Real-Time Detection of Tea Chrysanthemums in Complex Outdoor Scenes

Picking robots have emerged as a pivotal focus in smart agriculture, with precise target position detection being a key element for realizing robotic picking. However, detecting tea chrysanthemums presents various challenges, including their dynamic appearance, complex environmental conditions, susceptibility to occlusion and lighting variations, and the requirement for real-time embedded systems. When applied to complex outdoor scenes for Tea chrysanthemum targets, existing detection models suffer from drawbacks such as missed detections, low accuracy, and slow processing speeds. To address these challenges, we introduce YOLO-VTC (You Only Look Once-Varieties of Tea Chrysanthemum), a lightweight deep-learning model optimized for the Jetson Xavier NX platform. Detection experiments were conducted on image data capturing eight varieties of tea chrysanthemums in complex outdoor scenes collected in 2023. Results demonstrate that YOLO-VTC achieves an average accuracy of 86.60% and a peak accuracy of 97.10%, surpassing other lightweight networks with remarkable performance. Notably, the model effectively maintains a compact size of 6.58MB and has been successfully deployed on edge device platforms.