Self-organizing Map Classification Research Articles

Self organized mapping of data clusters to neuron groups

T. Kohonen's self organizing map (SOM) may be considered as a plausible structure for modelling pattern recognition processes in the brain. Neighborhood preservation corresponds closely to what is called somatotopy in the neurosciences, and the context specificity of mappings observed (e. g. in malfunctions of the brain) becomes easily explicable in the framework of the SOM. However, there are two features which impair the aptitude of the classical SOM for neurophysiological models: the adaptation procedure is explicitly time dependent and the procedure consumes the whole set of disposable neurons. Because of the latter property, a SOM cannot learn different tasks, adapting one subset of neurons to a data set Chi(1) and another to a subsequently presented data set Chi(2). The present paper describes a modified SOM which avoids the drawbacks mentioned above. Its adaptation procedure is time independent. When the training sequence consists of data from successive data clusters Chi(k) each cluster is mapped to a subset G(k) of the neuron set G while the other neurons are left almost unchanged. The behavior of the resulting DCNG-SOM is demonstrated in several experiments.

Multivariate Temporal Data Analysis Using Self-Organizing Maps. 1. Training Methodology for Effective Visualization of Multistate Operations

Multistate operations are common in chemical plants and result in high-dimensional, multivariate, temporal data. In this two-part paper, we develop self-organizing map (SOM)-based approaches for visualizing and analyzing such data. In Part 1 of this paper, the SOM is used to reduce the dimensionality of the data and effectively visualize multistate operations in a three-dimensional map. A key characteristic of multistate processes is that the plant operates for long durations at steady states and undergoes brief transitions involving large changes in variable values. When classical SOM training algorithms are used on data from multistate processes, large portions of the SOM become dedicated to steady states, which exaggerates even minor noise in the data. Also, transitions are represented as discrete jumps on the SOM space, which makes it an ineffective tool for visualizing multistate operations. In this Part 1, we propose a new training strategy specifically targeted at multistate operations. In the proposed strategy, the training dataset is first resampled to yield equal representation of the different process states. The SOM is trained with this state-sampled dataset. Furthermore, clustering is applied to group neurons of high similarity into compact clusters. Through this strategy, modes and transitions of multistate operations are depicted differently, with process modes visualized as intuitive clusters and transitions as trajectories across the SOM. We illustrate the proposed strategy using two real-case studies, namely, startup of a laboratory-scale distillation unit and operation of a refinery hydrocracker.

Objective Identification of Nonlinear Convectively Coupled Phases of Monsoon Intraseasonal Oscillation: Implications for Prediction

Abstract The nonlinear convectively coupled character of the summer monsoon intraseasonal oscillation (ISO) that manifests in its event-to-event variations is a major hurdle for skillful extended-range prediction of the active/break episodes. The convectively coupled character of the monsoon ISO implies that a particular nonlinear phase of the precipitation ISO is linked to a unique pattern of the large-scale variables. A methodology has been presented to capture different nonlinear phases of the precipitation ISO using a combination of a sufficiently large number of dynamical variables. This is achieved through a nonlinear pattern recognition technique known as self-organizing map (SOM) involving six daily large-scale circulation indices. It is demonstrated that the nonlinearly classified states of the large-scale circulation isolated at the SOM nodes without involving any information on rainfall are strongly linked to different phases of evolution of the rainfall ISO, including the active and break phases. While a lower SOM classification involving 9 different states identify the composite phases of the rainfall ISO, a higher SOM classification involving 81 states can identify different shades of composite phase of the rainfall ISO. The concept of isolating the nonlinear states, as well as the technique of doing so, is robust as almost identical phases of precipitation ISO are identified by the large-scale circulation indices derived from two different reanalysis datasets, namely, the 40-yr ECMWF Re-Analysis (ERA-40) and the NCEP–NCAR reanalysis. The ability of the SOM technique to isolate spatial structure and evolutionary history of nonlinear convectively coupled states of the summer monsoon ISO opens up a new possibility of extended-range prediction of summer monsoon ISO. This knowledge is used to develop an analog technique for predicting different phases of monsoon ISO. Skillful four-pentad lead prediction of rainfall over central India is demonstrated with the model using only large-scale circulation fields. A major strength of the model is that it can easily be used for real-time extended-range prediction of monsoons.

