Visual Udder Detection with Deep Neural Networks

Abstract

This paper discusses the techniques used to aid the work in progress artificial calf weaning system. The aim is to artificially wean the calves by deterring them to suckle the milk during the weaning phase. The proposed collar mounted micro-pneumatic deterring system is the work in progress. To deter the calves there should be a mechanism to visually detect the udder accurately. One possible way to do this is by detecting the udder from a certain distance with a collar mounted camera system and predict the suckling event. A pre-trained deep neural network model embedded in a device with camera mounted to the collar will detect the udder and predicts the suckling event which enables the deterring system to actuate. This method of detecting udders has been chosen to overcome collar’s varying oscillatory movements caused by the calf in an attempt to suckle and it is independent on variable visible light conditions.