10.18170/DVN/IKS9PZ慕, 号伟号伟慕0000-0002-9465-822X兰州交通大学周, 亮亮周0000-0003-3968-1897兰州交通大学Peking University Open Research Data Platform2019Earth and Environmental SciencesCrop estimation; Remote sensing; Convolutional neural network; Winter wheat; Northern winter wheat area慕, 号伟兰州交通大学2019-03-162019-08-2819149986application/zip1.0CC0 WaiverThe experiment constructed a set of winter wheat yield prediction model data based on CNN. Based on the histogram dimensionality reduction and data normalization method, the 12-band MODIS images of 2006-2016 were processed into 36*36*12 data samples. Convolutional neural network input layer and winter wheat yield data from prefecture-level cities in Henan, Hebei, Shandong, Shanxi, and Shaanxi are used as output layers to achieve a rapid and accurate prediction of large-scale winter wheat.The test selected MODIS image data of h26v04, h26v05, h27v04, h27v05 from 2006 to 2016, a total of 6376 scenes, of which MOD09A1 and MYD11A2 are 1584 scenes, MOD15A2H is 1580 scenes, MOD13A1, MYD13A1 are 792 scenes, MCD12Q1 is 44 scenes. The MOD15A2H image was missing on the 49th day of 2016. The mean value of the images on the 41st and 57th day was supplemented, and the MOD13A1 and MYD13A1 images were fused in time series to ensure the integrity of the sequence. Shell and GDAL are used for batch processing, and the projection of MODIS data is converted into UTM projection based on WGS-84 ellipsoid. At the same time, multiple types of images are extracted by the band, spliced, and then fused into 21600 sheets containing 12 bands. image. At the same time, the statistical yearbooks of Henan, Shandong, Shanxi, and Shaanxi from 2007 to 2017 and the winter wheat yield data of prefecture-level cities in Hebei's rural statistical yearbook were collected. On the basis of remote sensing data processing, combined with histogram dimensionality reduction and data normalization, the winter wheat yield prediction model data based on a convolutional neural network was constructed.