Hybrid Classification Method Research Articles

Assessment of the state of conservation of buildings through roof mapping using very high spatial resolution images

The assessment of the state of conservation of buildings is extremely important in urban rehabilitation. In the case of historical towns or city centres, the pathological characterization using traditional methods is a laborious and time consuming procedure. This study aims to show that Very High Spatial Resolution (VHSR) multispectral images can be used to obtain information regarding the state of conservation of roofs where, usually, building degradation starts. The study was performed with multispectral aerial images with a spatial resolution of 0.5 m. To extract the required information, a hybrid classification method was developed, that integrates pixel and object based classification methods, as well as information regarding the classification uncertainty. The proposed method was tested on the classification of the historical city centre of Coimbra, in Portugal, that includes over than 800 buildings. The results were then validated with the data obtained from a study conducted during 2 years by a nine element team from the University of Coimbra, using traditional methods. The study performed achieved a global classification accuracy of 78%, which proves that the state of conservation of roofs can be obtained from VHSR multispectral images using the described methodology with a fairly good accuracy.

Text Document Preprocessing with the Bayes Formula for Classification Using the Support Vector Machine

This work implements an enhanced hybrid classification method through the utilization of the naïve Bayes classifier and the Support Vector Machine (SVM). In this project, the Bayes formula was used to vectorize (as opposed to classify) a document according to a probability distribution reflecting the probable categories that the document may belong to. The Bayes formula gives a range of probabilities to which the document can be assigned according to a pre determined set of topics such as those found in the "20 newsgroups" dataset for instance. Using this probability distribution as the vectors to represent the document, the SVM can then be used to classify the documents on a multi – dimensional level. The effects of an inadvertent dimensionality reduction caused by classifying using only the highest probability using the naïve Bayes classifier can be overcome using the SVM by employing all the probability values associated with every category for each document. This method can be used for any dataset and shows a significant reduction in training time as compared to the LSquare method and significant improvement in classification accuracy when compared to pure naïve Bayes systems and also the TF-IDF/SVM hybrids.

PCA‐based land‐use change detection and analysis using multitemporal and multisensor satellite data

Remote‐sensing change detection based on multitemporal, multispectral, and multisensor imagery has been developed over several decades and provided timely and comprehensive information for planning and decision‐making. In practice, however, it is still difficult to select a suitable change‐detection method, especially in urban areas, because of the impacts of complex factors. This paper presents a new method using multitemporal and multisensor data (SPOT‐5 and Landsat data) to detect land‐use changes in an urban environment based on principal‐component analysis (PCA) and hybrid classification methods. After geometric correction and radiometric normalization, PCA was used to enhance the change information from stacked multisensor data. Then, a hybrid classifier combining unsupervised and supervised classification was performed to identify and quantify land‐use changes. Finally, stratified random and user‐defined plots sampling methods were synthetically used to obtain total 966 reference points for accuracy assessment. Although errors and confusion exist, this method shows satisfying results with an overall accuracy to be 89.54% and 0.88 for the kappa coefficient. When compared with the post‐classification method, PCA‐based change detection also showed a better accuracy in terms of overall, producer's, and user's accuracy and kappa index. The results suggested that significant land‐use changes have occurred in Hangzhou City from 2000 to 2003, which may be related to rapid economy development and urban expansion. It is further indicated that most changes occurred in cropland areas due to urban encroachment.

Refining forest classifications in the western Amazon using an intra‐annual multitemporal approach

