Inspection of teats by colour image analysis for automatic milking systems

In order to implement robotic milking it is necessary to automate the inspection and cleaning of cow teats prior to milking. This paper assesses the potential of detecting manure on teats using information from a colour image obtained using a standard colour CCD camera. Twenty-nine images were taken on clean and contaminated teats and a pixel-by-pixel analysis was then performed on the data from each image. It was shown that the colour information could not be used to unambiguously label each pixel as corresponding to clean teat tissue or manure. However, integrating the hue and saturation information from all the pixels did enable classification of the teats into clean and dirty categories which was in reasonable agreement with human assessment. It was found that all eleven of the images of dirty teats were correctly identified whereas four of the eighteen clean teats were incorrectly labelled.