Foreword
Advances in sensor technology have resulted in remote sensing datasets of improved spatial and spectral resolutions. The availability of such datasets is putting renewed focus on the development of efficient methods for image analysis and classification, like those that are based upon Artifi cial Intelligence techniques (AI). The goal of this special issue of PE&RS is to showcase recent developments in AI theory and methodologies for remote sensing data processing, thus demonstrating the diversity of approaches and applications.
The first paper by Lippitt et al. investigates machine learning methods in deciduous forest classifi cation. Multitemporal Landsat Enhanced Thematic Mapper-plus (ETM+) imagery from central Massachusetts is used to assess the performance of three Artifi cial Neural Networks --- Multi-Layer Perceptron, ARTMAP, Self-Organizing Map -- and two Classifi cation Tree splitting algorithms -- gini and entropy rules.\
Walton assesses machine learning techniques for subpixel estimation of urban forest canopy cover and impervious surface cover. Landsat 7 ETM+ imagery from Syracuse, New York is evaluated using Cubist, Random Forests and support vector regression. Each method is compared for its effectiveness and suggestions are made.
Our third paper, Zhao et al. focus on hyperspectral imagery. The authors present a kernel-based method, the Bayesian Gaussian processes for supervised learning. Their classifiers are compared with the K-nearest neighbor and Support Vector Machines as applied to an Airborne Visible/Infrared Imaging Spectroradiometer (AVIRIS) image from northern Indiana.
Qiu’s work also focuses on hyperspectral data analysis using a Hyperion image from Wuxi, China. A neuro-fuzzy system is discussed -- the Gaussian Fuzzy Learning Vector Quantization -- based on the synergistic integration of neural networks and fuzzy systems. The results obtained from the neuro-fuzzy system are compared to those from conventional statistics-based and endmember-based classifiers.
Eddy et al. investigate a machine learning classifi er for very high spatial resolution (1.25mm) ground-based hyperspectral images. A hybrid segmentation Artificial Neural Network method is compared to standard Maximum Likelihood Classification for improving crop/weed species discrimination in southern Alberta, Canada.
Providing a more theoretical insight into AI methods, Stathakis and Kanellopoulos examine the effect of pruning (node reduction) on neural networks in remote sensing applications. Different pruning methods are compared, such as Optimal Brain Damage and Optimal Brain Surgeon, with pruning based on a genetic algorithm. The authors fused four Landsat 7 ETM+ bands with DEM information to classify land use at the Lefkas Island in Greece.
The last paper by Shan et al. investigates the use of genetic algorithms to enhance the effi ciency of transition rule calibration in cellular automata urban growth modeling. The cellular automata model uses multitemporal satellite imagery and population density to predict urban growth in Indianapolis, Indiana.
The papers included in this special issue underwent a rigorous two-stage review process, and were selected from among numerous submissions, with the final acceptance rate being below 20%. We would like to thank our reviewers for their valuable contributions, and we hope that you find this thematically focused issue a valuable reference.
Dr. Anthony Stefanidis
Department of Geography and Geoinformation Science
George Mason