Hyperspectral target detection has irreplaceable advantages which can identify meticulous spectral characteristic difference between the target and background. At present, the rapid development of high spaial and hyperspectral remote sensing provides more data for target detection. Target detection accuracy is closely related to target size、spatial resolution、spectral resolution、band settings、spectral characteristics of the target and background, detection algorithms. Generally, with regard to specifically target and background, the finer the resolution is, the higher the accuracy will be. Yet the spatial and spectral resolution can barely meet the need simultaneously, for technical obstacles. A tradeoff between the two factors is needed for effective target detection and the choice of appropriate remotely sensed data for target detection has always been concentrated nowadays.
To address this concern, this paper researched on target detection mechanism, evaluated the suitability of different detection algorithms. On this basis, this paper using ground hyperspectral sensor FISS (Field Imaging Spectrometer System) data and aerial hyperspectral AVIRIS data，studied on the situations where the target was small and had a similar spectrum to the homogenously distributed or complex backgrounds. Through the down sampling processing of the high spatial hyperspectral images, the paper analyzed the relationship between the spatial and spectral resolution and the detection accuracy. And then it proposed the optimal spatial and spectral scale for target detection. This study focused on the quantification of the scale impact of spectral and spatial resolution on target detection precision. Results revealed that：
(i) RXD algorithm is applicable to small and outstanding targets, anomaly detection algorithm has a higher false alarm rate comparing to other algorithms which have prior knowledges; CEM can achieve high detection accuracy in the case of small targets; ACE algorithm can be used to homogenously distributed background which can be represented by multivariate normal distribution；OSP algorithm is sensitive to the input spectrals of the background, the results may be affected in the complex background.
(ii) With the decline of spatial resolution, the pure target to be detected turned into sub-pixel, the detection accuracy experienced three stages of descending rates: gently-dramatically-gently. The radio of the target size and the spatial resolution has positive correlation to detection accuracy. The corresponding spatial resolution before the second stage is the effective scale for detection. In the ground experiment, the required spatial resolution for camouflaged detection was about within twice the size of the target. Using AVIRIS data, the required spatial resolution for aircraft detection was also about twice the size of the target.
(iii) Suitable spectral scale is related to the spectral difference between the target and the background (significant reflection peaks distance). In the ground experiment, the reflection peaks differences, associated with the target and the background, were 20nm apart. When the spectral resolution was coarser than 40 nm, the differences of reflection peaks disappeared and the detection accuracy decreased. Using AVIRIS data, since the reflection peaks differences were 200 nm apart, the detection accuracy was essentially the same when the spectral resolution ranging from 10-60 nm.
(iv) In the appropriate range of spectral scales, the spatial resolution plays an more important role in detection accuracy. In the ground experiment, the detection accuracy changed little when the spatial resolution remained the same and the spectral resolution varied from 10 to 40 nm. However, when the the spatial resolution decreased, the accuracy had a clearly decline. The aircraft detection had the same result.
(v) To select the quite different characteristic bands or abandon the similar bands from the target and background spectral can improve the detection accuracy. Refer to the multispectral sensors, the accuracy increased when we got rid of similar bands(430 nm) in the ground experiment, and the optimal combination bands were 450-510 nm, 510-580 nm, 585-625 nm, 630-690 nm, 705-745 nm. In the aircraft detection experiment, the optimal combination bands were 450-510 nm, 510-580 nm, 630-690 nm, 770-895 nm, 2235-2285 nm, 2305-2365 nm, according to the OIF(Optimum Index Factor). In addition, removing 1200-1700 nm bands can improve the detection accuracy due to the variability of the target.
It was concluded that the quantitative analysis method and results of spatial and spectral scales for target detection would be of great significance for both data source selection and studing on other target-background combinations under similar conditions.
With the continuous development of mineral exploration technology, the surface mineral exploration work became more and more difficult. The main direction of the mineral investigation is toward the underground. Core as the final link in mineral exploration keep the records of vertical change geological information. Without undermining core surface and destroying the core integrity, remote sensing as a new technology have macroscopical, fast and informative advantages. It is an indispensable means of geological prospecting. Hyperspectral remote sensing technology has the ability to identify the different minerals and composition according to the different minerals have different spectrum feature. Geological experts can use remote sensing geological information to analysis the geological metallogenic conditions, find prospecting areas and delineate target goals which providing good information to carry out detailed geological work.
