Neural networks for robotic detection of mastitis in dairy cows: Netherlands and New Zealand perspectives

Abstract

This paper describes two parts of a continuing research project on developing neural network models for automated early diagnosis of mastitis in dairy cows milked by robotic milking systems. The justification for the project is that mastitis costs industry millions of dollars and severely compromises the health of cows. In the first part, robotic milking data from the Netherlands were used to develop Self Organising Map (SOM) networks providing 96% accuracy and revealing the nature of healthy and sick data regions. In the second part, New Zealand robotic data were used to map the development of mastitis from healthy, marginally ill through to ill stages. Models revealed that the characteristics of mastitis and healthy cases in terms of mastitis indicators are similar for the two countries.