Automatic classification of sheep behaviour using 3-axis accelerometer data
Automatic classification of sheep behaviour using 3-axis accelerometer data
复制标题
使用 3 轴加速度计数据对绵羊行为进行自动分类
DOI：
--
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
T. Niesler
中科院分区：
文献类型：
--
作者：
J. Marais;S. Roux;R. Wolhuter;T. Niesler
在难以接近的环境中工作时，监测动物行为可能具有挑战性。这个问题可以通过使用动物附着的加速度计和自动分类器来解决。本研究认为，使用专门设计的硬件捕获三维加速度计数据从羊，并随后自动分类的基础上，这些测量他们的行为的可行性。已经确定了五种常见的行为：躺，站，走，跑和吃草。线性判别分析（LDA）和二次判别分析（QDA）分类器的基础上训练的10个特征。一个贪婪的选择过程被用来确定哪些功能提供最高的分类精度。结果表明，这两个分类器可以自动识别的五个行为时，所有的功能都用于训练的高精度。LDA和QDA分类器的总体准确率分别为87.1%和89.7%。在两个分类器中，放牧被错误分类的最多，因为它与撒谎混淆了。考虑到与放牧和躺卧相关的原始加速度计数据之间的高度相似性，这一结果是预期的。当使用较少数量的特征时，QDA分类器显示出更大的改进。
Monitoring animal behaviour can prove challenging when working in inaccessible environments. This problem can be addressed by using animal attached accelerometers and automatic classifiers. This study considers the feasibility of using specially designed hardware to capture three-dimensional accelerometer data from sheep and to subsequently automatically classify their behaviour on the basis of these measurements. Five common behaviours have been identified: Lying, standing, walking, running and grazing. Linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) classifiers were trained based on 10 features. A greedy selection procedure was used to determine which features provide the highest classification accuracy. It is shown that both classifiers can automatically identify the five behaviours with high accuracy when all the features are used for training. The LDA and QDA classifiers achieved an overall accuracy of 87.1% and 89.7% respectively. Grazing was misclassified the most in both classifiers, because it was confused with lying. This result was expected considering the high similarity between the raw accelerometer data associated with grazing and lying. The QDA classifier showed larger improvements when using a smaller number of features.