Study On Behavior Recognition And Health Assessment System Of Laying Hens

The behavior performance of laying hens is closely related to the welfare status.Analyze the change law and distribution characteristics of daily behavior of laying hens,establish health evaluation indicators,and judge the health status of chickens.In this paper,7-week-old white green shell laying hens are taken as the experimental object.Using the built laying hens behavior monitoring system,a month’s flock activity video is collected.The single and multiple laying hens’ behavior is recognized by using contour features and multiple target detection algorithms.The change law,activity location and distribution of laying hens’ behavior are studied,and the laying hens’ behavior recognition and monitoring system software is compiled.The main research contents and conclusions are as follows:（1）The behavior monitoring system of laying hens was constructed.Hikvision webcam was selected to set up two monitoring areas,with single and eight test laying hens placed respectively.A month’s video of chicken herd activities was collected.The camera recording time was 7:00-18:00.（2）A behavior recognition method based on layer contour feature is proposed.Firstly,the layer image is extracted from the surveillance video,and the layer image is grayed and smoothed.The layer contour is obtained and fitted by the automatic threshold segmentation method,and seven features of the fitted contour are extracted.Then the features are arranged and combined to obtain four feature combinations.Combined with the training of limit learning machine（ELM）,the best feature combination of central column,perimeter,roundness and the length width ratio of circumscribed rectangle is extracted,Finally,a layer behavior recognition model is established.The average recognition time of this method is 1.54 s and the recognition rate is high.The recognition rates of feeding,standing,lying and modification of a single layer are 97.00%,94.46%,91.50% and 86.64% respectively.（3）Aiming at the poor recognition of layer image adhesion and occlusion in complex environment,this paper studies the chicken herd behavior recognition method based on deep learning.Firstly,establish the data set,randomly select 5D video files from the obtained surveillance video,intercept 1000 laying hens images with different light intensity and adhesion degree,maintain the balance of samples through data enhancement operation,and then conduct K-means clustering on the data set to obtain the anchor size suitable for the data set,use a variety of target detection algorithms for training and recognition,and compare the recognition effect of the algorithm on the test set image,It is concluded that the recognition effect of YOLOv5 and Faster R-CNN is better than the other three models.Finally,the recognition effect of the two algorithms on adhesive images is compared,and it is found that the recognition rate of YOLOv5 is higher than that of fast r-cnn3 5 percentage points,the recognition rates of feeding,standing,lying and dressing behaviors can reach 98.6%,95.4%,93.78% and 89.20%,the average detection accuracy（map）can reach 94.23%,and the detection frame rate of video can reach 25f/s,indicating that YOLOv5 can accurately identify the behavior of laying hens in the case of dispersion and adhesion,and achieve the effect of real-time detection.（4）This paper studies the change law,activity location and distribution of chicken herd behavior.This paper analyzes the video clips of chicken herd activity before and after feeding,obtains the behavior information and central coordinates of each individual in the video image through yolov5 algorithm,counts the change and location distribution of chicken herd behavior number with time,and obtains the proportion and distribution characteristics of chicken herd behavior number in different time periods,The behavior of healthy chickens has the following characteristics: before feeding,laying hens mainly carry out feeding and standing behavior,with less modification and lying behavior.Within 10 minutes after feeding,laying hens mainly carry out feeding behavior,and the other three behaviors become less.After feeding is complete,the other three behaviors begin to appear.The total number of four types of behaviors in a day is counted,and the behavior threshold of healthy chickens is set;The distribution of feeding and lying behavior of chickens is relatively concentrated,and the distribution of standing and modification behavior is relatively scattered.The results show that healthy chickens have more feeding behavior.After feeding,they will carry out activities and modification behavior in the fence.When the chickens rest,they are mostly distributed at the edge of the fence.The distribution index of chickens in different states is obtained.The health of chickens is evaluated based on the number and distribution index of laying hens.（5）The software of layer behavior recognition and monitoring system based on pyqt5 is compiled.The software can connect the camera to monitor the behavior of laying hens in real time,display the number and location of detected behaviors,and support the functions of video detection and result saving.