| 基于传感器技术的自由放牧羊行为识别研究进展 |
| |
| 引用本文： | 石红霄,高方馀,刘同海,王海,哈斯巴根,杨婷婷,袁闯闯.基于传感器技术的自由放牧羊行为识别研究进展[J].农业工程学报,2023,39(17):1-18. |
| |
| 作者姓名： | 石红霄 高方馀 刘同海 王海 哈斯巴根 杨婷婷 袁闯闯 |
| |
| 作者单位： | 中国农业科学院草原研究所, 呼和浩特 010010;中国农业科学院草原研究所, 呼和浩特 010010;天津农学院计算机与信息工程学院, 天津 300384 |
| |
| 基金项目： | 内蒙古自治区科学技术厅项目（2020GG0068）；天津市研究生科研创新项目（2022SKYZ257）；内蒙基金面上项目（2022MS03007） |
| |
| 摘 要： | 检测羊的行为能够充分反映其内在生理状态和健康水平，并有助于牧户掌握放牧草地状况，保证草地生态系统可持续发展。受空间尺度影响自由放牧条件羊的行为监测十分困难，传统人工观察效率低、主观性高。与室内集约化养殖模式相比，计算机视觉的方法虽然精度高，但野外环境恶劣且面积广阔不便于设备设施布设和数据采集。因此，越来越多的学者通过接触式传感器识别羊的行为、分析羊群与草地环境的关系，为畜情草况的监测提供了新的技术手段。该文详细介绍了现阶段常用于羊的行为监测的3种传感器技术，即三轴加速度传感器、声学传感器、定位项圈，通过分析数据获取、处理、特征提取、模型结果等过程，归纳与总结了关键技术中存在的问题和面临的挑战，如三轴加速度传感器的部署位置、时间间隔；声学传感器的噪声处理等问题，又结合羊的行为模式说明了关键技术在不同行为中的应用现状。最后指出接触式传感器的发展方向，即算法层次结合深度学习方法分析数据，挖掘深层次、更具辨别能力的特征信息；传感器方面选取多传感器组合的方式，构建羊只多源数据集，提供更全面的特征，以期为自由放牧下羊行为识别研究提供新思路。
|
| 关 键 词： | 传感器 行为识别 羊 放牧 生态环境 |
| 收稿时间： | 2023/5/8 0:00:00 |
| 修稿时间： | 2023/7/13 0:00:00 |
Research progress of the recognition of free-range sheep behavior using sensor technology |
| |
| SHI Hongxiao,GAO Fangyu,LIU Tonghai,WANG Hai,HASI Bagen,YANG Tingting,YUAN Chuangchuang.Research progress of the recognition of free-range sheep behavior using sensor technology[J].Transactions of the Chinese Society of Agricultural Engineering,2023,39(17):1-18. |
| |
| Authors: | SHI Hongxiao GAO Fangyu LIU Tonghai WANG Hai HASI Bagen YANG Tingting YUAN Chuangchuang |
| |
| Affiliation: | Grassland Research Institute, Chinese Academy of Agricultural Sciences, Hohhot 010010, China;Grassland Research Institute, Chinese Academy of Agricultural Sciences, Hohhot 010010, China;College of Computer and Information Engineering, Tianjin Agricultural University, Tianjin 300384, China |
| |
| Abstract: | An accurate and rapid recognition of the behavior of sheep can fully reflect the internal physiological state and health level for the better welfare of livestock breeding. At the same time, the health status of grazing grassland can also be analyzed to combine with the location characteristics. Then, the grass condition can be grasped in time to ensure the balance and sustainable development of grass and livestock in the grassland ecosystem. However, it is very difficult to monitor the behavior of free-grazing sheep under the influence of spatial scale. The computer vision with high accuracy is also inconvenient to arranging the equipment and facilities, and then collecting data under the harsh field environment and vast area, compared with the indoor intensive farming mode. The traditional manual observation cannot fully meet the requirement of large-scale production at this stage, due to the low efficiency and high subjectivity. Contact sensors can be expected to identify the behavior of sheep for the relationship between sheep and grassland environment, particularly with the rapid development of key technologies, such as communication and big data. Behavior monitoring included sustained behavior (grazing, walking, ruminating, running, and resting), and transient behavior (parturition, estrus, and urination) recognition. A systematic analysis has been implemented to determine the feed intake distribution, feeding intensity distribution, grazing intensity, and the spatiotemporal change of sheep in the field of grassland. A theoretical basis and technical means can be provided for the formulation of a grazing system in the monitoring of livestock and grass conditions for accurate animal husbandry. In this review, three commonly-used sensor technologies were introduced in sheep behavior monitoring at present. Data acquisition, processing, feature extraction, and modelling were also analyzed to summarize the existing methods and challenges in the key technologies. The research object was mainly for the large ruminants, where the research on sheep behavior monitoring was still in its infancy. There were many similarities between sheep and cattle, as the main grazing livestock in grassland. There were still some differences to need further exploration. Tri-Axis acceleration sensors were selected to compare the different deployment positions and time intervals. Appropriate positions and time intervals were then optimized, according to the target behavior combined with the grassland condition. However it is still lacking to consider the sampling frequency. The satellite signal was generally blocked by some obstacles in the condition of free grazing, which was easy to cause data loss. Current imaging data was collected from the artificial pastures with the small area, in order to simulate the living environment of sheep. But there were still differences from the real complex pastures. It was difficult to replace the equipment in the vast pasture area as the survival basis for sheep grazing. Frequent equipment replacement should be given special attention to reducing human participation in intelligent grazing management. In addition, the application status of key technologies in the different behaviors was illustrated in combination with the behavior pattern of sheep. Finally, the development direction of the contact sensor was proposed in the future. Firstly, the multi-sensor fusion system can be combined as complementary, according to the characteristics of the sensor, in order to more accurately infer the behavior of sheep, and then assess the health status of grazing pasture. Secondly, a deep learning network can be utilized to analyze the image data, in order to dig deeper and then distinguish the feature information. The manual extraction of features can be reduced to overcome the data imbalance and insufficient data for the recognition of complex patterns. A lightweight end-to-end network model was established for deployment in the embedded systems or mobile devices, fully meeting the practical applications. |
| |
| Keywords: | sensor behavior recognition sheep grazing ecological environment |
|
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
|
点击此处可从《农业工程学报》下载全文 |
|
