Zhang Liangpei

Current issues in high-resolution earth observation technology

This paper reviewed the developments of the last ten years in the field of international high-resolution earth observation, and introduced the developmental status and plans for China’s high-resolution earth observation program. In addition, this paper expounded the transformation mechanism and procedure from earth observation data to geospatial information and geographical knowledge, and examined the key scientific and technological issues, including earth observation networks, high-precision image positioning, image understanding, automatic spatial information extraction, and focus services. These analyses provide a new impetus for pushing the application of China’s high-resolution earth observation system from a “quantity” to “quality” change, from China to the world, from providing products to providing online service.

Unsupervised remote sensing image classification using an artificial DNA computing

In this paper, a spectral encoding and matching algorithm inspired by artificial DNA computing (ADC) is proposed to perform the task of unsupervised classification for hyperspectral remote sensing data. As a novel branch of biological computational intelligence, ADC has strong capabilities of pattern recognition, huge information memory, parallel and fast computation. Unsupervised classification for hyperspectral data is complicated pattern recognition problem with massive volume data. In this paper, unsupervised hyperspectral data classification task by ADC is attempted and the preliminary results are provided. The experiment was performed to evaluate the performance of the proposed algorithm compared with two known algorithms: K-means and ISODATA. It is demonstrated that our method is superior to the traditional algorithms and its overall accuracy and Kappa coefficient reach 80.96% and 0.7631 respectively.

Normalization of NDVI from Different Sensor System using MODIS Products as Reference

Medium Resolution NDVI(Normalized Difference Vegetation Index) from different sensor systems such as Landsat, SPOT, ASTER, CBERS and HJ-1A/1B satellites provide detailed spatial information for studies of ecosystems, vegetation biophysics, and land cover. Limitation of sensor designs, cloud contamination, and sensor failure highlighted the need to normalize and integrate NDVI from multiple sensor system in order to create a consistent, long-term NDVI data set. In this paper, we used a reference-based method for NDVI normalization. And present an application of this approach which covert Landsat ETM+ NDVI calculated by digital number (NDVIDN) to NDVI calculated by surface reflectance (NDVISR) using MODIS products as reference, and different cluster was treated differently. Result shows that this approach can produce NDVI with highly agreement to NDVI calculated by surface reflectance from physical approaches based on 6S (Second Simulation of the satellite Signal in the Solar Spectrum). Although some variability exists, the cluster specified reference based approach shows considerable potential for NDVI normalization. Therefore, NDVI products in MODIS era from different sources can be combined for time-series analysis, biophysical parameter retrievals, and other downstream analysis.

Linear Features Extraction From Remote Sensing Image Based on Wedgelet Decomposition

Linear feature extraction is an important problem for remote sensing image processing, and it is very difficult to extract those linear features embedded in strong noise or when the SNR (signal to noise) is low like the complicated environment of remote sensing image. In this paper, an algorithm based on wedgelet decomposition is proposed to extract linear features from remote sensing image. Firstly, beamlets can be generated by recursive dyadic partitioning, vertex marking and connecting in different scales, and beamlet transform is implemented as one important parameter to generate edge map of linear feature. Secondly, each dyadic square is split into two wedgelet segments, and wedgelet decomposition is implemented as the other important parameter to generate edge map of linear feature. The propose method can detect lines with any orientation, location and length in different scales. Experimental results show that the proposed method can extract linear features accurately from remote sensing image. It can be suited to remote sensing image processing and in practice it has surprisingly powerful and apparently unprecedented capabilities.

An Approach For Edge Detection Based On Beamlet Transform

Edge detection is very useful and important for image processing and computer vision, as it can locate significant variations of gray images. In this paper, an algorithm based on beamlet transform is proposed to detect edges in image. Beamlets can be generated by recursive dyadic partitioning, vertex marking and connecting, the beamlet transform is the collection of all line integrals formed by viewing the image as a piecewise constant object and integrating along line segment in the beamlet dictionary, for the maximal beamlet coefficient surviving the Canny criterion, draw a line segment depicting that beamlet, all these beamlets in different scales are fused to generate an edge map at the image pixel level. The propose method can detect lines with any orientation, location and length in different scales and avoids subjective setting. Experimental results show that the proposed method can detect edges accurately even from noise image and has a better performance. It can be suited to different images processing, in practice it has surprisingly powerful and apparently unprecedented capabilities.

The estimation and analysis of NPP from 1982 to 2014 in Yunnan province based on multi-source remote sensing data

Net primary productivity (NPP), as an important indicator for the carbon sequestration capacity of vegetation, has become one of the hotspots in global climate change research under the background of continuous increasing of CO2 concentration. Remote sensing data based model is an effective and widely used method to obtain NPP at regional and global scales. While the spatial and temporal resolution of estimated result is limited by the resolution of inputted NDVI data. In this paper, a new framework was proposed to simulate long time series and fine scale NPP based on multi-source remote sending data in Yunnan province, China. GIMMS3g NDVI datasets and MODIS NDVI products were integrated to construct the consistent and high quality monthly NDVI data from 1982 to 2014 with 1km spatial resolution. There processing steps were designed to reach this target, successively are reconstruction, normalization and multi-sensor fusion. Then the long term NPP were calculated by Carnegie-Ames-Stanford approach (CASA) model with the carefully interpolated meteorological data. The results showed a gradually decreased NPP from the southwest to the northeast in study area. Furthermore, the annual NPP presents a volatile upward trend in the past 33 years, consistent with the trend of temperature in.

Subspace clustering based on decision fusion strategy for hyperspectral imagery

In this paper, a novel hyperspectral subspace clustering algorithm based on decision fusion strategy (SCDFS) is proposed. Because the different clusters are contained in different subspace of the same hyper-dimensional data, the clustering processing in different subspace is conducted by genetic K-means algorithm (KGA). The clustering results from different subspace can be combined into decision string. The proposed subspace clustering based on decision fusion strategy is conducted on decision string. Considering the selection of subspace, the decision results may be inaccurate. So by the majority voting processing for different subspace, the steady subspace combination can be determined. Finally, the weighted strategy is introduced into SCDFS algorithm to evaluate the distance of different decision string, and determine the fusion clustering result.

An unstructured sub-pixel target detector for hyperspectral imagery

As the spatial resolution of hyperspectral imagery is usually limited, sub-pixel targets only occupy part of the pixel area. Unstructured detectors, such as adaptive cosine estimate (ACE), has shown promising performance in sub-pixel targets detection, which models the background with a distribution property. This paper proposes a unstructured detector in which selective endmembers are used according to different pixels to ensure that the true composition of endmembers in each pixel is utilized in both the subspace model and the statistical test. Experiments show its better performance for sub-pixel targets detection than other unstructured detector.

A target detection algorithm for urban areas using HSRH imagery

This paper presents an target detection algorithm focusing on making full use of both spatial and spectral features of the high spatial resolution hyperspectral (HSRH) imagery. It use the spatial relationship between pixels in the finite impulse filter with the low dimension data transferred from the original imagery. Experiments show it performs better than the method solely depending on spectral features.