Abstract. A major source of quantitative data, which contains information that can be used in the classification of land covers, is the Landsat series of low orbiting spacecrafts, which began in 1972 with Landsat-1. The later versions of this series, starting with Landsat-4 and -5 are equipped with Thematic Mapper (TM) sensor systems, which generate a vector for different intensity responses, of each pixel, in seven light and infrared spectral bands. With knowledge of different spectral responses of land covers it is possible to identify categories when analysing the vector data formats. This paper introduces a computerized procedure which is believed to be effective in identification of land covers. The method is particularly applicable to the Thematic Mapper system. It combines the linear analysis with the correlation procedures in specific formats, using a small number of reference identifiable categories in order to aggregate and pinpoint the pixel contents (with small probability of error). Fine identification of categories (such as the separation of corn or wheat in the vegetation covers) is the subject of further promising applicability using this described computerized technique
Abstract. This paper proposes an ontogenic neural network with application to the classification of remote sensing data. The purpose of this approach is the simultaneous design and training of a neural network classifier by means of genetic algorithm. The genetic algorithm, with its global search capability, finds a network structure with its weights that maximizes a classification fitness measure in one process, thus, avoiding the trial-and-error process of estimating the network structure. The idea of the presented classification method is to find a set of masks that encompasses all objects of a certain class, while screen out any objects of other classes. Before trying to find the suitable masks, the data is projected onto a set of significant principal axes and the projected data is used in the training process. Thus the data becomes more spread which makes it easy to partition into different classes. Moreover, this technique reduces the dimension of the used masks, and hence the neural network nodes. A Landsat thematic Mapper data is used to demonstrate the usefulness of the proposed method.
AbstractIn this paper, numerical exemplars are used in a training method to find the structure of a fuzzy-neural network. After this structure learning, a genetic algorithm is applied to determine the initial weights of the neural network, thereby guiding the neural network to a near optimal initialization. These well-initialized networks are then trained with backpropagation algorithm. Using this proposed approach, the local minimum phenomenon, which may cause the learning process to stagnate, can be avoided. Overall learning performance is, thus, significantly improved. The proposed method has been implemented and tested on the Thematic Mapper sensor system (TM) data to get the fractional representation of each class within a pixel. Results show the potential of the proposed method for this kind of applications.
Abstract. Texture analysis plays an essential and a major rule in image classification and segmentation in a wide range of applications such as medical imaging, remote sensing and industrial inspection. In this paper, we review the well known approaches of texture feature extraction and perform a comparative study between them. These approaches are namely gray level histogram, edge detection, and co-occurrence matrices, besides Gabor and Biorthogonal wavelet transformations. The feed forward artificial neural network (ANN) with back- propagation algorithm (BPA) is used as a supervised classifier. Experiments are conducted on two different datasets taken from multi-class engineering surfaces produced by six machining processes and from Brodatz (1966) textures album respectively. The classification accuracy is tested for both datasets, while the quality of estimation is tested for surface roughness parameters of the machined surfaces dataset only based on the roughness parameters evaluated from a contact measurement test.
Abstract. Texture classification is one of the most important clues of visual processing applications .In this paper, we present a comparison between the most two popular supervised texture classification methods based on the feed forward Artificial Neural Network (ANN) and the multi-class Support Vector Machine (SVM). Five of the most common used features extraction approaches were chosen in order to extract input vectors of different sizes for both classifiers. These approaches are namely gray level histogram, edge detection, and co-occurrence matrices, besides Gabor and Biorthogonal wavelet transformations. Experiments are conducted on two different datasets the first one is engineering surface textures produced by different machining processes, and the second was taken from Brodatz (1966) textures album. The classification accuracy rate is calculated for ANN and SVM in order to measure the efficiency of each technique based on the several features extraction methods. The results show that SVM with its linear and polynomial kernels is higher in classification accuracy and faster in training time.
Abstract: Most of the techniques used in the literature in clustering symbolic data are based on the hierarchical methodology, which utilizes the concept of agglomerative or divisive methods as the core of the algorithm. The main contribution of this paper is to show how to apply the concept of fuzziness on a data set of symbolic objects and how to use this concept in formulating the clustering problem of symbolic objects as a partitioning problem. Finally, a fuzzy symbolic c-means algorithm is introduced as an application of applying and testing the proposed algorithm on real and synthetic data sets. The results of the application of the new algorithm show that the new technique is quite efficient and, in many respects, superior to traditional methods of hierarchical nature.
Abstract: Most of the techniques used in the literature for hierarchical clustering are based on off-line operation. The main contribution of this paper is to propose a new algorithm for on-line hierarchical clustering by finding the nearest k objects to each introduced object so far and these nearest k objects are continuously updated by the arrival of a new object. By final object, we have the objects and their nearest k objects which are sorted to produce the hierarchical dendogram. The results of the application of the new algorithm on real and synthetic data and also using simulation experiments, show that the new technique is quite efficient and, in many respects, superior to traditional off-line hierarchical methods.
In this paper a new algorithm for subgraph isomorphism is proposed. The main idea of the new algorithm is to decompose the graphs to be matched into smaller subgraphs. The matching process is then done at the level of the decomposed subgraphs based on the concept of error-correcting transformations. The cost of matching two graphs is defined as the minimum of a weighted bipartite graph constructed from the decomposed subgraphs. The average computational complexity of the proposed algorithm is found to be O(N4). The results of the application of the new algorithm show that the new technique is quite efficient and, in many respects, superior to similar existing techniques in the literature. Besides, it overcomes most of the problems encountered in using tree search algorithms. The new algorithm is also suitable for parallel processing implementation
Abstract: In this paper a new predictive lossless coding scheme is proposed. The prediction is based on a cascaded peak to valley linear prediction method (PVLP). This method is based on simple linear prediction between the detected feature points. Experimental results on different types of music and songs show a new competitive compression ratio compared to the other algorithms of the lossless audio compression
Abstract: The growing hierarchal self-organizing map (GHSOM) is the most efficient model among the variants of SOM. It is used successfully in document clustering and in various pattern recognition applications effectively. The main constraint that limits the implementation of this model and all the other variants of SOM models is that they work only with vector space model (VSM). In this paper, we extend the GHSOM to work in the graph domain to enhance the quality of clusters. Specifically, we represent the documents by graphs and then cluster those documents by using a new algorithm G-GHSOM: graph-based growing merarchal SOM after modifying its operations to work with the graph instead of vector space. We have tested the G-GHSOM on two different document collections using three different measures for evaluating clustering quality. The experimental results of the proposed G-GHSOM show an improvement in terms of clustering quality compared to classical GHSOM.