The goal of this paper is design the core imaging spectrometer data catalog system including the interface and system function design. Firstly, in the condition of full consideration of the needs of users, the core catalog need stable, efficient, accurate, etc. This paper make a detail design on the core imaging spectrometer catalog system. Also this paper explore the spectral characteristics of rocks and minerals and study the key technologies of mineral identification. Then this paper work on the detailed design of the core imaging spectrometer system catalog fully considering three aspects: user habits, convenient and elegant interface master the software. The main results and conclusions are as follows:
1. Analyzed and summarized the common spectral characteristics of altered minerals. On the knowledge of various mineral spectral information, this paper studied the hyperspectral data extraction methods, such as the minimum distance matching, spectral angle matching, spectral absorption index etc.; Evaluated the spectral quality of core imaging spectral data using these methods, and investigated the accuracy of spectral matching technique. Quantified the spectral characteristics of minerals and used the mathematical methods in calculations to achieve unknown targets. Solved the core mineral identification problem, rapidly realized the translation and analysis of core information.
2. Based on the analysis and summarize the spectral characteristics of common alteration minerals, this paper explore the spectral information enhancement and various mineral extraction methods, such as spectral derivative, envelope removal, the minimum distance matching, spectral angle matching and spectral absorption index. Using those methods to make comprehensive statistical analysis of imaging spectrometer data core, the spectral characteristics can be quantified with the method of mathematical calculations to achieve the goal of determining the unknown targets. It solve the identified problems of the mineral core and achieve quick deciphering and analysis.
3. This paper amply design the interface and function modules of the core imaging spectrometer catalog system. The interface is divided into six part menu bar, toolbar, drilling display window, core catalog window, mathematical analysis window and parameter setting window. And design detailed function of each part to meet the basic need of each catalog functions. Also the interface strive to reasonably arrange space and make beautiful interface. From the user point of view the function module was design to automatically catalog to achieve the function of import of data, automatic cutting, automatic interpretation and editing and other geological data.
The retrieval of quantitative information of minerals has strong economic and scientific meanings, and is essential for the long term development of human beings. Hyperspectral remote sensing has unique advantage and great potential in quantitative analysis of minerals, but has not been fully exploited in practical applications. However, the primary cause is due to the uncertain factors in quantitative analysis of minerals using hyperspectral remote sensing, which includes the complex mechanism of mineral spectra, the accuracy of spectral unmixing, and the accuract extraction of mineral absorption features. To address this issue, this paper firstly systematically analyzed the mechanism of mineral spectra, and then investigated the uncertain factors in quantitative retrieval algorithms, which can be generally grouped into two classes: spectral unmixing and absorption feature extraction. As for spectral unmixing, the effects of spectral unmixing models and band position on mineral retrival accuracy were fully investigated. In the field of absorption feature extraction, a new absorption feature extraction method was proposed to retrieve more accurate absorption parameters. This dissertation will further serve as a guide for improved quantitative retrievals of minerals from spaceborne, airborne, or other hyperspectral remote sensing platforms, and pave the way to explore some important subjects using remotely sensed mineral contents, including the metallogenic prognosis, planetary geology evolution, mine environmental restoration, etc. The main conlusions and results are as follows.
(1) The mechanism of mineral spectra were systematically analyzed from five different aspects, including the mineral types, the chemical change, the physical propoties, spectral measurement conditions, and spectral mixing effect, and discussed about the correlations between these factors. Among them, spectral mixing effect is the main source and theory basis of quantitative analysis of minerals using hyperspectral remote sensing data.
(2) This dissertation summarized the concept of mixing reflectance reconstruction (MRR), and proposed a thorough method to determine the accuracies of spectral unmixing models based on MRR. The MRR error can be analyzed from 3 different dimensions, including the spectral dimension, the spatial dimension, and the total dimension. By measuring the spectra of proportionally mixed mineral powders, we were able to investigate and verify the relationship between MRR accuracy and spectral unmixing accuracy. This finding validates that when the actual fractions are not available, it is possible to estimate the spectral unmixing accuracy based on an MRR accuracy analysis.
(3) A newly developed LCR model was developed based on investigating the existing spectral unmixing algorithms and physical mechanism analysis. Existing models having typical applications related to mineral analysis, including the Linear model, the NL model, the CR model and the SH model, were also summarized. Experiments on the well-known AVIRIS data for Cuprite allowed us to evaluate the effects of the five unmixing models based on the MRR accuracy analysis. The results revealed that the LCR model yielded good results in nearly all aspects, and had considerable potential for practical application. By comparing the total MRR error of ATREM dataset and Flat Field dataset, the effects of atmospheric correction on spectral unmixing accuracy were verified, but the level of influence was different for different unmixing models. LCR model achieved the most outstanding results, which verified its great robustness.
