| The study described in this paper is based on the Project of Remote-sensing Geological Comprehensive Survey of Key Metallogenic Belts and the Project of the Studies on the Remote-sensing Geological Survey Application Technology for Dangerous Areas and Complex Regions in the West of the China Geological Survey, MLR. The topics of study were carried out under the technical background that more and more satellite sensors have been put into service these days and the remotely sensed data sources have made great progress in both spectral and spatial resolutions; how to give play to the respective advantages of different image data sources, improve the service efficiency and accuracy of data in the "Remote sensing market" and satisfy the new demands of geological works for remote sensing techniques in the new age.Previous studies have shown that each kind of rock mineral has its own characteristic spectra, which are represented by special absorbing or reflectance signatures in remote sensing images. The occurrence, scale, rhythmic structure and other characteristics of rock units form their unique spatial structural features, which are shown as texture features in the remote sensing images. Remote-sensing geology identifies their categories just based on the unique spectral features and texture features of rock units in their images, and thereby reflects the spatial distribution ranges and the abundance of rock minerals. It plays an important role in geological work such as lithologic classification and mineral survey. With the expanding of sensor response spectral domains and spatial resolutions, remote sensing images are providing more and more information. Fully exploiting and using the information is one of the important contents of remote-sensing geology studies. Higher requirements are presented on identification and classification of lithological units, determination of contact relations of geologic bodies and ore-forming information extraction using remote sensing. The limitation of single remote-sensing data source indicated in practical applications is prominent and it will persist for a long time. Synergistic use of multiple remote-sensing data is a new trend of excavating remote-sensing data and geoscience information under the Big Data tidal wave.The research ideas of this paper are as follows:analyze physical properties of different remote-sensing data sources and solve the problem of geometric registration before collaborative application of different data and the problem of their different scales of radiometric resolution;The spectral characteristics of rock minerals and lithological association characteristics of rock layers determine the spectral and structural characteristics of remote-sensing images. Texture information of high resolution images is their advantageous information. The shortwave infrared information of medium resolution images is the core and critical technology of the existence and the promotion of remote-sensing geology in the early stage. The study on imaging mechanism is the foundation laid for the data collaborative framework. Guided by the collaborative theory, the collaborative application framework and the application model for multi-source remote-sensing data are proposed based on the physical characteristics of the data itself and the model proposed has been tested and improved in practice.The core of studies in this paper is the practice of the conversion technology of geoscience knowledge and the collaborative model with a framework under the basis of the multiple remote-sensing data synergy. The overall objective is to use the multiple remote-sensing data to provide supporting technology for the regional geological survey and the mineral resources survey in the western complex and dangerous regions comprehensively.South Tianshan and Tarim regions with harsh natural conditions, high altitude, steep terrain and extremely inconvenient traffic conditions, belong to complex and dangerous regions in the west, where the conventional geological survey can hardly be carried out. The remote sensing technology can cross over obstructions to obtain macro, abundant and accurate information. In this paper, these regions are taken as examples to carry out studies on the similarity and difference of remote-sensing data of different types and different resolutions for the rocks, spectra and formation taking the high spatial resolution WV-2 images, Hyperspectral Hyperion images and medium resolution Landsat8 images as the main data sources to form the theory and method of remote-sensing collaborative application with different data sources and propose rock layer classification programs with remote-sensing data of different types and different resolutions so as to provide examples for remote-sensing geological survey and ore-prospecting in the dangerous and complex regions in the west.The full text is divided into eight chapters with the main contents and chapters as follows: Chapter I introduces the background and significance of the studies, summing up the progress in remote-sensing lithologic classification and relative studies on the multi-source remote-sensing data collaborative application research areas, current research findings, existing problems and future development trend, broadening the perspective of follow-up studies; Chapter II describes the profile of physical geographical and geological settings of Wuqia study area and Yingjisha study area and introduces the pretreatment process for various data sources based on their characteristics. Due to the differences in sensor performance, imaging time, imaging angle and other factors of various data sources, there are differences in different data sources for the same surface reflectance when carrying out multi-source data collaborative application. This chapter mainly introduces the study on the relative correction method