Sow Behavior Feature Recognition Based on Deep Camera Data

As pig farming in China gradually shifts from traditional small-scale breeding to intensive, large-scale operations, automated monitoring of pig behavior and health status has become increasingly crucial. However, due to significant variations in lighting conditions within pig houses throughout the day and night, accurately recognizing sow behaviors using only RGB images remains challenging. Therefore, this study proposes a dual-stream RGB-D Faster R-CNN model that combines RGB and depth images to recognize sow behavior characteristics. By utilizing RGB-D images captured by a depth camera, the model accurately identifies sow postures under varying lighting conditions, thereby achieving more precise behavior recognition. Experimental results demonstrate that the dual-stream RGB-D model significantly improves recognition accuracy compared to traditional single-stream models, contributing to the timely detection of sow postures and enhancing farm productivity and animal welfare.