Land-cover Classification Using ASTER Multi-band Combinations Based on Wavelet Fusion and SOM Neural Network

In this study, we developed a land-cover classification methodology using Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) visible near-infrared (VNIR), shortwave infrared (SWIR), and thermal infrared (TIR) band combinations based on wavelet fusion and the selforganizing map (SOM) neural network methods, and compared the classification accuracies of different combinations of ASTER multi-band data. A wavelet fusion concept named ARSIS (Amelioration de la Resolution Spatiale par Injection de Structures) was used to fuse ASTER data in the preprocessing stage. In order to apply the wavelet fusion method to ASTER data, the principal components of ASTER VNIR data were computed. The first principal component was used as the base image for wavelet fusion. In our experiments, the spatial resolution of ASTER VNIR, SWIR, and TIR data was adjusted to the same 15 m. SOM classification accuracy was increased from 83 percent to 93 percent by this fusion, and classification accuracy increased along with the increase of band numbers. Classification accuracy reaches the highest value when all 14 bands are used, but classification accuracy closely approached the highest value when three VNIR bands, three SWIR bands, and two TIR bands were used. A similar tendency was also obtained by the maximum likelihood classification (MLC) method, but the classification accuracies of MLC over all band combinations were considerably obviously lower than those obtained by the SOM method.

Combined Gesture-Speech Recognition and Synthesis Using Neural Networks

Sign languages such as Spanish Sign Language (LSE) are the primary communication way among members of the Deaf community. However this language is not widely known outside of this community. The techniques for automatic recognizing hand signs proposed in this paper allow creating systems which can help deaf people to communicate with others, by providing them with computer tools for assisted communication or potentially with a fully automatic portable sign-language-to-speech translation system. The aim of this research it to implement a self-organizing neural network based technique for hand sign recognition, using self organizing map (SOM) and learning vector quantization (LVQ) based algorithms. The two classifiers are then combined to make the final decision. SOM and LVQ classifiers are training for the 26 signs of the manual alphabet and the speech synthesizer concatenate the different phones, diphones and triphones stored in the database to generate the right words in artificial speech. Finally, the system is to contribute to the implementation of meaningful human machine interactions in a workstation, mainly for welfare applications.

Comparison of self-organizing maps classification approach with cluster and principal components analysis for large environmental data sets

Three classification techniques (loading and score projections based on principal components analysis (PCA), cluster analysis (CA) and self-organizing maps (SOM)) were applied to a large environmental data set of chemical indicators of river water quality. The study was carried out by using long-term water quality monitoring data. The advantages of SOM algorithm and its classification and visualization ability for large environmental data sets are stressed. The results obtained allowed detecting natural clusters of monitoring locations with similar water quality type and identifying important discriminant variables responsible for the clustering. SOM clustering allows simultaneous observation of both spatial and temporal changes in water quality. The chemometric approach revealed different patterns of monitoring sites conditionally named “tributary”, “urban”, “rural” or “background”. This objective separation could lead to an optimization of river monitoring nets and to a better tracing natural and anthropogenic changes along the river stream.

Changes in Antarctic net precipitation in the 21st century based on Intergovernmental Panel on Climate Change (IPCC) model scenarios

