基于机器视觉的猪个体身份和饮水行为识别方法

发布时间：2018-05-13 20:08

本文选题：特征区域 + 相似度计算　；参考：《江苏大学》2017年硕士论文

【摘要】：随着养猪业规模化和智能化水平的不断提高,智能视频监控技术正在被广泛地应用和研究。传统养猪业要求饲养员实时获取猪只状态信息,以便及时发现猪只异常,这不仅费时费力,还会干扰猪只的正常生长。针对此问题,本文提出了基于机器视觉的猪个体身份和饮水行为的识别方法,用此方法能够提高猪场的生产效益,并减少饲养员的工作量。首先,针对猪只的非刚体特性,提取俯视监控视频中具有稳定性和独特性的特征区域,并依次提取颜色信息熵、形状参数、Tamura纹理等多种特征,组合构成多维特征向量用于表征猪只身份,结合向量相似度计算方法,得到待识别猪只和训练样本猪只之间的相似性,从而实现猪个体的身份识别。其次,针对猪只饮水时姿态相对固定的特性,采用改进的Douglas-Peukcer多边形近似法对饮水区域内的猪只轮廓进行拟合,并提取角度和距离特征,构建具有尺度不变性和旋转不变性的二维特征量用于表征猪只饮水状态,利用匈牙利算法得到轮廓片段之间的最优匹配,再计算该匹配下的匹配代价,完成轮廓的匹配工作,从而实现猪只饮水行为的识别。最后,针对身份识别算法,通过测试不同背部特征区域边长下的识别率和单只猪的平均识别时间,选择最优边长,同时针对饮水行为识别算法,通过测试不同相似度阈值下的识别率,选择最优阈值。再利用MATLAB GUI设计图像处理界面,完成参数设置、身份识别、饮水行为识别等功能,实现猪只身份识别和饮水行为识别。实验结果表明,测试帧中猪个体身份的识别率为86.7%,识别单只猪的平均时间为1.9154s,相比于其他典型方法,在保证时间性能的前提下取得了较高的识别率,同时猪只饮水行为的识别率为94.05%,较好地区分了饮水状态和非饮水状态,达到了研究的预期效果。本文采用机器视觉技术,实现了猪只的身份和饮水行为的智能监测和识别,为今后对群养猪采食、排便等行为的识别研究打下了基础,同时为探索牲畜的身份及饮水行为识别提供了新思路。
[Abstract]:With the continuous improvement of the scale and intelligence of pig industry, intelligent video surveillance technology is widely used and studied. The traditional pig industry requires the keepers to obtain the status information of pigs in real time in order to find the abnormal pigs in time, which not only takes time and effort, but also interferes with the normal growth of pigs. In order to solve this problem, a machine vision based identification method for pig individual identity and drinking water behavior is proposed, which can improve the production efficiency of pig farm and reduce the workload of breeders. First of all, aiming at the non-rigid body characteristics of pigs, the stable and unique feature areas in the overhead surveillance video are extracted, and the color information entropy, the shape parameter Tamura texture and other features are extracted in turn. The combination of multi-dimensional feature vectors is used to represent pig identity. Combining the vector similarity calculation method, the similarity between the pig to be identified and the training sample pig is obtained, so that the identity of pig individual can be realized. Secondly, the improved Douglas-Peukcer polygonal approximation method is used to fit the profile of pigs in drinking water, and the angle and distance characteristics are extracted. Two dimensional characteristic quantities with scale invariance and rotation invariance are constructed to represent the drinking state of pigs. The optimal matching between contour segments is obtained by using Hungarian algorithm. Then the matching cost is calculated and the contour matching is completed. Thus, the recognition of drinking water behavior of pigs is realized. Finally, according to the identification algorithm, by testing the recognition rate of different back feature region side length and the average recognition time of single pig, the optimal side length is selected, and the drinking water behavior recognition algorithm is also used. The optimal threshold is selected by testing the recognition rate under different similarity thresholds. Then the image processing interface is designed by using MATLAB GUI, and the functions of parameter setting, identity recognition, drinking behavior identification and so on are completed to realize pig identification and drinking behavior recognition. The experimental results show that the identification rate of pig in the test frame is 86.7, and the average time of identifying a single pig is 1.9154s. Compared with other typical methods, a high recognition rate is obtained on the premise of ensuring the performance of the time. At the same time, the recognition rate of drinking behavior of pigs is 94.05, which can distinguish the drinking state from non-drinking state, and reach the expected effect of the study. In this paper, the machine vision technology is used to realize the intelligent monitoring and recognition of pig identity and drinking water behavior, which lays a foundation for the future research on the identification of feeding and defecation behaviors of pigs. At the same time, it provides a new idea for the identification of livestock and the identification of drinking water behavior.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S828;TP391.41

【参考文献】