SCFM-DETR: an enhanced transformer-based method for automated maize disease detection in field environments.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Maize diseases significantly reduce crop yields.
• Accurate disease identification is difficult in complex field conditions due to variable lighting and weather.
• Existing methods struggle with similar disease symptoms and background noise.

Purpose of the Study:

• To develop a robust and efficient maize disease detection model for complex environments.
• To improve the accuracy and reduce the computational load of maize disease identification.
• To facilitate intelligent crop monitoring and advance smart agriculture.

Main Methods:

• Proposed a novel detection model, SCFM-DETR, based on an improved Real-Time Detection Transformer (RT-DETR).
• Employed SimAM-StarNet as the backbone for feature extraction, optimizing multiscale feature fusion and reducing background noise.
• Integrated a newly designed CGLU-FasterBlock-MANet (CFM) module for enhanced adaptive feature fusion.

Main Results:

• SCFM-DETR achieved 96.7% average precision and 95.8% recall on a maize disease dataset.
• Outperformed the baseline RT-DETR-R18 model by 3.1% in precision and 6.0% in recall.
• Reduced model parameters by 47% and computational load by 49%, demonstrating a lightweight design.

Conclusions:

• SCFM-DETR offers a high-accuracy, lightweight framework for maize disease detection.
• The model is suitable for deployment in resource-constrained agricultural settings.
• This advancement supports intelligent crop disease monitoring and smart agriculture initiatives.