摘要
为了研究分类与猪只行为及精神状况相关的姿态,试验提出了一种基于几何参数特征与决策树支持向量机的猪只姿态分类方法,首先根据猪只行为学特征并借鉴前人的研究成果选择5种相关的姿态,包括躺卧、侧面抬头站立、侧面低头站立、侧面平视站立以及正面站立,并以此建立猪只姿态样本库,再针对所研究姿态的图像形状特点,提出利用圆形度、高宽比、伸长度、质高比、质左比、质心左右比、最高点左右比、左(右)夹角余弦与左(右)抬头度11个几何参数特征作为描述姿态的图像特征,最后利用决策树支持向量机(DT-SVM)进行猪只姿态分类。结果表明:与1-V-1多分类支持向量机(1-V-1 SVM)与Ada Boost分类算法相比较,DT-SVM的分类精度较高,耗时较短。说明所选取的特征可以有效地描述猪只姿态。
To study the classification of pig behavior and the postures related to pig mental state, a pig posture classification algorithm was put forward based on geometric parameter feature and decision - tree - based support vector machine （ DT - SVM）. Firstly, five postures related to pig mental state were selected according to the pig behavior characteristics and the research results of previous studies, including lying, side view of standing with raised head, side view of standing with lowered head, side view of standing horizontally＇, and front view of standing, and then a sample library of the postures was established. The use of eleven geometric parameter features as the image features in the description of postures was proposed according to the characteristics of the image shape, including circularity, aspect ratio, elongation, ratio of centroid height, ratio of centroid left, left- right ratio of centroid, left- right ratio of peak point, left （right） angle cosine and left （right） raised head degrees, and so on. Finally, DT - SVM was used to classify the pig postures. The results showed that DT - SVM had higher accuracy and shor- ter time - consuming in the classification compare with the 1 - V - 1 SVMs and AdaBoost classification algorithm. The result indicates that the selected features can effectively describe pig postures.
出处
《黑龙江畜牧兽医》
CAS
北大核心
2015年第10期21-25,共5页
Heilongjiang Animal Science And veterinary Medicine
基金
"863"国家高技术研究发展计划项目(2013AA102306)
山西省青年科技研究基金项目(2012021030-1)
关键词
猪只姿态分类
多维几何参数特征提取
决策树支持向量机(DT—SVM)
机器视觉技术
智能
监控
pig posture classification
extraction of multi - dimensional geometric parameter features
decision tree - support vector machine（DT- SVM）
machine vision technique
intelligent surveillance
