基于物联网和流式计算的蛋鸡设施养殖数字化初探

发布时间：2018-03-01 09:09

本文关键词： 蛋鸡物联网分布式流式计算蛋鸡行为决策分析　出处：《中国农业大学》2017年博士论文　论文类型：学位论文

【摘要】：本研究面向现代蛋鸡行业健康养殖的需求,探索如何将蛋鸡养殖从传统粗放管理转化成实际生产过程精细养殖,并对感知的蛋鸡生理、环境及生产过程数据进行实时处理,对照厂家提供的相关育种标准数据,对蛋鸡在禽舍的生存状态及生产经济效益予以实时监测,提高了蛋鸡养殖企业安全预警的水平及管理者的决策能力。针对上述目标,本研究研发了如下关键技术:基于物联网的蛋鸡养殖生产过程数据的实时获取与传输技术,基于分布式流式计算的蛋鸡养殖生产数据实时处理分析技术,基于机器学习的自动追踪蛋鸡行为算法,通过有效整合,研发了蛋鸡设施养殖数字化智能监测与远程管理系统,并在实际鸡场得到应用。该系统实现了对多个异地蛋鸡场的环境、生理及生产过程数据的实时监测、获取、处理和预警,提高了养殖场对蛋鸡养殖生产实时数据的存储、分析和管理的能力;实现了对蛋鸡行为视频的自动分析。该系统在蛋鸡场的应用实现了对养殖情况的精细感知,提升了我国蛋鸡养殖业的管理水平,降低人工劳动强度,对支撑我国蛋鸡行业的福利养殖、实时决策都有很大帮助。主要结论如下:(1)通过文献阅读和现场调研,本研究总结了蛋鸡养殖中的三类数据:环境、生理及生产过程数据。针对这三类数据的特点,使用物联网技术自动获取可精细感知的数据;对其它数据,改变原有纸质报表留存的方式,开发了基于WEB和APP的人工填报方式,实现了数字化,方便了现场工作人员的工作。应对蛋鸡养殖场一般地处偏僻,信息通讯难度大的特点,并基于节省流量和带宽的目标,本研究提出了异步传输机制,编写了监测数据更新的程序,最大程度的避免了传输过程中的数据丢失,并可显著降低通信运营成本。(2)为实现对蛋鸡养殖生产大数据的实时处理、计算和预警,自主研发了应用于蛋鸡设施养殖中的分布式流式计算框架Data-Canal。Data-Canal使用控制流集中、数据流分散的模型,以分布式文件系统为中间结果的存储,具有很强的可扩展性和可靠性。运行结果说明,使用Data-Canal计算框架可以解决蛋鸡养殖生产数据实时计算的问题,在部署8台机器的情况下,Data-Canal集群的处理能力峰值达到160 MB/s,延迟在分钟级别。(3)视频数据已经成为目前蛋鸡养殖生产大数据的主体。现有系统通过传感器设备收集了大量的现场视频图像数据,但往往缺乏应用分析手段,进而无法从中得到有价值的信息。基于此本研究提出了基于机器学习的自动追踪蛋鸡行为算法,实现了对蛋鸡行为视频数据的自动分析。该算法使用HOG特征描述方法提取鸡只样本特征,训练混合SVM模型计算鸡只最优位置,可自动识别小群体蛋鸡中任意鸡只的行为轨迹,计算其运动距离、速度等,量化鸡只的运动行为。结合设施内区域分布,自动统计鸡只对不同区域的使用情况,间接对采食/饮水行为进行量化。(4)本研究以规模化养殖蛋鸡为研究对象,在分布式架构下,使用物联网与流式计算技术,实现了对蛋鸡养殖生产过程实时数据的自动获取、异步传输和实时处理,并结合基于机器学习的自动追踪蛋鸡行为的算法,实现了对非结构化数据(视频图像)的自动处理,最后以数据分析服务的方式在实际蛋鸡场应用。数据分析成果主要有养殖经济效益分析、生产过程实时预警服务、环境数据分析、多维数据分析、生产信息管理服务等。
[Abstract]:This research is focused on the healthy culture of modern egg industry needs to explore how the laying hens are transformed from traditional extensive management to fine breeding process, and laying hens on physiological perception, environment and production process real-time data processing, control manufacturers to provide relevant breeding standard data to real-time monitoring of laying hens in poultry house living state the production and economic benefits, improve the level of enterprise management and laying hens safety warning decision makers'ability. According to the above objectives, this study developed the following key technologies: laying hens production process real-time data networking and transmission technology based on the analysis of real time processing technology of laying hens production data stream based on distributed computing. Automatic tracking behavior of layers algorithm based on machine learning, through the effective integration and development of laying hens breeding facilities and digital intelligent monitoring The remote management system, and has been applied in actual farms. The system implements the multiple remote egg farm environment, real-time monitoring, physiology and production process data, processing and early warning, improve the farm production of laying hens in real-time data storage, analysis and management ability; to realize the automatic analysis of behavior of layers video. The system realizes the fine perception of the situation in the application of breeding egg chicken, enhance the management level of China's poultry industry, reduce labor intensity, to support China's egg industry of animal welfare, real-time decision-making are of great help. The main conclusions are as follows: (1) through the literature reading and the field investigation, this paper summarizes three types of data cultivationoflayers: environment, physiology and production process data. According to the characteristics of these three kinds of data, can automatically obtain the fine sensing data to use networking technology; Other data, change the original paper statements retained, developed WEB and APP artificial reporting methods based on digitization, convenient on-site staff. To deal with the poultry farm general remoteness, difficulty of information communications, and based on the save traffic and bandwidth of the target, this study proposes an asynchronous the transmission mechanism, preparation of monitoring data to update the program, the maximum to avoid the loss of data in the transmission process, and can significantly reduce the operating costs of communication. (2) in order to realize the real-time processing of laying hens production data, calculation and early warning, Data-Canal.Data-Canal computing framework using control flow concentrated flow of independent research and development of distributed application in the laying hens breeding facilities, distributed data flow model, intermediate results to a distributed file system for storage, has strong scalability and reliability. The operation. The results illustrate that the use of Data-Canal computing framework can solve the real-time calculation of laying hens production data, for the deployment of the 8 machine case, the peak processing ability of Data-Canal cluster is up to 160 MB/s, the delay in the minute level. (3) the video data has become the subject of Laying Hens Production data. The existing collection of live video system a large number of image data through the sensor devices, but often lack the application of analysis method, and then to get valuable information from it. This research is put forward based on the behavior of layers automatic tracking algorithm based on machine learning, to realize the automatic analysis of the behavior of layers of video data. The algorithm uses HOG feature extraction method only samples the characteristics of chicken training, mixed SVM model to calculate the optimal position of chicken, chicken can track any automatic recognition of small groups of laying hens, calculate the moving distance, speed The degree of movement, quantitative chicken behavior. Combined with the regional distribution facilities, automatic statistics on the use of chickens in different areas, indirectly on foraging / drinking behavior was quantified. (4) based on the large-scale breeding hens as the research object, in the distributed architecture, the use of networking and stream computing technology to realize the automatic acquisition of real-time data in laying hens production process, asynchronous transmission and real-time processing, and the combination of automatic tracking behavior of layers of machine learning algorithm based on the realization of non structured data (video) automatic processing, and finally to the data analysis service in practical application. The data analysis results of hen the main analysis of the economic benefits of aquaculture, real-time early warning service production process, data analysis, multidimensional data analysis, production information management services.

【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.44;TN929.5;S831

【参考文献】