Abstract:Using remote sensing technology to rapidly and accurately differentiate the seed maize fields and grain maize fields is the urgent need of seed production and market supervision, and also is an important aspect of the research on the classification and planting mode of crops by using remote sensing to monitor. Based on the spectral and texture differences of seed maize and other crops in the high resolution remote sensing image, the multi-source remote sensing data were used, including GF-1 WFV multi-spectral image, Landsat8 OLI image and GF-2 PMS full-color image to extract the seed maize fields as research target, the vegetation index system of crop multi-temporal spectral characteristics was proposed, which multidimensionally reflected different spectral differences between crops;and adding the image rotation invariant processing before the texture detection, to solve the problem of crop field texture direction in remote sensing image;finally, the identification method system of seed maize fields based on multi-temporal spectral feature and LBP-GLCM texture feature in high spatial resolution remote sensing image were established. Qitai County, Xinjiang Uygur Autonomous Region was taken as the study area to verify, based on the above method and the random forest classifier, the overall accuracy was 90.57%, the Kappa coefficient was 0.79. The accuracy of the classification results of seed maize field was 99.20%, and the mapping accuracy was 86.68%, which basically satisfied the needs of seed maize recognition requirements. The research result not only provided a method for the monitoring of hybrid maize seed production in China, but also provided a technical reference for monitoring and supervision of hybrid seed field with the same planting system.
