To meet the requirements of millimeter wave circuits for high-current and high cutoff-frequency devices, a compact 4-finger InGaAs/InP single heterostructure bipolar transistor(HBT) was designed and fabricated successfully by using planarization technology. The results show that the width of the emitter fingers is as small as 1μm, the high Kirk current of 4-finger HBT reaches 110mA, and the current gain cutoff frequency is as high as 176GHz. The device is promising on the applications in the medium-power circuits operating at millimeter-wave range.

Techniques of assembly, packaging and reliability of infrared focal plane array(IRFPA) were systematically studied. Some theoretic and technical problems were solved, such as the simulation of assembly conformation, high precision allocation of the modules, the design of flat leading wire, low leakage rate of laser welding, the way to decrease outgassing of Dewar inner surface. For the IRFPA detector the tests of assembly reliability, high-energy particle and laser irradiations were carried out. A batch of practical infrared focal plane Dewar assembly and flat leading wire were obtained. These techniques have been implemented in project. The experimental results show that the thermal load of micro-Dewar is less than 250mW and the vacuum life of Dewar is more than 18 months.

The technology and performance of long-wave 2048-elements linear HgCdTe focal plane array were reported. The n-on-p photodiodes were fabricated by planar ion implantation and hybridized with readout circuit by indirect flip chip technique. The 2048-elements linear array consists of 8 segments of 256-element linear arrays that are “butted” on a single substrate in a staggered format. The cut off wavelength and R 0 A of the photodiode are 9.9μm and 10 Ωcm 2 , respectively. The average detectivity of the 2048-elements linear array is 9.3×10 10 cmHz 1/2 W -1 . The ununiformity of response is 8%. The array peak operability is larger than 99.5%.

The physical principle of laser induced breakdown spectroscopy (LIBS) was interpreted. How the pressure affects the laser induced plasma was analyzed. A LIBS experimental system for the low pressure condition was setup and the experiments at the different pressures were carried out. The experimental result expresses how the pressure affects the laser induced plasma quantitatively. The feasibility on the application of LIBS to lunar exploration was discussed.

Pseduo-color enhancement method of sub-band images of mid-wave infrared was presented based on both lαβ color space transform and wavelet packet transform. The color and luminance information of pseudo-color image obtained by RGB color mapping were separated by using lαβ color space transform. And the luminance part was replaced by the image fused from two sub-band images by wavelet packet transform, then pseudo-color enhancive image was obtained after inverse lαβ color space transform. The experimental results show that the local standard deviation and local entropy of enhancive image are increased by comparing with simple color mapping image, and the image enhancing has good perceptive effect.

Dust aerosol can cause the change of the land surface temperature(LST) by altering the thermal infrared radiative transfer of earth-atmosphere system. The influence of different winter and summer dust aerosol optical depth (AOD) on thermal infrared brightness temperature(BT) and LST retrieval was systematically studied and the inversion results influenced by water vapor in the atmosphere and surface factor were also studied. Thermal infrared radiative transfer calculation indicates that 1.brightness temperature difference is less than 0 under the influence of dust aerosol; 2.BT and LST decrease with the increase of AOD, in which the decreasing trend of LST is more obvious; 3.LST changes very little with different water vapor content; 4.LST mainly reflects the temperature of dust layer for bigger AOD. These results agree well with the instance analysis of typical dust in the north of China.

Noises and background produced by the absorption of water in spectra signal are the main interference for prediction accuracy of quantitative calibration in blood glucose noninvasive measurement. By utilizing wavelet transform, the spectra signal can be split into multi-scale approximation and detail component. According to uninformative variable elimination (UVE) criterion, it is possible to distinguish the high-scale approximation component which presents background and lowscale detail coefficients which present noise from original spectra, and then to eliminate the corresponding component, so the background and noise can be eliminated simultaneously. In this study, the pretreatment method of wavelet transform combined with UVE criterion was applied to the research of blood glucose noninvasive measurement based on near infrared spectroscopy, and the adaptability of this pretreatment method to the uncertainty complex spectra model was also discussed. The results indicate that this pretreatment method is effective to eliminate background and noise in near infrared spectra of blood glucose noninvasive measurement. It is helpful to improve the prediction accuracy of quantitative calibration for blood glucose noninvasive measurement.

