Hyperspectral imaging has the advantages of ultra-multiband, high spectral resolution (a few nm), narrow band (≤10-2λ), wide spectral range (200-2500nm) and the combination of spectra etc. The advantage is that the collected images Due to the "fingerprint" effect of the reflected spectrum of the object, the principle that different objects have different spectrums and same objects must be the same spectrum to distinguish different material information.
The spectral properties of an object are closely related to its intrinsic physico-chemical properties, which result in the selective absorption and emission of photons of different wavelengths within the material due to differences in the composition and structure of the material. A complete and continuous spectral curve can better reflect this difference between different substances The inherent differences in onlookers, which is exactly the physical basis of imaging spectroscopy to achieve the fine ground object detection.
The following experiment using Sichuan double-spectrum Technology Co., Ltd. imaging height spectrometer Carry out the test to conclude.
First, the spectral resolution is high: the spectral bandwidth of each band of the detector indicates the ability of the detector to detect the spectrum of the object, which includes the total width of the detection spectrum of the detector, the number of bands, the wavelength range and the interval of each band, The more bands detected by the detector, the smaller the wavelength range of each band, the smaller the band spacing, the higher the spectral resolution.The detector has a high spectral resolution, and the image obtained by the detector can well reflect the spectrum of the object Nature, differences between different objects can be well reflected in the image, the ability of the detector to detect features is strong.Hyperspectral remote sensing image data is an important feature of ultra-multi-band and large amount of data, its treatment It has become one of the key problems of its successful application.Generally, the spectral absorption peak width of a typical mineral is about 30 nm, which can only be detected by a sensor whose spectral resolution is less than 30 nm.
Second: the image spectrum of a unity:
Mainly reflected in the image and the spectrum are presented at the same time, and the material characteristic spectrum is continuous, the study of any part of the information can be analyzed through data modeling.
Figure Hyperspectral data processing and analysis
Minerals: With a stable chemical composition and physical structure, the spectrum of a mineral depends mainly on the characteristics of the spectral absorption. The deciding factor lies in the interaction between the electron and the crystal field in the substance and the vibration of the molecule.
Spectral properties of soils: The primary minerals in the soil are quartz, feldspar, muscovite, a small amount of amphibole, pyroxene, apatite, hematite, pyrite, etc. The gravel and sand in the soil are almost all Is composed of primary minerals, mostly quartz.Most of the powder is also composed of quartz and primary silicate minerals.The secondary minerals in the soil are mainly about the following categories: 1, simple salts, such as carbonated Salts, sulfates and chlorides, etc .; 2, water-containing oxides such as iron oxide, aluminum oxide, silicon oxide, etc .; 3, secondary stratiform aluminosilicates such as kaolinite, montmorillonite and mica Wait.
Soil moisture is an important part of the soil, soil reflectance decreases as the soil moisture content increases, especially at 1400 nm, 1900 nm, 2700 nm in all absorption bands of water. For plants and soils, Obviously, this phenomenon is apparently caused by the same reason that incident radiation is strongly absorbed by water at a specific absorption band of water.
Figure Low-moisture soil spectral curve
The spectral reflectance of the sandstone varies with time of water intrusion
Soil texture refers to the relative proportions of particles of various particle sizes in the soil.The influence of the soil spectral reflectance on soil properties is mainly manifested in two aspects:
1. Affect soil water holding capacity, thereby affecting soil spectral reflectance;
2. Soil particle size itself also has a great impact on soil reflectivity;
Due to its strong hygroscopicity, the water absorption band at 1400nm, 1900nm, 2700nm and so on is obviously abnormal for the clay part with smaller soil particle size. With the decrease of soil particles and the space between particles, the specific surface area increases Large, the surface tends to be smooth, so that the reflectivity of silt in the soil is higher than that of sand, but when the grain is fine to clay, it increases the water holding capacity of the soil, but reduces the reflectivity.
In addition, the soil texture affects the reflection characteristics of the factors not only particle size combinations and their surface conditions, but also with the composition of different particle size chemical composition is closely related.
RGB image of conglomerate R: 1112 G: 1322 B: 1533
Principal Component Analysis (PCA) is performed on the reflectivity-calibrated data when no conglomerate is added: PCA transforms the data to a new coordinate system using a linear transformation of the multi-band data to maximize data differences. This technique is useful for enhancing information content, isolating noise, and reducing the number of data dimensions.
Significance: Principal component analysis (PCA) is one of the most widely used linear dimensionality reduction methods, and is widely used in many dimensionality reduction methods. Principal component analysis (PCA) uses the variance as a measure of the amount of information, and considers that the larger the variance The more the information is, the less the information is provided on the contrary.The basic idea is to reduce the dimensionality of the data by preserving the large variance and information-rich components by linear transformation, Is a linear combination of the original variables, therefore, the principal component analysis method is essentially a linear dimension reduction method.The calculation steps are generally divided into the following four steps:
1) Standardize the collection of raw data samples.
2) Compute the covariance matrix of the normalized data matrix and orthogonalize it to obtain the principal components.
