YOLOv11-GSF: an optimized deep learning model for strawberry ripeness detection in agriculture

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Detecting strawberry ripeness in greenhouses is difficult due to dense clusters, occlusions, and lighting variations.
• Existing methods struggle with efficiency, computational cost, and accuracy for small, packed targets.

Purpose of the Study:

• To develop a real-time algorithm for accurate strawberry ripeness detection in complex environments.
• To improve upon existing detection methodologies by enhancing efficiency and accuracy.

Main Methods:

• Introduced YOLOv11-GSF, incorporating Ghost Convolution (GhostConv) for efficient feature mapping.
• Utilized a C3K2-SG module with self-moving point convolution (SMPConv) and convolutional gated linear units (CGLU) for detailed feature capture.
• Implemented a F-PIoUv2 loss function for accelerated convergence and optimized classification.

Main Results:

• YOLOv11-GSF achieved 97.8% average precision, 95.99% accuracy, and 93.62% recall.
• Demonstrated significant improvements over the original YOLOv11, with gains of 1.8% in precision, 1.3% in accuracy, and 2.1% in recall.
• Showcased superior recognition accuracy and robustness compared to alternative algorithms.

Conclusions:

• YOLOv11-GSF provides a practical and efficient solution for strawberry ripeness detection.
• The algorithm effectively addresses challenges posed by complex greenhouse environments.
• Offers enhanced performance for automated agricultural quality assessment systems.