Ecological observation station provides high-throughput canopy images for vegetation growth monitoring in karst ecological fragile area， however， there are few reports on the extractionof vegetation from bare rock and vegetation mixed underlying surface in karst area. In order to provide technical support for vegetation monitoring based on ground visible images， the canopy RGB images of the rocky desertification ecological observation station were used to study segmentation algorithm and growth monitoring index applicable to karst vegetation. The results show that： （1） The differentiation degree of light green vegetation in karst area is high for color space， nonlinear combination of color channels and machine learning， but its sensitivity to bare rock and dark green vegetation is various. The vegetation segmentation effects of the three segmentation methods were significantly different under strong light in sunny day and weak light in cloudy day. The machine learning algorithm has the best segmentation effect with the accuracy is over 80% under weak light in cloudy day and over 90% under strong light in sunny day. （2） The trend of vegetation growth reflected by GLA， NDYI， NGRDI and VARI was similar. In these indices， NDYI is more sensitive to the difference of vegetation growth. The compound sine function can simulate the daily dynamic changes of these four indices， moreover， the simulation accuracy of NGRDI trend is the highest with R² = 0.830.