Projections from 15 global climate models and 2 reanalysis products (National Center for Environmental Prediction (NCEP)/National Center for Atmospheric and Climate Research (NCAR) reanalysis (NNR) and European Centre for Medium‐Range Weather Forecasts (ECMWF) 40‐year reanalysis (ERA40)) were utilized to project changes in the net precipitation (P − E) over the Southern Ocean and Antarctica during the 21st century. Three time periods, 1979–2000, 2046–2055, and 2091–2100, of data were compared. The P − E was related to a classification of synoptic circulation patterns obtained using a neural network algorithm known as self‐organizing maps (SOMs). SOM classification was successfully used as a quality control tool to assess the simulated atmospheric circulation and model performance in P − E. The models predicted an increase of Antarctic P − E that averages 0.42 ± 0.01 mm year−1 for the coming hundred years based on the difference between 1979–2000 and 2091–2100. P − E changes of individual models ranged from 0.02 to 0.71 mm year−1. P − E integrated over the entire Antarctic ice sheet was forecast to increase more quickly from the end of the twentieth century until 2046–2055 than from 2046–2055 until 2091–2100. Contributions to the predicted change in P − E were evaluated for both thermodynamic and dynamic processes. The projected change in Antarctic P − E was primarily due to thermodynamic changes rather than circulation changes. The dynamic component of P − E change, associated with the circulation, was important at subcontinental scales, especially over the coastal regions. The role of dynamic changes was maintained until the end of the 21st century. Intermodel variation in predicted P − E changes and differences between models and reanalyses in the twentieth‐century simulations severely restrict the reliability of these projections and highlight the need for improved polar simulations in climate models.

Web mining based on Growing Hierarchical Self-Organizing Maps: Analysis of a real citizen web portal

This work is focused on the usage analysis of a citizen web portal, Infoville XXI ( http://www.infoville.es) by means of Self-Organizing Maps (SOM). In this paper, a variant of the classical SOM has been used, the so-called Growing Hierarchical SOM (GHSOM). The GHSOM is able to find an optimal architecture of the SOM in a few iterations. There are also other variants which allow to find an optimal architecture, but they tend to need a long time for training, especially in the case of complex data sets. Another relevant contribution of the paper is the new visualization of the patterns in the hierarchical structure. Results show that GHSOM is a powerful and versatile tool to extract relevant and straightforward knowledge from the vast amount of information involved in a real citizen web portal.

A new image classification technique using tree-structured regional features

Image classification is a challenging problem of computer vision. Conventional image classification methods use flat image features with fixed dimensions, which are extracted from a whole image. Such features are computationally effective but are crude representation of the image content. This paper proposes a new image classification approach through a tree-structured feature set. In this approach, the image content is organized in a two-level tree, where the root node at the top level represents the whole image and the child nodes at the bottom level represent the homogeneous regions of the image. The tree-structured representation combines both the global and the local features through the root and the child nodes. The tree-structured feature data are then processed by a two-level self-organizing map (SOM), which consists of an unsupervised SOM for processing image regions and a supervising concurrent SOM (CSOM) classifier for the overall classification of images. The proposed method incorporates both global image features and local region-based features to improve the performance of image classification. Experimental results show that this approach performs better than conventional approaches.

Combination of ICA and SOM for Classification of Machine Condition Patterns

Nonlinear independent component analysis (NICA) is a powerful method for analyzing nonlinear and nongaussian data. Artificial neural network (ANN), especially self-organizing map (SOM) based on unsupervised learning, is an excellent tool for pattern clustering and recognition. A novel multi-NICA network is proposed for feature extraction of different mechanical patterns, followed by a typical ANN that is one of Multi-Layer Perceptron (MLP), or Radial Basis Function Network (RBFN), or self-organizing map (SOM), which implements the final classification. Using NICA and appropriate strategies for further feature extraction, nonlinear and higher than second order features embedded in multi-channel vibration measurements can be captured effectively. Mechanical fault patterns can be recognized correctly. Results from the contrast classification experiments show that the new compound ICA-SOM classifier can be constructed in a simpler way and it can classify various fault patterns with high accuracy, both of which imply a great potential in health condition monitoring of machine systems.

A parallel growing architecture for self-organizing maps with unsupervised learning

Self-organizing maps (SOMs) have become popular for tasks in data visualization, pattern classification or natural language processing and can be seen as one of the major concepts for artificial neural networks of today. Their general idea is to approximate a high dimensional and previously unknown input distribution by a lower-dimensional neural network structure with the goal to model the topology of the input space as close as possible. Classical SOMs read the input values in random but sequential order one by one and thus adjust the network structure over space: the network will be built while reading larger and larger parts of the input. In contrast to this approach, we present a SOM that processes the whole input in parallel and organizes itself over time. The main reason for parallel input processing lies in the fact that knowledge can be used to recognize parts of patterns in the input space that have already been learned. This way, networks can be developed that do not reorganize their structure from scratch every time a new set of input vectors is presented, but rather adjust their internal architecture in accordance with previous mappings. One basic application could be a modeling of the whole-part relationship through layered architectures.

