To achieve timely monitoring of plant growth, rapid detection of differences in the distribution of catalase activity in melon leaves under different light intensities is essential. In this study, melon leaves were treated with different light intensities. Then the leaves were scanned using fluorescence hyperspectral imaging to extract the average spectral reflectance of 300 leaf samples, and the raw spectra were pre-processed and optimised by four pre-processing methods. Using interval Variable Iterative Space Shrinkage Approach (iVISSA), Competitive adaptive reweighted sampling (CARS), Genetic algorithm partial least squares algorithm(GAPLS), Iterative retained Information Variable(IRIV), and Variables Combination Population Analysis(VCPA) were used to extract the feature wavelengths. The partial-least-squares regression (PLSR) model screened the optimal feature wavelengths. Based on the preferred feature wavelengths, Principal component regression (PCR) model, Multiple linear regression (MLR) model, Convolutional Neural Network (CNN) model, Least Squares Support Vector Machine (LSSVM) model, and the results show that Baseline-IRIV-MLR model has the highest recognition accuracy, with an accuracy of 0.852 in both training and prediction sets. The results of this study provide a theoretical basis for applying fluorescence hyperspectral imaging technology in the quality evaluation of melon crops and technical support for the development of precision agriculture.