Research on Field Weed Detection Model Based on Improved RT-DETR
Abstract：Field weed detection is of great significance for improving yield and reducing pesticide use. However, factors such as small target size and large scale variation make existing methods difficult to balance accuracy and efficiency. A lightweight and efficient detection architecture is proposed based on RT-DETR. A lightweight multi-scale feature extraction and fusion module is designed to enhance the ability to detect weeds in complex environments. By partially extracting multi-scale features, diversified spatial features are obtained while reducing computational load; by enhancing cross-scale feature fusion, semantic consistency and convergence performance are improved. Experimental results show that the method significantly improves accuracy while maintaining a low parameter count and computational complexity, outperforming various mainstream models. It achieves a balance between accuracy and speed in lightweight models, providing new ideas for agricultural intelligent perception.
Keywords：Precision agricultureWeed detectionDeep learningRT-DETRSmart agriculture
Publication Date：2025-11-25
Online Publishing Date：2025-12-18(First online date of this platform, not the publication date of the document)
Pages：4( 39-42 )