Tomato disease object detection method combining prior knowledge attention mechanism and multiscale features.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Tomato disease detection faces challenges with dense objects, scale variations, and small object features in complex backgrounds.
• Existing object detection methods struggle with accuracy in these scenarios.

Purpose of the Study:

• To develop an accurate tomato disease object detection method addressing limitations of current approaches.
• To enhance the detection of small and multi-scale tomato disease objects.

Main Methods:

• Proposed a method integrating prior knowledge attention mechanism and multi-scale features (PKAMMF).
• Employed a prior knowledge attention mechanism for enhanced feature fusion.
• Introduced a new feature fusion layer in the Neck and a specialized prediction layer for small targets.
• Utilized an Adaptive Structured IoU (A-SIOU) loss function for bounding box regression optimization.

Main Results:

• The PKAMMF method achieved a mean average precision (mAP) of 91.96% on a self-built dataset.
• Demonstrated a 3.86% improvement in mAP compared to baseline methods.
• Showcased significant enhancements in detecting multi-scale tomato disease objects.

Conclusions:

• The proposed PKAMMF method effectively improves tomato disease object detection accuracy.
• The integration of attention mechanisms, multi-scale features, and specialized layers enhances performance on challenging datasets.
• The A-SIOU loss function contributes to better bounding box regression accuracy.