Seasonal dynamics in the north‐eastern Peruvian Amazon were assessed within a multitemporal LULC framework informed by Landsat 7 ETM+ imagery of the study area for 2001 that coincided with seasonal flooding dynamics. Three images (12 March, 31 May, and 20 September 2001) were classified separately using a hybrid classification method that combined unsupervised and supervised techniques, and attributed with a classification scheme consisting of 18 LULC classes. A multitemporal classification that included 25 LULC classes was created from the three single‐date classifications using a panel analysis technique. While some of the classes describe LULC ‘changes’ (land‐use and land‐cover change (LULCC)) that were stable over time (e.g. low sediment water during March, May, and September 2001), others were complex and included multiple trajectories of change. Panel analysis extracts pixel histories of change over three or more observations as a trend or trajectory, rather than segmenting those changes into piecemeal periods as normally done with from–to change detection. This technique was then assessed by testing hypothesized forest trajectories of LULCC. Traditional quantitative accuracy assessment techniques are less appropriate for panel analysis, and so a qualitative accuracy assessment was performed to evaluate the validity of the multitemporal classification. This study suggests that there may not be one typical year‐round behaviour for seasonal environments, and that population–environment interaction studies would benefit from incorporating this knowledge into future research. This analysis further demonstrates the effectiveness of a multitemporal remote‐sensing approach for gauging landscape fluctuations in seasonal environments.

Stratification of forest density and its validation by NDVI analysis in a part of western Himalaya, India using Remote sensing and GIS techniques

This study deals with the assessment of the status of forest density in the Kangra district of Himachal Pradesh, western Himalaya. A classified forest map of the area with an accuracy of 88.17% was produced using the hybrid classification method in Erdas Imagine. An IRS 1D LISS III satellite image was used for mapping and classification. Forest density was calculated in the ArcGIS environment by overlaying a mesh of uniform resolution cells (500 m×500 m) on a classified forest map. The forest density value of each cell was later used for the preparation of forest density contours. The forest density output was verified by Normalized Difference Vegetation Index (NDVI) analyses. The forest cover of the study area was found to be 34.3%. Baroh area had the highest (45.87%) forest density and Baijnath (18.65%) the lowest. Central and western regions of the district showed high‐value forest density contours (>50%). The derived NDVI values were compared against the forest density classes for assessing the accuracy of the results obtained. A positive correlation (r = 0.99) between NDVI values and forest density confirms the accuracy of the results.

A classification method using a hybrid genetic algorithm combined with an adaptive procedure for the pool of ellipsoids

This paper presents a hybrid classification method that utilizes genetic algorithms (GAs) and adaptive operations of ellipsoidal regions for multidimensional pattern classification problems with continuous features. The classification method fits a finite number of the ellipsoidal regions to data pattern by using hybrid GAs, the combination of local improvement procedures and GAs. The local improvement method adaptively expands, rotates, shrinks, and/or moves the ellipsoids while each ellipsoid is separately handled with a fitness value assigned during the GA operations. A set of significant features for the ellipsoids are automatically determined in the hybrid GA procedure by introducing "don't care" bits to encode the chromosomes. The performance of the method is evaluated on well-known data sets and a real field classification problem originated from a deflection yoke production line. The evaluation results show that the proposed method can exert superior performance to other classification methods such as k nearest neighbor, decision trees, or neural networks.

A hybrid classification method using error pattern modeling

This paper presents a new hybrid classification method using error pattern modeling to improve classification accuracy when the data type of a target variable is binary. The proposed method tries to increase prediction accuracy by combining two different supervised learning methods. That is, the algorithm extracts a subset of training cases that are predicted inconsistently by two methods, and the extracted data subset is used to learn when each method works better. The learned discrimination model is called error pattern models and is used to merge the prediction results of two different methods to generate final prediction. The proposed method has been tested using 13 real-world data sets. The analysis results show that the performance of the proposed method is superior to other hybrid methods and the single usage of existing classification methods such as artificial neural networks and decision tree induction. In particular when prediction inconsistency ratio of the two methods is high, the proposed hybrid method provides significant improvement of prediction accuracy.

