Abstract:Weed detection play an important role in variables spraying in precision agriculture. This paper presents a new SVM (support vector machine) method using decision binary tree to discriminate crop and weeds in visible/near infrared image. Vegetation is segment from soil according to spectral feature in near-infrared band based on threshold method. The multi-spectral reflectance features of vegetation canopy are combined with texture features and shape features. Then multiclass detection is achieved based on decision binary tree established by maximum voting mechanism. It was tested by discriminate maize seedling and its associated weeds. The validation tests indicated that SVM using decision binary tree could improve classification accuracy significantly, and meet real-time requirements of agricultural applications greatly. The proposed method has produced results superior to other approaches.

Abstract:MODIS V4 fire detection algorithm has problems of fire omissions at large scan geometries.In order to solve this problem, the brightness temperature changes of fire pixels, which consist of fire and background surface at different scan geometries, were analyzed. Active potential fire detecting thresholds were developed according to scan geometries.A fire detection algorithm for the HJ-1B-IRS based on MODIS V4 algorithm was presented. Using HJ-1B-CCD data, we validated the fire detection results of HJ-1B-IRS with the fire detection method based on smoke plumes. In comparison to MODIS fire products, HJ-1B-IRS fire product shows advantage on fire location and small fire detection.

Abstract:This paper, based on the rapid and nondestructive testing of hyperspectral reflectance compared with conventional methods, constructed hyperspectral models for retrieving soil heavy metal content in different historical period by correlation analysis and multivariable statistical method. The result shows that the correlation between content of Cd, Cr, Cu, Ni, Pb and hyperspectral reflectance of 400~550nm and 1000~2500nm is more distinguished than other bands, and that the correlation enhanced remarkably after the original spectra derivate transformed. The first derivate stepwise linear regression model is optimal for retrieving soil heavy metal content in different period, with a minimum determination coefficient of 0.92 and maximum RMSE of 1.83. The relationship of As with reflectance spectra was affected by ferriferous oxides, organic matter and carbonate mineral; of Cd, Cu and Ni, by ferriferous oxides, clay mineral and organic matter, respectively; of Cr, by many factors; and of Pb, by both ferriferous oxides and clay mineral.

Abstract:The purpose of the study was to approach the feasibility of the snow water equivalent retrieved with optical remote sensing data. Field measurement for separate layer snow density and other snow parameters was made in Feb 2009 and 2010. For different snow types, by using continuum removal method, spectral absorption characteristics of snow were analyzed. It comes to the conclusion that snow depth has a significant effect on its spectral absorption at near 1028nm, 1252nm, 1494nm and 1940nm. The deeper the depth, the smaller the absorption depth is. The image of moderate-resolution imaging spectroradiometer with spatial 500m resolution was used as the experimental optical remote sensing data in this study. Based on correlation analysis of the test sample, including reflectance of MODIS channel 5 and 6, elevation and snow pressure, the remote sensing model for the retrieval of snow pressure was built by statistical regression equations. The evaluation results of the model showed that its root mean square error is 0.075 and correlation coefficient between predicted and measured values is 0.72 when snow depth is less than 30 cm.

Abstract:A realistic computer model was used to simulate the tassel effect on the thermal emission directionaly of corn canopies, which was further validated by ground measurement. It is found that thetassel contributes little (mostly＜6%) on the total directional birghtness temperature. That means tassel can be omitted during modelling or inversion of the surface leaving thermal radiance.

Abstract:According to the unsatisfactory and lower efficiency of classical statistical models in leaf area index (LAI) estimation, a new inversion method combined with phenology-based data segmentation and principal component analysis was proposed in this paper. In the method, principal components of spectral data and differential (or difference) spectral data were chosen as independent variables, and phenology-based data segmentation was integrated into data processing in order to improve estimation accuracy. In addition, multi-scale was involved in modeling. Winter wheat was selected as experimental object for numerical simulation and comparative analysis. Results not only showed high precision in whole estimation and effectively improved data saturation, but also manifested stability and robustness under full scan.

Abstract:In hyperspectral unmixing, endmember signals are not independent with each other, which compromise the application of independent component analysis (ICA) algorithm. This paper presented a novel approach based on constrained ICA for hyperspectral unmixing to overcome this problem. By introducing the constraints of abundance nonnegative and abundance sum-to-one, the purpose of our algorithm was not to find independent components as decomposition results anymore. In order to accord with the condition of hyperspectral imagery, we developed an abundance modeling technique to describe the statistical distribution of the data. The modeling approach is capable of self-adaptation, and can be applied to hyperspectral images with different characteristics. Experimental results on both simulated and real hyperspectral data demonstrated that the proposed approach can obtain more accurate results than the other state-of-the-art approaches. As an algorithm with no need of spectral prior knowledge, our method provided an effective technique for the blind unmixing of hyperspectral imagery.

Abstract:A super resolution reconstruction method for dim point target detection in infrared image was presented. To achieve this, local descriptor based dense flow calculation is introduced first. Then, the high accuracy flow is used to registrate and fuse two adjacent images to generate the final high resolution image. Experiments show that, after the process, the resolution of small targets can be increased, local signal-to-noise ratio be increased and complex background be surprised.

