ChickCount: ML-Enhanced Visible Light Sensing for Smart Poultry Monitoring

Abstract:
Chicken counting and density estimation are critical aspects of poultry welfare management, directly impacting chicken health, growth, and economic returns for producers. Traditional techniques based on image and video processing, although extensively researched, often fall short in poultry environments due to issues such as occlusion and high computational costs. To address these limitations, we propose an automated, cost-effective, and low-complexity visible light sensing (VLS) system for accurately counting chickens on farms. The proposed method employs a random forest (RF) classifier using differential received signal strength (DRSS) features. Experimental results show that a 1-second time interval represents the bestcase scenario for ground-truth labeling of small chickens, while a 4-second interval is more effective for larger chickens. Furthermore, across an area of 1.32 m², our system achieved a counting accuracy of 90% and a one-off accuracy of 94% for small white chickens, and a counting accuracy of 83% and a one-off accuracy of 90% for large white chickens.