Detection Of Dairy Cow Mastitis Based On Thermal Infrared Image

The intelligent detection of dairy mastitis is of great significance to improve the economic benefits of large-scale ranch,to ensure the dairy source and dairy products safety,and to provide a powerful driving force for the development of intelligent dairy farming.The artificial physical and chemical detection method of dairy cow mastitis is time-consuming,laborious and easy to interfere with dairy cows,which is difficult to be popularized in large-scale farms.At present,dairy cow mastitis based on thermal infrared technology has been developed rapidly,but most of the research still needs to extract the temperature manually.Based on the research status of dairy cow mastitis detection at home and abroad,combined with the characteristics of dairy cow mastitis leading to breast temperature rise,the detection method of dairy cow mastitis was developed based on thermal infrared technology and deep learning in order to provide an intelligent,accurate and non-contact detection method for cow mastitis.The main work and conclusions are as follows:(1)A thermal infrared image acquisition system was built,and the methods of cow side thermal infrared image preprocessing and dairy cow target segmentation were studied.According to the actual environment of dairy farms and the route of dairy cows,a thermal infrared image acquisition system was designed and built.After screening,2300 thermal infrared images were obtained.In order to reduce the noise in the thermal infrared images,median filter,Gaussian filter and mean filter were used respectively.The results show that median filter has the best denoising effect and the highest signal-to-noise ratio The results show that the target segmentation algorithm can obtain the complete cow target in the thermal infrared image,which lays the foundation for the accurate detection of the key parts of the dairy cow.(2)The eye region detection method based on skeleton and eye position features and the breast region classification and recognition model based on xgboost model training were studied and proposed.Based on the skeleton tree model and skeleton feature points,the head of dairy cow is segmented,and the relative position feature between the head contour feature and the eyes is used to detect the eye region of dairy cow;the skeleton feature of dairy cow breast region is extracted,and the classification recognition model of dairy cow breast region is trained based on xgboost model,and the most representative feature factors are skeleton feature vector V1n-V8 n and region area SE.The best external ellipse long axis Lm,realized the detection of cow breast region.The results showed that the recognition accuracy rates of eyes were 95.6%,93.3% and 98.8% respectively,and the recognition accuracy rates of breasts were 87.35%.(3)A detection method of key parts of dairy cattle based on yolov4 model was studied and proposed.Through the expansion and labeling of the original thermal infrared image,a data set composed of 3000 images is obtained.The dataset were divided into training set(1800 images,30%),test set(600 images,10%)and verification set(600 images,10%).,The detection model of key parts of dairy cattle based on yolov4 was constructed based on the characteristics of dairy cattle.The test set test results showed that the accuracy rate of the model was 97.04%,the recall rate was 98.21%,the map rate was 97.69%,and the average recognition frame rate per second was 62 f/s.Using the same data set for comparative experiments,the results show that the AP value of yolov4 model in detecting the eyes of cows with low head posture is 3.6% higher than that of the algorithm in Chapter 3,and the AP value of breast detection of cows with three postures is 12.65% higher than that of the improved model proposed by other scholars,which shows that yolov4 model can accurately detect the breast and eyes of dairy cows.(4)Taking the temperature difference between breast and eye area as the grading threshold,the mastitis detection model was established.According to the location of the region obtained by the deep learning detection model,the temperature difference between the breast and the eye region was obtained,and the temperature thresholds of grade 1(invisible mastitis)and grade 2(clinical mastitis)were determined by statistical analysis.The model was established to detect the mastitis grade,and the detection results were compared with those obtained by somatic cell detection method.The results showed that the clinical mastitis model obtained detection accuracy of 91.4%,specificity of 80%,and sensitivity of 93.3%.Although the detection effect was good;the detection accuracy of invisible mastitis(85.3%)was lower than that of clinical mastitis,the sensitivity was 87.5%.The experiment verified that the detection model of cow mastitis proposed in this paper can realize the detection of cow mastitis.