Jianwen Ma

Land cover classification from MODIS EVI times-series data using SOM neural network

A high-dimensional dataset was built with time-series data of vegetation indexes derived from a Terra-Moderate Resolution Imaging Spectroradiometer (MODIS) sensor used for land use/cover classification. The self-organizing map (SOM) neural network technique can reduce the dimensionality of high-dimensional data, yet keep the same topological characters in the low-dimension space after dimension reduction. In this paper, we first employed the SOM neural network technique to classify land cover types using a 17-dimensional dataset that was generated from 16-day interval MODIS Enhanced Vegetation Index (EVI) data with a spatial resolution of 500 m in eastern China during the growing period of plants. Then, we defined an unlabelled class of neuron. Pixels matched to this type of neuron were regarded as unclassified land cover types, so that we could remove the poorly classified areas. Finally, the classification results were compared with those of the maximum likelihood classification (MLC) method. Comparison showed that the accuracy of the former exceeded that of the latter in classifying a high-dimensional dataset.

Land cover classification based on tolerant rough set

The tolerant rough set classifier (TRSC) was introduced for land cover classification. TRSC uses a tolerance relation to define the tolerant rough set of each object to be classified, and then classifies the object using the relative frequency of each class in the lower approximation or boundary of its tolerant rough set. According to the overall accuracy, the κ coefficient, the total normalized probability of misclassification (TNPM) and McNemars test, the result of TRSC was better than that of the minimum distance classifier (MDC), and similar to those of the maximum likelihood classifier (MLC) and the multiplayer perceptron (MLP).

Damage consequence chain mapping after the Wenchuan Earthquake using remotely sensed data

The Wenchuan Earthquake occurred on 12 May 2008 followed by frequent aftershocks. The earthquake suddenly caused heavy damage to roads, bridges, buildings and communication facilities. Two B4101 Citation II remote sensing airplanes from the Chinese Academy of Sciences took off to monitor the region soon after the earthquake and collected images using a synthetic aperture radar (SAR) device and an optical ADS40 sensor. Using images obtained from airborne surveys, a damage consequence chain was mapped that provided information about the type of damage and its extent. This proved to be of great help for rescue operations and after earthquake reconstructions. We present details of airborne survey, images and mapping of damages caused by the earthquake and the use of such information for a rescue operation.

Extraction of polymetallic mineralization information from multispectral Thematic Mapper data using the Gram-Schmidt Orthogonal Projection (GSOP) method

In order to accelerate the reconnaissance of mineral resources in Xingjiang Uygur Autonomous Region of China, multispectral Thematic Mapper (TM) data were used to delineate hydrothermal alteration related to polymetallic mineralization. In addition to the use of commercial image processing packages, new image processing software based on geological environment and spectral information have been developed. The Gram-Schmidt Orthogonal Projection (GSOP) method used in linear algebra for signal processing was adopted for our study. The kernel part of the algorithm is that the GSOP uses least mean square deviation of mathematical expectation in linear vector space (Hilbert space). In a vector space, one base axis is established and other transitional vectors are inputted to evaluate the effectiveness of GSOP. Satisfactory results have been achieved by applying this method to differentiate hydrothermal alterations in the Kangxiwa area of the Kunlun mountains, western China. The red colour in the image shows Pb-Zn-Cu-Au-Ag polymetallic mineralization zones, which were confirmed by field investigations and sample analyses in 1998.

Study of artificial neural network method for weather and AVHRR thermal data classification

In recent years, the Asian dust storm project was carried out. One of tasks was to study dust rising mechanism in dust source area. Surface temperature condition was regarded as one of the important factors for dust rise. In the study we retrieved surface temperature by using NOAA/AVHRR data. Basedon the published articles, traditionally, split window algorithm was use to deriving surface temperatures in the case of our study area mostly desert area, there was only three field observation data available in Talimu basin, at Dunhuang and Changwu. It was very difficult to validate the results. However, there were 52 county wearther observation stations in the area. The data might be used as import data in artificial neural network calculation. Most success examples of remote sensing data classification by using neural networks were in the condition of network training and classifying in the same types of data such as spatial data. For the use different data type collected by different techniques system such as satellite system and ground weather observation data to training, to find rule and to direct classification could be more impersonal which was one of the nature of artifical neural network method. In our case 52 weather temperature data were used from 52 observation stations where they were also the same positions for collecting AVHRR 1b data CH2, CH4, CH5 thermal data. Both groups of data were applied as fundamental import data in for artificial neural network calculation. Finally resultant rule was applied for classifying 15000 x 3 pixels in the whole area. The result was more reliable than that of split window not only because uncertainty caused by variations of topography but also it was very difficult to validate in field.

Use of wavelet high-frequency substitution fusion to increase remote sensing image spatial resolution

IHS transform was one of typical method for remote sensing data fusion. In recent years, newly developed method that combines advantages of IHS and Wavelet algorithms makes image fusion. In this case after the Wavelet substitution based on pixels or features, and then transforms inversely with IHS in Munsell color space. In this paper we introduce a high frequency substitution method to improve spatial resolutions of imagery. The procedure of the method introduced as flowchart, in which the dot line area is our newly added method. The resolution was greatly improved comparing original image. In cooperating with the demand of on going Minjiang river, Si Chuan, China. A 15m resolution PAN band and 30m resolution 7 bands of ETM data were selected for the method testing, the steps of method test showing in flow chart of this paper. In the future the dots area was our newly developed wavelet high frequency substitute. Improved NDVI imagery raised the quality for monitoring land cover change factor in the project of Return Farmland Back to Forest or Grassland.

Spectral and spatial feature integrated edge extraction method for high-resolution remote sensing image

With urban and township development and E-Government program promotion in China city remote sensing as base data has developed rapidly. The technique demands in accuracy and effective edge detection and extraction from higher resolution image become important focal area. In the current popular image processing software packages there are some existing edge detection convolution kernels suchc as Sobel, Robert, Prewitt, Kirsch, Gauss-Laplace kernels. In general the kernels all work based on algorithm of convolution kernel in spatial territory of the image. However, satellite sensors capture spatial and spectral signatures of surfaces at same time. Use of both spatial and spectral features to establish a edge detection process is a new notion for achieving more accuracy results. In the paper we introduce a spatial and spectral integrated method which is designed in four stages. The result suggests that four stages process can achieve more cleanly and accuracy edges of city construction than that results of using other algorithms. The procedure is summarized in figure 1.

GA-hyperplane segmentation method for MODIS data

For the traditional method of hyper-plane segmentation, the location of hyper-plane in data space was given by statistical method. In the case of the statistical value of regions is smaller than in the region, the statistical method was not effective. The character of genetic algorithm is global searching optimally. Taken this mathematical advantage the location of Hyper-plane could be located easily. In this paper, EOS/MODIS imagery data is used to test this method. The result is proved that Genetic Algorithms-Hyper-plane is better than MLC method by using same training data.