(4) This dissertation proposed a new spectral unmixing model based on derivative of ratio spectroscopy (DRS), which can directly interpret the correspondence between the target content and variations of mixing spectra, eliminate the influence of other substances in the mixture, and extract the bands which are more sensitive to the target information. Based on this model, the effects of spectral position on spectral unmixing accuracy were investigated using mineral powder mixtures. The results indicated that the absorption features have strong influence on the spectral unmixing accuracy, and the bands near the slopes of absorption feature valleys tend to have higher accuracy.
(5) Based on the study of the mechanism of background removal methods, represented by continuum removal, a new spectral fitting method was presented to obtain the background curve, and a novel background removal method named reference spectral background removal (RSBR) was given. RSBR retains the advantages of continuum removal, and when given the reference spectral background, RSBR can eliminate the influence of unwanted contribution factor, and extract the absorption feature of target contribution factor.
(6) Based experiments on both mineral powder mixtures and airborne hyperspectral data, RSBR was demonstrated to have the following advantages in extraction of absorption feature parameters: 1) RSBR can extract accurate absorption centers and absorption widths from mixing spectrum, independent of the variation in abundance; 2) absorption depths calculated from the RSBR spectra are strongly linearly correlated to the fractions of the component of interest; 3) the spectral waveform of the specific absorption factor can be well extracted by RSBR, and by using spectral matching methods, such as SAM, the mineral composition can be identified.
The physiological and biochemical parameters of vegetation has been considered as significant input parameters in the models of ecological process, including carbon cycling, nutrient cycling and rainfall interception. Due to its unique advantages, the technology of Hyperspectral remote sensing has been applied widely in the area of predicting vegetation physiological and biochemical parameters. However, it is extremely hard to obtain the image data simultaneous with high spatial, high time and high spectral resolution because the limitation of hyperspectral satellite sensors. Therefore, the issue about uncertainty in the different scale of space, time and spectrum was proposed when retrieving vegetation physiological and biochemical parameters quantitatively.
To address this issue, based on simulated and measured spectrum data, this paper was focused on the uncertainty issue caused by spectral scale and spectral mixture when estimating chlorophyll content (Ca+b) and leaf area index (LAI) of winter wheat in different growth stages. The main methods we used contains physical models and statistical analysis models of empirical/semi-empirical. Meantime, a new spectral reconstruction algorithm derived from UPDM principle was proposed in order to weaken the uncertainty, which resulted from spectral scale among different remote sensors. And then the research about uncertainty of spectral scale was further discussed in time scale. Main conclusions and results can be deduced as follows.
(1) The models for inversing physiological and biochemical parameters and the generated mechanism of uncertainty.
The models and the mechanism of uncertainty in the estimation of vegetation physiological and biochemical parameters were thoroughly reviewed and analyzed. Statistical analysis models including derivative spectrum, spectral indices, variables of spectral positions and continuum removed method, and physical models containing PROSPECT leaf model, SAIL canopy model and PROSAIL coupling model, which were used to predict chlorophyll content and leaf area index of winter wheat. These models contribute greatly to the retrieve of vegetation physiological and biochemical parameters by hyperspectral technique and also provide the basis theory supports for its remote estimates.
(2) The research on uncertainty issue caused by spectral scale in the retrieve of vegetation physiological and biochemical parameters based on radiative transfer theories.
On the basis of simulated spectrum data from vegetation radiative transfer models, traditional spectral indices and new pattern index VIUPD were selected and further compared with their sensitivities to Ca+b and LAI and uncertainty of spectral scale. It is demonstrated that index VIUPD stand first because of its high precision, stability and slight effect by spectral scale, can be viewed as the effective model in the estimation of Ca+b and LAI. In addition, followed by VIUPD, the spectral indices built by vegetation red-edge information has a comparatively good performance. This conclusion can provide the theoretical foundation to the study of improving and developing new kinds of spectral indices in the applications of inversing vegetation physiological and biochemical parameters.
(3) The research on uncertainty issue caused by spectral scale in the retrieve of physiological and biochemical parameters of winter wheat based on single-temporal measured spectrum.
