| According to application requirements of domestic GF-1 satellite data Agricultural project, we do research on the computing method of corn area with the domestic high resolution remote sensing satellite data, providing a technological support for the Agricultural remote sensing monitoring and evaluation system at the national level.The GF-1 satellite image has the richer structure information and the texture information, the phenomena of ‘the foreign matter same spectrum’ and ‘the same thing different spectrum’ is serious. The classical pixel-wise remote sensing image classification technique is poor in the processing efficiency and classification accuracy for the corn crop classification. In this paper, based on the object-based information extraction technology, combined the automatic classification and the man-computer interactive interpretation, researched a image segment algorithm which is suitable for GF-1 satellite remote sensing images, extracted the corn crop effectually, then finished statistical work for corn field area. Researched on the test-bed In Lantian county, Shanxi province, and jobs be done as following:(1)Analysis the characteristics of GF-1 satellite data, then accomplish Radiometric Correction, FLAASH Atmospheric Correction and Geometric Correction; fuse the GF-1 Low Spatial Resolution Multi Band and High Spatial Resolution Pan band to Enhance image clarity and interpretability; crop the image for the main cultivated land area, which blocks vegetation that can be confused with corn crop on the non-cultivated land. The above provides basic data for the statistical methods of corn crop area.(2)Research the segmentation technique to form homogeneous image objects on the basis of object-based image classification. Through the analysis of the irregular structure features on the corn field’s edge, enhance the contrast of edge areas by building a fuzzy distribution curve, then introduce the near-rectangle guided threshold function into the graph-based segmentation algorithm. Complete the segmentation of corn field remote sensing image effectively.(3)The theory of sparse representation is introduced to regard the statistic of corn field area as the classification of Corn class with the others. Simultaneously model corn class and the others to the sample data and the over-segmentation superpixels. Determine the category of segments using sparse representation of the segment over the superpixels. Since the proposed algorithm does not require to Set feature attribute values, it has the strong robustness. This can efficiently extracts the corn crop.(4)For extracting the corn class effectively and accurately, I select some discrepant features to finish the image classification using the model proposed in this paper, at last make a comprehensive multi-feature decision about each feature classification result. Evaluate the effect of this method through comparing and experimenting with other method.(5)The system software is designed and developed for the corn crop classification in VC++ platform combining with Opencv. The modular design benefited the extending of follow-on function has been got with modular design. |
