Unsupervised Classification Algorithm Research Articles

MDCGen: Multidimensional Dataset Generator for Clustering

We present a tool for generating multidimensional synthetic datasets for testing, evaluating, and benchmarking unsupervised classification algorithms. Our proposal fills a gap observed in previous approaches with regard to underlying distributions for the creation of multidimensional clusters. As a novelty, normal and non-normal distributions can be combined for either independently defining values feature by feature (i.e., multivariate distributions) or establishing overall intra-cluster distances. Being highly flexible, parameterizable, and randomizable, MDCGen also implements classic pursued features: (a) customization of cluster-separation, (b) overlap control, (c) addition of outliers and noise, (d) definition of correlated variables and rotations, (e) flexibility for allowing or avoiding isolation constraints per dimension, (f) creation of subspace clusters and subspace outliers, (g) importing arbitrary distributions for the value generation, and (h) dataset quality evaluations, among others. As a result, the proposed tool offers an improved range of potential datasets to perform a more comprehensive testing of clustering algorithms.

Classification of North Africa for Use as an Extended Pseudo Invariant Calibration Sites (EPICS) for Radiometric Calibration and Stability Monitoring of Optical Satellite Sensors

Pseudo invariant calibration sites (PICS) have been extensively used for the radiometric calibration and temporal stability monitoring of optical satellite sensors. Due to limited knowledge about the radiometric stability of North Africa, only a limited number of sites in the region are used for this purpose. This work presents an automated approach to classify North Africa for its potential use as an extended PICS (EPICS) covering vast portions of the continent. An unsupervised classification algorithm identified 19 “clusters” representing distinct land surface types was used; three clusters were identified with spatial uncertainties within approximately 5% in the shorter wavelength bands and 3% in the longer wavelength bands. A key advantage of the cluster approach is that large numbers of pixels are aggregated into contiguous homogeneous regions sufficiently distributed across the continent to allow multiple imaging opportunities per day, as opposed to imaging a typical PICS once during the sensor’s revisit period. This potential increase in temporal resolution could result in increased sensitivity for the quicker identification of changes in sensor response.

Unsupervised seismic facies classification applied to a presalt carbonate reservoir, Santos Basin, offshore Brazil

Mapping of seismic and lithological facies is a very complex process, especially in regions with low seismic resolution caused by extensive salt layers, even when only an exploratory view of the distribution of the reservoir facies is required. The aim of this study was to apply multi-attribute analysis using an unsupervised classification algorithm to map the carbonate facies of an exploratory presalt area located in the Outer high region of the Santos Basin. The interval of interest is the Barra Velha Formation, deposited during the Aptian, which represents an intercalation of travertines, stromatolites, grainstones and spherulitic packstones, mudstones, and authigenic shales, which were deposited under hypersaline lacustrine conditions during the sag phase. A set of seismic attributes, calculated from a poststack seismic amplitude volume, was used to characterize geological and structural features of the study area. We applied k-means clustering in an approach for unsupervised seismic facies classification. Our results show that at least three seismic facies can be differentiated, representing associations of buildup lithologies, aggradational or progradational carbonate platforms, and debris facies. We quantitatively evaluated the seismic facies against petrophysical properties (porosity and permeability) from available well logs. Seismic patterns associated with the lithologies helped identify new exploration targets.

Unsupervised PolSAR image classification based on sparse representation

ABSTRACTA novel unsupervised image classification algorithm which based on the sparse representation theory for polarimetric synthetic aperture radar (PolSAR) image is introduced in this paper. The algorithm conjunctively uses sparse representation-based classification (SRC) theory, dictionary updating method, and label smoothness constraint to update class labels. The unsupervised H//A Wishart classification method is introduced to provide the preliminary classification result, from which the initial dictionary and class labels can be extracted. An energy function is defined, and it contains two terms. The first term is based on the sparse representation theory. It reflects the cost of assigning different class labels to a pixel. The second term is label smoothness constraint. It constrains that class labels of neighbouring pixels in flat regions should be the same. By alternately minimizing the energy function, two unknown variables, dictionary and class labels are updated. Optimized class labels are the outputs to compose the final classification result. Extensive experimental results for three PolSAR datasets are analysed to verify the validity of the proposed method. Comparison with other unsupervised/supervised classification methods indicates its superiority.

Online image classification and analysis using OGC web processing service

The online geoprocessing and analysis using state-of-the-art technology are offering automated analytical tools to a large group of users. This paper describes an online geoprocessing and analysis framework, which allows an inexperienced user to perform unsupervised classification of remote sensing data. The parallel computing based geoprocessing and analysis framework adopted in this work has been implemented using Free and Open Source Software for Geospatial (FOSS4G). Web Processing Service (WPS) based geoprocessing framework facilitates the deployment of unsupervised classification algorithm on the web in a standardized way. However, it is dynamic in nature to deploy other geoprocessing algorithms. The developed system describes how to process remote sensing data (Sentinel-2, Landsat-8, etc.) for classification and sharing the interoperable results in a distributed environment. To validate the classification results, a prototype architecture based on participatory GIS is developed for field data collection, accuracy assessment and online dissemination. The accuracy assessment (i.e., overall accuracy, Kappa coefficient) is performed to validate the derived classification results using collected field data.

Intertidal area classification with generalized extreme value distribution and Markov random field in quad-polarimetric synthetic aperture radar imagery

Classification of intertidal area in synthetic aperture radar (SAR) images is an important yet challenging issue when considering the complicatedly and dramatically changing features of tidal fluctuation. The difficulty of intertidal area classification is compounded because a high proportion of this area is frequently flooded by water, making statistical modeling methods with spatial contextual information often ineffective. Because polarimetric entropy and anisotropy play significant roles in characterizing intertidal areas, in this paper we propose a novel unsupervised contextual classification algorithm. The key point of the method is to combine the generalized extreme value (GEV) statistical model of the polarization features and the Markov random field (MRF) for contextual smoothing. A goodness-of-fit test is added to determine the significance of the components of the statistical model. The final classification results are obtained by effectively combining the results of polarimetric entropy and anisotropy. Experimental results of the polarimetric data obtained by the Chinese Gaofen-3 SAR satellite demonstrate the feasibility and superiority of the proposed classification algorithm.

A ring in a shell: the large-scale 6D structure of the Vela OB2 complex

Context. The Vela OB2 association is a group of ∼10 Myr stars exhibiting a complex spatial and kinematic substructure. The all-sky Gaia DR2 catalogue contains proper motions, parallaxes (a proxy for distance), and photometry that allow us to separate the various components of Vela OB2. Aims. We characterise the distribution of the Vela OB2 stars on a large spatial scale, and study its internal kinematics and dynamic history. Methods. We make use of Gaia DR2 astrometry and published Gaia-ESO Survey data. We apply an unsupervised classification algorithm to determine groups of stars with common proper motions and parallaxes. Results. We find that the association is made up of a number of small groups, with a total current mass over 2330 M⊙. The three-dimensional distribution of these young stars trace the edge of the gas and dust structure known as the IRAS Vela Shell across ∼180 pc and shows clear signs of expansion. Conclusions. We propose a common history for Vela OB2 and the IRAS Vela Shell. The event that caused the expansion of the shell happened before the Vela OB2 stars formed, imprinted the expansion in the gas the stars formed from, and most likely triggered star formation.

Hyperspectral Image Classification Using Parallel Autoencoding Diabolo Networks on Multi-Core and Many-Core Architectures

One of the most important tasks in hyperspectral imaging is the classification of the pixels in the scene in order to produce thematic maps. This problem can be typically solved through machine learning techniques. In particular, deep learning algorithms have emerged in recent years as a suitable methodology to classify hyperspectral data. Moreover, the high dimensionality of hyperspectral data, together with the increasing availability of unlabeled samples, makes deep learning an appealing approach to process and interpret those data. However, the limited number of labeled samples often complicates the exploitation of supervised techniques. Indeed, in order to guarantee a suitable precision, a large number of labeled samples is normally required. This hurdle can be overcome by resorting to unsupervised classification algorithms. In particular, autoencoders can be used to analyze a hyperspectral image using only unlabeled data. However, the high data dimensionality leads to prohibitive training times. In this regard, it is important to realize that the operations involved in autoencoders training are intrinsically parallel. Therefore, in this paper we present an approach that exploits multi-core and many-core devices in order to achieve efficient autoencoders training in hyperspectral imaging applications. Specifically, in this paper, we present new OpenMP and CUDA frameworks for autoencoder training. The obtained results show that the CUDA framework provides a speed-up of about two orders of magnitudes as compared to an optimized serial processing chain.

The GALAH survey: a catalogue of carbon-enhanced stars and CEMP candidates

Swan bands - characteristic molecular absorption features of the C$_2$ molecule - are a spectroscopic signature of carbon-enhanced stars. They can also be used to identify carbon-enhanced metal-poor (CEMP) stars. The GALAH (GALactic Archaeology with Hermes) is a magnitude-limited survey of stars producing high-resolution, high signal-to-noise spectra. We used 627,708 GALAH spectra to search for carbon-enhanced stars with a supervised and unsupervised classification algorithm, relying on the imprint of the Swan bands. We identified 918 carbon-enhanced stars, including 12 already described in the literature. An unbiased selection function of the GALAH survey allows us to perform a population study of carbon-enhanced stars. Most of them are giants, out of which we find 28 CEMP candidates. A large fraction of our carbon-enhanced stars with repeated observations show variation in radial velocity, hinting that there is a large fraction of variables among them. 32 of the detected stars also show strong Lithium enhancement in their spectra.

A COMPARATIVE STUDY OF ADVANCED LAND USE/LAND COVER CLASSIFICATION ALGORITHMS USING SENTINEL-2 DATA

Abstract. Land Use/ Land Cover (LU/LC) is a major driving phenomenon of distributed ecosystems and its functioning. Interpretation of remote sensor data acquired from satellites requires enhancement through classification in order to attain better results. Classification of satellite products provides detailed information about the existing landscape that can also be analyzed on temporal basis. Image processing techniques acts as a platform for analysis of raw data using supervised and unsupervised classification algorithms. Classification comprises two broad ranges in which, the analyst specifies the classes by defining the training sites called supervised classification where as automatically clustering of pixels to the defined number of classes namely the unsupervised classification. This study attempts to perform the LU/LC classification for Paonta Sahib region of Himachal Pradesh which is a major industrial belt. The data obtained from Sentinel 2A, from which the stacked bands of 10 m resolution are only used. Various classification algorithms such as Minimum Distance, Maximum Likelihood, Parallelepiped and Support Vector Machine (SVM) of supervised classifiers and ISO Data, K-Means of unsupervised classifiers are applied. Using the applied classification results, accuracy assessment is estimated and compared. Of these applied methods, the classification method, maximum likelihood provides highest accuracy and is considered to be the best for LU/LC classification using Sentinel-2A data.

An Unsupervised Classification Algorithm for Multi-Temporal Irrigated Area Mapping in Central Asia

Sound water resources planning and management requires adequate data with sufficient spatial and temporal resolution. This is especially true in the context of irrigated agriculture, which is one of the main consumptive users of the world’s freshwater resources. Existing remote sensing methods for the management of irrigated agricultural systems are often based on empirical cropland data that are difficult to obtain, and that put into question the transferability of mapping algorithms in space and time. Here we implement an automatic irrigation mapping procedure in Google Earth Engine that uses surface reflectance satellite imagery from different sensors. The method is based on unsupervised training of a pixel-by-pixel classification algorithm within image regions identified through unsupervised object-based segmentation, followed by multi-temporal image analysis to distinguish productive irrigated fields from non-productive and non-irrigated areas. Ground-based data are not required. The final output of the mapping algorithm are monthly and annual irrigation maps (30 m resolution). The novel method is applied to the Central Asian Chu and Talas River Basins that are shared between upstream Kyrgyzstan and downstream Kazakhstan. We calculate the development of irrigated areas from 2000 to 2017 and assess the classification results in terms of robustness and accuracy. Based on seven available validation scenes (in total more than 2.5 million pixels) the classification accuracy is 77–96%. We show that on the Kyrgyz side of the Talas basin, the identified increasing trends over the years are highly significant (23% area increase between 2000 and 2017). In the Kazakh parts of the basins the irrigated acreages are relatively stable over time, but the average irrigation frequency within Soviet-era irrigation perimeters is very low, which points to a poor physical condition of the irrigation infrastructure and inadequate water supply.

Solving Classification Problems Using Projection-Based Learning Algorithm with Fuzzy Radial Basis Function Neural Network

Radial basis function (RBF) is combined with fuzzy C-means algorithms and its learning process made by projection-based learning (PBL) has been proposed in this paper, which is pointed out as PBL-fuzzy radial basis function (PBL-FRBF). The proposed method PBL-FRBF is producing good performances by selecting appropriate center and its width in order to achieve it by unsupervised classification algorithms instead of random selection. The PBL decreases the learning time, finds optimum output weight by its energy function and prefers smallest amount of samples for testing. Performance analysis is evaluated by benchmark datasets for classification problem taken from the UCI machine learning repository. The performance of the proposed PBL-FRBF has produced superior results when compared with FRBF and RBF for classification problems.

Merging of Classifiers for Enhancing Viable vs Non-Viable Tissue Discrimination on Human Injuries.

Non-invasive optical imaging techniques have been recently proposed for distinguishing between different types of tissue in burns generated in porcine models. These techniques are designed to assist surgeons during the process of burn debridement, to identify regions requiring excision and their appropriate excision depth. This paper presents a machine learning tool for discriminating between Viable and Non- Viable tissues in human injuries. This tool merges a supervised (QDA) with an unsupervised (k-means clustering) classification algorithms. This combination improves the Non-Viable tissue detection in 23.7% with respect to a simple QDA classifier.

Unsupervised Classification Algorithm for Early Weed Detection in Row-Crops by Combining Spatial and Spectral Information

In agriculture, reducing herbicide use is a challenge to reduce health and environmental risks while maintaining production yield and quality. Site-specific weed management is a promising way to reach this objective but requires efficient weed detection methods. In this paper, an automatic image processing has been developed to discriminate between crop and weed pixels combining spatial and spectral information extracted from four-band multispectral images. Image data was captured at 3 m above ground, with a camera (multiSPEC 4C, AIRINOV, Paris) mounted on a pole kept manually. For each image, the field of view was approximately 4 m × 3 m and the resolution was 6 mm/pix. The row crop arrangement was first used to discriminate between some crop and weed pixels depending on their location inside or outside of crop rows. Then, these pixels were used to automatically build the training dataset concerning the multispectral features of crop and weed pixel classes. For each image, a specific training dataset was used by a supervised classifier (Support Vector Machine) to classify pixels that cannot be correctly discriminated using only the initial spatial approach. Finally, inter-row pixels were classified as weed and in-row pixels were classified as crop or weed depending on their spectral characteristics. The method was assessed on 14 images captured on maize and sugar beet fields. The contribution of the spatial, spectral and combined information was studied with respect to the classification quality. Our results show the better ability of the spatial and spectral combination algorithm to detect weeds between and within crop rows. They demonstrate the improvement of the weed detection rate and the improvement of its robustness. On all images, the mean value of the weed detection rate was 89% for spatial and spectral combination method, 79% for spatial method, and 75% for spectral method. Moreover, our work shows that the plant in-line sowing can be used to design an automatic image processing and classification algorithm to detect weed without requiring any manual data selection and labelling. Since the method required crop row identification, the method is suitable for wide-row crops and high spatial resolution images (at least 6 mm/pix).

An Electronic Registration for Undergraduate Students with Department Selection Based on Artificial Neural Network

The objective of the present research is to facilitate the administrative procedures associated with student registration process, and to ensure equal opportunities for all applicants in college. It aims to assist students in identifying appropriate alternatives available to the departments electronically anytime and anywhere saving time and effort for the student. For this purpose, an intelligent e-government system named "An Electronic Intelligent Registration with Department Selection" (E-IRDS) is designed as one of the intelligent e- services in Iraqi e-governance by using many tools and programming languages which are (PHP, MYSQL, HTML, CSS, XML, NOTPAD++, C#). Artificial neural networks (ANNs) technology is applied, notably Kohonen's self-organizing map (SOM) as one of the important unsupervised classification algorithms of machine learning for classifying and distributing the students automatically into the college academic departments based on their desires, their total degrees, and according to scientific plan for each department, in addition to the specific and personal student information. The applied results based on international standards demonstrated the accuracy of Kohonen's SOM algorithm in classification and distribution methods at least time and possible learning ratio. The system test and assessment results confirmed that it is characterized with a very high security and reliability and accuracy. It is also distinguished with very high efficiency and transparency as well as flexibility and high performance speed. The results also emphasized the ease and availability of the system to all students, besides the possibility of troubleshot and correct errors easily.

Understanding of the Geomorphological Elements in Discrimination of Typical Mediterranean Land Cover Types

Quantification of geomorphometric features is the keystone concern of the current study. The quantification was based on the statistical approach in term of multivariate analysis of local topographic features. The implemented algorithm utilizes the Digital Elevation Model (DEM) to categorize and extract the geomorphometric features embedded in the topographic dataset. The morphological settings were exercised on the central pixel of 3x3 per-defined convolution kernel to evaluate the surrounding pixels under the right directional pour point model (D8) of the azimuth viewpoints. Realization of unsupervised classification algorithm in term of Iterative Self-Organizing Data Analysis Technique (ISODATA) was carried out on ASTER GDEM within the boundary of the designated study area to distinguish 10 morphometric classes. The morphometric classes expressed spatial distribution variation in the study area. The adopted methodology is successful to appreciate the spatial distribution of the geomorphometric features under investigation. The conducted results verified the superimposition of the delineated geomorphometric elements over a given remote sensing imagery to be further analyzed. Robust relationship between different Land Cover types and the geomorphological elements was established in the context of the study area. The domination and the relative association of different Land Cover types in corresponding to its geomorphological elements were demonstrated.

River Levels Derived with CryoSat-2 SAR Data Classification—A Case Study in the Mekong River Basin

In this study we use CryoSat-2 SAR (delay-Doppler synthetic-aperture radar) data in the Mekong River Basin to estimate water levels. Compared to classical pulse limited radar altimetry, medium- and small-sized inland waters can be observed with CryoSat-2 SAR data with a higher accuracy due to the smaller along track footprint. However, even with this SAR data the estimation of water levels over a medium-sized river (width less than 500 m) is still challenging with only very few consecutive observations over the water. The target identification with land–water masks tends to fail as the river becomes smaller. Therefore, we developed a classification approach to divide the observations into water and land returns based solely on the data. The classification is done with an unsupervised classification algorithm, and it is based on features derived from the SAR and range-integrated power (RIP) waveforms. After the classification, classes representing water and land are identified. Better results are obtained when the Mekong River Basin is divided into different geographical regions: upstream, middle stream, and downstream. The measurements classified as water are used in a next step to estimate water levels for each crossing over a river in the Mekong River network. The resulting water levels are validated and compared to gauge data, Envisat data, and CryoSat-2 water levels derived with a land–water mask. The CryoSat-2 water levels derived with the classification lead to more valid observations with fewer outliers in the upstream region than with a land–water mask (1700 with 2% outliers vs. 1500 with 7% outliers). The median of the annual differences that is used in the validation is in all test regions smaller for the CryoSat-2 classification results than for Envisat or CryoSat-2 land–water mask results (for the entire study area: 0.76 m vs. 0.96 m vs. 0.83 m, respectively). Overall, in the upstream region with small- and medium-sized rivers the classification approach is more effective for deriving reliable water level observations than in the middle stream region with wider rivers.

