In the heart of China’s tea-growing regions, a technological revolution is brewing, one that could redefine pest management and bolster the tea industry’s resilience. Researchers have developed an innovative detection method for Ectropis grisescens larvae, a notorious defoliating pest that threatens tea crops worldwide. Published in the journal *智慧农业*, this study combines the power of YOLO (You Only Look Once) object detection models with diffusion models to create a robust, real-time monitoring system for tea plantations.
Ectropis grisescens, commonly known as the tea looper, can cause significant damage to tea plants by feeding on their leaves. Traditional manual detection methods are labor-intensive, time-consuming, and often inaccurate. “The manual methods are not only time-consuming and labor-intensive but also suffer from low efficiency, high costs, and considerable subjectivity,” said lead author LUO Xuelun, a researcher at Zhejiang University. This new approach aims to address these challenges by leveraging advanced machine learning techniques.
The research team designed a hierarchical three-level detection system to capture the varying morphological characteristics of E. grisescens larvae across their four distinct instar stages. This system includes full-instar detection, grouped-stage detection, and fine-grained detection, each tailored to different aspects of larval identification. To overcome the challenges posed by limited and imbalanced training data, the researchers introduced a semi-automated dataset optimization strategy. This strategy enhances data quality and improves class representation, leading to more accurate and reliable detection models.
One of the most innovative aspects of this study is the use of a controllable diffusion model to generate high-resolution, labeled synthetic images of E. grisescens larvae. These synthetic images emulate real-world appearances under diverse environmental conditions, providing a rich and varied dataset for training the detection models. “The incorporation of the controllable diffusion model led to a universal performance boost across all YOLO variants,” noted LUO. This enhancement is statistically significant, as confirmed by a paired t-test (p < 0.05), suggesting that the synthetic images effectively enriched the training data and improved model generalization.The experimental results demonstrated the strong and consistent performance of the YOLO series models across various detection tasks. In the full-instar detection task, the best-performing YOLO model achieved an impressive average mAP@50 of 0.904, indicating a high level of detection precision. For the grouped instar-stage detection task, the highest mAP@50 recorded was 0.862, reflecting the model's ability to distinguish developmental clusters with reasonable accuracy. Even in the more challenging fine-grained individual instar detection task, the best mAP@50 reached 0.697, showcasing the feasibility of detailed stage-level classification.Among all evaluated models, YOLOv9 achieved the best overall performance, with top mAP@50 scores of 0.909, 0.869, and 0.702 in the full-instar, grouped-stage, and fine-grained detection tasks, respectively. When averaged across all tasks, YOLOv9 reached a mean mAP@50 of 0.826, accompanied by a macro F1-Score of 0.767, highlighting its superior balance between precision and recall.The commercial implications of this research are substantial. By enabling early and accurate detection of E. grisescens larvae, this technology can help tea growers implement timely and targeted pest management strategies. This can reduce economic losses, enhance crop quality, and promote sustainable tea cultivation practices. "This study demonstrated that the integration of a controllable diffusion model with deep learning enabled accurate field-level instar detection of Ectropis grisescens, providing a reliable theoretical and technical foundation for intelligent pest monitoring in tea plantations," said LUO.The lead author and their team hail from prestigious institutions, including the College of Biosystems Engineering and Food Science at Zhejiang University, the Department of Nutrition & Food Science at the National Research Centre in Egypt, the State Key Laboratory of Tea Plant Germplasm Innovation and Resource Utilization, and the Tea Research Institute of the Chinese Academy of Agricultural Sciences.As the tea industry continues to grow, driven by the increasing popularity of tea-based beverages, the need for efficient and sustainable pest management solutions becomes ever more critical. This research not only addresses these needs but also sets the stage for future developments in the field of agricultural technology. By combining advanced machine learning techniques with real-world data, this study paves the way for smarter, more resilient tea plantations and a thriving agricultural sector.