Abstract: This study aimed to rapidly and non-destructively evaluate the levels of polysaccharides and flavonoids in A. roxburghii leaves under various photoperiods (10, 12, 14, 16, 18, and 20 h/d). Hyperspectral imaging was employed to acquire pixel-level spectral data from the leaves, and discriminant models for content levels were developed using traditional machine learning methods (PCA-LDA and PCA-SVM) and deep learning approaches (1D CNN and its optimized version). The findings revealed that the 1D CNN model outperforms the PCA-LDA and PCA-SVM models in terms of discrimination accuracy on the training, validation, and independent test sets, achieving 99.99% and 99.89%, 99.98% and 99.78%, and 91.62% and 87.92%, respectively. The introduction of a Dropout layer in the 1D CNN model enhances its generalization capability, increasing the discrimination accuracy for polysaccharide and flavonoid content levels on the independent test set to 98.92% and 95.67%, respectively. Additionally, visualization images depicting the discrimination results for different compound levels were constructed, providing an intuitive representation. This study validates the feasibility of hyperspectral imaging in evaluating polysaccharide and flavonoid levels in A. roxburghii leaves cultivated under various photoperiods, and the research results can provide technical support for the quality control of A. roxburghii.