SLFCNet: an ultra-lightweight and efficient strawberry feature classification network.

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Automated strawberry detection, classification, and harvesting are advancing agricultural technology.
• Existing object detection methods face challenges with computational demands, resource utilization, and efficiency, hindering edge device deployment.
• Suboptimal user experiences arise from the limitations of current strawberry detection systems.

Purpose of the Study:

• To develop a lightweight model for real-time strawberry fruit detection and classification.
• To address the computational and efficiency limitations of existing object detection methods for agricultural applications.
• To enable seamless deployment of strawberry classification on edge devices for improved user experience.

Main Methods:

• Developed the Strawberry Lightweight Feature Classify Network (SLFCNet), a novel lightweight model.
• Incorporated a lightweight encoder and a custom feature extraction module (Combined Convolutional Concatenation and Sequential Convolutional - C3SC).
• Evaluated the model using a high-resolution strawberry dataset with image augmentation and compared results with manual counts.

Main Results:

• SLFCNet achieved 98.9% mAP@0.5, with 94.7% precision and 93.2% recall.
• The model boasts a compact size of 3.57 MB and processes images in 4.1 ms on a GTX 1080 Ti GPU.
• Demonstrated real-time performance suitable for edge device deployment.

Conclusions:

• SLFCNet provides a novel, efficient solution for automated strawberry harvesting and management.
• The model's lightweight design and high performance make it suitable for real-time applications on edge devices.
• This research contributes to advancing precision agriculture through efficient AI-powered fruit classification.