Abstract:Rayleigh quotient quadratic correlation filter (RQQCF) is an important technique for target detection. Since it operates directly on image data, satisfying results can’t be always achieved when it is used in infrared target detection. Higher-order statistical properties of the image can effectively suppress the noise and clutter and improve the detection results which can be realized by means of kernel method in kernel space. In this paper, kernel Rayleigh quotient quadratic correlation filter (KRQQCF) was developed by extending RQQCF to the higher-dimensional space, i.e., the kernel space. Though the derivation was completed, this kernel filter couldn’t be achieved directly. Kernel feature extraction method was proposed to settle this problem. The algorithm was used to detect infrared targets, and kernel principal component analysis(KPCA) was adopted to obtain this KRQQCF in experiments. Experimental results using real-life infrared images confirm the excellent performance of KRQQCF, and that KRQQCF outperforms RQQCF significantly in infrared target detection. Consequently, KRQQCF is an effective method for infrared target detection and can achieve accurate detection results.

Abstract:The scope of this paper addressed the problem of detecting a dim moving target from a sequence of multispectral IR cubes. The detection problem was formulated in a general framework, assuming unknown target amplitude, position and velocity. This composite hypothesis testing problem was approached by means of the generalized likelihood ratio test (GLRT) theory. The detector structure and its actual implementation based on velocity filters were discussed in detail. Approximated expressions of the false alarm and detection probabilities were obtained and validated by means of simulation. To test the effectiveness of the detection algorithm, the detection results obtained on a set of synthetic multispectral IR image sequences were presented and discussed. These results indicate that the algorithm proposed can obtain a good performance on dim target detection with low SNR.

Abstract:The sparse representation based on over-complete dictionary is a new image representation theory. The redundancy of over-complete dictionary can enable it effectively to capture the geometrical characteristics of the images. In this paper, a novel detection method based on image sparse representation was introduced. The over-complete target dictionary is first constructed with atoms which are produced by two-dimensional Gaussian model. Then the sub-image blocks of the test image are extracted successively and the corresponding coefficients are calculated with the constructed over-complete target dictionary. There is a significant difference between the coefficients of objective and background. Whether the sub-image block contains small target or not can be determined by the index of sparse concentration. Experimental results demonstrated the effectiveness of the proposed method.

Abstract:A new background suppression method based on combined shearlet transform and Bayesian mechanism was proposed to solve the problem which is dim and small target detection contained complex sky clouds and ground background clutter for infrared search and tracking system. Firstly, according to difference of distributed charateristiscs between target and background clutter,in infrared image. the shearlet transform was adopted to decompose the origimal infrared image into multi-scale and multi-direction, which extracts multi-scale and multi-direction detail features of origimal image. Then, Gaussian scale mixture (GSM) model was introduced to separate dim, small target and background clutter from infrared image for suppression background. Finally, target image was obtained by using classical adaptive thresholding segmentation technique and target detection implemented. When compared with two dimensional least mean square (TDLMS) method, several groups of experimental results demonstrate that the proposed method can suppress complicated background in dim small target image effectively.

Abstract:Synthetic Aperture Radar (SAR) images reveal serious geometric distortions that are caused by terrain undulations. This paper presents a newly developed method on SAR image geometric correction which is the dual-aspect geometric correction based on digital elevation model (DEM) to overcome the inherent shortages of SAR image such as foreshortening, shadow and layover and correct the distorted or lost backscatter coefficient values in mountainous area. The Radarsat-2 SAR images were used in this study. The results show that this method can effectively eliminate the effect of geometric distortions and compensate the lost or distorted backscatter coefficients, especially is useful for eliminating layover and shadow distortions in SAR images. This method solves the geometric correction problem that cannot be solved with single SAR image.

Abstract:An unsupervised technique for detecting change area between two SAR images was proposed. The detection process is based on distribution property of the joint intensity histograms and need not distribution hypothesis. The algorithm uses adaptive edge detection to get training data. The joint intensity histograms in different levels are used to decide the membership degree of unlabeled points through Fisher classifier. The fusion model which considers the context relationship and inter-scale information improves the sensitivity. The simulation results of two real SAR images show that the algorithm is effective and has better detection results.

Abstract:Active imaging LADAR (Laser Detection And Ranging) stands out from other passive imaging systems for its function of acquiring range image of targets. Compared with traditional optical image, range image has more particular statistical characteristics, it can obviously show multi-peak structures when multiple targets exist in the range image. Target Extraction of range image could be realized by extracting peaks of its stat-histogram; and targets could be classified by the size and rectangle degree obtained by the use of minimum enclosing rectangle. Processing a real range image by using this method has obtained the anticipated result.

Abstract:The problem of natural gas pipeline leak becomes increasingly serious. A practical airborne natural gas pipeline inspection laser radar which is based on tunable diode laser spectroscopy was developed. The operating principle of this instrument was introduced, the system configuration and method of design were discussed. The key technologies including high-accuracy laser frequency stabilization and real time calibration were further discussed. The analysis results of ground and airborne test proved that the instrument can be applied in practical project.

Abstract:The visible image of FY-2 Geostationary Meteorological Satellite is produced by multiple sensors in the same time, so image has higher resolution. The images produced by multiple sensors are not exactly the same because of the sensors’ differences. This paper introduced several methods to calculate the difference and to adjust the satellite by remote commands and decrease stripes of multi-sensor imagery. The results show that it is correct and the image quality is greatly improved after the adjustment. These methods are also useful to decide when and how to adjust the VISSR while the satellite is on orbit.