Dyah Erny Herwindiati

Robust Multivariate Outlier Labeling

A criterion for robust estimation of location and covariance matrix is considered, and its application in outlier labeling is discussed. This method, unlike the methods based on MVE and MCD, is applicable to large and high-dimension data sets. The method proposed here is also robust and has the same breakdown point as the MVE- and MCD-based methods. Furthermore, the computational complexity of the proposed method is significantly smaller than that of other methods.

Performance of robust two-dimensional principal component for classification

The robust dimension reduction for classification of two dimensional data is discussed in this paper. The classification process is done with reference of original data. The classifying of class membership is not easy when more than one variable are loaded with the same information, and they can be written as a near linear combination of other variables. The standard approach to overcome this problem is dimension reduction. One of the most common forms of dimensionality reduction is the principal component analysis (PCA). The two-dimensional principal component (2DPCA) is often called a variant of principal component. The image matrices were directly treated as 2D matrices; the covariance matrix of image can be constructed directly using the original image matrices. The presence of outliers in the data has been proved to pose a serious problem in dimension reduction. The first component consisting of the greatest variation is often pushed toward the anomalous observations. The robust minimizing vector variance (MW) combined with two dimensional projection approach is used for solving the problem. The computation experiment shows the robust method has the good performances for matrix data classification.

The New Measure of Robust Principal Component Analysis

Principal Component Analysis (PCA) is a technique to transform the original set of variables into a smaller set of linear combinations that account for most of the original set variance. The data reduction based on the classical PCA is fruitless if outlier is present in the data. The decomposed classical covariance matrix is very sensitive to outlying observations. ROBPCA is an effective PCA method combining two advantages of both projection pursuit and robust covariance estimation. The estimation is computed with the idea of minimum covariance determinant (MCD) of covariance matrix. The limitation of MCD is when covariance determinant almost equal zero. This paper proposes PCA using the minimum vector variance (MVV) as new measure of robust PCA to enhance the result. MVV is defined as a minimization of sum of square length of the diagonal of a parallelotope to determine the location estimator and covariance matrix. The usefulness of MVV is not limited to small or low dimension data set and to non-singular or singular covariance matrix. The MVV algorithm, compared with FMCD algorithm, has a lower computational complexity; the complexity of VV is of order O(p 2).

Robust Kurtosis Projection Approach for Mangrove Classification

Mangroves are coastal vegetations that grow at the interface between land and sea. It can be found in tropical and subtropical tidal areas. Mangrove ecosystems have many ecological roles spans from forestry, fisheries, environmental conservation. The Indonesian archipelago is home to a large mangrove population which has enormous ecological value. This paper discusses mangrove land detection in the North Jakarta from Landsat 8 satellite imagery. One of the special characteristics of mangroves that are distinguishing them from another vegetation is their growing location. This characteristic makes mangrove classification using satellite imagery non trivial task. We need an advanced method that can confidently detect the mangrove ecosystem from the satellite images. The objective of this paper is to propose the robust algorithm using projection kurtosis and minimizing vector variance for mangrove land classification. The evaluation classification provides that the proposed algorithm has a good performance.

Sugarcane Land Classification with Satellite Imagery using Logistic Regression Model

This paper discusses the classification of sugarcane plantation area from Landsat-8 satellite imagery. The classification process uses binary logistic regression method with time series data of normalized difference vegetation index as input. The process is divided into two steps: training and classification. The purpose of training step is to identify the best parameter of the regression model using gradient descent algorithm. The best fit of the model can be utilized to classify sugarcane and non-sugarcane area. The experiment shows high accuracy and successfully maps the sugarcane plantation area which obtained best result of Cohens Kappa value 0.7833 (strong) with 89.167% accuracy.

Combining ground-based data and MODIS data for rice crop estimation in Indonesia

In this study, ground based data from spectroradiometer International Light type ILT900 combined with remotely sensed data from MODIS (Moderate Resolution Imaging Spectrometer) sensor of experimental farmland of the Ministry of Agriculture Republic of Indonesia in Sukamandi, Subang, West Java were used as input data for rice crop estimation using regression analysis. We chose four spectral bands (1-4) of MODIS data and four spectral bands of spectroradiometer data with same (the most similar) wavelength with chosen MODIS data. In addition to the spectral reflectance measurements, we also measured rice production data from several 7 × 20 plot areas that contain different rice varieties and different fertilizer compositions. The data from spectroradiometer then used for estimating regression model based on two approaches, Principal Component Regression (PCR) and Partial Least Square Regression (PLSR). The evaluation on ground-based data shows that PCR and PLSR give good accuracy with r2 = 0.968 and 0.984 respectively.

Robust discriminant analysis for classification of remote sensing data

This paper discusses the classic and robust discriminant analysis algorithm applied to the classification of rice fields, water, buildings, and bare land areas. Discriminant Analysis for multiple groups is often done. This method relies on the sample averages and covariance matrices computed from the training sample. Since sample averages and covariance matrices are not robust, it has been proposed to use robust estimators and covariance instead. In order to obtain a robust procedure with high breakdown point for discriminant analysis, the classical estimators are replaced by Feasible Solution Algorithm (FSA). The input data is a time-series of Landsat 8 Normalize Difference Vegetation Index (NDVI). The classification process is guided over two steps, training and classification. The purpose of the training step is to produce discriminant functions using FSA estimators, and the purpose of the classification step is to classify rice fields, water, buildings and bare land areas. The aim of this paper is to measure the accuracy of Classic and Robust Discriminant Analysis to classify the rice fields, water, buildings and bare land areas from Landsat 8 NDVI time series.

An efficient and effective robust algorithm for the classification of Jakarta vegetation area

This paper discusses an efficient and effective robust algorithm applied to the classification of vegetation areas in the Jakarta Province. The input data is remote sensing data from the Landsat 7 Satellite. The classification process is guided over two steps, training and classification. The purpose of the training step is to determine the reference spectra of the vegetation area, and the purpose of the classification step is to classify Jakarta areas as either vegetation or nonvegetation. An efficient robust approach is used to classify the Jakarta area using the anomolous digital number resulting from a failed instrument. This paper discusses the application of an efficient and effective robust method to classify the remote sensing data with anomolous or inconsistent observations. The aim is to propose a new efficient subset robust approach - the subset minimum vector variance - to classify the vegetation area of Jakarta. The minimum vector variance (MVV) is a robust method having a minimum of the square of the length of a parallelotope diagonal.