Deep Learning Based Keypoints Detection Of Cows

As a very important research direction in the field of computer vision,pose estimation has become a hot research direction in recent years,and it also plays a major role in the agricultural field.Taking cow breeding as an example,the purpose of positioning the key points of the cow is to use the computer to automatically locate the coordinate position of the specific joint point of the cow in the image or video,to realize the description and understanding of the cow’s posture,and to help the breeder better understand.The behavior of cows can also detect multiple cows more accurately and efficiently,greatly reducing the input of manpower and material resources.Despite this,the estimation of the attitude of the animal is still a challenge,no matter the animal’s body shape and posture characteristics are very different from peopleThis article will use cows as research objects to construct a deep learning model and carry out research on the joint location of dairy cows.The details are as follows(1)Through the photographic equipment such as mobile phones and video cameras,photographing and collecting data sets for the joint positioning of cows in different environments and different climates,and making a special labeler to mark the key points of the data set(2)Based on the analysis and comparison of the existing deep learning model of human body pose estimation,this experiment selected PyraNet,Stacked Hourglass and ResNet18 in the form of full convolutional network as the contrast experiment.The Stacked Hourglass network is the human body pose estimation.A high-performance network model,PyraNet is based on the network structure of Stacked Hourglass and replaces the network of new modules on the basis of it.ResNet is also a very classic neural network,and most of the neural networks appearing afterwards are used for reference.The design idea of ResNet Therefore,this experiment will select these three networks for comparison,so as to select the key point detection model that is most suitable for this data set(3)Based on the comparison of the effects of these three models,this project constructs a system for positioning the joint points of dairy cows.The system is divided into two parts,one part can select the deep learning model to train the data set of this experiment,and can process the related annotation file to simplify the training operation;the other part can be loaded by loading a trained model.Zhang reads the picture of the incoming cow to make joint point positioning,and visually reflects the positioning result in the form of drawing points on the picture.