚中国农业科学院国家南繁研究院南繁专项(YDLH01,YDLH07,YBXM10) |
|
| Research progress in intelligent livestock and poultry breeding management and disease diagnosis based on visual perception |
|
HE Peitong1,ZHANG Jianhua1,2*,ZHANG Ning1,XIA Xue1,CHAI Xiujuan1
|
| (1.Agricultural Information Institute of Chinese Academy of Agricultural Sciences, Beijing 100081, China;2.National Nanfan Research Institute, Chinese Academy of Agricultural Sciences, Sanya 572024, China) |
| Abstract: |
| In order to gain an in-depth understanding of the current research status of intelligent livestock and poultry breeding management and disease diagnosis based on visual intelligent perception, in this study, “deep learning”, “individual detection”, “identification of livestock and poultry”, “body temperature detection”, “body size and weight assessment”, “epidemic diagnosis”, “behavior recognition”, and other keywords were used to search for published literature from 1990 to 2022 in the Web of Science core collection, Science Direct, CNKI, and other databases. And the research methods and technologies for intelligent livestock and poultry breeding management and disease diagnosis based on visual intelligent perception from five aspects were summarized and analyzed. The results showed that: 1)Livestock and poultry identity recognition is mainly achieved through facial recognition, and designing unconstrained methods for single frame livestock and poultry facial data is the future research direction. 2)In the research of intelligent evaluation of livestock and poultry body size and weight, high-precision and rapid measurement technology based on three-dimensional point cloud is the focus of research. 3)Due to the scarcity of livestock and poultry epidemic data sets, the identification technology of livestock and poultry epidemic based on small samples is the key to breaking through the epidemic diagnosis. 4)The key to livestock and poultry body temperature detection is the accurate positioning of livestock and poultry heat windows in complex breeding environments, and to accurately detect images in thermal infrared mode through detection segmentation algorithms. 5)The main difficulties in daily behavior recognition are long-term detection and tracking of livestock and poultry intensive targets, and calculating their behavior trajectory and characteristics; Abnormal behavior is identified through context relationships between consecutive frames. The main difficulties are the scarcity of data on abnormal behavior of livestock and poultry and the imbalance of positive and negative samples. This review describes the current research status, difficulty analysis, and future trends of livestock and poultry temperature detection, body size and weight assessment, behavioral recognition, and disease diagnosis techniques based on visual intelligent perception, providing a reference direction for the technological evolution and application development of visual perception technology in livestock and poultry breeding. |
| Key words: smart farming visual perception identification disease detection behavioral recognition body size assessment weight assessment |
