Automatic Detection of Mounting Behavior in Cattle using Semantic Segmentation and Classification

In cattle farming sector, the accurate detection of estrus plays a vital role because incorrect timing for artificial insemination affects the cattle business. The noticeable sign of estrus is the standing heat, where the cattle standing to be mounted by other cows for a couple of seconds. In this paper, we proposed cattle region detection using deep learning semantic segmentation model and automatic detection of mounting behavior with machine learning classification methods. Based on the conducted experiment, the results show that a mean Intersection of Union (IoU) of 98% on the validation set. The pixel-wise accuracy for two classes (cattle and background) was found to be both 98%, respectively. For the classification, the proposed method compares the four supervised machine learning methods which can detect with the accuracy rate of Support Vector Machine, Naïve Bayes, Logistic Regression and Linear Regression are 87%, 96%, 90%, and 80% respectively. Among them, Naïve Bayes algorithm perform the best. The novelty of this work noticeably implies that deep learning semantic segmentation could be effectively employed as a preprocessing step in segmenting the cattle and background prior to using various classification models.