Objective This study aims to identify mattic epipedon using smartphone images and hyperspectral data.
Method A total of 163 soil profile images and corresponding hyperspectral data were collected in the northeastern Qinghai-Tibet Plateau. After preprocessing, color features from four color models (RGB, HSV, YIQ, and Lab) and six texture features (roughness, contrast, orientation, linearity, regularity, and roughness) were extracted from the soil profile images. Additionally, 38 spectral characteristic bands were obtained from the preprocessed hyperspectral data. Three feature schemes were utilized, including color features alone, color features combined with texture features, and color features combined with texture and spectral features. These were used to establish models using the K-Nearest Neighbor (KNN), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT) algorithms for identifying mattic epipedon.
Result The recognition performance of the three feature schemes showed that the combination of color, texture, and spectral features outperformed the combination of color and texture features, as well as the single color feature scheme. Among the three machine learning models, the performance of the GBDT model was slightly better than the RF model and significantly better than the KNN model. Specifically, the GBDT model using the combined color, texture, and spectral features achieved the best recognition results, with an overall classification accuracy of 0.929, an F1 score of 0.928, and an AUC value of 0.975.
Conclusion Compared to traditional methods that rely on expert experience and laboratory measurements, the identification of mattic epipedon using smartphone image features and hyperspectral data offers significant advantages in cost and efficiency. The GBDT model coupled with the combination of color, texture, and spectral features, enables high-precision identification of mattic epipedon, demonstrating that this integrated approach is reliable for identifying mattic epipedon.
DownLoad: