AUTOMATIC DETECTION AND TRACKING OF MOUNTING BEHAVIOR IN CATTLE USING A DEEP LEARNING-BASED INSTANCE SEGMENTATION MODEL

In precision livestock farming, estrus detection in cattle is particularly important for cattle breeding management. With accurate estrus detection, artificial insemination can be administered, which proportionally affects the productivity of livestock farms. Most estrus behaviors can be successfully detected by recognizing the mating postures of cattle. Therefore, in this paper, we propose an estrus detection approach that tracks and identifies cattle mating postures individually based on video inputs. To achieve precise identification and to obtain individual cattle information, segmenting each cattle from its background is a vital step. To solve pixel-level segmentation masks for the cattle in an outer ranch environment, an instance segmentation approach based on a Mask R-CNN deep learning framework is also proposed. In this paper, individual cattle segmentation for detecting the mounting behaviors is carried out first. This is followed by a lightweight tracking algorithm as a post-processing step which is our study innovation. The training data were collected by installing surveillance cameras at a livestock farm, and for the testing data, various datasets from different camera placements were used. The proposed approach achieved 95.5% detection accuracy in identifying the estrus behaviors of cattle.