Abstract: 【Objective】 Rapid and non-destructive monitoring of SPAD value has important guiding significance for field management measures. 【Methods】 In this study, the DJI M350 RTK drone equipped with multispectral sensors was used to obtain multi-temporal canopy remote sensing images, calculate the multispectral vegetation index, screen the features significantly related to SPAD value, and four machine learning algorithms including limit learning machine, random forest regression, support vector regression and multiple stepwise regression were combined to construct a SPAD value estimation model for cotton at each growth stage. 【Results】 The results showed that there was a significant positive correlation between vegetation index and cotton SPAD value at each growth stage, and there was a high correlation between CCCI (canopy chlorophyll content index) and CIrededge (red edge chlorophyll index) and SPAD value, with the highest correlation coefficients of 0.81 and 0.78, respectively. Comparing the accuracy of the model at different growth stages, it was found that the model at flowering stage had the highest accuracy, with the best estimation model being ELM, R² being 0.741, RMSE being 1.447, rRMSE being 0.023, ELM being the best estimation model at budding stage and flocculation stage (R² being 0.656 and 0.587, respectively), and RFR (R² being 0.577) at full boll stage. 【Conclusion】 This study shows that the optimal growth period for SPAD value estimation of cotton leaves is atthe flowering stage, the optimal model is ELM, and the model accuracy R² is the highest 0.741.