New Adaptive Color Quantization Method Based on Self-Organizing Maps

Color quantization (CQ) is an image processing task popularly used to convert true color images to palletized images for limited color display devices. To minimize the contouring artifacts introduced by the reduction of colors, a new competitive learning (CL) based scheme called the frequency sensitive self-organizing maps (FS-SOMs) is proposed to optimize the color palette design for CQ. FS-SOM harmonically blends the neighborhood adaptation of the well-known self-organizing maps (SOMs) with the neuron dependent frequency sensitive learning model, the global butterfly permutation sequence for input randomization, and the reinitialization of dead neurons to harness effective utilization of neurons. The net effect is an improvement in adaptation, a well-ordered color palette, and the alleviation of underutilization problem, which is the main cause of visually perceivable artifacts of CQ. Extensive simulations have been performed to analyze and compare the learning behavior and performance of FS-SOM against other vector quantization (VQ) algorithms. The results show that the proposed FS-SOM outperforms classical CL, Linde, Buzo, and Gray (LBG), and SOM algorithms. More importantly, FS-SOM achieves its superiority in reconstruction quality and topological ordering with a much greater robustness against variations in network parameters than the current art SOM algorithm for CQ. A most significant bit (MSB) biased encoding scheme is also introduced to reduce the number of parallel processing units. By mapping the pixel values as sign-magnitude numbers and biasing the magnitudes according to their sign bits, eight lattice points in the color space are condensed into one common point density function. Consequently, the same processing element can be used to map several color clusters and the entire FS-SOM network can be substantially scaled down without severely scarifying the quality of the displayed image. The drawback of this encoding scheme is the additional storage overhead, which can be cut down by leveraging on existing encoder in an overall lossy compression scheme.

Expanding Self-Organizing Map for data visualization and cluster analysis

The Self-Organizing Map (SOM) is a powerful tool in the exploratory phase of data mining. It is capable of projecting high-dimensional data onto a regular, usually 2-dimensional grid of neurons with good neighborhood preservation between two spaces. However, due to the dimensional conflict, the neighborhood preservation cannot always lead to perfect topology preservation. In this paper, we establish an Expanding SOM (ESOM) to preserve better topology between the two spaces. Besides the neighborhood relationship, our ESOM can detect and preserve an ordering relationship using an expanding mechanism. The computational complexity of the ESOM is comparable with that of the SOM. Our experiment results demonstrate that the ESOM constructs better mappings than the classic SOM, especially, in terms of the topological error. Furthermore, clustering results generated by the ESOM are more accurate than those obtained by the SOM.

Self-organizing maps with recursive neighborhood adaptation

Self-organizing maps (SOMs) are widely used in several fields of application, from neurobiology to multivariate data analysis. In that context, this paper presents variants of the classic SOM algorithm. With respect to the traditional SOM, the modifications regard the core of the algorithm, (the learning rule), but do not alter the two main tasks it performs, i.e. vector quantization combined with topology preservation. After an intuitive justification based on geometrical considerations, three new rules are defined in addition to the original one. They develop interesting properties such as recursive neighborhood adaptation and non-radial neighborhood adaptation. In order to assess the relative performances and speeds of convergence, the four rules are used to train several maps and the results are compared according to several error measures (quantization error and topology preservation criterions).

자기구성 지도를 이용한 인터넷 FAQ의 자동응답 및 개념적 브라우징

Though many services offer useful information on internet, computer users are not so familiar with such services that they need an assistant system to use the services easily In the case of web sites, for example, the operators answer the users e-mail questions, but the increasing number of users makes it hard to answer the questions efficiently. In this paper, we propose an assistant system which responds to the users questions automatically and helps them browse the Hanmail Net FAQ (Frequently Asked Question) conceptually. This system uses two-level self-organizing map (SOM): the keyword clustering SOM and document classification SOM. The keyword clustering SOM reduces a variable length question to a normalized vector and the document classification SOM classifies the question into an answer class. Experiments on the 2,206 e-mail question data collected for a month from the Hanmail net show that this system is able to find the correct answers with the recognition rate of 95% and also the browsing based on the map is conceptual and efficient.

