Development of an on-line mastitis detection system within an integrated knowledge-based system for dairy farm management support.
Some aspects of automated clinical mastitis detection and diagnosis are discussed. Knowledge representation techniques for the different steps in the diagnostic process are presented. The main focus of this paper is on automated early detection, based on data that are automatically collected in the milking parlour. Principal component analysis, logistic regression and back-propagation neural networks were used in the analysis of the automatically collected data. The 3 techniques did not differ greatly in performance. All the techniques performed better when data from milking with observed clinical signs were used, compared with data from milking before clinical symptoms were noticed. Healthy quarters were mostly correctly classified by all techniques. It seems unlikely that all clinical mastitis cases can be detected at milking before visible clinical signs occur.