Kim Hwa Lim

Spatial data mining: clustering of hot spots and pattern recognition

Spatial data mining is the extraction of implicit knowledge, spatial relations or other patterns not explicitly stored in spatial database. The focus of this paper is placed on the information derivation of spatial data. Geographical coordinates of hot spots in forest fire regions, which are extracted from the satellite images, are studied and used in the detection of likely fire points. False alarms can occur in the derived hotspots. While this false information can be identified by comparing the radiance detected at several bands, we introduce a different approach to remove some of the false alarms. We use clustering and a Hough transformation to determine regular patterns in the derived hotspots and classify them as false alarms on the assumption that fires usually do not spread in regular patterns such as in a straight line. This project demonstrates the application of spatial data mining to reduce false alarms from the set of hotspots derived from NOAA images.

Hyperspectral land cover classification of EO-1 Hyperion data by principal component analysis and pixel unmixing

In this paper, we attempt to perform land cover classification using hyperspectral data acquired by the EO-1 Hyperion instrument over two test sites in the tropical region: one in Singapore and the other one in coastal Jambi on the Sumatra island of Indonesia. Atmospheric correction on the hyperspectral imagery was first performed using a commercial package. Principal component decomposition was then performed and an unsupervised ISODATA classification was carried out on the dominant components to produce a land cover classification map for each test site. Classification using the pixel unmixing method as implemented in the ENVI package was also performed. The results of classification were compared with existing land cover maps.

Modeling the reflectance spectra of tropical coastal waters

This work aims to extract optical parameters of coastal sea water with high sediment and chlorophyll content by fitting reflectance spectra to a model. Sea-truth water sampling campaigns were carried out from Dec 1996 to Dec 1999 in coastal waters around Singapore. In-situ reflectance spectra were acquired using a portable spectro radiometer. Laboratory measurements of the total suspended solid (TSS) and chlorophyll-a (Chl-a) were made from the water samples. The Chl-a concentration ranges from 1 to 90 mg/m3, with an average value of about 10 mg/m3. The three component model for sea water was used to model the reflectance spectra. The model also includes chlorophyll fluorescence and surface reflection due to skylight. Each reflectance spectrum is fitted to the model by finding a set of the fitting parameters that best fits the reflectance curve to the mode. The downhill simplex method is employed as the optimization procedure. The chlorophyll absorption coefficient at 440 nm is retrieved and is found to relate to the measured chl-A concentration by a power law relation.

Implmentation of a covariance-based principal component analysis algorithm for hyperspectral imaging applications with multi-threading in both CPU and GPU

Principle component analysis (PCA) [1] is widely utilized in hyperspectral image analysis [3, 4, 5]. There are three major approaches of principle component analysis: singular value decomposition (SVD) [2], covariance-matrix and iterative method (NIPALS) [6, 7]. In our previous work [9], we have demonstrated the advantage of the GPU implementation of covariance method for medium-sized hyperspectral images. In this paper, we present an improvement which combines the multithreading in CPU, GPU and CUDAs graphics interoperability [8]. It is found that this combined framework approaches real-time processing much further.

Retrieval of subpixel fire temperature and fire area in Moderate Resolution Imaging Spectrometer

The ability to retrieve information at subpixel spatial resolution from a satellite with more than one channel in the thermal infrared spectral range has benefited the monitoring of forest fires. With better sensors, the severity of fire and the degree of damage can be estimated by retrieving fire fraction and fire temperature using the Dozier procedure. A sensitivity analysis was also performed to estimate the uncertainty in the retrieval of fire fraction and fire temperature as a result of inaccurate determination of emissivity and background temperature.

Web-based processing for remotely sensed data

Remote sensing data are usually acquired at large file size and its processing is sequential. As most of the processes performed on the image file are usually repetitive, the processing time is directly proportional to its file size. We adopt the client-server processing technique to process NOAA data in a distributed manner, thereby reducing the program elapsed time. A JAVA socket based on the Transmission Control Protocol/Internet Protocol (TCP/IP) is used. Our implementation platform is the Internet with six Intel Pentium-III computers connected by TCP/IP. The objective of the computer program is to process the NOAA level-1B file so as to detect hotspots in a geographic area. Our experiments show that Web-based processing is a viable approach to reduce the time required for hotspot detection. However, there is an optimal point where additional processors beyond the optimum will not further contribute to the efficiency.

Retrieval of chlorophyll absorption spectra from remote sensing reflectance of turbid coastal waters

The phytoplankton absorption coefficients of turbid coastal sea water with high sediment and chlorophyll content is retrieved by modeling of the remote sensing reflectance measured at the water surface. Sea-truth water sampling campaigns were carried out from Dec 1996 to Dec 1999 in coastal waters around Singapore. In-situ reflectance spectra and laboratory measurements of the total suspended solid (TSS) and chlorophyll-a (Chl-a) were made from the water samples. The phytoplankton absorption at 440 nm is retrieved by fitting the reflectance spectra to a theoretical model.

Implementation of a covariance-based principal component analysis algorithm with a CUDA-enabled graphics processing unit

There are three major approaches of principle component analysis (PCA [1]): singular value decomposition (SVD [2]), covariance-matrix and iterative method (NIPALS). This paper implemented these methods for medium-sized hyperspectral images [3, 4, and 5] in NVIDIA CUDA and compared the performance between them and their CPU counterparts. It is found that the covariance-matrix approach has a great potential of reaching a real-time performance.

Development of a MODIS direct broadcast reception and processing system

Since the successful launching of the TERRA satellite in late 1999, CRISP has been preparing to receive MODIS data in Direct Broadcast mode. A multi-mission demodulator with SPOT, ERS, TERRA and AQUA capabilities has been purchased and integrated into CRISPs antenna system. The output of the demodulator is directly recorded on the existing direct ingest system (disk arrays). A reformatting program, with the option of performing Reed Solomon error correction, has been written to convert the recorded signal data into MODIS level-0 formatted data. This level-0 data is then fed into a level I processing program. Two versions of the level 1 processing software have been tested IMAPP by the University of Wisconsin and NASA GSFC MODIS level 1 Processing Software. One of the inputs for the NASA software is the construction record generated by a Windows-based program - Sorcerer. In order to integrate all our processing on the Unix platform, a small program that is able to generate the construction record has been written. We have found that the IMAPP software runs faster but the geolocationing is not very accurate. As for the NASA software, the geolocationing is fairly accurate, but a shortcoming is that its calibration process, MOD02, can only handle a granule size of 300 seconds. Therefore, we replace the geolocation produced by IMAPP by NASAs geolocation and further calibrate the results into level 1b using IMAPP software. A georectification code has also been developed to georectify the level 1b products onto a map.