RESUMO
Automated milking systems (AMSs) already incorporate a variety of milk monitoring and sensing equipment, but the sensitivity, specificity, and positive predictive value of clinical mastitis (CM) detection remain low. A typical symptom of CM is the presence of clots in the milk during fore-stripping. The objective of this study was the development and evaluation of a deep learning model with image recognition capabilities, specifically a convolutional neural network (NN), capable of detecting such clots on pictures of the milk filter socks of the milking system, after the phase in which the first streams of milk have been discarded. In total, 696 pictures were taken with clots and 586 pictures without. These were randomly divided into 60/20/20 training, validation, and testing datasets, respectively, for the training and validation of the NN. A convolutional NN with residual connections was trained, and the hyperparameters were optimized based on the validation dataset using a genetic algorithm. The integrated gradients were calculated to explain the interpretation of the NN. The accuracy of the NN on the testing dataset was 100%. The integrated gradients showed that the NN identified the clots. Further field validation through integration into AMS is necessary, but the proposed deep learning method is very promising for the inline detection of CM on AMS farms.
RESUMO
Mastitis-associated milk losses in dairy cows have a massive impact on farm profitability and sustainability. In this study, we analyzed milk losses from 4 553 treated mastitis cases as recorded via treatment registers at 41 AMS dairy farms. Milk losses were estimated based on the difference between the expected and the actual production. To estimate the unperturbed lactation curve, we applied an iterative procedure using the Wood model and a variance-dependent threshold on the milk yield residuals. We calculated milk losses both in a fixed window around the first treatment day of each mastitis case and in the perturbations corresponding to this day, at the cow level as well as at the quarter level. In a fixed time window of day -5 to 30 around the first treatment, the absolute median milk losses per case were 101.5 kg, highly dependent on the parity and the lactation stage with absolute milk losses being highest in multiparous cows and at peak lactation. Relative milk losses expressed in percentage were highest on the first treatment day, and full recovery was often not reached within 30 days from treatment onset. In 62 % of the cases, we found a perturbation in milk yield at the cow level at the time of treatment. On average, perturbations started 8.7 days before the first treatment and median absolute milk losses increased to 128 kg of milk per perturbation. Mastitis is not expected to have equal effects on the four quarters so this study additionally investigated losses in the individual udder quarters. We used a data-based method leveraging milk yield and electrical conductivity to project the presumably inflamed quarter. Next, we compared losses with the average of presumably non-inflamed quarters. Median absolute losses in a fixed 36-day window around treatment varied between 50.2 kg for front and 59.3 kg for hind inflamed quarters compared to respectively 24.7 and 26.3 kg for the median losses in the non-inflamed quarters. Also here, these losses differed between lactation stages and parities. Expressed proportionally to expected yield, the relative median milk losses in inflamed quarters on the treatment day were 20 % higher in inflamed quarters with a higher variability and slower recovery. In 86 % of the treated mastitis cases, at least one perturbation was found at the quarter level. This analysis confirms the high impact of mastitis on milk production, and the large variation between quarter losses illustrates the potential of quarter analysis for on-farm monitoring at farms with an automated milking system.
