Despite the enormous amounts of waste textiles produced by the world’s textile industry’s explosive growth, resource utilization rates are still poor. Cotton/polyester blended waste fabrics make up a sizable share, and sorting them precisely is essential to increasing recycling value and promoting the circular economy in the textile industry. Traditional mechanical and human sorting techniques are ineffective and inaccurate; current spectral analysis algorithms mainly concentrate on quantitative composition prediction and are insufficiently capable of differentiating between waste fabrics with comparable content gradients. To address these challenges, this paper proposes an improved 1DCNN model (Dual-1DCNN-Residual-SE) integrated with Near-Infrared (NIR) hyperspectral imaging technology. This model takes raw spectral data and Savitzky-Golay (SG) smoothing data as dual-channel inputs, introducing residual connections to capture subtle spectral differences between similar fabric categories, and employs SE attention mechanisms to adaptively enhance key features. Comparative experiments with four traditional algorithms—KNN, RF, SVM, and PLS—demonstrate that the proposed model achieves a classification accuracy of 95.94%, surpassing the best traditional algorithm SVM (88.12%) by 7.82%. Ablation experiments confirm each enhanced module’s efficacy. This study achieves high-precision classification of cotton/polyester blended waste fabrics, providing technical support for intelligent sorting of industrial waste fabrics.

High-value recycling of waste textiles requires accurate fiber-composition identification, yet similar blends often exhibit overlapping near-infrared absorption bands and substantial within-class variation. Using 291 waste-textile swatches, this study develops ConvPatchTST, a convolution-enhanced Patch Transformer that integrates Raw and SNV (SG) spectra as complementary input channels. Overlapping convolutional patch embedding captures local continuous absorption features, while a Transformer encoder models cross-band dependencies. Spectral bands spanning 958.5–1646.6 nm were selected for model training and analysis. Under a single fabric-swatch-grouped internal train-validation split with pixel-level evaluation, ConvPatchTST achieves an accuracy of 0.966 and a macro F1 of 0.970, outperforming several baseline models. The Raw + SNV (SG) representation also surpasses single-channel, three-channel, and alternative dual-channel inputs. Pairwise analysis identifies statistically discriminative and highly attributed intervals for four similar material pairs. For polyester/polysp, the intervals near 1369.0–1379.6 nm show clear statistical-model agreement. These results indicate that complementary spectral representations and convolution-enhanced Patch modeling support accurate classification and provide band-level clues for interpreting model decisions. Generalization to external samples, batches, and instruments requires further validation.