Detection Of Dairy Cow’s Respiratory Behavior Based On Video Analysis

Respiratory behavior is one of the most basic physiological activities of dairy cows,which contains abundant health information.Accurate and automatic detection of respiratory behavior of dairy cows is the technical prerequisite for intelligent diagnosis of respiratory-related diseases such as heat stress based on the “cross of medicine and engineering” technology,which is not only conducive to timely treatment of dairy cows’ health problems,but also conducive to optimizing feeding schemes,improving milk yield and farm economic benefits.To improve the deficiency of high cost and low accuracy in manual detection of dairy cows’ respiratory behavior in modern large-scale farms,the dairy cows in Yangling Keyuan Clone Co.,Ltd.were took as the research object in this study,and the research on the detection method of dairy cows’ respiratory behavior based on video analysis technology was carried out,so as to provide some technical supports for intelligent detection of dairy cows’ respiratory behavior.The research contents involved and the main conclusions reached are as follows:(1)Aiming at solving the problems of abdominal occlusion,complex illumination,large calculation amount of the traditional algorithm and slow execution speed in the detection of respiratory behavior of lateral decubitus dairy cows in natural scenes,a respiratory behavior detection method based on LK sparse optical flow algorithm was proposed to detect the motion law of the edge of the abdominal speckles.Moreover,the experimental result were compared with those of the global LK optical flow method,the global Horn-Schunck optical flow method,and the Horn-Schunck optical flow method based on the speckle boundary.The result showed that that average accuracy of the respiratory behavior detection of the proposed algorithm was 98.58%,the frame processing time was consume in 0.10-0.13 s,the detection accuracy was improved by1.26%-2.36% compared with the comparison algorithm,the execution time of the algorithm was shortened by about 34.05%-39.31%,and the comprehensive performance was better than the three comparison algorithms.The method could lay a technical foundation for the diagnosis of respiratory diseases of the lateral decubitus dairy cows,and provided a certain theoretical support for the follow-up research on the standing and multi-target dairy cows’ respiratory behavior detection method.(2)According to the characteristics of weak respiratory movement,more occlusion interference and difficulty in observation and detection,a method of respiratory movement detection for standing cows based on Deeplab V3+ and PBVM motion amplification algorithm was proposed.Moreover,it was compared with original video LK optical flow method,object segmentation video LK optical flow method and original video motion amplification LK optical flow method.The test results showed that the algorithm combined the object segmentation based on Deeplab V3+ with PBVM motion amplification processing,which largely avoided irrelevant background interference and enhanced the range of dairy cows’ breathing motion,and the respiratory behavior detection accuracy of standing dairy cows’ was 93.04%,which was 45.82%,17.72% and22.94% higher than the three comparison algorithms,respectively.It showed that the detection of standing dairy cows’ respiratory behavior based on this algorithm was effective and feasible.(3)Considering the problems of complex and changeable motion interference,many irrelevant interference and difficult detection of different dairy cows’ respiratory behavior in natural scenes,a multi-target dairy cows’ respiratory behavior detection method based on deep learning and video analysis was proposed.The results indicated that the algorithm could accurately realize multi-target dairy cow recognition and segmentation,and recognize dairy cow behaviors with higher accuracy.The average accuracy of multi-target detection of dairy cows’ respiratory behavior could reach 93.56%,which indicated that the algorithm could realize the multi-target dairy cows’ respiratory behavior detection.(4)Based on the development and design foundation of lying and standing of single-target cow and the detection algorithm of multi-target cows’ respiratory behavior,the demonstration software of cows’ respiratory behavior detection in unstructured environment was designed by employing MATLAB Graphical User Interface toolbox.The software has completed the integrated demonstration of the main processes of the three respiratory behavior detection algorithms proposed in this study.The test results showed that the software could realize the functions of detecting the breathing behavior of cows in different scenes and saving the results based on the above three algorithms.The software was characterized by complete and scientific interface,quick and simple operation,intuitive and concrete image display and good operation fluency.