舍饲环境下母羊产前典型行为识别方法研究

发布时间：2018-01-08 18:00

本文关键词：舍饲环境下母羊产前典型行为识别方法研究　出处：《内蒙古农业大学》2017年博士论文　论文类型：学位论文

【摘要】：肉羊产业作为内蒙古地区的特色产业,对内蒙古的经济发展起到至关重要的作用。近年来,为恢复草原生态实现可持续发展,实施了轮牧禁牧政策,肉羊的饲养模式由传统的放养逐渐转变为设施养殖,规模化、集约化养殖成为未来肉羊产业的主要发展趋势。随着舍饲饲养羊只密度的增加,养殖环境无法得到保障,导致羊只处于亚健康或不健康状态,羊的发病率大幅度上升;而孕期母羊的健康状况又会直接影响仔羔的生存状态,同时与饲养者的经济效益密切相关,因此利用自动化、智能化方法实时、准确的监测母羊产前行为信息,为提高母羊的生产力和繁殖效率及母羊和羔羊的健康水平具有十分重要的意义。本文以小尾寒羊母羊为研究对象,获取了母羊产前站立、行走、趴卧和刨地四种典型运动行为及采食和饮水行为信息,并对母羊产前运动行为信息进行了小波阈值去噪、特征参数的提取以及行为分类识别模型的建立。取得了如下结论:(1)设计了可穿戴式的、以三轴加速度传感器MPU6050为核心的母羊产前运动行为实时监测装置,以及以红外传感器为核心的母羊产前采食及饮水行为监测装置,实现了对母羊产前运动、采食及饮水行为信息的连续、实时监测。采用LabVIEW软件设计了上位机软件,实现母羊产前典型行为信息的采集、显示与存储。试验证明了装置的准确性、可靠性和有效性。(2)利用小波阈值去噪方法对母羊产前四种运动行为三轴加速度数据进行去噪,并提出改进的小波阈值函数去噪方法。通过MATLAB软件平台实现运动行为加速度信号的去噪处理。试验表明,改进后的小波阈值函数去噪算法能够取得较好的去噪效果。(3)对去噪后的母羊产前运动行为加速度数据进行特征提取,选取了方差、主峰频率及频率能量多个表征母羊行为的加速度数据特征,采用主元分析方法对特征值进行降维。试验表明,经过特征降维之后的母羊行为识别率提高了 9.8%。(4)针对母羊运动行为中趴卧与站立行为分类识别率低的问题,提出在MATLAB软件平台上,采用K-means聚类算法对母羊产前趴卧行为进行识别,经验证此算法对趴卧行为的识别率能够达到99%;利用BP神经网络算法对剩余的三种运动行为:站立、行走、刨地行为进行识别。试验结果表明,神经网络分类算法对母羊产前运动行为平均识别率达到78.93%;针对神经网络识别算法对母羊产前非标准运动行为如站立时蹭栏杆、挠耳朵等行为识别效果差的问题,提出了采用递阶遗传算法对BP神经网络的结构和参数进行改进。经验证,改进后的识别算法泛化能力有了明显的改善,对母羊产前运动行为平均识别率达到了 89.12%。(5)针对母羊产前采食及饮水行为监测结果中存在的误差,提出了通过母羊的运动行为状态对母羊产前采食及饮水行为进行校正。经验证,校正后的母羊产前采食及饮水行为识别率提高了 16%,完全能够满足母羊行为采食及饮水行为识别的要求。(6)本文对母羊不同分娩时期运动行为、采食及饮水行为分配的时间及行为节律进行了统计分析,所得行为规律与国内外其他研究学者的结论相一致,进一步证明了所设计的母羊产前行为识别算法模型的准确性及有效性。
[Abstract]:The sheep industry is a pillar industry of Inner Mongolia area, play an important role in the economic development of Inner Mongolia. In recent years, for the restoration of grassland ecological sustainable development, the implementation of the grazing sheep grazing policy, feeding mode has gradually changed from the traditional stocking for breeding facilities, large-scale, intensive farming has become a main trend for the development of the sheep industry with the increase of sheep feeding breeding density, breeding environment can not be guaranteed, resulting in sheep in sub healthy or unhealthy, the incidence rate of sheep increased significantly; and the health status of pregnant ewes and lambs will directly affect the offspring's survival condition, at the same time with the breeders is closely related to the economic benefit, therefore the use of automation, intelligent method for real-time monitoring, accurate prenatal ewe behavior information, to improve the health level of ewe productivity and reproductive efficiency and the ewe and lamb There is very important significance. In this paper, small tail Han sheep as the research object, the ewe prenatal standing, walking, lying and digging, four kinds of typical behavior and feeding and drinking behavior information, and the information of the ewe prenatal exercise behavior of wavelet threshold denoising, feature parameter extraction and behavior classification model. The results are as follows: (1) the wearable design, ewe prenatal exercise real-time monitoring device behavior with three axis acceleration sensor MPU6050 as the core, and the infrared sensor as the core of the ewe prenatal feeding and drinking behavior monitoring device, the realization of ewes feeding and prenatal exercise. Drinking behavior information for continuous, real-time monitoring. The upper computer software is designed by LabVIEW software, realize the information collection of prenatal behaviors typical of ewes, display and storage. The test indicates that the accuracy of the device, reliability And effective. (2) using wavelet threshold denoising method of four kinds of prenatal exercise behavior ewe three axis acceleration data denoising, and puts forward the improved wavelet threshold function denoising method. Achieve the denoising motion acceleration signal through the MATLAB software platform. The results show that the improved wavelet threshold function denoising the algorithm can obtain better denoising results. (3) of the ewes prenatal exercise acceleration data after denoising behavior for feature extraction, selection variance, peak frequency and frequency characteristic of acceleration data of multiple energy Behavior Characterization of ewes, using principal component analysis method of characteristic values of dimensionality reduction. The results show that after ewe behavior after recognition feature reduction rate increased by 9.8%. (4) according to the classification of lying and standing behavior in the low rate of ewe behavior problems, put forward on the MATLAB software platform, using K-means clustering algorithm Method of prenatal ewe lying behavior recognition. The experiment prove that the algorithm for recognition of lying behavior rate can reach 99%; three kinds of residual motion behavior using BP neural network algorithm: standing, walking, digging behavior recognition. The experimental results show that the neural network classification algorithm average recognition on ewe prenatal exercise behavior reached 78.93%; the neural network recognition algorithm for non standard behavior such as prenatal ewe standing against the railing, the effect of recognition of the problem of poor behavior such as scratching his ear, put forward the hierarchical genetic algorithm structure and parameters of the BP neural network was improved. After verification, the improved generalization ability of recognition algorithm is significantly improved. The average recognition rate of ewe prenatal exercise behavior reached 89.12%. (5) according to the errors of ewes feeding and drinking behavior of prenatal monitoring results, put forward dynamic behavior through ewe transport State Food and drinking behavior correction to prenatal ewes. After validation, the correction rate increased by 16% ewes after prenatal feeding and drinking behavior recognition, can fully meet the behavior of ewes feeding and drinking behavior recognition requirements. (6) according to different delivery period of ewe exercise behavior, and behavior and foraging time rhythm the distribution of drinking behavior was analyzed, the behavior is consistent with other domestic and foreign scholars' research conclusion, further proves the accuracy and effectiveness of ewe prenatal behavior identification algorithm design model.

【学位授予单位】：内蒙古农业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP274;S826

【相似文献】