The spectral indices method was introduced to analyze the uncertainty of spectral scale produced in the estimation of Ca+b and LAI of winter wheat based on various satellite sensors. Results indicate that there are lots of uncertainties of every spectral index for their retrieve performance, due to the difference of spectral response characteristics of different remote sensors. The results we also found that spectra index VIUPD established by universal pattern decomposition method showed its high precision and best stability when predicting Ca+b and LAI.
(4) The research on algorithms about weakening uncertainty of spectral scale in the retrieve of physiological and biochemical parameters of winter wheat based on satellite remote images.
The algorithm of universal pattern decomposition method which was used to process multiple and hyperspectral data with its advantage of sensor-independent, was introduced to this paper. The simulated MODIS image generated from original HJ1A-CCD2 data based on spectral reconstruction principle using UPDM algorithms. And then the difference of spectral indices and the uncertainty of retrieve Ca+b exist in original HJ data, original MODIS and simulated MODIS were analyzed and compared. It is obvious that UPDM not only possess an excellent spectral scale adaptability, but also reduces the uncertainty caused by different setting of spectral channels among various remote sensors.
(5) The research on uncertainty issue caused by spectral scale in the retrieve of physiological and biochemical parameters of winter wheat based on multi-phase measured spectrum.
The methods including variables of derivative spectrum, spectral indices, variables of spectral positions and continuum removed method were selected and help to establish six kinds of spectral parameters (SP). In the entire and single growth periods of winter wheat, we devoted to analyze the sensitivities of SP to Ca+b and LAI, and discuss the spectral scale uncertainty which existed in spectral indices itself and the inversion of Ca+b and LAI. The results showed that chlorophyll index CIgreen, height of green peak (GH), absorption depth of red band (RD) and area of red edge (REA) were found to be slightly affected by spectral scale and can be considered as the effective index for inversing Ca+b and LAI of winter wheat. In addition, the situation of uncertainty also presented certain difference when taking the diverse growth period into consideration, which demonstrate that the generation of uncertainty issue were also affected by time scale.
(6) Based on the research about spectral scale uncertainty of inversing physiological and biochemical parameters of winter wheat, we can get another conclusion that the retrieve precision was not proportional to the spectral resolution. Certain noise could be found in the reflectance results due to the effects of environment, background, instruments and other possible factors when measuring hyperspectral data. Although the vegetation spectrum got from wide-band range would lost some details about its characteristics, it can avoid or partly eliminate the noise and redundant information among the narrow bands. Therefore, the selection of appropriate spectral scales and analyze methods must be taken into consideration in the estimation of vegetation physiological and biochemical parameters.
(7) The research on uncertainty issue caused by spectral scale in the retrieve of physiological and biochemical parameters of winter wheat based on simulated and measured spectrum.
In this part, VIUPD and GH which presented the best performance in the research about spectral scale uncertainty, were selected to explore their ability in the estimation of LAI of winter wheat with the influence of mixture spectral. Under the assumption that one pixel only contains vegetation and soil, we investigate the uncertainty issue of estimating LAI under the different linear mixture patterns. Results show that the addition of soil background into pure leaf vegetation resulted in the uncertainty when predicting LAI. It can be further found that the accuracy of LAI retrieve describes a generally declining trend with increase of the proportion of soil background, but not coming into a uniform rules. Meanwhile, the uncertainty generated from measured spectrum were greater than the simulated spectrum. And there are still some difference of uncertainty under various growth periods of winter wheat. Based on different satellite sensors, the uncertainty caused by spectrum mixing of vegetation and soil were found to be diverse in the estimation of LAI.
This study provides the basic theory for the inversion of vegetation physiological and biochemical parameters by hyperspectral remote sensing techniques and also contribute greatly to the development of hyperspectral sensors. While making the above achievements, there are some disadvantages as following:
(1) The spatial resolution of MODIS is quite different from HJ. Compared to HJ, MODIS data has too low spatial resolution. As a result, surface heterogeneity will generate significant effect on the spectral reflectance of pixel when examining the uncertainty issue of spectrum scale in the inversion of chlorophyll content.
(2) Only three growth periods including jointing, filling and milky stage were selected in this paper, the time nodes are less than normal. The further study can be conducted throughout the whole growing season and covered more growth stages. And a systematic study about the uncertainty of spectral scale would be executed under the more complete time sequence.