Multi‐level Land Cover Mapping of the Twin Cities (Minnesota) Metropolitan Area with Multi‐seasonal Landsat TM/ETM+ Data

Land cover maps, especially vegetation maps, are of increasing interest and use to resource agencies. This paper describes a three‐stage hybrid classification method for regional‐scale multi‐level land cover mapping. The first stage involves an unsupervised classification and stratification. The second stage includes supervised classification of forest types, rule‐based clustering of non‐forested vegetation, and estimation of percent impervious area with a regression model. The third stage is final map generation and post processing. Landsat TM/ETM+ images of three (spring, summer, fall) dates were used to classify land cover of the seven‐county Twin Cities Metropolitan Area of Minnesota into three levels of the modified Minnesota Land Cover Classification System. The overall accuracies for Level‐1 and Level‐2 classes were 95% and 89%, respectively, and the agreement between the estimation of percent impervious surface in Level‐3 classification and the measurements from digital ortho photographs was 96%.

Urban Land-Cover Change Analysis in Central Puget Sound

A methodology was developed to interpret and assess land cover change between 1991 and 1999 in Central Puget Sound, Washington at several scales (landscape, sub-basins, and 90 m grid window) relevant to regional and local decision makers. Land cover data are derived from USGS Landsat (Thematic Mapper and Enhanced Thematic Mapper � ) images of Central Puget Sound. Landsat data were registered, intercalibrated, and corrected for atmosphere and topography to ensure accuracy of land cover change assessment. We apply a hybrid classification method to each image to address the spectral heterogeneity of urbanizing regions. The method combines a supervised classification approach with a spectral unmixing approach to produce seven classes: � 75 percent impervious, 15 to 75 percent impervious, forest, grass, clear cut, bare soil, and water. Land cover change is identified using the direct spatial comparison of classified images derived independently for each time period. We assess that the overall accuracy of each classified image was 91 percent for 1991 and 88 percent for 1999 respectively, which produces an accuracy of 85 percent for the change analysis. Our results show that urban growth over the last decade has produced an overall 6.7 percent increase in paved area.

하이브리드 분류기법을 이용한 위성영상의 분류에 관한 연구

본 논문에서는 위성영상의 분류에 대한 성능 개선을 위하여 ISODATA 클러스터링, 퍼지 C-Means 알고리즘, 베이시안 최대우도 분류기법을 통합한 하이브리드 분류기법을 제안하였다. 본 연구에서는 분석자에 의하여 분류항목별 학습 데이터를 선정한 후 이를 ISODATA 클러스터링을 이용하여 각각의 분류항목별로 분광특징에 따라 학습 데이터를 세분화하여 새로운 학습 데이터를 선정하였다. 새롭게 선정된 학습 데이터를 이용하여 퍼지 C-Means 알고리즘을 이용하여 분류를 수행하고 그 결과를 베이시안 최대우도 분류기의 사전확률로 적용하여 분류를 수행하였다. 그 결과 분석자가 선정한 분류항목별 훈련데이터의 분광적인 특징에 관계없이 분류를 수행할 수 있었으며 위성영상의 분류의 성능을 개선할 수 있었다. 제안된 기법은 Landsat TM 위성영상을 이용하여 그 적용성을 시험하였다. This paper presents hybrid classification method to improve the performance of satellite images classification by combining Bayesian maximum likelihood classifier, ISODATA clustering and fuzzy C-Means algorithm. In this paper, the training data of each class were generated by separating the spectral signature using ISODATA clustering. We can classify according to pixel's membership grade followed by cluster center of fuzzy C-Means algorithm as the mean value of training data for each class. Bayesian maximum likelihood classifier is performed with prior probability by result of fuzzy C-Means classification. The results shows that proposed method could improve performance of classification method and also perform classification with no concern about spectral signature of the training data. The proposed method Is applied to a Landsat TM satellite image for the verifying test.

