Detection Of Pig Dietary Behavior Based On Deep Learning

Aiming at the current low level of automatic detection of pigs’ dietary behavior in the pig house environment,For specific monitoring scenarios(such as the pig pen in the experiment),the eating and drinking behaviors of the pigs in the pen are defined according to the experimental requirements,and the eating behavior of the pigs is classified;according to the definition and according to the real breeding environment and complex conditions A specific data acquisition model was established for the following eating behaviors.Through format conversion,frame interception,data filtering,data enhancement,data calibration and other work on the original video data,a data set of 2500 training data images and 500 test image data was established.By inputting pig eating and drinking behavior image data into the deep neural network for automatic feature extraction and classification,using the deep feature extraction and highprecision detection and classification characteristics of YOLOv4 neural network,the defined eating behavior of pigs can be accurately detected,For different scenarios,targeted training can be made to improve its applicability.and provide targeted and adaptable technical support for intelligent pig breeding and management.The experimental results show that the pig eating behavior detection model based on the YOLOv4 network structure has high accuracy.The average detection accuracy of the established test data set is 95.5%,which is 2.8% higher than the YOLOv3 model and 3.6% higher than the Tiny-YOLOv4 model.1.5% higher than the Faster R-CNN model,5.9% higher than the Retina Net model,and 5% higher than the SSD model;in the stuck state,the detection accuracy of the YOLO-DBP model based on YOLOv4 is much higher than other models,The detection accuracy of pig eating behavior reached 85.8%,which was 52.5 points and 42.2 points higher than the same series of YOLOv3 and TOLOv4-tiny models,and 24.3,44.8,and 31.1 percentage points higher than Faster R-CNN,Retina Net,and SSD models.The method in this paper can accurately predict the eating behavior of pigs and provide technical support for intelligent pig breeding and management.