(3) In this work, the function of Gaussian spectral response was used to simulate the spectrum data of different spectral scale based on different Full Width Half Maximum (FWHM). However, the simulated data was not able to be fully accordant with the actual remote sensors. So there are still some limitation in the results we got. The further research we should pay close attention to more typical actual sensors in order to practically investigate the spectral scale uncertainties in the estimation of vegetation physiological and biochemical parameters.
(4) Linear mixing of vegetation and soil in different proportions was to analyze the uncertainty resulted from the inversion of LAI. In fact, the combinations of surface features are often in the non-linear way and the pixel is consist of more than just vegetation or soil. Therefore, the effects of variety of feature types with diverse combinations including linear and nonlinear should be considered in the application of estimating vegetation physiological and biochemical parameters.
Aiming at the problems mentioned above, the further investigation and improvement would be done in future.
For a long time, using traditional technology to obtain remote sensing data such as satellite-based or manned space-based fails to provide timely and effective information when it comes to sudden natural disasters or dynamic monitoring of daily tasks, which roots in its low spatial resolution, long revisit cycle, time-poor, airspace control and so forth. Fortunately, this problem has been addressed by the arise of unmanned aerial vehicle (UAV) remote sensing. However, the UAV platform will inevitably result in tilting, shaking which leads to rotation and projection deformation in the acquired image, thus geometric correction is needed before analyzing and applying the UAV remote sensing data. At present, research on the geometric correction technology of UAV onboard aerial photography and aerial multispectral camera has been sophisticated from home and abroad, while relatively few studies focuses on the geometry processing of the rotary-wing UAV platform equipped with imaging spectrometer system. This thesis proposes a set of comprehensive geometric processing procedure aimed at rotary-wing UAV imaging spectrometer system. The feasibility of this process was implemented and verified through hyperspectral data which is attained by aerial imaging spectrometer system (AISS). An initial workflow of data acquisition, data preprocessing, selection of geometric approach and accuracy assessment of rectification results was formed, which provides a reference standard for the geometry of rotary-wing UAV imaging spectrometer system. The main work of this paper can be deduced as follows:
(1) Describe the composition of the rotary-wing UAV remote sensing systems, and the flight tests was conducted using the developed AISS, which equipped with a series of comprehensive systems including data acquisition and storage to meet the high-speed transmission , massive data, and light-weight design requirements.
(2) Analyze the imaging geometric principles of imaging spectrometer system, then specific to the rotary-wing UAV remote sensing platforms, design the mathematical model to qualitatively explore the impacts of systematic and non-systematic geometric distortion on the rotary-wing UAV remote sensing image and its corresponding ground points.
(3) Propose two types of geometric correction methods according to different combinations of ancillary data sets: control points based and POS data based. In the process of control points based approximate geometric correction, analysis and comparison of two kinds of detecting techniques to eliminate error control points was done under the premise of satisfying the selection criteria and selection method to get the ground control points. The appropriate error control point detection method was chosen in accord with the characteristics of the experimental data. The results show that the proposed method can effectively eliminate the error control points and improve the accuracy of geometric correction which guarantees the precision of the constructed control point database and provides strong support for automatic extraction of control point within this region; in the process of POS data based geometric correction, two fast search methods were proposed in the indirect method for determining the optimal scan lines, and according to the characteristics of the test data, best search strategy was selected to effectively improve the search efficiency and accuracy.
本文以自主研发的岩心组分成像光谱系统（Imaging Spectrometer System for Analying Core components）所采集的岩心成像光谱数据作为数据源，最终目标是建立岩心成像光谱数据编录系统，主要开展了以下两个方面的研究工作：一是进行岩心成像光谱数据编录系统的分析与设计。二是进行岩心成像光谱数据编录系统关键技术研究并实现岩心成像光谱数据编录原型系统。论文的主要成果和结论如下：
Mineral exploration becomes more difficult with the depth of the exploration while the cost increases drastically. Because geological prospecting techniques seem serious lag, people began to continuously find new technologies and new methods for exploring deep mines. How to give full play to the advantages of hyperspectral remote sensing technology based on drilling engineering to quickly identify deep mineralization information and realize the fine detection of deep geological structure and material composition is the current key study of remote sensing technology for mineral exploration. Using hyperspectral remote sensing technology to conduct drill core catalog and analysis of spectral features is the new research direction of information acquisition and information mining of the core. The core imaging spectrometer data catalog system is the latest stage of hyperspectral core catalog system, which uses core imaging spectrometer data to catalog and can integrate visualization of hyperspectral data’s spectral and image, and will provide effective technical support for deep prospecting.
