The study of sodium content in plants is crucial for the improvement of saline-alkali soil. Existing metal element detection methods pose challenges because they are complicated and time-consuming. In this study, we propose a quantitative detection model, FusionNet, that integrates Laser-Induced Breakdown Spectroscopy (LIBS) and Near-Infrared Hyperspectral Imaging (NIR-HSI) to realize the detection of Na element content in sorghum roots. To address the small-sample dataset, A Generative Adversarial Network (GAN) was employed to increase the diversity of the samples. The results indicated that data augmentation effectively enhanced the diversity of the original dataset and improved model performance. The modeling results from the FusionNet network achieved R2cv of 0.9915 and RMSECV of 0.7418, while R2p and RMSEP were 0.9808 and 0.6693. Compared to training with LIBS data alone, FusionNet achieved improvements of 4.94 % in R2cv and 5.61 % in R2p. This study provides a new method for detecting metal elements in plants.