Design and testing of a target-oriented spraying system for an all-day cabbage-spraying robot

To address the problem of low detection accuracy caused by the degradation of image quality due to uneven lighting, strong light, or weak light in all-day conditions for open-field cabbages. This study focuses on field cabbage and investigates a deep learning-based target online detection method and model, which effectively reduces redundant information in both RGB and infrared images (IR) while efficiently extracting complementary features. The network was integrated with YOLOv8n to construct an all-day cabbage detection model that fuses RGB and IR modalities. Test results demonstrate that the dual-modality fusion significantly enhances all-day cabbage detection performance, achieving the recall of 0.951—an increase of 19.8% and 9.3% compared to single-modality IR and RGB models, respectively. The model was deployed on the Nvidia Jetson AGX Orin platform, and an all-day target-oriented spraying system was developed. Field tests show that the target spraying position errors in the daytime mode ( L 1 = 2.8 cm, L 2 = 3.6 cm) and nighttime mode ( L 1 = 2.4 cm, L 2 = 3.6 cm) both maintain centimeter-level accuracy. The target detection accuracy reaches 0.955 during the day and 0.965 at night. The all-day detection model and target-oriented spraying system proposed in this study provide effective technical support for the intelligent plant protection in field cabbage production.