Learnable Residual Local Binary Patterns: A Pretraining Preserving Architecture for Cotton Percentage Estimation in RGB Fabric Images

Automated cotton-percentage identification underpins sustainable textile recycling, but established near-infrared and ATR-FTIR spectroscopy systems cost USD 10,000–25,000 per unit and remain inaccessible to small recyclers. We address this on the CottonFabricImageBD dataset (1,300 RGB originals, 13 ordinal cotton classes from 30% to 99%) and report three contributions. First, the Learnable Residual LBP stem, which retains the pretrained ResNet50 first convolution intact and adds a fully differentiable Local Binary Pattern branch as an additive contribution gated by a single learnable scalar α initialized to zero, ensuring the model is numerically equivalent to the baseline at initialization (verified to a maximum absolute logit difference below 10⁻⁴). Second, a controlled six-variant comparison (vanilla baseline, CLBP, LBP-Conv, LBP-Residual, LBP+SVM, LBP+ANN) under identical stratified 5-fold cross-validation on the 1,300 dataset originals. Third, the isolation of pretraining preservation as the dominant architectural variable: the 7.08 pp top-1 gap between LBP-Conv (43.77%) and LBP-Residual (50.85%), both embedding the identical learnable LBP module, is statistically significant (p = 0.004, paired t-test, df = 4) and consistent across all five folds. Classical LBP+SVM and LBP+ANN baselines reach 31.85% and 34.46% top-1, confirming genuine but limited cotton-density signal in hand-crafted descriptors. Compared to the concurrent triplet-architecture approach of Wiedemann et al. (2025), which achieves 48.15% top-1 accuracy and 14.77% RMSE on the same dataset under identical 5-fold cross-validation, LBP-Residual attains 50.85% top-1 and an RMSE equivalent of ≈7.2% (computed under a different convention and thus not strictly comparable) using a single lightweight backbone rather than an ensemble of three. These results support the design principle: augment, do not replace.