Deep learning-based analysis of daily activity patterns of farmed dromedary camels

Abstract

Introduction This study addresses the need for automated monitoring solutions to evaluate the daily activity patterns of camels, which is critical for improving animal welfare and farm management practices. By leveraging advanced deep learning techniques, this research aims to identify and analyze five key daily activities—sleeping, sitting, standing, eating, and drinking—using video recordings from a camel farm in Fujairah, United Arab Emirates. Methods The dataset was collected over two 7-day phases in November and December 2022. In Phase 1, video recordings were analyzed to monitor the activities of two camels and measure the duration of each activity. In Phase 2, the study expanded to include six camels, enabling an evaluation of individual behavioral variations. The YOLOv7 object detection algorithm was used to train and validate the model on images extracted from the recordings, achieving high accuracy in detecting and classifying the defined activities. Results The results showed notable variations in activity patterns between Phases 1 and 2. Average standing time decreased from 9.8 hours (40.8%) to 6.0 hours (25.1%), and sleeping time dropped from 4.3 hours (18.0%) to 2.8 hours (11.7%). Conversely, sitting time increased from 6.2 hours (25.8%) to 9.9 hours (41.5%), and eating time rose from 3.1 hours (12.8%) to 4.6 hours (19.2%). Drinking time remained consistent at an average of 37 minutes (2.6%) across both phases. Activity peaks were observed during early mornings and after 16:00, with midday hours dominated by resting in shaded areas. Evening and nighttime activities primarily included sitting, minimal head movements, and occasional standing or walking. Discussion The established deep learning framework demonstrated reliable performance in detecting and analyzing camel activity patterns, offering a practical solution for continuous monitoring and improved farm management. However, further research is recommended to validate the model’s performance across different seasons and environmental conditions to enhance its robustness and adaptability.