Related Experiment Video
Updated: Jul 16, 2026
08:47
Computer Vision-Based Biomass Estimation for Invasive Plants
Published on: February 9, 2024
A semi-supervised domain adaptation framework with attention-enhanced ResNet50 for LIBS-based soil classification
Yihua He1, Yu Ding2, Xiangchu Li2
1School of Automation, Nanjing University of Information Science & Technology, Nanjing, 210044, China; School of Mechanical and Electrical Engineering, Anhui Jianzhu University, Hefei, 230601, China.
Analytica Chimica Acta
|March 11, 2026
Summary
This study introduces a semi-supervised domain adaptation framework to improve soil classification using Laser-Induced Breakdown Spectroscopy (LIBS) by addressing spectral shifts. The method achieves high accuracy, enhancing LIBS analysis for diverse soil samples.
Area of Science:
- Analytical Chemistry
- Machine Learning
- Geoscience
Background:
- Laser-Induced Breakdown Spectroscopy (LIBS) is a rapid, non-destructive technique for soil classification.
- Deep learning integration enhances LIBS soil classification, but spectral shifts challenge model performance.
- Addressing spectral distribution shifts is crucial for accurate LIBS soil analysis.
Purpose of the Study:
- To develop an effective method for mitigating spectral distribution shifts in LIBS soil classification.
- To improve the adaptability and performance of LIBS analysis across different soil domains.
- To enhance the application potential of LIBS technology in soil analysis.
Main Methods:
- A semi-supervised domain adaptation (SSDA) framework was proposed for cross-domain LIBS spectral classification.
- Gramian Angular Field (GAF) technique encoded 1D spectra into 2D images for enhanced feature representation.
- A deep convolutional neural network (SE-ResNet50) with attention mechanisms was integrated with SSDA, combining model-based transfer learning (MBTL) and pseudo-label learning (PLL).
Main Results:
- The proposed SSDA framework achieved a classification accuracy of 97.67% on the target domain test set.
- The method significantly outperformed comparative models in cross-domain soil classification.
- Experimental results validated the framework's effectiveness in mitigating spectral distribution shifts.
Conclusions:
- The study successfully addressed spectral distribution shifts caused by variations in soil properties.
- The developed SSDA framework significantly improved LIBS performance in cross-domain soil classification.
- The proposed method demonstrates substantial application potential for LIBS-based soil analysis.