EDDet: efficient deep-fusion and dynamic optimization for small target detection in eggplant diseases

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Smart agriculture and global population growth necessitate advanced disease detection in vegetable production.
• Traditional methods struggle to identify small diseased areas in crops like eggplant, impacting yield and quality.
• Accurate disease detection is crucial for optimizing eggplant cultivation and ensuring food security.

Purpose of the Study:

• To develop an improved deep learning model for detecting small diseased spots in eggplant disease identification.
• To enhance the accuracy and efficiency of disease detection in smart agriculture settings.
• To address limitations in traditional methods for capturing subtle disease indicators.

Main Methods:

• Proposed the Efficient Deep-fusion Detection Model (EDDet) incorporating a Pinwheel Fusion Feature Extractor (PFFE) with Pinwheel Convolutions (PConv).
• Introduced a Cross-layer Attention Module (CAM) for efficient multi-scale feature fusion, including Cross-layer Channel Attention (CCA) and Cross-layer Spatial Attention (CSA).
• Implemented Scale-based Dynamic Loss (SD Loss) to stabilize bounding box regression and improve localization accuracy for small targets.

Main Results:

• EDDet achieved a mean Average Precision at IoU threshold 0.5 (mAP50) of 85.4%, a 2.8% improvement over the baseline.
• The model demonstrated high efficiency with 2.75M parameters and 9.1 GFLOPs.
• EDDet achieved a superior inference speed of 288.3 FPS, 37.5 FPS faster than the baseline.

Conclusions:

• EDDet significantly improves the detection accuracy and efficiency of small diseased areas in eggplant.
• The model's lightweight design and high inference speed make it suitable for real-time deployment in smart agriculture.
• This research offers a promising solution for precise and rapid disease monitoring in complex agricultural environments.