Abstract
INTRODUCTION: Hyperspectral imaging (HSI) is a powerful, non-destructive technology that enables precise analysis of plant nutrient content, which can enhance forestry productivity and quality. However, its high cost and complexity hinder practical field applications. METHODS: To overcome these limitations, we propose a deep-learning-based method to reconstruct hyperspectral images from RGB inputs for in situ needle nutrient prediction. The model reconstructs hyperspectral images with a spectral range of 400-1000 nm (3.4 nm resolution) and spatial resolution of 768×768. Nutrient prediction is performed using spectral data combined with competitive adaptive reweighted sampling (CARS) and partial least squares regression (PLSR). RESULTS: The reconstructed hyperspectral images enabled accurate prediction of needle nitrogen, phosphorus, and potassium content, with coefficients of determination (R²) of 0.8523, 0.7022, and 0.8087, respectively. These results are comparable to those obtained using original hyperspectral data. DISCUSSION: The proposed approach reduces the cost and complexity of traditional HSI systems while maintaining high prediction accuracy. It facilitates efficient in situ nutrient detection and offers a promising tool for sustainable forestry development.