Remote Sensing Monitoring Of Soil Salinization Based On Fusion And Classification Of Radar And TM Image

Soil salinization not only causes the destruction of resources and immense decline of agricultural productivity, but also threatening the biosphere and the constitution of ecological environment. As a primary form of soil degradation, salinization has already become the global environmental problem.Fusion of multi-sensor remote sensing images could improve the analyzing, understanding and target recognizing ability of image. Though Synthetic Aperture Radar images have much superiority such as day and nigh imaging, all wehather imaging, high spatial resolution etc, its efficiency is restricted by the effect of its strong speckle niose, whereas TM images have abundant spectral information, which is more applicable to many practical applications like classification and change detection. Therefore, the fusion of SAR and visible spectral image can utilize its mutual complementary information, can acquire earth's multilayer characteristics, and can unveil its intrinsical characteristics.There are many classification methods of remote sensing images, but there is not any method which is universally applicable and efficient, this is because of the complexity of Remote Sensing itself. It is extremely important that how to extract more information, how to improve classification efficacy and classification accuracy by utilizing the characteristics of SAR image more efficiently.This thesis carried through fusion for SAR (Radarsat image) image and visible spectrum remote sensing image (TM image) of the Delta Oasis of Weigan and Kuqa Rivers area. Adopted new classification method of remote sensing image—Support Vector Machines (SVM). TM, SAR and fusion image of SAR and TM are carried out monitoring classification of salinization information respectively, and then analyzed and evaluated fusion result and fusion precision quantitatively. Lastly, research carried on the typical application problem of monitoring of soil salinization information in arid area by fusion and classification of Radar remote sensing data and visible spectrum data.Main research contents, results and initiatives in this thesis include as following:1. SAR and TM images are carried on several common fusion processing, including PCA, HIS, Brovey, Gram-Schmidt and wavelet transforming fusion.2. The qualitative and quantitative evaluation of the fusing result indicates that: the image of wavelet transforming fusion gained enhancement in both image's spectrum information maintenance and increasing image's spatial detailed information. Comparing with other typical fusion methods, wavelet transforming fusion is more applicable to SAR and TM images of the studying area. Furthermore it provided effective fusion information for the future monitoring of salinization.3. This paper adopted a new kind of remote sensing classification method—SVM—which is based on statistical learning theory, and explicated the advantages of SVM classification method in comparison with those of conventional methods. Then Cross-Validation method which is based on statistical learning is introduced for confirming the best classification parameter.4. The best parameter which is obtained by Cross-Validation method used for training whole training samples, and attained SVM classification model. TM, SAR and wavelet transforming fusion image of SAR and TM are classified respectively by the SVM classification model. For validating SVM classification's performance and efficiency, this article also adopted conventional classification methods like minimum distance classification and maximum likelihood classification. Finally the classification results of each image carried accuracy evaluation and comparatively analyzed qualitatively and quantitatively.5. The analyzing of classification results indicates that:①When using SVM classification, TM image's extraction precision of each class is increased in different degree in comparison with those of minimum distance classification and maximum likelihood classification. Classification overall accuracy is reached 88.52%, while the classification overall accuracy of maximum likelihood classification and minimum distance classification is reached 87.04% and 84.36% respectively. Mixing, missing samples phenomenon of high, medium, lower salinized soil is reduced remarkably, achieved somewhat reasonable classification result.②Although the spatial resolution of SAR image is high, its SVM classification accuracy is rather low (60.81%). The traditional classification's classification accuracy (low than 60%) of SAR image is lower than that of SVM.③The classification result of wavelet transforming fusion image of SAR and TM is relatively ideal, it almost attained preferable classification result. The SVM classification accuracy of SAR and TM fusion image is higher than that of single TM image; overall accuracy is increased to 91.08% from 88.52%. The extraction accuracy of salinized soil had certain improvement as a whole; particularly the improvement of medium salinized lands is more remarkable.To sum up, this research carried on general evaluation on the application of Remote Sensing monitoring of soil salinization based on fusion and classification of Radar and TM image. The result indicates that the image which is based on wavelet transforming fusion obtained higher distinguishing ability, difference between salinized land and other ground objects widened greatly, its result, in comparison with optical remote sensing image, got improvement in information extraction accuracy. It indicates that SAR image has its predominance in extraction of salinized soil information. This thesis utilized Radar remote sensing image as a main data source, integratedly used superiority of other remote sensing data and spatial data, enhanced monitoring accuracy of salinized soil information, accordingly explored and evaluated salinized soil of arid and semi-arid zone in many prospects, developed application of microwave remote sensing technics in extraction of soil salinization special information. Subsequently it provided one more efficient technical means for remote sensing monitoring of soil salinization in arid area.