for multi-source data reflectivity; Chapter Ⅲ introduces the study on the image conformation mechanism of the lithological units. After summing up and analyzing the spectra generation mechanism of the rock minerals, it discusses the absolute control of composition and structural features of the lithological units on the spectral signatures and texture features of the remote sensing images; Chapter IV carries out in-depth analysis and compares detection capabilities of different data sources (spectra, space and radiation) and establishes the multi-source remote-sensing data collaborative application framework and application model preliminarily based on the synergetic theory; Chapter V introduces the study on the "1+1" collaborative model of multi-source remote-sensing data and example of lithologic classification, synergestic methods such as wave band superposition, wave band superposition texture, gap inserting superposition and wave band optimization superposition are proposed to carry out experimental studies on lithologic classification; Chapter VI introduces the study on the examples of the multi-source remote-sensing data fusion synergestic lithologic classification model, proposing the pixel level fusion synergy and the feature level fusion synergy. In different levels of fusion, the scale is always an important factor affecting image quality of synergy; Chapter Ⅶ introduces the collaborative model based on the multi-classification method. Under the multi-classification method collaborative model, it discusses the application method of the multi-source remote-sensing data of SAM and object oriented classifier synergy taking WV-2 and Hyperion images as examples. It introduces the study on the neural network based multi-source remote-sensing data reorganization model taking WV-2 and Landsat8 images as examples; Chapter Ⅷ is the conclusion and outlook. The contents from Chapter Ⅲ to Chapter VII are the focus and core of this paper, belonging to independent research findings of the author. The main academic achievements are as follows:1. The reflectance relative correction method is proposed for the differences in reflectivity that occurred for the same ground object caused by sensor performance, imaging time, imaging angle and other factors when carrying out multi-source data imaging.(1) For the characteristics of WV-2 image and hyperspectral Hyperion images, some typical rock samples are taken from Wuqia study area to count their average spectral curve characteristics. Through comparative analysis, the blue-violet light wave ranges of the two types of data sources are rejected, and finally No.2, No.5 and No.7 of WV-2 wave band, and No.13, No.31 and No.48 of Hyperion wave band are selected to calculate the average reflectance of the typical rock samples. The reflection matching factor of Wuqia study area is finally determined using the reflection factor ratio of the typical rock samples;(2) For the characteristics of WV-2 images and Landsat8 images, some typical rock samples are taken from Yingjisha study area and their average spectral curve characteristics are counted. Through comparative analysis, the average reflectance of the typical rock samples is calculated by selecting No.2 and No.5 of WV-2 wave range as well as No.2 and No.4 of Landsat8 wave range with close wavelength ranges. The reflectance matching factor for Yingjisha study area is finally determined using the reflectance ratio of the typical rock samples.2. It focuses on discussing the relevant theory and method of the multi-source remote-sensing data synergy for the purpose of the remote-sensing geology field-oriented application. A basic framework and general research method are preliminarily established for multi-source remote-sensing data synergy.(1) The advantages and disadvantages of WV-2, Hyperion and Landsat8 remote-sensing data sources on spectral, spatial, radiation detection capabilities are analyzed and compared, giving huge application demands for multi-source remote-sensing data synergy.(2) Under the guidance of the synergestic theory combined with its application in the fields of remote sensing technology and remote-sensing geology, the basic theoretical system of multi-source remote-sensing data synergy is preliminarily established.(3) The multi-source remote-sensing data synergy models can be divided into three levels, which are the "1+1" synergy model, the fusion synergy model and the classification method based synergy model from low to high. The "1+1" synergy model includes the synergestic methods of wave range simple superposition, wave range superposition texture, wave range gap inserting superposition and wave range optimization superposition; The fusion synergy model is divided into the synergestic methods of pixel level fusion synergy and feature level fusion synergy; The classification method based synergy model is divided into the single classification method and the multi-source data synergy for the joint application of the multi-classification method.3. The synergestic method under the "1+1" synergy model is practiced and the quality of collaborative data is evaluated taking the accuracy of lithologic classification as the standard.(1) Taking WV-2 and Landsat8 as collaborative data sources, the wave range simple superposition, wave range superposition texture, wave range gap inserting and wave range optimization synergestic methods are adopted. Lithologic classification tests are carried out based on various collaborative data and original data of Yingjisha study area. The results show that the lithologic classification of collaborative data has obtained a higher accuracy.