Dairy Cattle's Information Perception And Behavior Analysis Based On Machine Vision

Precision dairy farming is an important part of intelligent agriculture.With the development of modern agriculture,precision dairy farming has gained great attention by the researchers around the world,and become one of the most important research topics of modern agriculture.To solve the problems of obtaining accurate cow information in precision dairy farming,hardware and software systems were developed based on machine vision for automatic information perception and behavior analysis of individual cow,which provided the basis for precision dairy farming.Based on domestic and international research,and aiming at improving the automation of individual cow information acquisition,automatic systems and methods were studied and developed for cow moving target detection,non-contact identification,lameness detection and locomotion score,body parts segmentation,respiratory monitoring,body condition score.The main work and innovation of this paper are as follows:(1)The processing of videos captured in farming scenes often suffers from a complex background.A new target detection method was proposed to detect the moving target accurately for cows based on background subtraction.The bounding rectangle of cows was calculated using the frames difference method to extract the local background in frames,which were averaged and spliced into one image as the entire background image.The body area was located and tracked through the video.The summation coefficients on RGB channels were adjusted to improve the contrast between the target and background images.Taking the body area in every frame as reference area,the performance of target detection was evaluated by the reference area to determine the optimal summation coefficients on RGB channels,and then background subtraction was processed again to finish the detection.A total of 129 videos was used to test the detection algorithm,and the accuracy of the algorithm was 88.34%,which was 24.85% higher than the classical background subtraction method.The study shows that the algorithm in this paper is feasible to detect the target accurately and timely when cows are walking straight in the farming environment under natural light,and this method can improve the detection performance and is an extension to the classical background subtraction method.(2)An accurate and efficient system that can recognize individual information from cows in feeding environment using a non-contact sensor is studied.Individual dairy cattle was discriminated using the body images based on convolutional neural networks with video analysis method.After graying,resizing and normalizing,the body image of cow was transferred into a matrix sized 48×48 as the input of the network with a structure of 4c-2s-6c-2s-30 o.30 cows were captured 12 times for each totaling in 360 sets of videos were obtained,60000 training frames,21730 testing frames and 90 testing videos were selected from the 360 videos randomly.90.55% of the testing frames and 93.33% of the testing videos were recognized correctly,respectively.The results suggested that the methods in this research were feasible to discriminate the individual dairy cattle.This study proved that image processing technique had great potential to be used for discrimination of animals.(3)An automatic system was developed to analyze leg swing using computer vision techniques and to develop an automatic and continuous system for scoring locomotion of cows to detect and predict lameness with high accuracy and practicability.The motion curve was plotted by extracting the position of the moving leg by image processing,and the motion curve was analyzed to generate six features referring to the gait asymmetry,speed,tracking up,stance time,stride length,and tenderness.The data set included 621 videos from 98 cows.A box-plot of the features within 3 classes showed that the dataset was nearly linear and separable under the six features and that the cows had different lameness indicators in different lameness stages.The Decision Tree classifier was applied to the dataset,and 2-,3-,and 10-fold cross validation was used to verify the performance of the algorithm.The accuracy of the classification was 90.18%,and the averages of sensitivity and specificity were 90.25 and 94.74%,respectively.This research demonstrates the feasibility of classifying dairy cow lameness based on the six motion features extracted by leg swing analysis.(4)A real-time video capturing system for dairy was designed,and methods for automatic monitoring respiration rate and abnormal behaviors of cows were studied based on the system.The optical flow method was used to calculate the velocity of each pixel.Breathing points was located by looping Otsu operation according to the magnitude of velocity.By analyzing the curve of velocity orientation,the respiration rate was calculated,and the abnormal behaviors were detected by the duration of each breath.Videos lasting 360 minutes with 72 breathing cows was used to evaluate the method.The results demonstrated that the accuracy of calculated respiration rate and abnormal behaviors detection are 95.68% and 89.06%,respectively,and the error detection of abnormal behaviors is 2.53 times per-minute.This method can be used to detect respiration rate and abnormity of dairy cattle.(5)A system was developed for segmenting a cow's body parts,including the head,neck,body,forelimbs,hind limbs,and tail,with high accuracy on the basis of depth image processing and machine learning.The local binary patterns of each pixel under several sampling radii were used as the features with which the filtering rules were designed,and a decision forest was trained and tested to classify the pixels into six groups.288 depth images from 30 cows were used for training and testing the model.The results showed that when the number of sampling radii and training layers were 30 and 20,respectively,the recognition rate reached 95.15%.Compared with the typical depth image features,the new feature employed in this study can extract the details of a cow's body and recognizing complex parts more accurately with fewer parameters and a simple model.(6)An system was developed to automatically determine body condition scores(BCSs)by using the depth image of the back of cows.By background subtracting,the back-contour area of the cow was extracted from the depth image.The thurl,sacral ligament,hook bone,and pin bone were detected and evaluated by analyzing visibility and curvature,and four features were generated to measure their fatness.A training dataset was randomly selected from the entire dataset which contained 4,824 depth images.Decision tree learning,linear regression,and Backpropagation network models were built based on the training dataset and tested on the entire dataset.Over 99% of the predicted scores were within half-step of the BCS point from the manual scores based on the result of all three models,and 95.48%,86.14%,and 91.68% of the predicted scores were within quarter-step of the BCS point based on result of the decision tree learning,linear regression,and BP network models,respectively.On average,the BP network model scored all cows within quarter-step of BCS point from their manual scores during the study period.The result of BP network model showed that the average of mean absolute error(MAE)and standard deviation(STD)among cows were 0.11 and 0.069,respectively.The measurements of visibility and curvature have high correlations with BCS and can be used to assess BCS with high accuracy.