Computer-Aided Diagnosis of Masses with Full-Field Digital Mammography

Rationale and Objectives The authors developed and evaluated a method of computer-aided diagnosis (CAD) for mass detection with full-field digital mammography (FFDM). Materials and Methods The new CAD method for FFDM employs adaptive, nonlinear multiscale processing and hybrid classification methods. The major strategies are ( a) to “standardize” the mammographic image before it is input to the analysis modules, ( b) to adapt the segmentation of suspicious regions adapt to accommodate different characteristics of masses and mammograms, and ( c) to use combined “hard” and “soft” decision making in discriminating between mass and normal tissue regions. Two data sets of diagnostic FFDM mammograms were used. The training data set includes 36 normal and 24 abnormal mammograms (34 masses), and the testing data set includes 24 normal and 10 abnormal mammograms (10 masses). The tumors in this diagnostic database were more subtle and difficult to detect than those in screening databases the authors have used before. Results With the limited database and a partial optimization, a sensitivity of 91% was obtained in training, with a false-positive rate of 3.21 per image. At this trained operating point of the CAD system, six of 10 subtle masses were detected in testing. Conclusion The CAD algorithms developed in screen-film mammography can be modified for FFDM. More data analysis and system optimization and evaluation will be needed before CAD can be integrated efficiently into the performance of FFDM.

A hybrid classification method: discrete canonical variate analysis using a genetic algorithm

This paper describes a novel, hybrid multivariate classification method: discrete canonical variate analysis (DCVA), which is integrated in the present implementation with a genetic algorithm (GA). DCVA transforms a multivariate data set into a set of discrete scores of lower dimensionality, intended specifically to act as classifiers of observations into one out of multiple pre-defined groups. The condition for selecting the DCVA loadings is maximization of the ratio of the between-groups to within-groups variance of the scores, but unlike conventional CVA, there is a non-linear, discontinuous relationship between the scores and loadings. The performance of the DCVA method is compared with that of two competing classification methods, Artificial Neural Networks (ANNs) and Mahalanobis distance-based Linear discriminant analysis (LDA) using six example problems. In all cases, internal (leave-one-out) cross-validation was used, and classification success rates retained from both the training and test segments. Of the methods studied, DCVA clearly performed the best in training, producing the highest mean success rates for four out of the six example data sets. For the test segments, DCVA produced the best performance for two of the data sets, and equalled that of LDA and ANN for a third. However, LDA produced the best performance from the remaining three data sets. This is suggestive of a greater tendency of DCVA, like other search-based methods, to overfit.

Hybrid consensus theoretic classification

Hybrid classification methods based on consensus from several data sources are considered. Each data source is at first treated separately and modeled using statistical methods. Then weighting mechanisms are used to control the influence of each data source in the combined classification. The weights are optimized in order to improve the combined classification accuracies. Both linear and nonlinear optimization methods are considered and used in classification of two multisource remote sensing and geographic data sets. A nonlinear method which utilizes a neural network gives excellent experimental results. The hybrid statistical/neural method outperforms all other methods in terms of test accuracies in the experiments.

Nontraditional approaches to statistical classification: Some perspectives on L_p-norm methods

The body of literature on classification methods which estimate boundaries between the groups (classes) by optimizing a function of the L p -norm distances of observations in each group from these boundaries, is maturing fast. The number of published research articles on this topic, especially on mathematical programming (MP) formulations and techniques for L p -norm classification, is now sizable. This paper highlights historical developments that have defined the field, and looks ahead at challenges that may shape new research directions in the next decade. In the first part, the paper summarizes basic concepts and ideas, and briefly reviews past research. Throughout, an attempt is made to integrate a number of the most important L p -norm methods proposed to date within a unified framework, emphasizing their conceptual differences and similarities, rather than focusing on mathematical detail. In the second part, the paper discusses several potential directions for future research in this area. The long-term prospects of L p -norm classification (and discriminant) research may well hinge upon whether or not the channels of communication between on the one hand researchers active in L p -norm classification, who tend to have their roots primarily in the decision sciences, the management sciences, computer science and engineering, and on the other hand practitioners and researchers in the statistical classification community, will be improved. This paper offers potential reasons for the lack of communication between these groups, and suggests ways in which L p -norm research may be strengthened from a statistical viewpoint. The results obtained in L p -norm classification studies are clearly relevant and of importance to all researchers and practitioners active in classification and discriminant analysis. The paper also briefly discusses artificial neural networks, a promising non-traditional method for classification which has recently emerged, and suggests that it may be useful to explore hybrid classification methods that take advantage of the complementary strengths of different methods, e.g., neural network and L p -norm methods.