[Hyperspectral Inversion of Soil Organic Matter Content in Mountainous Farmland Based on ResNet-MHAM Model]

Area of Science:

• Agricultural Science
• Remote Sensing
• Soil Science

Context:

• Karst mountainous regions present unique challenges for soil analysis.
• Accurate soil organic matter (SOM) prediction is crucial for agricultural management.
• Existing hyperspectral remote sensing methods lack accuracy and generalization.

Purpose:

• To develop an accurate and generalizable hyperspectral model for predicting SOM content.
• To address limitations in current SOM prediction techniques in Guizhou's karst soils.
• To introduce the ResNet-MHAM model for enhanced soil analysis.

Summary:

• A novel ResNet-MHAM model was developed using hyperspectral reflectance data.
• The model combines a 50-layer ResNet structure with MHAM for optimized performance.
• Validation demonstrated superior accuracy (R²=0.9172, RMSE=7.4549 g·kg⁻¹) over traditional models.

Impact:

• Provides an effective approach for SOM prediction in complex mountainous terrains.
• Enhances the application of hyperspectral remote sensing in soil science.
• Supports sustainable agricultural practices through improved soil monitoring.