Mineral pigments are widely used in the ancient Chinese painting. Classification and identification of mineral pigments are important for cultural heritages conservation. As a non-destructive method, hyperspectral classification is based on the knowledge that different mineral pigments have distinct reflection spectra. This study acquired the hyperspectral images of 38 mineral pigments and established a reflection spectral library. Then Spectral Angle Mapper (SAM) was used to classify the test data and 0.20 was selected as the optimal threshold. For pigments having similar color and spectra, SAM was unable to classify them correctly. Therefore, decision tree, a machine learning method, was applied to the classification of the pigments misclassified by SAM. For each pigment, 7500 samples were randomly selected as training data and 2500 samples were selected as test data. Though the ID3 algorithm, a decision tree for pigment classification was learned. Then test data was classified by the decision tree. Compared with SAM, the accuracy of classification observed from decision tree was obviously improved. For most pigments, the accuracy of decision tree reached 94%. The results revealed that the SAM combined with decision tree could effectively achieve a discrimination of all the 38 mineral pigments in the experiment, thus providing a new approach for mineral pigments classification.
A newly developed real-time infrared signal processing system based on the heterogeneous multi-processor system on chip (MPSoC) is proposed in this paper. The architecture, hardware configuration, image pre-processing algorithms used in the system and the experimental result are presented. Compared to the infrared signal processing system in being, Xilinx Zynq-7000 All Programmable SoC has been used in the proposed system which is more portable, integrated, and has excellent performance during its signal processing.
Haze always exists in hyper-spectral remote sensing imagery, and it is a key reason that influences the effective information extraction of hyper-spectral images. Specially, when the faint haze covers part of the target in remote sensing images, the target still can be detected but not clear. So, how to remove the influence of the haze and improve the applicable efficiency of hyper-spectral images is a popular research point. This paper proposes a dehazing method for hyper-spectral images based on linear unmixing. First, a popular hyper-spectral unmixing method called FUN is used to get the signature of all the end-members and their corresponding abundance. And then, the abundance of the haze end-member is removed and the abundances of the rest end-members are adjusted to satisfy the sum-to-one and non-negative constraint. Lastly, the new abundance and the signature of the end-members are linearly mixed to get the dehazed hyper-spectral images. The experiment result shows that the dehazed hyper-spectral images exhibit better target information and details. The method is effective and available.
We propose an approach to correct the data of the airborne large-aperture static image spectrometer (LASIS). LASIS is a kind of stationary interferometer which compromises flux output and device stability. It acquires a series of interferograms to reconstruct the hyperspectral image cube. Reconstruction precision of the airborne LASIS data suffers from the instability of the plane platform. Usually, changes of plane attitudes, such as yaws, pitches, and rolls, can be precisely measured by the inertial measurement unit. However, the along-track and across-track translation errors are difficult to measure precisely. To solve this problem, we propose a co-optimization approach to compute the translation errors between the interferograms using an image registration technique which helps to correct the interferograms with subpixel precision. To demonstrate the effectiveness of our approach, experiments are run on real airborne LASIS data and our results are compared with those of the state-of-the-art approaches.
Due to the manufacturing technique, some kinds of CCD, such as the back illuminated CCD, have the problem of spectral response nonuniformity. The near infrared light passing through the substrate and gates and is reflected back into the substrate for a second pass resulting in increased response. For the Fourier transform imaging spectrometer, it adds stripe pattern error to the interferogram and distorts the reconstructed spectrum. The nonuniform response is wavelength dependent due to changes in reflectivity of metal and the cavity formed by silicon and metal with transparent dielectric, so it adds difficulty to the correction of the error of the reconstructed spectrum.
In order to reduce the error of the reconstructed spectrum, in this paper, a calibration method and a correction method to correct the error caused by the CCD spectral response nonuniformity was developed, basing on analysis of the property of the CCD spectral response nonuniformity. Firstly, a calibrated monochromater was used to measure the CCD spectral response nonuniformity and the property and affect of the CCD spectral response nonuniformity were analyzed. Method to correct the error of the reconstructed spectrum caused by the stripe pattern error was developed. Secondly, to calibrate the CCD spectral response nonuniformity, the spectral response coefficient and the spatial response nonuniformity coefficient was measured and computed. Finally, we took data with a Fourier transform imaging spectrometer, and got the correction results of the reconstructed spectrums. The results showed that the distortion of recovered spectrum was evidently reduced and the effect of the calibration and correction method was proved.
The principle of Fourier transform spectrometer is based on the relationship of Fourier-Transform between interferogram and spectrum. The spectral information of Fourier transform imaging spectrometer (FTIS) reconstructed from raw interferogram by data processing. So there are two kinds of signal-to-noise ratio (SNR) to evaluate instrument performance, one regarding interferogram and the other regarding reconstructed spectrum. Because the raw interferogram is intuitive, the interferogram SNR is studied usually. On the contrary, the spectrum SNR is studied less because of the complexity of the data processing from interferogram to spectrum. The research about the effect of the interference fringe visibility on the spectrum SNR is especially few. This paper present a research work on the relations between the interference fringe visibility and the spectrum SNR. Firstly, the reduction of fringe visibility caused by imaging lens defocus was analyzed. Secondly, the changes of the average spectrum signal and noise caused by the reduction of fringe visibility were calculated. For average spectrum signal, the math deductions are done base on Fourier transform theory. The average noise with different input signal optic-electrons number are simulated. the results show that the average spectrum signal is directly proportional to the fringe visibility, and the effect of fringe visibility on the noise related to signal can be ignorable. Finally, In order to demonstrate above results, the imaging experiment was done with white-light source, using LASIS (Large aperture static imaging spectrometer) based on Sagnac Interferometer. The average spectrum SNRs under different fringe visibility are calculated and analyzed. The experimental results show that: the average spectrum SNRs increase from 42.7 to 62.4.along with the fringe visibility increasing from 0.5051 to 0.687. the reconstructed spectrum SNR is directly proportional to the fringe visibility. As a result, the interferogram fringe visibility can be used to estimate the reconstructed spectrum SNR, and evaluate the performance of FTIS before data processing.