摘要
本文介绍了我们在动物行为的智能分析研究中,针对青海湖野鸟监控获得的视频数据,基于数字图像处理及机器学习的方法对斑头雁形态样本进行分类的方法。我们首先采用尺度不变特征变换提取样本的特征点,选择不同的聚类中心对训练样本的特征点进行聚类,为每个样本生成特征向量来对样本进行描述,然后采用K最近邻算法建立模型,对斑头雁形态进行分类。对提出的方法进行了实验验证。通过恰当的特征选取,测试数据的分类准确率达到了73.75%。实验表明,本文提出的方法可以有效地从视频数据中实现斑头雁形态的自动分类。
For Animal behavior analysis, we present a classification method to study the behavior of bar-headed geese based on digital image processing and machine learning from the video data obtained in Qinghai Lake. Firstly, using SIFT（Scale Invariant Feature Transform）,by extracting feature points in the samples and choosing different cluster centers for the feature points, we can describe each sample with one feature vector. Then, the Bar-headed Goose morphology is classiifed by the model based on K-nearest neighbor algorithm, and we can predict and analyze the behavior of the Bar-headed Goose from the classiifcation＆amp;nbsp;results with its biology background and relevant information from the video. Through proper selection of the features, we achieved an accuracy of 73.75%for the identiifcation of the test sets. Our experiment shows that the method is effective for automatic morphology classification and identification from the video data.
出处
《科研信息化技术与应用》
2014年第3期87-94,共8页
E-science Technology & Application
关键词
视频分析
形态分类
SIFT
KNN
video content analysis
morphology classiifcation
SIFT
KNN
