Classification of Strawberry Maturity Using Swin Transformer V2: Integrating Near-Infrared Spectral, Chemical, and Optical Properties

• Swin Transformer V2 achieves perfect accuracy in strawberry maturity assessment. • Novel multi-modal dataset combines visual, chemical, and spectral measurements. • Brix-to-acidity ratio effectively distinguishes three maturity stages. • Optimized NIR preprocessing reveals distinct maturity-based clustering. • Visual features strongly correlate with chemical indicators of fruit maturity. This study presents a comprehensive approach to automated strawberry maturity classification by integrating computer vision techniques with chemical and spectral analyses. We developed and evaluated three deep learning models—RegNet_Y_800MF, RegNet_Y_8GF, and Swin Transformer V2 Tiny—for classifying strawberry maturity using a novel dataset of 270 images from 90 strawberry samples. The Swin Transformer V2 Tiny achieved perfect classification accuracy (100%) on the test set, outperforming both RegNet architectures while maintaining a balanced computational footprint. Chemical analysis revealed that the Brix-to-acidity ratio provided the most distinct separation between maturity stages (4.92 ± 0.80 for immature, 6.18 ± 0.98 for semi-mature, and 8.12 ± 2.40 for mature fruits). Near-infrared spectral analysis, optimized through 1,220 preprocessing pipelines, achieved a Silhouette score of 0.5744 using localized standard normal variate with seven windows and flat-window smoothing filtering. Strong correlations were demonstrated between visual characteristics, spectral properties, and chemical indicators of maturity, validating the effectiveness of computer vision-based approaches for automated strawberry maturity assessment. This integrated approach advances the development of reliable and efficient fruit quality assessment systems for commercial applications.