Automated color detection in orchids using color labels and deep learning

Area of Science:

• Botany and Computer Vision
• Machine Learning for Biological Classification

Background:

• Flower color is a key feature for plant identification and image classification.
• Existing flower image datasets lack descriptive text, hindering detailed analysis.
• Automated color detection in flowers is crucial for developing plant recognition systems.

Purpose of the Study:

• To develop an automated color detection model for flower images, specifically orchids.
• To create a novel dataset combining flower images with human-friendly textual descriptions.
• To evaluate deep learning models and transfer learning strategies for flower color recognition.

Main Methods:

• Developed a new dataset of orchid images with detailed textual descriptions.
• Applied transfer learning using five neural network architectures (VGG16, Inception, Resnet50, Xception, Nasnet).
• Investigated multi-class, combined binary, and ensemble classifiers for color detection.

Main Results:

• The proposed model accurately detects flower and labellum colors without requiring image segmentation.
• The ensemble classifier demonstrated the best overall performance in color detection tasks.
• Transfer learning with various layer unfreezing schemes was evaluated across different architectures.

Conclusions:

• The developed method provides a robust approach for automated flower color detection.
• This research lays the foundation for explainable, image-based plant recognition systems.
• The novel dataset and model advance the field of computer vision in botany.