Recognizing mastitis using temperature distribution from thermal infrared images in cow udder regions

Abstract: Mastitis is one of the most common infectious diseaases resulting in the persistent and inflammatory response to the udder tissue of the dairy cow. This infection of microorganisms has posed a great threat to the milk yield, quality, and even be fatal to the cow. The severe losses have significantly been triggered to hinder the sustainable development of the dairy farming industry. Therefore, it is highly urgent to promptly and accurately monitor the udder health of cows. In this study, a new mastitis identification was presented to real-time measure the temperature distribution in the udder region of cows using a thermal infrared imaging system. A sample of 189 Holstein cows was collected in sites. A California Mastitis Test (CMT) was performed on the 142 infected and 47 uninfected cows. The mammary gland and the mammary gland pool areas were selected as the feature regions of interest (ROI), due to the anatomical structure of the udder and external factors. A regional temperature measurement toolbox of AnalyzIR was utilized to acquire the average, the minimum, and the maximum temperature of the feature ROI. An optimal time before milking was determined to collect the data during mastitis identification. A single-factor analysis of variance was also conducted to evaluate the temperature difference before and after milking, and the temperature measurement on both uninfected and infected cows. Meanwhile, the temperature differences demonstrated that the temperature of the mammary gland pool area in the uninfected cows was much lower than that of the mammary gland area, whereas, there was a much higher temperature reading of the mammary gland pool area in the infected cows, compared with the mammary gland area. Furthermore, a lines section was employed to divide the left and right udder regions of cows, particularly ranging from the mammal gland area to the mammal gland pool area. Every pixel on the section of the lines corresponded to a specific temperature value. These values were then acquired to find out the distribution tendency. As such, the temperature values on the section of the lines were well fitted to statistically analyze the associated slopes. The slope analysis revealed that 91.9% of healthy cows shared a slope of less than 0, with a slope range of -0.083 to -0.001, whereas, 92.1% of diseased cows presented a slope of greater than 0, with the values ranging from 0.001 to 0.093. Thus, automatic identification of mastitis in cows was achieved using the positive or negative slope of the temperature fitted lines. Three segmentations of gray-scale threshold were selected to calculate the rates of detection, false, and accuracy during imaging processing, including the fixed threshold, the iterative, and the OTSU algorithm. The fixed threshold was then determined as the optimized algorithm to recognize the right or left udder. Correspondingly, the dairy cow mastitis was identified to combine with the line section and the positive or negative of the fitting line slopes. In addition, the thermal infrared images of the 139 cows were collected before and after the milking happened. A comparison was then made on the measurement and automatically diagnostic data of CMT. It was found that the average diagnostic accuracies were 76% and 75%, respectively, under the left and right milk areas of uninfected and infected dairy cows. Finally, the proposed recognition can be widely expected to effectively identify dairy cow mastitis in real applications. The finding can offer a promising potential framework to monitor cow mastitis in a contactless manner.