Visual Classification of Mixtures Composed of Different Soil Types: An SVM-Based Approach

Soil type affects the bearing capacity of soils, which in turn affects the stability of civil engineering structures. Therefore, the determination of soil types is an important issue in geotechnical engineering. The conventional methods for soil classification are laborious, human error-prone, and long processes. To mitigate these effects, this study focuses on the effectiveness of Support Vector Machine (SVM) models in visual soil classification from images. A new dataset of binary soil mixtures was created using six different soil samples. After image preprocessing, SVM models with Linear, Radial Basis Function (RBF), and Polynomial kernels were trained and tested. Their performance is compared statistically by one-way repeated-measures ANOVA and evaluated with 10-fold cross-validation. The results showed that kernel choice has a clear effect on accuracy, recall, and stability. Among the models, the Polynomial and Linear kernels gave the most consistent and balanced outcomes.