Research on Potato Eye Detection Method Based on Improved RT-DETR

Accurate identification of potato bud eyes is crucial for seed potato cutting, quality grading, and automated planting. To address challenges such as small target size, inconspicuous features, and low detection accuracy in potato bud eyes, this study proposes an improved RT-DETR-based detection method. The approach first employs ResNet18 as the backbone network to extract multi-scale features, then incorporates a Channel Cross-Integration Transformer (CCFT) module to enhance multi-scale feature fusion, and finally introduces an Adaptive Feature Attention Module (AFAM) in the feature fusion network to optimize feature representation. Experimental results demonstrate that the improved model achieves a mean absolute precision (mAP) of 95.2% on the self-built potato bud eye dataset, representing a 3.5 percentage point improvement over the original RT-DETR. With a detection speed of 72 frames per second (FPS), the model meets real-time detection requirements. This research provides effective technical support for intelligent potato production.

Hali tarjima qilinmagan