Soil moisture classification is essential for precision agriculture, influencing irrigation management, plant health monitoring, and resource efficiency. The Soil Digital Twin (DT) framework, previously validated for RGBY LED illumination, has demonstrated reliable classification aligned with real-world soil images. However, alternative wavelengths remain underexplored. This study introduces Cyan, Magenta, Ultraviolet, and White (CMUW) LEDs to evaluate spectral feature extraction and classification accuracy. A multispectral imaging system captured soil images under controlled conditions, followed by feature extraction and classification using Random Forest (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANNs). Statistically significant improvements (p < 0.05) were observed for clay (6.53%) and silt (6.31%), with moderate improvement for loam (3.45%). Sand classification remained essentially unchanged (1.11%) due to high reflectance. CMUW LEDs enhanced spectral differentiation, addressing RGBY’s limitations in challenging soil textures. CMUW LEDs show promise in autonomous soil monitoring with mobile robots, a significant practical implication of this research.