Development of a Potato Seedling Detection Method and Platform for Field Emergence Monitoring

Cell Death & Disease · Springer Nature · 2026-09-21 · DOI: 10.1007/s12230-026-10078-2

Аннотация

Real-time monitoring of seedling emergence is of great importance for crop management and yield estimation. However, existing detection methods generally suffer from insufficient accuracy due to the complex field environment, the small size of potato seedlings, and their inconspicuous texture features. To address these issues, this study proposes an efficient potato seedling detection method based on the Real-Time Detection Transformer (RT-DETR). The original model is improved from three aspects. First, a lightweight backbone network, termed High-Frequency Feature Enhancement Network (HFNet), is designed to enhance the perception capability for small targets. Second, an efficient global feature integration module, namely Efficient Additive Attention (EAA), is introduced to improve the representation of seedling edge and texture features. Third, to reduce the computational cost in high-resolution image processing, a Shift-based Enhancement Block (SEB) is proposed to strengthen multi-scale modeling capability. Experiments conducted on the constructed potato seedling image dataset demonstrate that the optimized model (HES-DETR) achieves excellent performance across all evaluation metrics, with mAP@0.5 reaching 97.6%, Precision reaching 96.6%, and Recall reaching 94.0%. Compared with the original RT-DETR model, mean Average Precision (mAP@0.5) is improved by 3.6% points, the number of parameters is reduced by 42.2%, Floating Point Operations (FLOPs) are reduced by 29.5%, and the inference time is shortened by 2.7 ms. In addition, a Personal Computer (PC)-based graphical detection platform was developed in this study, supporting batch image import, detection, and result export, which further verifies the practicality and deployability of the proposed method. The results indicate that the improved model has significant advantages in enhancing the accuracy and resource efficiency of potato seedling detection, providing strong technical support for intelligent crop monitoring and field management.

Читать у издателя →

Похожие статьи