Infrared hollow fibers were developed for spectroscopic gas sensing. The hollow core fiber has a silver film and a dielectric film on its inner surface to enhance the reflection rate, and thus to low the transmission loss. The chemical silver mirror-reaction method and the liquid-phase coating technique were used to form the silver and the dielectric film. Low-loss properties were achieved from the visible to the mid-infrared wavelength regions by optimizing the dielectric material and the film thickness. The low-loss hollow core fiber can be used not only as the absorption cell for gas sensing but also as the transmission medium for infrared light. Primary experimental results show that the hollow fiber is a good alternative for gas cell for miniaturization in spectroscopic gas sensing.

A new method for FTIR spectral quantitative analysis was presented. The new method couples kernel partial least squares(KPLS) feature extraction with support vector regression machine(SVR) to improve the quantitative analysis accuracy and speed of seven-component alkane gas mixtures composed of methane, ethane, propane, iso-butane, n-butane, isopentane, and n-pentane, whose feature absorption spectra are cross each other and overlapped seriously. Firstly, the KPLS was employed to extract feature components from the FTIR spectra of above-mentioned seven-component gas mixtures. And then, the extracted feature components were fed into SVR to create the quantitative analysis model of seven component gases. The quantitative analysis results of calibration gas mixtures show that the prediction accuracy by KPLS-SVR model is higher than that by SVR model without feature extraction processing. Meanwhile, the predicting time by KPLS-SVR model is only half of that by SVR model. The study indicates that KPLS approach can effectively extract the latent nonlinear features implied in the spectra and component concentration, eliminate the noise of FTIR spectral data, and reduce the dimension of the spectral data. Coupling with SVR, KPLS feature extraction can improve the accuracy of FTIR spectral analysis, shorten the predicting time. KPLS-SVR is a very effective method for infrared spectral quantitative analysis.

Rough topographic data of granite surface were measured by atomic force microscope (AFM), and the characteristics of this real surface were analyzed according to the topographic data. In the range of geometric optics approximation, the bi-directional reflectance distribution function(BRDF) of this real surface was modeled by Monte Carlo method based on the surface topographic data, where the shadowing effect of incidence and the multiple reflections were considered. By comparing the modeling results of BRDF with its measured results which were measured at the wavelength of 0.6328μm and at different incidence angles, the results show that there is a reasonable agreement between them within a certain error range.

Suspended matter concentration is an important parameter of water quality and water environment evaluation. The field experiments including water quality analysis and spectrum measurements were carried out in 74 stations of Lake Tai during 14 days from 8 th Nov. 2007 to 21 th Nov. 2007. After analyzing the correlations between remote sensing reflectance and suspended matter concentrations, the results show that remote sensing reflectance in the range of 400～900nm wave bands is highly and moderately related to total suspended matter (TSM) and inorganic suspended matter (ISM) concentrations, and the biggest Pearson coefficients for TSM and ISM all appear at 725nm, and they are 0.883 and 0.869 respectively. And remote sensing reflectance is't related to organic suspended matter concentration. Neural network models of retrieving suspended matter concentrations were established by using remote sensing reflectance at sensitive wave bands. As to TSM concentration retrieval, a neural network model with 6 nerve cells in connotative layer shows best, whose R 2 is 0.948 and RMSE is 4.947; but as to ISM, another model with 4 nerve cells in connotative layer is the best one, whose R2 is 0.956, and RMSE is 5.104. Additionally, error analysis of neural network model and empirical model were conducted by using test samples. Based on the above analyses, the conclusion is that neural network models with hyper-spectrum remote sensing reflectance are more suitable for retrieving suspended matter concentrations of TSM and ISM than empirical models.

The object emissivity characteristics and the theory of heat radiation were studied to detect sand dust storm by remote sensing technique. The results show that the material nature of sand dust, ground surface and clouds is different when the sand dust happens. The ratio between the ground and sand dust in pixel images is different when the strength of sand dust changes. And these differences lead to the variations of material nature in pixels and object emissivity. At last, by a combination of theoretical and experimental researches, it was found that the degree of approach between ε 29 and ε 31 was different among cloud area, ground surface, weak sand dust and strong sand dust areas. And in a whole range of 0～1 of ε 31 , the order of ε 29 is cloud>strong sand dust>weak sand dust>ground surface. According to the results of the above, it was established a stable sand dust index (DSI) model to identify the sand dust strength . And the model can more accurately detect the range and intensity of sand dust after several rounds of sand dust examples.

