MFDN: an efficient detection method for Alstroemeria Genus flowers based on multi-scale feature fusion

Area of Science:

• Precision Agriculture
• Computer Vision
• Deep Learning

Background:

• Alstroemeria Genus Morado is a significant ornamental plant.
• Automatic detection of flower maturity is crucial for precision agriculture.
• Challenges include diverse morphologies, complex environments, occlusion, and lighting variations, with limited research.

Purpose of the Study:

• To propose a deep learning-based object detection framework for Alstroemeria Genus Morado flower maturity.
• To enhance the detection of small targets and improve classification accuracy.

Main Methods:

• Developed the Morado Flower Detection Network (MFDN), comprising backbone and head networks.
• Introduced novel modules like C3k2_PPA for multi-branch fusion and attention mechanisms.
• Utilized CARAFE for upsampling, Concat for feature combination, and an optimized C2f module for accelerated processing.

Main Results:

• MFDN demonstrated outstanding performance on the morado_5may dataset in Precision, Recall, and F1-score.
• Achieved a mean Average Precision (mAP) 1.3%-5.8% higher than YOLO-series models.
• Exhibited strong generalization ability.

Conclusions:

• MFDN offers a robust solution for precise detection and classification of Alstroemeria Genus Morado flower maturity.
• The framework is expected to significantly improve efficiency and automation in agricultural production.