[Cotton identification and extraction using near infrared sensor and object-oriented spectral segmentation technique]

Area of Science:

• Agricultural Science
• Remote Sensing
• Computer Vision

Context:

• Precision agriculture relies on accurate crop identification for scientific management.
• Traditional pixel-based methods struggle with complex image processing and objective identification.
• High-resolution visible-near infrared imagery is crucial for detailed crop analysis.

Purpose:

• To develop and evaluate an object-oriented segmentation technique for precise cotton identification.
• To overcome limitations of traditional pixel-based methods in crop identification.
• To integrate spectral, shape, and topological features for accurate crop classification.

Summary:

• Visible-near infrared images of cotton were acquired using a high-resolution sensor.
• An object-oriented segmentation technique generated image objects with spatial/spectral features.
• A nearest neighbor classifier utilized these features for precise cotton identification, achieving 96.33% overall accuracy and a KAPPA coefficient of 0.9267.

Impact:

• The developed method provides a reliable foundation for scientific crop management in precision agriculture.
• It meets the demands for automatic management and decision-making in modern farming.
• This approach enhances the efficiency and accuracy of crop identification systems.