Weakening the effect of heat island by environment construction is one of the key aspects for city development. To assure the reasonable planning of city construction, it is helpful analyzing the effect of land use types on ground heat distribution to get scientific data. In this study, by using Landsat ETM+ daytime remote sensing data of Shanghai, the inversion of ground bright temperature (GBT), the compose of color map and the classification of land use were made. On this basis, the relationships between daytime GBT distribution and land use types of city were analysed qualitatively and quantitatively. The results show that there are close relations between GBT distribution and land use types. By taking the GBT of resident sections (except that haven't landscapes of vegetation and water) as reference temperature, the water sections play a role of decreasing GBT both in summer and in winter, the soil and vegetation sections can decrease GBT in summer, but increase GBT in winter, and the other city establishment sections will increase GBT both in summer and in winter. The statistical formula obtained in this study may be used in programming of city environment and forecasting of heat island effect.temperature(GBT)

Polarization remote sensing is a new method with several advantages in detection. It is very important to study atmosphere and targets on the ground. To investigate the polarimetric scattering characteristics of camouflage targets, polarimetric parameters of three color pattern camouflage net were measured in visible and infrared waveband. By the analysis of data, it was found that polarimetric scattering characteristics of camouflage net were mainly influenced by incidence angle and reflectance of surface under a definite detecting condition. The polarization degree of scattering light goes up with the increase of incidence angle and falls when the reflectance of surface rises. The imaging of polarization degree can enhance the brightness and contrast of low reflectivity pattern, and the imaging of polarization angle is very sensitive to the outline of camouflage net. Polarization remote sensing is very useful to the detection of low reflectivity targets in complex backgrounds and threatens the existence of camouflaged military targets badly.

SVM model optimal multi-parameter selection method for imbalanced data recognition was proposed. First, the connotation and necessity of SVM model multi-parameter selection were theoretically analyzed. Then, a multi-parameter selection criterion based on F-measure was given, which can represent the recognition performance completely. The genetic algorithm was used in search of optimization of multi-parameter in parallel parameter optimal selection. The proposed method can get the global optimal solutions of SVM model multi-parameter, and avoid the local optimal solution caused by gradient method, and can also decrease the computational complexity of experiential selection method. The experimental results of the benchmarks and radar HRRP data sets reveal that the proposed multi-parameter selecting method can get the global optimal solution of SVM model. The recognition performance of the optimal SVM model can achieve much improvement.

In recent years, C-V model has been extensively used in image segmentation, but the computation speed is the key factor restricting the use of the method, especially for the large high resolution remote sensing imagery with complex scene, since the consuming time is very long. This research presented the method of combining C-V model and wavelet transform, which not only can improve the speed but also can achieve the multi-resolution segmentation, and has good antinoise performance. Experiment results show that our method can improve the speed 1～2 times compared to the C-V model in the premise of segmentation quality assurance.

When utilizing polarization remote sensing for earth observation, or atmosphere scattering, the polarization effect of atmosphere is always higher than the land surface's. It is usually believed that the polarization effect of the atmosphere is the main information resource of the remote sensing sensor. Thus the land surface's polarization is often ignored in remote sensing retrieval. This conclusion is based on the assumption that the reflection of land objects is diffuse reflection. However, in most cases, land objects are not standard Lambertian. Sometimes the polarization effect of those objects is stronger than the atmosphere's. In this study, the reflection mechanism of the land surface and the interpretation model of polarization spectrum were deduced. Our results show that the imaging spectrum experiment has the land surface polarization effect which can even reach to 50% or more in some cases. The polarization signal of sensor will be greatly affected by the land surface. So the polarization effect of land surface is an important factor which can not be ignored during the study of quantitative remote sensing.

A segmentation algorithm based on wavelet domain hidden Markov tree model was improved. The pixel level segmentation result can not be obtained because of choosing wavelet coefficients as training feature directly in traditional methods. At the same time, traditional methods ignore the feature of labeling maps at different scales by using one single context to all scales in fusion phase. Hence, this study dealt with the initial parameters set problem and chose better feature for training. In this way, the fine pixel level segmentation can be acquired directly in the raw segmentation step, and in multiscale fusion phase, the characteristics of labeling maps at different scales are used sufficiently. Among them, both the information from coarse-scale segmentation and the one from fine-scale segmentation were considered. Experiments show that the visual effects of our algorithm are the best compared with the HMTseg method proposed by Choi and the WD-HMTseg algorithm of remote sensing image segmentation presented by Sun Q.