This paper addresses the problem of automated monitoring of behavior and physiological state of dairy cows. The research aims to develop and apply neural network models using the YOLO architecture for video data analysis, which will automate and enhance the accuracy of monitoring animal behavior and condition. Key states of cattle on a dairy farm are identified in this study. As part of creating an effective system for analyzing cow behavior based on neural network models, video data was collected and prepared for training. A successful attempt was made to identify the “heat” state (COW_IS_ON_HEAT) after model fine-tuning. Based on the experiments conducted, it was concluded that the proposed approach allows models to better adapt to different conditions and visual signals, thereby increasing the accuracy and reliability of classification of animal states. The practical significance of this research lies in the potential application of the developed neural network models to improve conditions, increase productivity, and enhance animal health on dairy farms.