Unsupervised Classification Techniques Research Articles

A segmentation approach to classification of remote sensing imagery

In this paper we propose a new approach for land cover classification of remote sensing imagery. It is a two-stage technique, where in the first stage the global feature-based technique of histogram thresholding generalized to multidimensional cases developed by Khotanzad and Bouarfa (1990) is used, and in the second stage a local feature-based region growing technique is generalized to grow multiple non-contiguous regions in parallel. The Khotanzad and Bouarfa technique has the advantage of being a non-iterative unsupervised classification technique, but suffers from a failure to detect regions of small spatial extent which may have high local contrast but low weightage in the global feature space. Our proposed technique divides the image into blocks of suitable size so that regions of small spatial extent are detected in the block's histogram, and they are grown across neighboring blocks. The proposed technique is illustrated with actual remote sensing imagery. A number of choices of feature space for the first stage, and different measures of similarity for the second stage were investigated on remote sensing data, both visually as well as quantitatively in terms of classification accuracy. It was found that the xyz colour space (Ohta et al. 1980) for the first stage, and the J-M distance for the second stage similarity measure, gave the best results in terms of classification accuracy. Though the approach is unsupervised and non-iterative in nature, it has given a classification accuracy of better than 91 per cent for a five-class landcover classification.

Fuzzy and isodata classification of landform elements from digital terrain data in Pleasant Valley, Wisconsin

Numerical classification methods may provide an alternative to manual landform delineation using aerial photographs, a subjective process that requires much knowledge of the landscape in question. Continuous classification (fuzzy set) methods and unsupervised (ISODATA) classification techniques were used to classify the landscape of a study area in southwestern Wisconsin, USA. Each pixel of a 10-m resolution digital elevation model (DEM) was grouped according to its membership in a continuous landform class. These classes were determined by the natural clustering of the data in attribute space. Attributes used for the classification were elevation, slope, profile and tangent (related to plan) curvature, compound topographic (wetness) index, and incident solar radiation. The ISODATA classification assigned pixels to one, and only one, landform class while the continuous classification allocated relative class memberships to each pixel. The resulting classifications roughly follow subjective manual delineation lines but give more detailed results. These classification methods may prove useful for statistical analyses and determination of sample schemes.

Neural-fuzzy classification for segmentation of remotely sensed images

An unsupervised classification technique conceptualized in terms of neural and fuzzy disciplines for the segmentation of remotely sensed images is presented. The process consists of three major steps: 1) pattern transformation; 2) neural classification; 3) fuzzy grouping. In the first step, the multispectral patterns of image pixels are transformed into what we call coarse patterns. In the second step, a delicate classification of pixels is attained by applying an ART neural classifier to the transformed pixel patterns. Since the resultant clusters of pixels are usually too keen to be of practical significance, in the third step, a fuzzy clustering algorithm is invoked to integrate pixel clusters. A function for measuring clustering validity is defined with which the optimal number of classes can be automatically determined by the clustering algorithm. The proposed technique is applied to both synthetic and real images. High classification rates have been achieved for synthetic images. We also feel comfortable with the results of the real images because their spectral variances are even smaller than the spectral variances of the synthetic images examined.

Use of Landsat TM Imagery in Determining Important Shorebird Habitat in the Outer Mackenzie Delta, Northwest Territories

Landsat Thematic Mapper (TM) imagery was examined to determine important habitats for shorebirds in the outer Mackenzie Delta, Northwest Territories. In June and July 1991 and 1992, 89 ground plots (200 X 200 m) in different habitats were censused for breeding shorebirds. Habitat type in ground plots was determined by observation and compared to the type identified at the site by an unsupervised Landsat classification technique. The most common species of shorebirds breeding in the area were red-necked phalaropes (Phalaropus lobatus) and common snipe (Gallinago gallinago), followed by semipalmated sandpipers (Calidris pusilla), stilt sandpipers (C. himantopus), pectoral sandpipers (C. melanotos), whimbrel (Numenius phaeopus), Hudsonian godwits (Limosa haemastica), lesser golden plovers (Pluvialis dominica), and semipalmated plovers (Charadrius semipalmatus). Long-billed dowitchers (Limnodromus scolopaceus) were rarely seen. Most species were concentrated in areas of low-centre polygons, sedge, and "low terrain" upland tundra (damp and tussocky). However, snipe were most common in dense willow habitat, and semipalmated plovers were found breeding only on sparsely vegetated gravel. Average density of breeding shorebirds in low-centre polygon or "pure" sedge habitat was 82 pairs per sq km in 1991 (SD=73.8), and 49 in 1992 (SD=49.5). Although the Landsat TM imagery analysis used here correctly identified habitat types near the original, intensively surveyed ("ground-truthed") area, it often misidentified habitats at some sites 10 to 30 km away, probably because of irregular flooding and subtle year-to-year differences in water levels in the active outer delta, and edge habitats too narrow to be distinguished by the satellite imagery. However, the technique can identify potential shorebird habitat roughly, and at least eliminate obviously unsuitable areas in large regions of the Arctic.