This study aims to establish the core imaging spectrometer data catalog system based on the imaging spectrometer data collected by Spectrometer System for Analying Core components. This study includes two parts:1) to analyze and design of the prototype system of Core imaging spectrometer data catalog system; 2) to conduct Key technology research and implement the prototype system of core imaging spectrometer data catalog system.
Main results and conclusions are summarized as follows:
1. Core data storage model whose key is GeoRaster Object is proposed. Extensive researches were carried out on data imported and exported method based on procedure. The experimental results indicate that this model work well at the catalog and Visualization of core imaging spectrometer data.
2. Researches about the visualization of the Core imaging spectrometer data and the visualization of spectral curve are based on Qt and C++. The result shows that flawless combination of data obtaining method, image visualization method and spectral visualization method can support the integrated visualization of spectrum and image of hyperspectral data effectively.
3. On the basis of a successful study of logging technology,image and spectrum curve display ,visualization technologies and information extraction technology, Core imaging spectrometer data catalog prototype system was achieved .The system basically provides preprocessing, core imaging spectrometer data catalog, visualization and information extraction functions.
In recent years, China's coastal sea level rise is faster than the global average; the upward rising trend will continue to develop in the future. Sea level rise will destroy the local ecological environment, and also it will adversely affect the social and economic development. Therefore, it is of great significance to master the sea level variation, and to forecast the trend of sea level change. In present study, the analysis of sea level change often needs dozens or even hundreds of years of data to look for patterns, most of the observation sequence is difficult to meet the requirements. GM(1,1) grey prediction model can overcome the shortage, it can predict possible future trends by digging the intrinsic relation through a small amount of data. However, using the conventional grey GM(1,1) model which is the exponential function for sea level change prediction can reflect the trend of accelerated change, but the exponential change is far bigger than the accelerated trend of real sea level change. And the single exponential function can't reflect the linear trend of sea level change, it needs to be improved.
This thesis took Yangtze river estuary area as the research area to predict sea level change. Firstly, used the long time satellite altimetry data from the T/P and Jason-1 altimetry satellites to discuss tidal correction based on the data itself and reaserch the sea level change. And also the sequence of absolute sea level changes year by year in the study area in 1993-2010 was calculated，which is used for predicting future sea-level change research. Then, in the study of prediction model, the thesis improved the conventional grey GM(1,1) model with the method of slope equal, the absolute linear improvement method, and the twice incremental linear improvement method and metabolism improvement method. Use the global average sea level change data provided by the organization of Permanent Service for Mean Sea Level as the experimental data to evaluate the improved results. Finally, this paper takes the time series of sea level change from 1993 to 2010 in the Yangtze River estuary area which is obtained by satellite altimetry data as basic data， and then uses the absolute linear improved grey model to predict the change in the future. The main conclusions are as follows： (1) The harmonic analysis method can effectively separate tidal information from satellite altimeter data. Then The results of analyzing the sea level change of time and space used the sea surface height data, from which tidal information were removed, show that: on the spatial distribution, it has two opposite trends of sea level fluctuations change at the same time from coastal to the depths in Yangtze river Estuary and its adjacent sea, and also the sea level fluctuations have north-south gradient which is related to latitude approximately; on the time scale, there exists two stable periodic oscillation signals consist of years cycle and 2 months cycle. (2) It is shown from the comparison of the results that the absolute linear improved model is the best one no matter in the data fitting effect or in the prediction of the future effect. The absolute linear improved model takes the actual situation that sea level changes always contain linear trend into consideration, it is more rationality than either the single grey model or the single linear model. The results show that the absolute linear improved grey model can be applied to the predictions of sea level change. (3) The absolute sea level of the area in 2020 will increase by33.8 mm, compared with 2010, which is similar to the research results of other researchers. And also the result of analyzing the response of sea level change on temperature show that the trend of sea level change is consistent with the change of temperature.
As one of the most important data source of detailedvegetation classification(DVC), hyperspectral remote sensing(HRS)data make some classes distinguishable that cannot be separated in multispectral images, due to its advantage of imaging spectrum. HRSdata therefore have widespread application prospects in identification of crop species, monitoring of invasive species and precision agriculture. However, there are also some problems of HRSdata in DVC: Firstly, classification results based on only spectral information cannot meet the application requirements, as the classes become more and more sophisticated. Secondly, the increasing amount of spectral bands requires not only a huge number of calculations but also the corresponding growth of training samples in supervised classification. Thirdly, as the spatial resolution of hyperspectral sensor improves, the applications of classification are seriously affected by the salt and pepper noise in the classification results.
