Abstract: Crop phenology period is an important feature of the agriculture eco-system. Using remote sensing technology to monitor crop phenology accurately and timely, which plays an important supporting role in effective assessment of crop growth trends, improving the information management level of agricultural conditions and providing technical support for precision agriculture. Normalized difference vegetation index (NDVI) can well describe the growth process of different types of vegetation, which is the most frequently used data in crop phenology. In this paper, NDVI is extracted from 8-day synthetic data of MODIS in 2016. The time series of NDVI is not continuous in time and space due to the influence of air pollution, it is necessary to smooth the remote sensing time series data which represent the vegetation growth process before the phenology study. Then according to the characteristics of vegetation growth curve, the phenology information is extracted after removing noise of time series data. Maximum value composite (MVC) is widely used in the initial de-noising process of NDVI because of its simple calculation and convenient use, but it is prone to large errors for continuous multi-day cloudy weather. An improved MVC is proposed for the preprocessing of MODIS NDVI time series data in this paper, which is very convenient and does not require additional parameters. The new NDVI time series data can be constructed by extracting NDVI from positive and reverse sequence on growth time series in a fixed interval and then synthesizing it. The reconstructed NDVI time series data are filtered by S-G filter to further eliminate the noise, and then the growth curve of summer maize is reconstructed by logistic function fitting. Finally, the jointing and maturity stages of summer maize are extracted by curvature, and the emergence and tasseling stages of summer maize are extracted by dynamic threshold. Compared with the observed results, the absolute errors of different phenology starting time of summer maize obtained by improved MVC method is less than that of traditional MVC method. Especially the emergence stage, the absolute error of improved MVC method is reduced by 4.5 d. The absolute errors of phenology on summer maize using improved MVC are 3.72, 5, 1.06 and 1.26 d at emergence, jointing, tasseling and maturity stages, respectively. The absolute errors of that by traditional MVC method in subsequent phenology periods aere 8.22, 5.72, 2.78 and 5 d, respectively. From the spatial distribution maps of different phenology periods of summer maize, it can be seen that the starting time of summer maize emergence stage in the study area is relatively concentrated, generally starting from 166 d to 170 d base on the day of year (DOY), namely 15 June to 18 June. Jointing stage generally starts from 185 d to 190 d, namely 3 July to 8 July. Most of the areas in the study area enter the stage of tasseling from 208 d to 212 d, that is, from 25 July to 29 July. A large area of summer maize begins to enter the maturity stage from around 251 d, that is, on September 7. In spatial distribution, the northwest of the canal head in the study area is slightly ahead of the southeast on phenology period. It can be said that using improved MVC to extract NDVI time series data of crops can effectively remove the impact of continuous cloud and fog on vegetation index, improving the accuracy of monitoring crop phenology, providing support for precision agriculture.