Impacts of hurricane Hugo on a coastal forest: Assessment using Landsat TM data

Abstract Landsat Thematic Mapper (TM) data were used to characterize the extent and intensity of coastal forest damage resulting from Hurricane Hugo (September 22, 1989) at Hobcaw Barony, South Carolina, USA. Low altitude vertical color infrared aerial photography and field observations provided the ground truth data necessary for validation. Bands 4 (0.76 ‐ 0.90 μm) and 7 (2.08 ‐ 2.35 μm) and three unsupervised classification techniques were used for forest damage classification of an image immediately following Hurricane Hugo. Similarly, a ratio of bands 7/4 was used as an index for mapping wind and storm surge damage and for change detection. The three unsupervised classification methods and the change detection analysis were effective at identifying the extent of forest damage. Healthy and heavily damaged stands were identified most accurately. Classification of damage intensity, especially for intermediate levels of damage, was more difficult and was affected by soil characteristics and understory ve...

Assessment of 'subcortical dementia' in patients with Huntington's disease, Parkinson's disease, multiple sclerosis and AIDS by a neuropsychological screening battery.

Four diagnostic groups categorized as 'subcortical dementia' were studied using a neuropsychological screening battery. The specificity of the battery in disclosing characteristic patterns of impairment in subgroups within the category, and the correspondence between natural subgroups generated by an unsupervised statistical classification technique and the diagnostic labels were investigated. All cognitive affected subgroups demonstrated impairment on tests of sensomotor function, cognitive efficiency and/or memory, and were characterized in terms of severity of impairment as well as a progressive involvement of functional areas. The diagnostic groups were not evenly represented in all natural subgroups. Affected AIDS-patients seemed to resemble patients with Huntington's disease more than the other diagnostic groups. The screening battery could give a general characterization of the pattern of 'subcortical dementia' and different levels of impairment, but more specific process-oriented tasks are needed to contrast diagnostic groups within the concept of 'subcortical dementia'.

Classification of variety musts by statistical analysis of their electrophoretic protein pattern

Unsupervised classification techniques (Principal Components Analysis and Cluster Analysis) have been applied to electrophoretic analytical data from seventeen grape must samples of six different cultivars. A good grouping of samples according to their cultivar has been obtained.

Satellite Data and Geographic Information System for Land Use Classification

Landsat Multispectral Scanner System (MSS) data in conjunction with the unsupervised classification technique of Earth Resources Laboratory Application Software (ELAS) were used to determine land use/land cover classifications. The Landsat computer compatible tapes (CCT) for February 9, 1976 were analyzed to spectrally classify unique land use/land cover conditions within the Econlockhatchee (Econ) River basin, Florida. The results showed that the scatter diagrams of Band 4 (0.5-0.6μm)‐Band 7 (0.8-1.1μm) and Band 4‐Band 6 (0.7-0.8μm) can be utilized as well as the traditional Band 5 (0.6-0.7μm)‐Band 7 scatter diagram approach for classifying land use/land cover. Both the zoom transfer scope and the geographic information system have been demonstrated to be very useful tools for land use/land cover classification of Landsat data. The Landsat data can be successfully used to monitor the USGS land use/land cover Level I, and the cost and time requirements for the Landsat approach are much less than for the conventional method. Another benefit of Landsat data is that new data are periodically available for updating the land use/land cover information.

An adaptive machine learning algorithm for color image analysis and processing

In this paper, a new adaptive machine learning algorithm for analyzing and processing color images of natural scenes is presented. The eventual goal of this research is to obtain a mathematical training algorithm to guide the operation of an unsupervised pattern recognition and classification technique for detecting and extracting the image modes or clusters in a selected or constructed feature space. For this purpose, the peak modality of one-dimensional (1-D) image histograms is selected as the mathematical training criterion. Area, mode dispersion, approximated curvature and steepness are some of the measured quantities for a modality test. Linear discriminant function is then used to extract the detected image clusters in the feature or measurement space.

Delineating salt-affected soils in the ganges plain by digital analysis of landsat data

A study has been conducted to determine the feasibility of delineating salt-affected soils using computer-aided analysis of Landsat-1 data in an area of the Ganges Plain. The multispectral scanner data have been obtained on 2nd December 1972 Landsat pass. Both supervised and unsupervised classification techniques are used. Four spectral classes of salt-affected soils have been separated. The results indicate that Landsat data can be successfully used for differentiating salt-affected soils.