Research On Pig Body Measurement And Behavior Recognition Based On Machine Vision

With the continuous development of the breeding industry,automatic pig farming has become the development trend of the pig industry.Aiming at the low efficiency of manually measuring the body size of pigs and observing the behavior of pigs manually,automatic monitoring based on machine vision can replace manual work to improve detection efficiency,which has become one of the current research hotspots.In order to improve the image utilization and body measurement efficiency of pig body measurement,and to monitor the behavior of pigs in real time and detect abnormalities in time,this study proposes a research method of pig body measurement and behavior recognition based on machine vision.The main contents and conclusions are as follows:(1)Algorithms design of ideal posture frame.In this algorithm,the minimum external rectangles were computed to adjust the level of the pig’s body.Head and tail positions were identified by projection and difference methods.Boundary signature was used to determine whether part of the ears was missing.Image skeleton algorithm and Hough transform algorithm were applied to judge whether the pig body was skewed.(2)Body measurement.Based on finding the ideal posture frame,a pig body measurement algorithm was designed.The algorithm needs to accurately find the ear,tail and scapula positions of the pigs.After finding the approximate positions of ear and tail roots by using the difference method,the representation of ear and tail root positions in the difference curve is not accurate.It is necessary to further confirm their positions and find the measurement point position Then,the pig’s body width,height,and body length can be determined.This paper took Landrace and Large White as the researches object.On this basis,algorithms for measuring pig body size were designed.The top view and side view of video had 52 016 frames,respectively.These frames of 103 sets of video data were tested by the posture detection algorithm and body size measuring algorithm.2 592 frames of ideal posture frames were screened out.It produced high false negatives(432 frames)and very low false positives(0 frames).The results showed that the absolute deviation of the body length is small.The body length deviation of each frame was less than 2.3%,and the consistency of the measurement results was high.The average accuracy of body width was 95.5%,the average accuracy of body height was 96.3%,and the average accuracy of body length was 97.3%.(3)Behavior recognition.In order to automatically monitor the behavior of pigs,an algorithm can be used to automatically identify the behaviors of the pigs,find out their behavior laws,and find abnormalities in time.This paper proposes a method to automatically identify sow behaviors(feeding,drinking,standing,prone lying,side lying and sitting).The first step of this method is to identify sows’ eating and drinking behaviors according to the feature analysis,and then classify the unrecognized images using support vector machine.The infrared thermal camera was used to monitor the sows before farrowing,and the collected images are recognized by the algorithm in this paper,and the results of algorithm recognition are compared with those of manual recognition.The classification recognition accuracy of the algorithm was 96.3%for feeding,92.7%for drinking,92.8%for standing,82.4%for prone lying,97.0%for side lying,and 91.1%for sitting.