Ryuei Nishii

Maximum likelihood principle and model selection when the true model is unspecified

Suppose that independent observations come from an unspecified unknown distribution. Then we consider the maximum likelihood based on a specified parametric family which provides a good approximation of the true distribution. We examine the asymptotic properties of the maximum likelihood estimate and of the maximum likelihood. These results will be applied to the model selection problem.

Enhancement of low spatial resolution image based on high resolution-bands

Thermal infrared measurements of Band 6 acquired by Landsat TM sensor have lower spatial resolution than those of the other six bands. The authors propose a statistical approach to enhance the resolution of low spatial resolution image by using remaining bands. They employ a multivariate normal distribution for the seven band values. The values of Band 6 are predicted by the conditional expectations. Validity of their procedure is examined by mean squared errors based on actual images.

Accuracy and inaccuracy assessments in land-cover classification

Several measures assessing accuracy of land-cover classification are available, e.g., overall and class-averaged accuracies. Also the kappa statistic is widely used for this purpose. The authors discuss properties of these criteria and point out that the kappa statistic has an unfavorable feature. They propose an alternative coefficient based on Kullback-Leibler information. A test statistic for significance difference between coefficients is also established. Further, the Bayes risk, which takes types of misclassifications into account, is discussed. Their assessment measures are examined through actual error matrices by Z. Ma and R. L. Redmond (1995).

Supervised image classification by contextual AdaBoost based on posteriors in neighborhoods

AdaBoost, a machine learning technique, is employed for supervised classification of land-cover categories of geostatistical data. We introduce contextual classifiers based on neighboring pixels. First, posterior probabilities are calculated at all pixels. Then, averages of the log posteriors are calculated in different neighborhoods and are then used as contextual classification functions. Weights for the classification functions can be determined by minimizing the empirical risk with multiclass. Finally, a convex combination of classification functions is obtained. The classification is performed by a noniterative maximization procedure. The proposed method is applied to artificial multispectral images and benchmark datasets. The performance of the proposed method is excellent and is similar to the Markov-random-field-based classifier, which requires an iterative maximization procedure.

Hyperspectral Image Classification by Bootstrap AdaBoost With Random Decision Stumps

We consider a supervised classification of hyperspectral data using AdaBoost with stump functions as base classifiers. We used the bootstrap method without replacement to improve stability and accuracy and to reduce overtraining. We randomly split a data set into two subsets: one for training and the other one for validation. Subsampling and training/validation steps were repeated to derive the final classifier by the majority vote of the classifiers. This method enabled us to estimate variable relevance to the classification. The relevance measure was used to estimate prior probabilities of the variables for random combinations. In numerical experiments with multispectral and hyperspectral data, the proposed method performed extremely well and showed itself to be superior to support vector machines, artificial neural networks, and other well-known classification methods.

A Markov random field-based approach to decision-level fusion for remote sensing image classification

A method is proposed for the enhancement of the quality of a classification result by fusing this result with remote sensing images, based on a Markov random field approach. The classification accuracy is estimated by a modified posterior probability, which is used for choosing the optimal classification result. The procedure is applied to a benchmark dataset for discrimination provided by the IEEE Geoscience and Remote Sensing Society Data Fusion Committee, and it shows an excellent performance. The classified result won the competition of the data fusion contest 2001 held by the same committee.

More precise tables of Srivastava-Chopra balanced optimal 2m fractional factorial designs of resolution V, m≤6

Abstract More precise tables of balanced trace-optimal 2m fractional factorial designs of resolution V for m=4,5,6, which were originally obtained by Srivastava and Chopra, are presented.

Modeling and inference of forest coverage ratio using zero-one inflated distributions with spatial dependence

This paper explores statistical modeling of forest area with two covariates. The forest coverage ratio of grid-cell data was modeled by taking human population density and relief energy into account. The likelihood of the forest ratios was decomposed into the product of two likelihoods. The first likelihood was due to trinomial logistic distributions on three categories: the forest ratios take zero, or one, or values between zero and one. The second one was due to a logistic-normal regression model for the ratios between zero and one. This model was applied to real grid-cell data and it fit better than zero-inflated beta regression models.

Analytical estimation of the convergence point of populations

We propose methods of estimating the convergence point for the moving vectors of individuals between generations or evolution paths and show that the estimated convergence point can be useful information for accelerating evolutionary computation (EC). As the first stage of this new approach, we do not combine the proposed methods with EC search in this paper, but rather evaluate how power an individual the the estimated convergence point is by comparing fitness values. Through experimental evaluations, we show that the estimated point can be a powerful elite for unimodal fitness landscapes and that clustering moving vectors according to the aimed points is the next research target for multimodal fitness landscape.

Verification of deforestation in East Asia by spatial logit models due to population and relief energy

Deforestation is a result of complex causality chains in most cases. But identification of limited number of factors shall provide comprehensive general understanding of the vital phenomenon at a broad scale, as well as projection for the future. Only two factors -- human population size (N) and relief energy (R: difference of minimum altitude from the maximum in a sampled area) -- were found to give sufficient elucidation of deforestation by nonlinear logit regression models, whose functional forms were suggested by step functions fitted to one-kilometer square high precision grid-cell data in Japan (n=6825). Likelihood with spatial dependency was derived, and several deforestation models were selected for the application to East Asia by calculating relative appropriateness to data. For the measure of appropriateness, Akaikes Information Criterion (AIC) was used. Logit model is employed so as to avoid anomaly in asymptotic lower and upper bounds. Therefore the forest areal rate, 0 < F < 1. To formulate East-Asian dataset, landcover dataset estimated from NOAA observations available at UNEP, Tsukuba for F, gridded population of the world of CIESIN, US for N, and GTOPO30 of USGS for R, were used. The resolutions were matched by taking their common multiple of 20 minutes square. It was suggested that data of full forest coverage, F=1.0, which were not dealt in calculations due to logit transformation this time, should give important role in stabilizing parameter estimations.