A Multi-Scale Dense Perception and Scale-Adaptive Approach for Blueberry Ripeness Detection

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Accurate blueberry ripeness detection is vital for intelligent harvesting.
• Complex orchard environments with dense fruit clusters pose challenges for current detection methods, leading to missed detections and ripeness misjudgments.

Purpose of the Study:

• To develop an improved object detection model, BBYOLOv12, to enhance blueberry ripeness detection accuracy in challenging agricultural settings.
• To address limitations in existing models regarding missed detections and ripeness misjudgments in dense fruit clusters.

Main Methods:

• Integration of a lightweight RepGhost backbone for efficient multi-scale feature extraction.
• Modification of the SimAM attention mechanism to improve feature capture in dense fruit regions.
• Implementation of an improved WIoU loss function for optimized small object localization.

Main Results:

• BBYOLOv12 achieved high performance metrics: mAP@0.5 of 98.97%, mAP@0.5:0.95 of 83.55%, precision of 97.55%, and recall of 97.27%.
• The model demonstrated superior performance compared to baseline and other lightweight models.
• Achieved high accuracy with a reduced model complexity (2.36 million parameters and 5.59 GFLOPs).

Conclusions:

• BBYOLOv12 offers an effective solution for multi-scale, dense perception tasks in agricultural applications, particularly for blueberry harvesting.
• The developed model provides a practical tool with a Graphical User Interface for real-time detection and analysis.
• This research contributes to advancing intelligent agricultural systems through improved computer vision techniques.