3) Calculate the cumulative contribution of each principal component, select the principal component according to the required contribution rate threshold.
4) For the selected principal components to establish the principal component equation, calculate the principal component value.
PCA maps original variables into a few latent variables with the most information, and then uses linear least-squares method to determine the coefficients of these latent variables. After establishing the regression equation of latent variables and dependent variables, PCA converts them into original variables and dependent variables The regression equation has a high efficiency of compressing independent variables, but its mapping process has nothing to do with the dependent variable, so its prediction accuracy is also difficult to reach very high.
Savitzky-golay (digital smoothing and filtering)
The data with different weights, to obtain more effective data smoothing, based on the principle of least squares, can retain the useful information analysis signal, to eliminate random noise effective data smoothing method, the use of high-order polynomial data smoothing, in fact Is a kind of deconvolution operation.
Because the spectral signal collected by the spectrometer contains both the experimental and useful information, at the same time due to instrument Precision and other reasons to bring random noise, the most commonly used method to eliminate noise Savitzky-Golay (SG) convolution smoothing method, through the polynomial to move the window data for polynomial least squares fitting signal smoothing, both to eliminate noise and retain Spectral profile. Due to the uneven distribution among samples, the different sample sizes, the scattering of the sample surface and the change of the optical path length, etc., the scattering effects will be generated. The use of multiplicative scatter correction (MSC) can effectively eliminate these scattering In addition, the derivative spectrum can effectively eliminate the baseline and other background interference, identify overlapping peaks, improve the sensitivity and resolution.For the spectral source of possible noise during the acquisition, respectively, using MSC and SG convolution method of different guidance The combination of spectral data and processing.
After obtaining and comparing the following time periods, the conglomerate's overall image changes after the conglomerate absorbs moisture.The water shows strong absorption in the rock or conglomerate in the 900nm-2500nm band of the near infrared band, which has been described above.
In this experiment, the spectral band of the camera is 900nm-1700nm, and the absorption band of water is mainly concentrated in the vicinity of 1400nm. The spectral characteristics of mineral rock are mainly concentrated in the range of 2000nm-2400nm.
The figure below shows the conglomerate is treated with PCA, Savitzky-golay (digital smoothing and filtering) and multiplicative scatter correction (MSC) when the conglomerate is not in contact with water. Component PC-3 in the state of the image performance.
Figure conglomerate without water (PC3)
Similarly, 10 samples of the entire conglomerate data collection were selected for 11:08, 11:17, 11:20, 11:27, 12:00, 12:30, 12:58, 13:36, 13 : 57, 14:35 A total of 10 hyperspectral images were compared with uncoated conglomerate images.
The conglomerates show the same overall state when they are not absorbing water, and when the conglomerates begin to absorb moisture, starting from the point where they come into contact with water until the absorption appears to contain, the conglomerate behaves A very obvious state of water absorption.
In the picture below, from right to left, we can see the change of conglomerate over time. The conglomerate continuously absorbs moisture and the deeper the color, the more it shows that the more obvious the water is absorbed in this area, the more information about the performance of conglomerate Yes, the entire conglomerate absorbs moisture to varying degrees, rather than a single granular component.
The data analysis report you provided earlier is based on one of the particles to analyze its characteristic changes in the spectra with the change of the moisture content of the same particles over time.
11: 08 11: 17
11: 20 11: 27
12: 00 12: 30
12: 58 13: 36
13: 57 14: 35
Its spectral changes, I will not be here for a detailed analysis, because every pixel on conglomerate or each conglomerate particle can be effectively studied in the hyperspectral.
This is a very important technical advantage of hyperspectral: the combination of images and spectra can be a lot of data acquisition, very effective for the study of the overall macro-performance of objects or substances, but also very intuitive .Other technologies There may be some advantages in spectral accuracy, but they can not collect so many data and images at once, and the analysis of precision can be provided by means of analysis.
Spectral reflectance of conglomerate varies with time of water intrusion
A: Conglomerate without water PC status image: 20160408 11: 08
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Second: PC status of conglomerate water Image: Time interval: 20160408 11:17
newrawfile-conglomerate-1 water-1_ref
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Three: conglomerate water state of PC image: time interval: 20160408 11:20
newrawfile-conglomerate -1- clamp water -10_ref
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Four: conglomerate water state of the PC image: time interval: 20160408 11: 27
20160408112703 test:
RGB image
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Five: conglomerate water state of the PC image: time interval: 20160408 12: 00
20160408120014
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
VI: PC status of conglomerate with water Image: Time interval: 20160408 12: 30
newrawfile20160408123032_ref
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Seven: conglomerate water state of the PC image: Time interval: 20160408 12: 58
newrawfile20160408125820
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Eight: conglomerate water state of the PC image: time interval: 20160408 13: 36
newrawfile20160408133617
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Nine: PC status of conglomerate water Image: Time interval: 20160408 13: 57
newrawfile20160408135737_ref
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
Ten: conglomerate water state of the PC image: time interval: 20160408 14: 35
newrawfile20160408143523_ref
PC-1 PC-2
PC-3 PC-4
PC-5 PC-6
