Improving C3K2 with RepConvBlock for high-accuracy and fast pig behavior object detection

Pig behavior objct detection is of great significance for swine health management and the improvement of breeding efficiency. However, existing methods suffer from low detection accuracy and insufficient efficiency in complex farming environments. To address these limitations, this study proposes an improved YOLOv11 algorithm by optimizing the C3K2 module with RepConvBlock, thereby enhancing the network's ability to extract pig behavior features. Experimental results demonstrate that the enhanced model achieves significant performance improvements, with a computational cost (GFLOPs) of 6.1 (an increase of 0.2 compared to the baseline model), an inference speed (GPS) of 627.98 (a 273.68 improvement over the baseline), and a mean average precision (mAP⁵⁰) of 0.9857. This method effectively balances computational cost and detection accuracy while significantly improving detection efficiency. It provides a high-performance and precise technical solution for intelligent pig farming monitoring, contributing to the digital transformation and precision management of swine breeding in smart agriculture.