Research On Cow Body Condition Score Evaluation Based On Deep Learning

Cow body condition evaluation is an important tool to assess the health of dairy cows,it could assess energy reserves of dairy cows and predict their diseases and make the management more conveniently.Cow body condition is the evaluation of obesity of cows,low score means thin and weak,high score means fat.Suffering long-term obesity or long-term emaciation can cause a variety of physiological or psychological diseases,which will make the production of milk become less and reduce the quality of milk,has a negative impact on economic benefits of animal husbandry farm.Artificial scoring is the most common method of cow body condition evaluation up to now,which wastes time and energy,suffers low accuracy and low efficiency.Checking the cow body condition score regularly is very important,so the development of automatic cow body condition evaluation system that could eliminate subjective error and liberate human resources is very important.This dissertation evaluates cow body condition using deep learning,which is a data-drive technology and needs lots of data.This dissertation captures a larger number of videos through webcam,and selects lots of images as dataset in detail.This dissertation gets cow body condition score using image classification and object detection technology,the research content and innovation are as follows:(1)This dissertation gets cow body condition score using image classification,which sees different scores as different classes and classifies images using convolutional neural networks.This dissertation trains cow body condition evaluation dataset with nine great classification networks and improves their performance through transfer learning and data enhancement.By balancing accuracy and efficiency,this dissertation gets the most suitable classification network to evaluate cow body condition.(2)This dissertation gets cow body condition score using object detection,which extracts the key part of cow firstly and then get the body condition score by classifying the key part.This dissertation evaluates cow body condition using seven object detection networks,selects the most suitable object detection network by comparing their accuracy and speed.(3)This dissertation analyses the merits and demerits of classification network and object detection network and gets the most suitable network for cow body condition evaluation,and develops the automatic cow body condition evaluation system by it.