| The breeding scale of large-scale pastures in China is expanding year by year,and the use of automatic monitoring equipment in pastures is increasing year by year,which provides great convenience for cow oestrus,simultaneous breeding and grouping management in production management.However,the continuous monitoring data in the production management system is not fully used for pastures production practice,and most of the data still need to be deeply mined with the help of computer technology to serve production practice.This experiment is based on the continuous monitoring data of cows suffering from four common diseases(reproductive diseases,digestive diseases,hoof diseases and respiratory diseases)20 days before diagnosis under the automatic monitoring system of large-scale pasture 1,including of milk yield,activity and rumination time as the training set.Through the descriptive statistics and significance analysis of the data and the correlation analysis of diseases,the risk factors of various diseases are determined.The early warning model of common diseases of dairy cows is established by binary logistic regression analysis,and the classification model of common diseases of dairy cows is established by discriminant analysis.The model is verified by pasture 1 test set and pasture 2 test set.The results were as follows:1.Dairy cow mastitis disease in dairy cow reproductive diseases is based on the data of pasture 1 training set.Through descriptive statistics and significance analysis,the early warning model is established by binary logistic regression analysis.The average milk yield in the first two days(P < 0.001),the average activity in the first three days(P = 0.032)and The rumination time on the 3rd day before diagnosis(P = 0.007)were significantly lower than that of healthy dairy cows,which showed that the greater the decline,the higher the risk of dairy cows suffering from mastitis disease.The sensitivity,specificity,AUC and accuracy of the early warning model were63.11%,97.02%,0.8786 and 89.62%.The mastitis early warning model was verified by the data of ranch 1 test set.The results showed that the sensitivity was 66.67%,the specificity was 94.83%,the area under the curve(AUC)was 0.9470 and the accuracy was 85.88%.The mastitis early warning model was verified by the data of ranch 2 test set.The results showed that the sensitivity was 92.11%,the specificity was 94.64%,the area under the curve(AUC)was 0.9807 and the accuracy was 93.62%.Using the data of ranch 1 training set,through descriptive statistics and significance analysis,and using binary logistic regression analysis to establish an early warning model,it can be concluded that the average activity of cows with metritis on the 13 th and 14 th day before diagnosis(P = 0.006)is significantly lower than that of healthy cows,indicating that cows with endometritis have a negative impact on the activity of these two days.The sensitivity,specificity,AUC and accuracy of the early warning model were 60%,99.19%,0.8799 and 90.51%.The dairy cow metritis early warning model was verified by the data of pasture 1 test set.The results showed that the sensitivity was 80%,the specificity was 98.31%,the area under the curve(AUC)was 0.9492 and the accuracy was 94.59%.The dairy cow metritis early warning model was verified by the data of pasture 2 test set.The results showed that the sensitivity was 72.22%,the specificity was 98.55%,the area under the curve(AUC)was 0.9090 and the accuracy was 93.10%.2.Using the data of pasture 1 training set,through descriptive statistics and significance analysis,and using binary logistic regression analysis to establish an early warning model,it can be concluded that the average milk yield(P = 0.017)and rumination time(P < 0.001)of dairy cows with digestive diseases in the two days before diagnosis are significantly lower than those of healthy dairy cows.The sensitivity,specificity,AUC and accuracy of the early warning model were 60%,99.19%,0.8799 and 90.51%.The dairy cow digestive diseases early warning model was verified by the data of pasture 1 test set.The results show that the sensitivity is 84.62%,the specificity is 100%,the area under the curve(AUC)is 0.9815 and the accuracy is 97.92%.The dairy cow digestive diseases early warning model was verified by the data of pasture 2 test set.The results show that the sensitivity is 64.71%,the specificity is 98.55%,the area under the curve(AUC)is 0.9455 and the accuracy is 93.75%.3.Using the data of pasture 1 training set,through descriptive statistics and significance analysis,and using binary logistic regression analysis to establish an early warning model,we can get the average milk yield of dairy cows from the 14 th day to the 12 th day before diagnosis(P =0.006),the average milk yield of dairy cows from the 7th day before diagnosis(P = 0.006)The mean rumination time on the 14 th and 13 th day before diagnosis(P = 0.013)and the rumination time on the first day before diagnosis(P = 0.012)were significantly lower than that of healthy dairy cows,indicating that the greater the decline,the higher the risk of dairy cows suffering from hoof disease.Using binary logistic regression analysis to establish the early warning model,the sensitivity is 78.95%,the specificity is 98.84%,the area under the curve(AUC)is 0.9694 and the accuracy is 95.24%.The dairy cow hoof disease early warning model was verified by the data of pasture 1 test set.The results show that the sensitivity is 75.00%,the specificity is 100%,the area under the curve(AUC)is 0.9701 and the accuracy is 97.47%.The dairy cow hoof disease early warning model was verified by the data of pasture 2 test set.The results showed that the sensitivity was 70.00%,the specificity was 98.70%,the area under the curve(AUC)was 0.9455 and the accuracy was 95.40%.4.Using the data of pasture 1 training set,through descriptive statistics and significance analysis,and using binary logistic regression analysis to establish an early warning model,it can be concluded that the milk yield of dairy cows with respiratory diseases one day before diagnosis(P < 0.001)is significantly lower than that of healthy cows.Binary logistic regression analysis was used to establish the early warning model,with sensitivity of 77.78%,specificity of 97.75%,area under curve(AUC)of 0.9501 and accuracy of 94.39%.The dairy cow respiratory diseases early warning model was verified by the data of pasture 1 test set.The results showed that the sensitivity was 77.78%,the specificity was 98.61%,the area under the curve(AUC)was 0.9783 and the accuracy was 96.30%.The dairy cow respiratory diseases early warning model was verified by the data of pasture 2 test set.The results show that the sensitivity is 90%,the specificity is 100.00%,the area under the curve(AUC)is 0.9844 and the accuracy is 98.85%.5.The classification model of common diseases of dairy cows was established based on the data of sick lactating cows in pasture 1 training set.The overall accuracy of the classification model was 73.0%,81.6% of mastitis,71.4% of metritis,63.2% of digestive diseases,55.6% of hoof disease and 74.1% of respiratory diseases.The classification model is verified by the data of ranch 2 test set,and the overall accuracy is 79.6%,mastitis 78.9%,metritis 94.4%,digestive diseases 64.7%,hoof disease 80.0%,respiratory diseases 80.0%.In summary,based on the data of milk yield,activity and rumination time of lactating cattle continuously monitored under the pasture automatic monitoring system,the changes and differences between sick cattle and healthy cattle can be obviously observed through descriptive statistics and significance analysis.The common disease early warning model and classification model have good sensitivity,specificity and accuracy.They can predict the occurrence of common diseases in the dairy farm and classify the diseases,which improves the use value of large-scale pasture automatic monitoring system. |
PDF Full Text Request