IAT-YOLO: A black pig detection model for use in low-light environments

In the large-scale intelligent pig farming industry, it is crucial to achieve real-time and accurate identification and positioning of each pig. However, the existing mainstream pig detection algorithms are mainly designed for white pigs, and the detection effect of black pigs is poor in low-light environments. In addition, when black pigs gather and adhesion and occlusion occur, the complexity of detection is further increased. Although strong light irradiation in the pigpen can improve the detection accuracy, it will have an adverse effect on the normal growth of pigs. To ensure accurate detection of black pigs in low-light environments without interfering with the normal growth of pigs, this research presents the IAT-YOLO model, derived from YOLOv7. It combines the brightness adaptive module IAT (Illumination-Adaptive-Transformer), the CA (Coordinate Attention mechanism) and the improved SPPCSPC-R module, which can effectively identify and locate the position and number of black pigs. Tests on the low-light black pig dataset reveal that the proposed model attains 96.8% accuracy, 83.8% recall, 91.8% mAP@0.5, and 54.5% mAP@0.5:0.95, which demonstrates higher accuracy and provides technical support for the intelligent detection of black pigs in such environments.