Texture-Driven Rolling Dynamics Characterization via Lightweight SCConv and Multi-Modal Gabor-FFT Analysis for Tire Tread Assessment

Tire defect detection is critical for vehicle safety, yet existing methods inadequately correlate surface texture with rolling dynamics behavior. This paper proposes an integrated framework combining adaptive multi-model ensemble learning with Spatial-Channel Reconstruction Convolution (SCConv) for tire tread classification and texture-dynamics mapping. Four CNNs (ResNet18, ResNet50, VGG16, DenseNet121) are fused via diversity-weighted soft voting, achieving 96.8% accuracy on cracked/normal binary classification. SCConv reduces computational complexity by 42% while preserving texture-critical features through spatial-channel redundancy suppression. Multi-modal texture analysis—Gabor filter orientation energy, FFT spatial frequency, edge detection—establishes empirical correlations with rolling mechanics: friction coefficient μ, slip ratio s, and contact stiffness kc. Experimental validation demonstrates that orientation-dependent Gabor energy predicts friction anisotropy (μ∥/μ⊥ = 1.20), FFT spectral peaks correlate with slip (R2 = 0.82), and edge density quantifies stiffness degradation (12-18%). This texture-informed approach enables predictive maintenance and physics-guided tire design optimization.