基于时空轨迹数据和声信号的放牧羊群牧食行为研究

Grassland grazing management is the main way of sheep flocks grazing in Inner Mongolia. In recent years, due to the herdsmen's blind expansion of the livestock population, the conflict between grass and livestock has been aggravated. The grass and animal husbandry management department also lack of innovative management methods of real-time monitoring herdsman grazing behavior, so it has happened things like the excessive grazing on the same pasture, failing to transfer pastures in time, even finding herds in prohibited grazing areas, etc. This has impact on the animal husbandry management departments to make scientific decision on projects of returning grazing to grassland, rotation grazing and grassland ecological compensation. Therefore, the project is going to develop the intelligent collar for detecting sheep behavior, and build sheep flocks behavior monitoring software and hardware platform. Besides, the Beidou satellite positioning technology is used to obtain the data of the spatial and temporal trajectory of grazing herds, and the feature extraction and classification of sheep behavior such as feeding, ruminant and tourism will be studied based on acoustic signal. So, the dynamic monitoring of the spatial-temporal distribution of herd grazing intensity and feed intake in a wide range is realized quickly. In addition, in order to build the prediction model of sheep feed intake, and estimate the stocking capacity under livestock balance, the massive spatial-temporal trajectories data mining will be carried out. The implementation of this project can provide decision-making basis for pasture-grassland construction of grass-livestock management department and realize the benign development of grassland grazing monitoring, utilization and construction, which has important significance and application value for promoting animal husbandry economy and sustainable development of grassland ecosystem.

内蒙古草原羊群主要采用草地放牧管理，近年来，由于牧民盲目扩增牲畜数量，草畜矛盾加重，而草畜管理部门缺乏创新的管理方法，无法实时监控和掌握牧民的放牧行为，存在同一块草场过度放牧，未能及时转场，以及羊群出现在禁牧区等情况，影响草畜管理部门对退牧还草、划区轮牧、草原生态补偿等项目的科学决策。本项目研制羊只牧食行为检测的智能项圈，搭建羊群牧食行为监测软硬件平台，采用北斗卫星定位技术获取放牧羊群的时空轨迹数据，并对羊只采食、反刍和游息等牧食行为声信号进行特征提取和分类识别研究，实现快速、大范围的羊群放牧强度和采食量时空分布动态监测，并对海量时空轨迹数据进行挖掘，建立羊群采食量的预测模型，估测草畜平衡下的合适载畜率。该项目的实施可为草畜管理部门的牧场草地建设提供决策依据，实现草地放牧监测、利用和建设的良性发展，对促进畜牧业经济与草地生态可持续发展有重要的意义和应用价值。

内蒙古草原羊群主要采用草地放牧管理，近年来，由于牧民盲目扩增牲畜数量，草畜矛盾加重，而草畜管理部门缺乏创新的管理方法，无法实时监控和掌握牧民的放牧行为，存在同一块草场过度放牧，未能及时转场，以及羊群出现在禁牧区等情况，影响草畜管理部门对退牧还草、划区轮牧、草原生态补偿等项目的科学决策。本项目研制羊只牧食行为检测的智能项圈，搭建羊群牧食行为监测软硬件平台，采用北斗卫星定位技术获取放牧羊群的时空轨迹数据，并对羊只采食、反刍和游息等牧食行为声信号进行特征提取和分类识别研究，实现快速、大范围的羊群放牧强度和采食量时空分布监测，并对时空轨迹数据进行挖掘，建立羊群采食量的估测模型。. 项目在乌兰察布市四子王旗试验草场基地和鄂尔多斯市鄂托克旗试验草场基地进行，（1）利用声信号特征提取技术和参数优化方法，构建了牧食行为声信号事件自动识别算法，能自动切分牧食声信号片段；（2）将识别的牧食行为事件分类为咬食、采食咀嚼、反刍咀嚼，对比深度神经网络、卷积神经网络、循环神经网络模型，分析最佳的分类模型；（3）根据牧食声信号，分析草地的牧草生长状况；（4）在咀嚼声信号中提取多个解释变量，准确地估测羊只的采食量；（5）建立BP神经网络模型和线性回归模型，将轨迹段的持续时间、羊只行走距离、羊只的体重、试验区的坡度、坡向、高程、温度、植被指数及其天气状况作为特征向量，分析羊的采食量时空分布;（6）提出一种基于遗传算法优化的长短时记忆神经网络的牧群采食量估测模型，该模型具有较好的估测性能和较强的泛化能力；（7）以放牧羊群的时空轨迹数据为基础，结合时空匹配、叠加处理等分析方法构建放牧羊群采食量时空分布模型。..该项目的实施可为草畜管理部门的牧场草地建设提供决策依据，实现草地放牧监测、利用和建设的良性发展，对促进畜牧业经济与草地生态可持续发展有重要的意义和应用价值。