[Objective] Defoliation rate is an important basis for evaluating defoliation and ripening effect of machine-picked cotton. In this study, 14 kinds of color vegetation index were extracted from RGB images to establish a fast and accurate estimation model for cotton defoliation rate, which provides a theoretical and technical basis for timely harvesting of machine-picked cotton. [Methode] Different cotton varieties were set up. The data of cotton defoliation rates of different defoliant concentrations and spraying times were collected, and the canopy RGB images were collected by unmanned arerial vehicle(UAV). The correlation between color vegetation index and cotton defoliation rate was analyzed. Then, the estimation model of cotton defoliation rates was constructed by using the methods of simple linear regression (SLR), multivariate linear regression (MLR) and partial least square regression (PLSR). Meanwhile, the model was evaluated. [Result] The results showed different cotton defoliation rates after different treatments. There was a remarkable correlation between the defoliation rate and the visible light vegetation index; especially, the correlation coefficient between the triangular greenness index (TGI) and cotton defoliation rate is up to 0.81. The results showed that the model based on TGI index was the best in the linear regression model, coefficient of determination (R2) = 0.66, root mean squared error (RMSE) = 10.44%, relative RMSE (rRMSE) = 12.87%; the model based on excess blue index (ExB), green leaf index (GLI), TGI and excess green index (ExG) was outstanding in the multiple linear regression models (R2 = 0.70, RMSE = 10.26%, rRMSE = 12.65%). In the PLSR models, the one with ExB, GLI, TGI, ExG, Comprehensive 2 and Comprehensive 1, had higher accuracy, R2= 0.70, RMSE = 10.02%, rRMSE = 12.22%. The external verification showed that there was a good fitting relationship between the measured values and the predicted values of each model. [Conclusion] The model established by MLR and PLSR has high accuracy and great fitting degree. Therefore, considering of the weight computation and complexity, the cotton defoliation rate estimation model established with ExB, GLI, TGI and ExG has excellent performance.