Extraction Of Tea Garden With High Resolution Remote Sensing Image Combined With Spectral And Texture Information

The spectral characteristics of tea garden are very complex.They often overlap with other woody vegetation and form confusion.An important means to overcome spectral confusion in tea gardens is to introduce texture features based on spatial information.Due to the complexity of texture features,it is still not possible to determine the goodness or the badness of a texture extraction and construction method at present.Usually,the most suitable texture feature construction method is given for different situations.High-resolution remote sensing images provide richer and clearer representations of ground information,and they have great advantages in extracting spatial information such as textures.Therefore,based on the high-resolution remote sensing images,this paper carries out the research on the tea garden extraction method combining the characteristics of the spectrum and texture.The main contents and conclusions are as follows:(1)Image feature construction of tea garden.? Optimal combination of spectral information.Based on the NDVI index and the MNDVI index,the DNDVI index is constructed to enhance the spectral information and eliminate the partial shadow at the same time.The best combination band is selected by the best index factor(OIF),and the main spectral information is preserved while the data is reduced.?The texture features suitable for WV-2 and GF-2 images are constructed by using the gray level co-occurrence matrix(GLCM),the Gabor filter and the local two value mode(LBP).In addition,a texture construction method,which combines LBP and Gabor,is proposed,and LBP GABOR texture is constructed to better express the tea garden information of the image.(2)Optimization of extraction methods for tea garden.Comparing the precision of five different spectral and texture feature combinations to extract tea gardens,the most suitable tea garden extraction method was found.The results show that in combination with spectral information and LBP GABOR texture(B0IF +DNDVI + LBP_GABOR)classification method,the overall accuracy and Kappa coefficient of WV-2 images can reach 92.34%and 0.843 respectively.The overall accuracy and Kappa coefficient of GF-2 images can reach 90.78%and 0.816 respectively,and the results are best.(3)Verification of regional applicability of the method.The data of GF-2 in the south of Changkeng Township(covering an area of 2600mx 1790m)and Lantian Township(coverage of 12992m×11067m)in Anxi County were selected for the more complex topographical features,and tea plantation was extracted to verify the effectiveness of the method.The overall precision and Kappa coefficient of the southern part of the county of Anxi county were 88.74%and 0.756 respectively.The overall accuracy and the Kappa coefficient of Lantian Township in Anxi county were 89.11%and 0.843 respectively,which proved that the method of this paper could be better suitable for the extraction of tea garden information.