Remote sensing technology has evolved gradually from aerial photogrammetry and has gone through three stages of development:

1. Aerial Photogrammetry Development Stage

The earliest aerial photograph still preserved is that of the city of Boston, which was taken from a balloon by J.W. Black in 1860. The application in geology began in 1913, someone in an airplane with a camera photographic imaging of the Bengtson oil field in Libya, Africa, and used this set of dirty aerial photographs to prepare a geological map of the Bengtson oil field. Aerial photography remote sensing mainly takes an airplane or a balloon as a means of transportation, uses an aerial camera to obtain information about the target, and then goes through the process of negative and positive to get the final aerial photographs. Aerial photography uses the visible light panchromatic band of electromagnetic wave, and uses photographic film to accept the sunlight reflected from the target to be photographed, imaging, the general photographic range of the photographic film is 0.3 to 0.9 μm. Most of the cases of aerial photography are vertical photography, i.e., the main axis of aerial camera is kept in the direction of the plumb line to take pictures; in special cases, the use of special cameras for oblique tilting photography. Aerial photography according to the use of electromagnetic wave bands, the corresponding photographic film and the characteristics of the image, divided into four kinds, namely: aviation visible light panchromatic black and white images; aviation visible light true color images: aviation infrared false color images: aviation infrared black and white images. Among them, aviation visible full-color black-and-white images and aviation infrared false-color images are most commonly used, and they mainly use the wide-band reflective intensity characteristics of the feature spectrum.

2. Multi-spectral Satellite Remote Sensing Stage

Digital satellite imaging first began with meteorological satellites, and in 1960 the TIROS-1 meteorological satellites provided very coarse satellite images, which were mainly used to show cloud patterns. Then, in the 1970s, the U.S. National Oceanic and Atmospheric Administration (NOAA) launched the Very High Resolution Radiation Sensor (AVHRR) for weather forecasting, which has a ground resolution of 1.1km, and we see the cloud maps it obtains on TV weather forecasting programs. Meanwhile, from the 1970s onwards, a number of satellites carrying higher resolution sensors have been launched one after another. For example, on July 23, 1972, the U.S. National Aeronautics and Space Administration (NASA) launched the first Earth Resources Technology Satellite (ERTS-U), which was specifically designed to monitor and map the Earth's surface, and in 1975 was renamed Landsat. In Landsatl-3 are equipped with multispectral scanner (MSS), the scanner has four bands, namely, green, red and two infrared bands, the ground resolution of about 80 m. In 1982, Landsat4 carried a thematic mapper (TM), which has seven bands, covering a wider spectral range than the MSS, and the width of the bands is divided into finer, more reflective of the reflectance spectral characteristics of the ground objects The most typical feature of multispectral remote sensing is the ability to use multiple bands to obtain multiple spectral features of the same target at the same time. This greatly improves the ability of remote sensing to recognize features. Subsequently, countries have followed suit, the spectral range of the sensor from the visible light, infrared up to microwave bands, the scope of application is also expanding.

3. Imaging Spectral Remote Sensing Technology Development Stage

Imaging Spectral Remote Sensing Technology is a leap forward in the development of multispectral technology.Hunt's research results show that the absorption width of the characteristic minerals is about 20-40nm, and the spectral resolution of the multispectral remote sensing data (e.g., MSS and TM) is only about 100nm, so remote sensing scientists began to study the spectral resolution and spatial resolution of the remote sensing of the high spectral resolution and spatial resolution of the remote sensing. In 1981, a Shuttle Multispectral Infrared Radiometer (SMIRR) was used with the U.S. Space Shuttle Columbia to make a limited-band observation of the Earth's surface, and for the first time it was possible to identify carbonates and clay kaolinite minerals directly from space by high spectral resolution remote sensing. Clay kaolin minerals, thus opening a new chapter in imaging spectral remote sensing lithology identification. Following the successful development of JPL's AIS-1 and AIS-2 and AVIRIS aerial imaging spectrometers, Canada has also successfully developed several imaging spectrometers, such as FIL/PML, CAS1 and SFSI (Tong et al., 1993). Others are: HIRIS (high resolution imaging spectrometer) imaging spectrometer, there are 192 spectral bands in the range of 0.4-2.5 μm, with a ground resolution of 30 m. The spectral resolution in the wavelength range of 0.4-1.0 μm is 9.4 nm, and in the range of 1.0-2.5 μm, it is 11.7 nm ( Goetz & Herring 1989; Kerekes & Landgrebe, 1991). The 63-channel imaging spectrometer (GER) of the Geophysical and Environ-mental Research Corporation (GER) was designed specifically for geological remote sensing studies and has been used many times for lithologic mapping (Lanfen Zheng et al. 1992; Bamaby W rockwell 1997). ). In addition to the airborne imaging spectrometer, the United States and the European Space Agency (ESA) have made plans to develop spaceborne imaging spectrometers, among which the U.S. Moderate Resolution Imaging Spectroradiometer (MODIS) has been added to the Earth Observing System (EOS) for launching into orbit to achieve periodic high spectral resolution remote sensing observations of the Earth. ESA's Moderate Resolution Imaging Spectroradiometer (MERIS) will also be launched at the same time (Tong et al., 1993).

