Above ground biomass (AGB) is an important parameter that characterizes crop life activities and is particularly critical for crop growth monitoring and yield prediction. Therefore, obtaining AGB information quickly and accurately is of great significance for monitoring crop growth, guiding agricultural management and improving yield. Using UAV as a platform to carry digital camera sensors, due to the advantages of strong maneuverability, low price and high spatial resolution, and to estimate crop AGB timely and accurately has become one of the hotspots in remote sensing estimation research. As the accuracy of the AGB estimation model for digital images of different flying heights with different resolutions of UAV is different, this study tried to set up 5 types of flying heights at 10, 20, 30, 40, and 50 m during the potato tuber growth period to obtain digital images of different resolutions, and to explore its influence on the accuracy of building AGB model based on spectral information, texture features and spectral information + texture features. Firstly, based on the digital image of UAV, the spectral information and texture features are extracted separately, the vegetation index from the spectral information and texture features constructed, combined with the measured AGB obtained by ground experiments respectively for correlation analysis, and the top 10 image indexes and the top 8 texture features with larger absolute values of correlation coefficients are selected separately. Then, three variables integration variance inflation factor (VIF) are used to perform principal component analysis (PCA) dimensionality reduction processing, and the best principal components are obtained and multivariate linear regression (MLR) constructs AGB estimation model. Finally, compare the AGB estimation model precision of digital images with different resolutions with three variables and the same resolution with the same variable. The results show that: (1) When the image resolution changes between 0.43 and 2.05 cm, the correlation between texture features and potato AGB is weaker than that of vegetation index, but both reach a very significant level of correlation (p<0.01). With image resolution is reduced, its correlation is significantly different. (2) Under the same resolution image, spectral information+texture features have the best precision in estimating AGB, followed by a single texture feature model, and a single spectral model has the worst performance. (3) As digital images’ resolution increases, the accuracy of estimating AGB from spectrum information, texture information, and spectrum + texture information gradually improves.