Classification of copper stress levels in oilseed rape leaves using Raman spectroscopy and CS-CARS feature selection

Copper stress affects the biochemical status of oilseed rape leaves and may alter their Raman spectral responses. However, rapid and nondestructive classification remains challenging because leaf Raman spectra are high-dimensional and contain redundant variables, while class-sensitive differences may be confined to narrow spectral regions. This study developed a Raman-spectroscopy-based framework for classifying copper stress levels in oilseed rape leaves. Raman spectra were collected from 600 leaves of 120 plants under five copper treatments by averaging five adaxial-surface measurements taken away from major veins and margins. To retain variables informative for particular stress levels while reducing redundancy, a class-specific competitive adaptive reweighted sampling (CS-CARS) method was proposed. Unlike conventional CARS, CS-CARS incorporates class relevance, class differentiation, and inter-variable redundancy into initial candidate-pool construction, candidate-subset evaluation, iterative weight updating, and final feature-subset selection. Feature selection and model optimization were conducted exclusively using the training data. Among the preprocessing methods evaluated, multiplicative scatter correction (MSC) provided the best overall classification performance. CS-CARS retained 124 of the 961 processed spectral variables. The resulting support vector machine (SVM) model achieved test-set Accuracy, macro-Precision, macro-Recall, and macro-F1 values of 97.78%, 97.81%, 97.78%, and 97.79%, respectively. Compared with SPA, conventional CARS, and iVISSA, CS-CARS achieved a more favorable balance between feature compression and classification accuracy. These results demonstrate that class-specific information in the CARS competition process improves the retention of Raman variables associated with different copper stress levels and provides an effective strategy for rapid, nondestructive classification of copper stress in oilseed rape leaves.