Segmentation of Individual Cattle in the Feedlot Using Mask R-CNN

ABSTRACT

Segmentation of individual cattle in the feedlot has been a herculean task for computer vision algorithms with great implications for animal farming. In this paper, segmentation of individual cattle in the feedlot was conducted by employing the algorithm of Mask R-CNN. Intersection over union (IOU) threshold of 0.5, average precision (AP) and mean average precision (mAP) were employed as metrics for the performance evaluation of the proposed model. The result of the experiment shows that the proposed model achieves an accuracy of 95%, which is an affirmation of the potential of Mask R-CNN model to perform competitively with the comparative object detection and instance segmentation models for individual cattle segmentation in the feedlot for the purpose of inventory, health and behavioural monitoring.

Keywords: Cattle, Feedlot, Mask R-CNN, Segmentation