The Supervised Network Self-Organizing Map for Classification of Large Data Sets

Complex application domains involve difficult pattern classification problems. The state space of these problems consists of regions that lie near class separation boundaries and require the construction of complex discriminants while for the rest regions the classification task is significantly simpler. The motivation for developing the Supervised Network Self-Organizing Map (SNet-SOM) model is to exploit this fact for designing computationally effective solutions. Specifically, the SNet-SOM utilizes unsupervised learning for classifying at the simple regions and supervised learning for the difficult ones in a two stage learning process. The unsupervised learning approach is based on the Self-Organizing Map (SOM) of Kohonen. The basic SOM is modified with a dynamic node insertion/deletion process controlled with an entropy based criterion that allows an adaptive extension of the SOM. This extension proceeds until the total number of training patterns that are mapped to neurons with high entropy (and therefore with ambiguous classification) reduces to a size manageable numerically with a capable supervised model. The second learning phase (the supervised training) has the objective of constructing better decision boundaries at the ambiguous regions. At this phase, a special supervised network is trained for the computationally reduced task of performing the classification at the ambiguous regions only. The performance of the SNet-SOM has been evaluated on both synthetic data and on an ischemia detection application with data extracted from the European ST-T database. In all cases, the utilization of SNet-SOM with supervised learning based on both Radial Basis Functions and Support Vector Machines has improved the results significantly related to those obtained with the unsupervised SOM and has enhanced the scalability of the supervised learning schemes. The highly disciplined design of the generalization performance of the Support Vector Machine allows to design the proper model for the particular training set.

Using Self-Organizing Maps to Investigate Extreme Climate Events: An Application to Wintertime Precipitation in the Balkans

Abstract This paper examines some of the physical mechanisms and remote linkages associated with extreme wintertime precipitation in the Balkans. The analysis is assessed on daily timescales to determine the role of the circulation and atmospheric moisture on extreme events, and also at intraseasonal and interannual timescales to find possible linkages with the North Atlantic Oscillation (NAO) and the Arctic Oscillation (AO) patterns. A nonlinear classification known as the self-organizing map (SOM) is employed to obtain the climate modes and anomalies that dominated during the 1980–93 period. An artificial neural network (ANN) is also used to derive daily precipitation at gridpoint scale and at local scale in Bucharest, Romania. Of the predictors used, 500–1000-hPa thickness, 700-hPa geopotential heights, and 700-hPa moisture are the most important controls of daily precipitation. These results are substantiated with the climate states from the SOM classification, which show strong meridional flow over c...

A multiscale approach to automatic medical image segmentation using self-organizing map

In this paper, a new medical image classification scheme is proposed using self-organizing map (SOM) combined with multiscale technique. It addresses the problem of the handling of edge pixels in the traditional multiscale SOM classifiers. First, to solve the difficulty in manual selection of edge pixels, a multiscale edge detection algorithm based on wavelet transform is proposed. Edge pixels detected are then selected into the training set as a new class and a multiscale SOM classifier is trained using this training set. In this new scheme, the SOM classifier can perform both the classification on the entire image and the edge detection simultaneously. On the other hand, the misclassification of the traditional multiscale SOM classifier in regions near edges is greatly reduced and the correct classification is improved at the same time.

Handwritten numeral recognition using self-organizing maps and fuzzy rules

Handwritten numeral recognition using combined self-organizing maps (SOMs) and fuzzy rules is presented in this paper. In the learning phase, the SOM algorithm is used to produce prototypes which together with corresponding variances are used to determine fuzzy regions and membership functions. Fuzzy rules are then generated by learning from training patterns. In the recognition stage, an input pattern is classified by a fuzzy rule based classifier. An unsure pattern is then re-classified by an SOM classifier. Experiments on a database of 20,852 handwritten numerals (10,426 used for training and a further 10,426 for testing) show that this combination technique achieves satisfactory results in terms of classification accuracy and time, and computer memory required.