Study On Extraction Of Mango Forest With High Resolution Remote Sensing Image

In recent years,the mango planting industry in China had been developing rapidly,but the mango spatial distribution information also relied on the slow and inefficient artificial census,so it is urgent to study the remote sensing extraction method of mango forest.Abundant studies focused on forest or agriculture information extraction with remote sensing techniques,but studies of extracting mango forest were rare.Taking into consideration the cloudy and rainy areas often cause the absence of optical satellite remote sensing images and the characteristic of land use is fragmentation and structural dispersion in the south of China,we studied the extraction of mango forest based on GF-2 satellite imagery.In our study,we discussed the applicability of different image fusion methods and different image spatial resolution in the extraction of mango forest information.And we also discussed the way that combined spectral information and spectral index and texture feature parameters to extract mango forest information.(1)There was no image fusion method that was generally suitable for every terrain and all data sources.Every fusion method had its specific applicability and limitations.In view of this problem,we carried out image fusion research which was based on GF-2 data and applied to the extraction of mango forest.In our study,three fusion algorithms(PCA fusion,Gram-Schmidt fusion and NNDifuse fusion)were utilized to fuse the study area and a small area of mango forest.And the quality of fusion result was assessed by qualitative and quantitative evaluation methods.The results showed that all of the fusion method enhanced the image's spatial resolution,and the line spacing of mango forest could be well reflected in the fusion image.We used entropy,the average gradient and the average deviation as the quantitative evaluation index,and the study showed that: NND fusion algorithm was better than the other two methods in the index of the average gradient and entropy,although the mean deviation of NND fusion algorithm was high in the whole study area,but it had the smallest value in the small area of mango forest,which was beneficial to the identification of mango forest.Based on the analysis of the three methods.NND fusion algorithm was the most suitable algorithm for the extraction of mango forest.(2)Different feature extraction had unique optimal spatial resolutions.If the resolution was too small,it would result in mixed pixel and the too high resolution would make the details of Mango tree and soil background information too much.All of this would narrow the accuracy of Mango forest's extraction.In view of this problem,in our study,we considered the spacing characteristics of mango forest and the resolution characteristics of high resolution satellite in China.Based on the 1m resolution image,we got 2 m,4 m and 8 m resolution images.The results showed that the spatial range of mango forest had little change at 1 m and 2 m spatial resolution images.Under the spatial resolution of 2 m,the extraction of mango forest information had the highest accuracy.Under the spatial resolution of 4 m,extraction accuracy of mango forest was reduced,and cultivate land was the main factor to which lower the extraction precision of mango forest.Under the spatial resolution of 8m,mango forest and the secondary forest and cultivate land were mixed serious.Therefore,when we only use the spectral information,the 2-meter spatial resolution is the best resolution for the extraction of mango forest spatial information.(3)The problem that it was difficult to distinguish surface features by remote sensing classification which was caused by “same object with different spectra” and “same spectra with different object”.In our study,we introduced the spectral index and texture features for the extraction of mango forest.The study showed that,we could not get high accuracy to extract the information of mango forest by using the spectral information only,the vegetation index could improve the extraction accuracy of mango forest,but the overall classification accuracy was low.Compared with spectral information,texture information was regarded as better factors to extract mango forest.Compared with the classification which only relied on spectral information,the producer's accuracy and the user's accuracy of mango obtained from the classification which integrated with texture information increased 17.04% and 7.08% respectively.In texture information,contrast is the most important information for distinguishing between mango forests and other vegetation.The results also showed that the spatial distribution of mango forest extracted by combination of spectral,texture and vegetation index had the highest accuracy,with the producer's accuracy was 89.29%,and the user's accuracy was 97.40%.