From 1990 to 1995, Roger N. Clark et al. successively used AVIRIS data to identify and map minerals and lithologies in Nevada, U.S.A., at the Capulet Proving Ground, and they found that the Imaging Spectroradiometer not only distinguishes the overall brightness and slope differences in the surface emission spectra (the basis of the multispectral techniques MSS, TM, and SPOT for distinguishing between features), but also derives information for identifying special landforms.

Shanghai Institute of Technical Physics, Chinese Academy of Sciences (SITP, CAS) is the main development organization of imaging spectrometer in China.In 1983, the first 6-channel infrared subdivision spectral scanner working in the short-wave infrared spectral region (2.05-2.5 μm) was successfully developed with a spectral resolution of 30-50 nm.In 1987, driven by the state and the Chinese Academy of Sciences (CAS) mission of gold prospecting, the instrument was developed to 12 channels, and its band position tends to be more consistent with the absorption bands of terrestrial clay minerals and carbonatite minerals, thus providing greater capability in geologic lithology identification (Tong Qingxi et al., 1993). There are also thermal infrared multispectral scanners (TIMS), 19-band multispectral scanners (AMSS) and 71-band multispectral airborne imaging spectrometers (MATS). The data from these spectrometers are mainly used in remote sensing of oil and gas resources (Zhu Zhenhai, 1993) and mineral mapping (Wang Jinnian et al., 1996), etc. Progress has been made to varying degrees in data processing techniques and theoretical research on mineral identification (Li Tianhong, 1997).

To summarize the acquisition of remote sensing spectral data, there are several new developments:

①Extending the spectral range of application and increasing the spectral bands; ②Improving the spectral and spatial resolution; ③Having the function of obtaining stereo image pairs, breaking the ability of stereo image pairs that can be available only in aerial photographs (e.g., SPOT images); ④Improving the detector performance or detector devices, i.e., line and surface array CCD devices; ⑤ improved image data accuracy; ⑥ vertical development of application areas, such as the use of TM image data can directly identify hematite, acicular iron ore and other minerals.

At the end of the 20th century and the beginning of the 21st century, space hyperspectral imaging satellites have become an important frontier technology in remote sensing earth observation, playing an increasingly important role in the study of Earth's resources, monitoring the Earth's environment.

The development of hyperspectral resolution remote sensing technology is one of the major technological breakthroughs made by mankind in earth observation in the last two decades of the late 20th century, and it is a frontier technology of remote sensing at present and even at the beginning of the 21st century. The images of the earth's surface obtained through hyperspectral imaging contain rich spatial, radiometric and spectral triple information. In the late 1990s, along with a series of basic problems in the application of hyperspectral remote sensing, such as the calibration and quantization of hyperspectral imaging information, visualization and multidimensional expression of imaging spectral image information, image-spectral transformation, information processing of large data volume, etc., hyperspectral remote sensing has been gradually shifted to the stage of practical application by the stage of experimental research, and as a hot spot in the application of hyperspectral remote sensing is the focus on The focus of the hot spot of hyperspectral remote sensing application is the improvement of hyperspectral data information mining technology and the expansion of the application field closely connected with it.

The main feature of hyperspectral remote sensing data is that it integrates the traditional image dimension and spectral dimension information into one, and obtains the continuous spectral information of each feature while acquiring the spatial image of the ground surface, thus realizing the inversion of the feature composition information and feature recognition based on the spectral features of the features. It consists of the following three parts:

(1) Spatial image dimension

In the spatial image dimension, the hyperspectral data is similar to the general image. General remote sensing image pattern recognition algorithms are applicable information mining techniques.

(2) Spectral dimension

A "continuous" spectral curve can be obtained from each pixel of the hyperspectral image, and the "spectral matching" technique based on the spectral database can realize the purpose of identifying the features. At the same time, most of the features have typical spectral waveform features, especially the spectral absorption features are closely related to the chemical composition of the features, and the extraction of spectral absorption parameters (absorption wavelength position, absorption depth, absorption width) will become the main aspect of hyperspectral information mining.

(3) Feature Space Dimension

Hyperspectral images provide a super-dimensional feature space, and the mining of hyperspectral information requires a deep understanding of the characteristics and behavior of the distribution of features in the two-dimensional feature space formed by hyperspectral data, and the study found that: the high-dimensional space of the hyperspectral is quite empty, and the distribution of the data is uneven, and tends to focus on the corner end of the super-dimensional cube space, and the typical data variability, can be mapped to a series of low-dimensional subspaces, so there is an urgent need to develop effective feature extraction algorithms to discover the low-dimensional subspaces that maintain important variability, so as to effectively realize information mining.