(2) WV-2 and Hyperion are collaborative data sources. The wave range optimization superposition synergestic method is adopted. Lithologic classification tests based on collaborative data and various original data are carried out. The results indicate that a high classification accuracy is obtained under the conditions of reducing the dimensions of hyper-spectral data, the amount of data and data processing time effectively.4. The synergestic method under the fusion synergy model is practiced and the quality of collaborative data is evaluated taking the accuracy of lithologic classification as the standard.Learning from the idea of the homologous data fusion and transplanting its theory and method to the heterologous remote-sensing data sources, the multi-source remote-sensing data fusion synergy is divided into the pixel level fusion synergy and the feature level fusion synergy.(1) WV-2 and Hyperion are collaborative data sources. The pixel level based G-S fusion method is adopted to study the resolution ratio of using the multi-source remote-sensing data fusion and the method of panchromatic image construction. Taking the lithologic classification accuracy of Wuqia study area as the assessment criteria, the classification accuracies of collaborative data are studied with resolution ratios of 1:2,1:3,1:4,1:6 and 1:15. It is proposed that the optimum resolution ratio fitting for lithologic classification in the study area is 1:3 and the optimal spatial scale is 10 meters for the classification.(2) Taking WV-2 and Landsat8 as collaborative data sources, the wavelet based multi-scale decomposition multi-features fusion method is adopted. Extracting variance texture features of WV-2 panchromatic images to carry out feature level fusion with Landsat8 multi-spectral data, using the Daubechies wavelets which have the function of extracting the features of edges, and complementary to the texture features of variance. Taking the lithologic classification accuracy of Yingjisha study area as an assessment criteria, the classification accuracy of collaborative data with decomposition scales of 2,3,4,5,6,7 and 8 is studied, and the optimal decomposition scale for lithologic classification in the study area is proposed as 6.5. The multi-classification method is practiced combined with the synergy models. The lithologic classification accuracy is taken as a standard to evaluate the quality of collaborative data.(1) Hyperion and WV-2 are collaborative data sources, and the object oriented classification method synergy is based to carry out classification for the lithology studied using both data sources. Using the outstanding spectrographic detection capability of the Hyperion Hyperspectral data to carry out preliminary classification for the study area and obtain samples of different types of lithology from the study area. The WV-2 high spatial resolution image of the study area is split using the object oriented classifier to obtain the minimum basic units of the follow-up classification. The lithologic classification of the study area is completed based on Hyperion extraction samples, putting Hyperion SWIR range, WV-2 multi-spectral and panchromatic wave ranges into the nearest neighboring classifiers.(2) Taking Landsat8 and WV-2 as collaborative data sources and using the neural network classification method to carry out lithologic classification for Landsat8 data and WV-2 data respectively. The results of classification indicate that the spectral information of Landsat8 data has high impact on the accuracy of lithologic classification, and the spatial information of VW-2 data also has some impact on the lithologic classification. Therefore, the synergestic principle using the collaborative spectral information of Landsat8 data and the spatial information of WV-2 data as the main part assisted by the spectral information of WV-2 data is adopted finally to reconstruct the collaborative data and complete iithologic classification for the study area. The test results indicate that when the ratios between the spectral information of Landsat8 data, the spatial information of WV-2 data and the spectral information of WV-2 data of the study area are 5:4:1, the accuracy of lithologic classification for the study area is high.The innovations of this paper are as follows:(1) The basic framework theory, the collaborative models and methods for multi-source remote-sensing data synergy are established, preliminarily solving the problem of locating for freshwater in the ocean of data;(2) On the cooperative mechanism, the two key problems of inconsistent radiometric resolution and large differences in spatial resolution in data collaborative application are solved;(3) Three models, namely the superposition collaborative model, the fusion collaborative model and the multi-classification method joint collaborative model are proposed. The superposition collaborative model has 4 collaborative methods, namely wave range simple superposition, wave range superposition texture, and wave range gap inserting and wave range optimization methods. Empirical evidence indicates that the thematic classification effect of collaborative data is significantly improved;(4) The new classification method of combining SAM with the object oriented classification method is proposed. The key technology is sample selection for the high resolution data which is based on the spectral angle classification result. The SWIR data from the Hyperspectral data is utilized when establishing and optimizing the feature space.The remote-sensing lithologic classification is a cross-disciplinary research field covering multidisciplinary theories, methods and techniques of geology, petrology & spectroscopy, image processing etc., with interdisciplinary and forward-looking features. In practical situations, there are still a number of theoretical problems and critical technologies which need further studies, especially the mechanism of increasing the recognition capability of collaborative remote-sensing data, i.e., how is the synergistic effect generated, needs further in-depth studies. |
PDF Full Text Request