MYCOLab Ai Logo - gold and army green

Improved Real-Time Detection Transformer with Low-Frequency Feature Integrator and Token Statistics Self-Attention for Automated Grading of Stropharia rugoso-annulata Mushroom

Summary

This research presents an improved artificial intelligence system for automatically grading Stropharia rugoso-annulata (wine cap) mushrooms based on their size and quality. The new system uses advanced computer vision techniques to analyze mushroom images in real-time, achieving 95.2% accuracy while being efficient enough to run on smaller computing devices used in food processing facilities. By combining wavelet analysis for capturing overall mushroom shape with a streamlined attention mechanism, the system successfully grades mushrooms faster and more consistently than manual sorting, potentially reducing labor costs in industrial mushroom production.

Background

Manual grading of Stropharia rugoso-annulata mushrooms is inefficient and subjective, accounting for significant labor costs in industrial production. Existing detection models struggle to balance accuracy, real-time performance, and deployability on resource-constrained edge devices. This study addresses the gap between maintaining global structural awareness and sensitivity to local details for large-target detection.

Objective

To develop an improved Real-Time Detection Transformer (RT-DETR) specifically tailored for automated grading of Stropharia rugoso-annulata mushrooms. The model aims to achieve high accuracy while maintaining real-time performance and suitability for deployment on edge devices used in post-harvest operations.

Results

The improved model achieved 95.2% mAP@0.5:0.95 at 262 FPS, outperforming APHS-YOLO and original RT-DETR in both accuracy and efficiency. LFFI alone improved mAP by 3.0%, while TSSA reduced FLOPs by 31.9% and parameters by 11.9%. The combined model reduced computational overhead by 27.3% compared to original RT-DETR while eliminating the need for NMS post-processing.

Conclusion

The improved RT-DETR offers an efficient framework for automated grading of large-target crops like Stropharia rugoso-annulata, balancing global structural awareness with local detail sensitivity. The synergistic integration of LFFI and TSSA modules enables practical deployment on resource-constrained edge devices while maintaining high accuracy. The model’s end-to-end architecture and reduced computational requirements make it suitable for industrial-scale post-harvest processing applications.
  • Published in:Foods,
  • Study Type:Algorithm Development and Experimental Validation,
  • Source: PMID: 41154117
Scroll to Top