| 摘要: |
| 针对奶牛行为判别自动化水平不足、准确率低的问题，采用惯性测量单元(IMU)和卷积神经网络(CNN)，对细粒度奶牛行为判别进行研究。结果表明:1)在KNN、SVM、BPNN、CNN和LSTM 5个模型中，CNN模型在奶牛行为分类测试集上的准确率最高。2)含有三轴加速度计、陀螺仪和磁力计的IMU更加适用于奶牛行为分类，其分类效果优于含一种传感器的IMU。3)传感器频率与分类模型的性能相关，频率越高，正确率越高，当传感器频率设置为25 Hz时，奶牛行为判别效果最好。4)在1、2和4 s这3种时间窗中，使用4 s时间窗的奶牛行为分类模型性能最好。5)采用最优配置时，卷积神经网络模型能够有效的判别奶牛站立、躺卧2种状态，正确率为99%；可以对奶牛卷食、咀嚼、站立反刍、躺卧反刍、躺卧休息、站立休息6类行为进行判别，正确率为85%。采用IMU和卷积神经网络算法，可以有效的对细粒度奶牛行为进行判别，为奶牛养殖的自动化、智能化管理提供支撑。 |
| 关键词: 奶牛 行为判别 卷积神经网络 IMU |
| DOI：10.11841/j.issn.1007-4333.2022.04.15 |
| 投稿时间：2021-08-05 |
| 基金项目:国家自然科学基金项目(32102600)；国家重点研发计划项目(2017YFD0502006)；中国农业科学院科技创新工程(CAAS-ASTIP-2016-AII)；中央级公益性科研院所基本科研业务费专项(JBYW-AII-2020-42,JBYW-AII-2021-33) |
|
| Fine-grained cows' behavior classification method based on IMU |
|
CHENG Guodong,WU Jianzhai,XING Liwei,ZHU Mengshuai,ZHANG Jianhua,HAN Shuqing
|
| (Agricultural Information Institute/Key Laboratory of Agricultural Big Data of Ministry of Agriculture and Rural Affairs, Chinese Academy of Agricultural Sciences, Beijing 100081, China) |
| Abstract: |
| In order to realize the automatic recognition of dairy cows' behaviors, an fine-grained cow behavior classification method based on inertial measurement unit(IMU)and convolutional neural network(CNN)was proposed. The results showed that: 1)Among the five classification models of KNN, SVM, BPNN, CNN and LSTM, the CNN model has the highest accuracy on the cow behavior classification test set. 2)The IMU with triaxial accelerometer, gyroscope and magnetometer is more suitable for cow behavior classification, and its classification effect of cow behavior is better than IMU with a single type of sensor. 3)Sampling frequency was related to the performance of classification models. The higher the frequency, the higher the accuracy. When the sampling frequency was set to 25 Hz, the performance of classification model was the best. 4)Among the three time windows of 1, 2 and 4 s, the performance of behavior classification model of cows with the time window of 4 s is the best. 5)When the optimal configuration was adopted, the CNN model can effectively distinguish the standing and lying states of dairy cows with a correct rate of 99. 08%; The six behaviors of cows' feeding, chewing, standing ruminating, lying ruminating, lying resting, standing resting can be identified, and the accuracy is 85. 19%. In conclusion, the proposed method can be used to distinguish the fine-grained behaviors of dairy cows effectively and support the automatic and intelligent management of dairy farming. |
| Key words: cow behavior identification convolutional neural networks IMU |
