Enhanced Garlic Crop Identification Using Deep Learning Edge Detection and Multi-Source Feature Optimization with...

Area of Science:

• Agricultural remote sensing
• Deep learning applications in agriculture
• Geospatial analysis of crop production

Background:

• Accurate identification of garlic cultivation is vital for agricultural management and economic planning.
• Traditional crop identification methods struggle with accuracy and spatial fragmentation in diverse agricultural landscapes.
• Precision agriculture demands advanced techniques for reliable crop mapping.

Purpose of the Study:

• To develop an integrated technical framework for accurate garlic identification.
• To enhance garlic cultivation area mapping in Kaifeng City, Henan Province.
• To overcome limitations of traditional methods in complex agricultural settings.

Main Methods:

• Utilized deep learning edge detection (DexiNed) with high-resolution satellite data for field boundary extraction.
• Integrated multi-source features (Sentinel-1 SAR, Sentinel-2 multispectral, vegetation indices) and optimized using random forest and recursive feature elimination.
• Applied spatial constraints via field boundaries to refine pixel-level classification and generate field-scale products.

Main Results:

• Feature optimization improved overall accuracy from 0.91 to 0.93 and Kappa coefficient from 0.8654 to 0.8857.
• The DexiNed network achieved a 94.16% F1-score for precise field boundary extraction.
• Spatial optimization effectively reduced noise, validating successful garlic identification in Kaifeng.

Conclusions:

• The integrated framework offers a robust solution for accurate garlic crop identification.
• Deep learning and multi-source data fusion significantly enhance precision agriculture capabilities.
• The developed method provides reliable field-scale crop identification products for agricultural resource management.