基于跨视角机器视觉的羊体尺测量及形态评价研究

Inner Mongolia and the surrounding western area is the main producing areas of sheep industry, and the sheep varieties is essential for the normal development of the sheep industry. the research and analysis of growth characteristics of sheep, were often required a number of body size parameter to detection performance and to evaluate their genetic characteristics. But this work is mostly done manually, only one person is difficult to complete accurate measurement, and prevent infectious diseases. In the project, first, a remote networking environment monitoring system for livestock based on WSN (wireless sensor networks) was designed, and CCD cameras as the image acquisition device was used to acquire the top view image and side view image of the sheep. then the body sizes of Sunit sheep were non-contact acquired in different gender, different age, and different fed scenarios。For the surface body size, such as heart girth, multi-source data fusion was done. Following,correlation analysis was done based on body sizes and body weigh to select the characteristic parameters,and to estimate the growth characteristics of the sheep. Finally, based on statistical analysis and neural network approach to body size data do depth data mining, evaluation of physical characteristics was done used the method of multivariate statistical analyses and neural network network. Overall, this project can provide a theoretical basis for the improvement of sheep breeding, feeding, and the research results can also be extended to other industries such as cattle. The project has universal significance.

内蒙古及周边西部地区是我国养羊业的主产区，羊的优良品种并正常发育对养羊业至关重要。在研究和分析羊的生长发育特性、生产性能及其遗传特性时常常需要检测羊的多项体尺参数，并作出评价。但目前这项工作大都是人工完成的，不但一人难以完成、不准确，而且也不利于防疫。本项目通过基于无线传感网络的远程网络化视觉监测系统采集羊的体貌图像（俯视图、侧视图），对不同性别、不同月龄、不同饲喂情景下的苏尼特羊的体尺信息进行无接触测量，并对空间体尺参数做多源数据融合；根据体尺数据参照体重信息做关联分析、选择特征参量，从而评价羊的体尺生长特性；基于统计分析方法及神经网络方法对体尺数据做深度数据挖掘，从而给出基于体尺参数的羊体貌特征评价，为羊的选育改良、饲养等提供理论基础，其研究成果也可推广到养牛等其他行业，具有普遍意义。

内蒙古及周边西部地区是我国养羊业的主产区，羊的优良品种并正常发育对养羊业至关重要。研究表明羊的体尺参数可以反映羊的生长发育特性、生产性能及其遗传特性，在集约化养羊生产过程中，长期、连续监测羊只体尺，对指导实际生产具有重要意义。然而，传统的体尺测量方法不仅测量工作量大，对羊的站姿要求高，且需要直接接触羊体，羊的应激反应大，对羊产生严重的不良影响。因此，课题组提出基于结构化限位装置及机器视觉技术的无接触肉羊生长参数测量；应用超像素图像分割及模糊C均值聚类算法自动提取前景；以羊只体尺生物学测量位置为基础，提出基于曲线拟合、波峰检测、曲率检测、距离检测等测点检测算法；针对羊体俯视图提出基于图像骨架的柔性对称体对称中心提取算法；实现羊只体高、体长、体宽3类9种体尺参数的无接触测量；并讨论无接触方法获取的体尺数据与羊只生长特性的关系。试验结果表明基于视觉方法的体尺测量具有较高的稳定性、准确性，可有效提升工作效率；基于视觉的形态评价可以精准的体现羊只生长特性，是长期、实时监测羊只生长的实用方法，对推动精准、福利化养羊具有重要意义。