Connected attribute morphology for unified vegetation segmentation and classification in precision agriculture

Area of Science:

• Agricultural Science
• Computer Vision
• Image Processing

Background:

• Precision agriculture requires accurate crop and weed discrimination.
• Traditional segmentation methods often use global thresholding, limiting local detail.
• Existing techniques struggle with fine segmentation of plant regions.

Purpose of the Study:

• To develop a novel image processing pipeline for segmenting and classifying vegetation.
• To improve crop/weed discrimination in precision agriculture using attribute morphology.
• To enable local decision-making in image segmentation for enhanced detail preservation.

Main Methods:

• A novel image processing pipeline based on attribute morphology.
• Utilizing connected components within a max-tree hierarchical structure.
• Attribute filtering for local segmentation and feature extraction for Support Vector Machine (SVM) classification.
• Application to Normalized Difference Vegetation Index (NDVI) images.

Main Results:

• The proposed method achieves local segmentation of fine plant details, outperforming global thresholding.
• The segmentation data structure provides discriminative features for classification.
• Competitive classification rates for crop/weed discrimination were achieved.
• Successful application demonstrated on onion and sugar beet datasets.

Conclusions:

• Attribute morphology offers a powerful approach for precise vegetation segmentation in agriculture.
• The pipeline enables efficient and accurate crop/weed discrimination.
• This method advances automated analysis in precision agriculture.