The Sliding Singular Spectrum Analysis: A Data-Driven Nonstationary Signal Decomposition Tool

Singular spectrum analysis (SSA) is a signal decomposition technique that aims at expanding signals into interpretable and physically meaningful components (e.g., sinusoids, noise, etc.). This paper presents new theoretical and practical results about the separability of the SSA and introduces a new method called sliding SSA. First, the SSA is combined with an unsupervised classification algorithm to provide a fully automatic data-driven component extraction method for which we investigate the limitations for components separation in a theoretical study. Second, the detailed automatic SSA method is used to design an approach based on a sliding analysis window, which provides better results than the classical SSA method when analyzing nonstationary signals with a time-varying number of components. Finally, the proposed sliding SSA method is compared to the empirical mode decomposition and to the synchrosqueezed short-time Fourier transform, applied on both synthetic and real-world signals.

Aspect-based opinion mining framework using heuristic patterns

The aspect-based online opinions expressed by users on social media sites have become a popular source of information for consumers regarding their purchase decisions as well as for companies seeking opinions on their products. Therefore, it is important to develop aspect-based opinion mining applications with an emphasis on extracting and classifying the aspect-based opinions expressed by users about products in a given review. Previous studies have used a limited set of heuristic patterns for aspect extraction with both supervised (annotated-dataset-based) and unsupervised (lexical-resource-based) aspect-related sentiment classification algorithms. However, the present study proposes an integrated framework comprising of an extended set of heuristic patterns for aspect extraction, a hybrid sentiment classification module with the additional support of intensifiers and negations, and a summary generator. The performance evaluation of the proposed aspect-based opinion mining system using state-of-the-art methods shows that the proposed system outperforms the alternative methods in terms of better precision, recall and F-measure, since it achieves an average precision of 85%, an average recall of 73% and an average F-measure of 0.78. The comparative results indicate that the proposed technique provides more efficient results for the aspect-sentiment extraction, classification and summary generation of online product reviews.

Identification of selected CITES-protected Araucariaceae using DART TOFMS

Determining the species source of logs and planks suspected of being Araucaria araucana (Molina) K.Koch (CITES Appendix I) using traditional wood anatomy has been difficult, because its anatomical features are not diagnostic. Additionally, anatomical studies of Araucaria angustifolia (Bertol.) Kuntze, Araucaria heterophylla (Salisb.) Franco, Agathis australis (D.Don) Lindl., and Wollemia nobilis W.G.Jones, K.D.Hill & J.M.Allen have reported that these taxa have similar and indistinguishable anatomical characters from A. araucana. Transnational shipments of illegal timber obscure their geographic provenance, and therefore identification using wood anatomy alone is insufficient in a criminal proceeding. In this study we examine the macroscopic appearance of selected members of the Araucariaceae and investigate whether analysis of heartwood chemotypes using Direct Analysis in Real Time (DART) Time-of-Flight Mass Spectrometry (TOFMS) is useful for making species determinations. DART TOFMS data were collected from 5 species (n =75 spectra). The spectra were analyzsed statistically using supervised and unsupervised classification algorithms. Results indicate that A. araucana can be distinguished from the look-alike taxa. Another statistical inference of the data suggests that Wollemia nobilis is more similar and within the same clade as Agathis australis. We conclude that DART TOFMS spectra can help in making species determination of the Araucariaceae even when the geographic provenance is unknown.