The semi-supervised classification algorithmsdeveloped in recent years play a key roleinHRSclassificationwith limited samples. The semi-supervised algorithmscould perform theclassification process by usinga small amount of labeled samples and a large number of unlabeled samples. However, most of the current semi-supervised classification algorithmsof HRSdataonly take advantage of the statistical characteristics of ground feature spectrum.On the basisof full investigation of current overseas and domestic research status, this study focuses on both vegetation feature band set construction and optimizationand semi-supervised classification method based on support vector machine (SVM) for limited labeled samples, to overcome the problem that labeled samples are difficult to be acquired in hyperspectral data classification. Some classification experiments of field imaging spectral data and airborne hyperspectral data with high spatial resolution are also performed, in order to verify the effectiveness of the proposed methods of sophisticated vegetation classification both on ground and airborne scales.
The main results and conclusions of this study are presented as follows.
(i) A DVCstrategy based on vegetation feature band set(FBS)construction and optimization is proposed: besides the spectral and texture feature of original images, we add 50 spectral indices that are sensitive to chlorophyll, carotenoid, anthocyaninand nitrogencontent to the vegetationFBS. Results show that this strategy is able to effectively improve the separability between different vegetation classes.
(ii) A spectral dimension optimization algorithm ofFBSbased on class-pair separablity (CPS) is proposed. This method focuses on the separablity of different vegetation classes at different feature bands, i.e. CPS. Itpreserves the original bands, texture features and spectral index features respectively that have the largest Bhattachryya distance of each CPthrough the iteration, and calculates the Jeffries-Matusitadistance to make sure that each CPmaintains a good separabilitythroughthe spectral dimension optimization of the FBS.Then Optimum Index Factor is employed to reduce the feature bands with high correlation.This proposed method can reduce the redundant data and improve the classificationefficiency.
(iii) A spatial dimension optimization algorithm of FBSbased on neighborhood pixels' spectral angle distance(NPSAD)is proposed, considering that in general the probability of adjacent pixels being the same class is relatively high. This method can set thresholds automatically according to training samples if there are, otherwise users have to set thresholds basedon prior knowledge. From the comparison of classificationresults, the proposed method could remove the salt and pepper noise from the classification results while avoiding the “edge effect” and keeping details at the same time, which can help to increasethe classification accuracy.
(iv) A progressive transductive support vector machine method based on the discrimination of both Spectral Angle Distanceand Euclidian distance(SAD/ED-PTSVM)is proposed. On the basis of traditional PTSVM method, this methoddiscriminates the unlabeled samples by respectively calculating SAD and ED, and implements “automatic label” for the unlabeled samples according to their distances to the separating hyperplane borders. SAD/ED-PTSVMmakes good use of spectral information of HRSdata, reducing the risk of incorrectly labeling and thus the time cost of label reset later. This method also effectively simplifies the parameter set of traditional PTSVM, reducing the amount of time spent on parameter optimization and thus improving classifying efficiency.
(v) A support vector machine classification method based on active learning using spectral unmixing technology(SUAL-SVM)is proposed. This method combines spectral unmixing technology in hyperspectral study and active learning strategy in machine learning field and fully utilizes the abundance of each pixel for better classification. Meanwhile, weighting factors are set to adjust the ratio of the most homogeneously mixed pixels and the most easily misclassified pixels. During active learning samples are added according to the distinguishing complexity of class-pairs, which makes the newly added samples more targeted. This method greatly improves identifying accuracy of the classes with small distribution, acquiring higher overall accuracy with less labeled samples and reducing not only the workload of labeling samples but also the time of samples training.
(vi) The results of detailedvegetation classification experiments on ground and airborne scales show that the classification method based on vegetation FBSconstruction and optimization can increase the classification accuracy of different types of crops/weeds on ground/airborne scale and extract more complete blade/crop plots information. SAD/ED-PTSVM and SUAL-SVMcan effectively improve the classification accuracy and efficiency, reaching relatively high accuracy when there are only 25 labeled samples for each class. The results of experiments illustrate that the proposed methods in this study have great potentialand broad application prospects in DVC, both on ground and airborne scales.