SGSNet: a lightweight deep learning model for strawberry growth stage detection.

Area of Science:

• Agricultural Technology
• Computer Vision
• Deep Learning

Background:

• Accurate detection of strawberry growth stages is vital for optimizing crop production and management.
• Greenhouse environments and dense planting pose challenges for traditional monitoring methods.
• Lightweight, portable detection technologies are essential for modern precision agriculture.

Purpose of the Study:

• To develop a lightweight deep learning model for fast and accurate strawberry growth stage detection.
• To create a comprehensive dataset covering the entire strawberry growth cycle for model training and validation.
• To design a resource-efficient model suitable for deployment on portable devices.

Main Methods:

• Developed SGSNet, a lightweight deep learning model featuring the GrowthNet backbone for efficient feature extraction.
• Implemented DySample adaptive upsampling for improved detection of objects at various scales.
• Utilized the RepNCSPELAN4 module with an iRMB lightweight attention mechanism for multi-scale feature fusion.
• Applied the Inner-IoU optimization loss function to enhance convergence and accuracy.

Main Results:

• SGSNet achieved high performance with 98.83% precision, 99.45% recall, 99.14% F1 score, and 99.50% mAP@0.5.
• The model demonstrated superior performance compared to Faster R-CNN, YOLOv10, and RT-DETR.
• SGSNet boasts low computational cost (14.7 GFLOPs) and a small parameter count (5.86 million), ensuring resource efficiency.

Conclusions:

• SGSNet offers superior detection accuracy with significantly reduced computational requirements, making it suitable for portable agricultural applications.
• The model's efficiency and accuracy pave the way for advanced smart agricultural management systems.
• Future work includes extending SGSNet for growth stage detection in other crop types.