基于三轴加速度传感器的山羊行为特征识别研究

发布时间：2018-08-22 18:33

【摘要】：近年来，随着信息技术的不断发展，我国对山羊畜群的健康养殖、安全生产、质量监管及质量溯源的要求越来越严格，但是对山羊畜群行为特征监测依然要依赖饲养员的直觉和经验来判断，不但耗费大量人力，而且工作效率也低，已不再适合规模化、集约化的现代家畜养殖业。为了快速并精准的判定动物行为与疫病诊断的关系，各国研究人员引入了机器视觉技术、视频或图像处理技术、声学探测等行为特征识别技术，，但是这些识别技术对环境光线、摄像机的固定位置及角度有较高要求。如果想要连续不间断地记录动物（特别是散养放牧的家畜动物）的行为数据，就必须依赖于传感器信息采集与处理技术。为此本论文设计的核心就是利用三轴加速度传感器对山羊典型日常行为特征进行识别研究。主要内容包括：（1）本文有针对性的对山羊的三轴加速度数据进行采集并分析，通过K-means均值聚类算法和SVM算法分别对采集的山羊三轴加速度数据集进行迭代训练，分析比较两种算法对分类模型精度的影响，并结合动作发生的时间识别并验证山羊的4种典型日常行为对应的加速度数据模型。研究结果显示：依据K-means均值聚类算法对山羊的躺卧、站立或慢走、采食、跨跳等典型日常行为识别的准确率达87.76%，但经过SVM算法优化之后的典型行为识别率提高了近6个百分点。研究结果表明该算法对山羊典型日常行为的特征分类与识别能够达到较高的分类精度，具有更高的寻优效率，在行为识别领域有非常高的应用价值。（2）本文对山羊的生活习性进行了详细分析，研究了三轴加速度传感器在山羊身体部署位置与试验结果的关系，得出了三轴加速度传感器最佳部署位置。（3）本文还研究了传感器记录间隔时间对分类精度及数据处理的影响，结果表明当记录间隔时间选定为2s时，可以在不影响分类精度的前提下，减少数据处理量，提高数据处理速度。（4）本文通过建立山羊典型日常行为加速度数据模型，可以进一步判定山羊的异常行为特征，改善山羊生产管理系统中的动物福利，为判定山羊典型日常行为与疾病关系以及山羊疾病预测模型的建立提供基础。
[Abstract]:In recent years, with the continuous development of information technology, the requirements of healthy breeding, safe production, quality supervision and quality traceability of goat herds in China are becoming more and more stringent. However, the monitoring of goat herd behavior still depends on the intuition and experience of the breeder, which not only consumes a lot of manpower, but also has low working efficiency, so it is no longer suitable for the large-scale and intensive modern livestock breeding industry. To quickly and accurately determine the relationship between animal behavior and disease diagnosis, researchers from all over the world have introduced behavioral feature recognition techniques such as machine vision, video or image processing, acoustic detection, etc. The fixed position and angle of the camera have high requirements. If we want to continuously record the behavior data of animals, especially those in bulk grazing, we must rely on the technology of collecting and processing information from sensors. Therefore, the core of this paper is to identify the typical daily behavior of goats by using three-axis acceleration sensor. The main contents are as follows: (1) this paper collects and analyzes the goat's triaxial acceleration data, and trains the goat's triaxial acceleration data set by K-means mean clustering algorithm and SVM algorithm. The effects of the two algorithms on the accuracy of the classification models are analyzed and compared, and the acceleration data models corresponding to the four typical daily behaviors of goats are identified and verified in combination with the time of action. The results show that the recognition accuracy of typical daily behaviors such as lying down, standing or walking, feeding and jumping is 87.76% according to the K-means mean clustering algorithm, but the recognition rate of typical behavior optimized by SVM algorithm is improved by nearly 6 percentage points. The results show that the algorithm can achieve higher classification accuracy and better optimization efficiency for the classification and recognition of goat typical daily behavior. It has very high application value in the field of behavior recognition. (2) the life habits of goats are analyzed in detail, and the relationship between the position of three-axis acceleration sensor in goat body and the experimental results is studied. The optimal deployment position of triaxial acceleration sensor is obtained. (3) the effect of recording interval time on classification accuracy and data processing is also studied. The results show that when the recording interval time is chosen as 2 s, Without affecting the classification accuracy, we can reduce the amount of data processing and improve the speed of data processing. (4) by establishing the acceleration data model of typical daily behavior of goats, we can further determine the abnormal behavior characteristics of goats. The improvement of animal welfare in goat production management system provides a basis for judging the relationship between goat typical daily behavior and disease and the establishment of goat disease prediction model.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S827;TP212

【参考文献】