Abstract: Abstract: With the enhancement of living conditions, the demand for milk is increasing rapidly, the quality of milk is paid more and more attention, and the improvement of the quality of milk has already become an important issue. However, subclinical mastitis in dairy cows is the most dangerous and costly disease which is difficult to control in dairy farm. In recent years, about 1/3 cows of the world are suffering from mastitis, especially subclinical mastitis in dairy cattle. Among them, the incidence of subclinical mastitis is 40%-80% in China, which is seriously harmful to the healthy development of dairy industry. In order to solve the problem of rapid detection of subclinical mastitis in dairy cows, a fast test system based on the computer vision technology of subclinical mastitis was proposed in this paper. Firstly, 25 dairy cows were selected randomly in the experiment, including 5 dairy cows with recessive mastitis, 5 dairy cows with severe mastitis and other 15 healthy dairy cows. Each cow has 4 breasts, so there were 100 sets of data in total. The Foss 5 000 milk somatic cell counts detector was used to obtain the number of somatic cells per sample. At the same time, the samples were dropped on the pH test paper, whose images were collected by USB (Universal Serial Bus) camera connected with the computer. The collected milk pH test paper images were changed into 500 × 500 pixels, and transformed from RGB (red, green, blue) color space to HSV (hue, saturation, value) color space. According to the color characteristics of the pH test paper, the threshold value was selected and the collected images were binarized. On the other hand, the segmented image was processed by morphological processing to remove the segmentation error and edge burr. Finally, the segmentation results were achieved by fusing the 2 results. Linear regression, power regression, quadratic regression, and principal component regression were used to establish estimation models using 75 sets of data. Those models were compared using the remaining 25 sets of data. The power regression of the principal component had a higher correlation coefficient, a lower standard error, and the highest determination coefficient (R2) of 0.970. System function and user interface were designed based on Android programming technology. The second experiment was carried out in the cattle farm to validate the favorable model by using the designed mobile terminal equipment which was connected with the USB camera. Using the 20 sets of data to validate the model, the correlation coefficient of the estimated milk somatic cell counts and the measurement value was 0.970, the estimated average relative error was 3.67%, and the standard deviation was 1.88%. The established estimation model of milk somatic cell counts using R and G indices estimated the milk somatic cell counts better than the model using only one index and the model combining 3 indices. Through the model comparison using the 100 sets of data and the validation in the real farm, the detection system of milk somatic cell count is more accurate, and can be used for the rapid detection of subclinical mastitis in dairy cows.