Palma Blonda

Multispectral classification by a modular neural network architecture

Deals with the application of a modular neural network system to classification of remote sensed data characterized by a high number of spectral bands. The classification task was separated into two phases: (i) unsupervised data compression by a linear propagation network (LPN); (ii) supervised feature classification by a multi layer perceptron (MLP). In this work the unsupervised LPN module has been introduced to speed up the training phase of the MPL module. The performance of the MLP classifier trained, respectively, with the original uncompressed data and with the data preprocessed by the LPN module have been compared in terms of classification accuracy and computation time. The experimental results prove that even though the overall classification accuracy is comparable in both the experiments, the convergence time spent in the MLP training with the compressed data has been significantly reduced.<<ETX>>

Neural network ensemble and support vector machine classifiers for the analysis of remotely sensed data: a comparison

This paper presents a comparative evaluation between a classification strategy based on the combination of the outputs of a neural (NN) ensemble and the application of Support Vector Machine (SVM) classifiers in the analysis of remotely sensed data. Two sets of experiments have been carried out on a benchmark data set. The first set concerns the application of linear and non linear techniques to the combination of the outputs of a Multilayer Perceptron (MLP) neural network ensemble. In particular, the Bayesian and the error correlation matrix approaches are used for coefficient selection in the linear combination of the networks outputs. A MLP module is used for the non linear outputs combination. The results of linear and non linear combination schemes are compared and discussed versus the performance of SVM classifiers. The comparative analysis evidences that the nonlinear, MLP based, combination provides the best results among the different combination schemes. On the other hand, better performance can be obtained with SVM classifiers. However, the complexity of the SVM training procedure can be considered a limitation for SVMs application to real-world problems.

Detailed comparison of neuro-fuzzy estimation of subpixel land-cover composition from remotely sensed data

Mixed pixels, which do not follow a known statistical distribution that could be parameterized, are a major source of inconvenience in classification of remote sensing images. This paper reports on an experimental study designed for the in-depth investigation of how and why two neuro-fuzzy classification schemes, whose properties are complementary, estimate sub-pixel land cover composition from remotely sensed data. The first classifier is based on the fuzzy multilayer perceptron proposed by Pal and Mitra: the second classifier consists of a two-stage hybrid (TSH) learning scheme whose unsupervised first stage is based on the fully self- organizing simplified adaptive resonance theory clustering network proposed by Baraldi. Results of the two neuro-fuzzy classifiers are assessed by means of specific evaluation tools designed to extend conventional descriptive and analytical statistical estimators to the case of multi-membership in classes. When a synthetic data set consisting of pure and mixed pixels is processed by the two neuro-fuzzy classifiers, experimental result show that: i) the two neuro- fuzzy classifiers perform better than the traditional MLP; ii) classification accuracies of the two neuro-fuzzy classifiers are comparable; and iii) the TSH classifier requires to train less background knowledge than FMLP.

RBF two-stage learning networks exploiting supervised data in the selection of hidden unit parameters: an application to SAR data classification

Radial basis function (RBF) classifiers, which consist of an hidden and an output layer, are traditionally trained with a two-stage hybrid learning approach. This approach combines an unsupervised (data-driven) first stage to adapt RBF hidden layer parameters with a supervised (error-driven) second stage to learn RBF output weights. Several simple strategies that exploit labeled data in the adaptation of centers and spread parameters of RBF hidden units may be pursued. Some of these strategies have been shown to reduce traditional weaknesses of RBF classification, while typical advantages are maintained. In the field of remotely sensed image classification, the authors compare a traditional RBF two-stage hybrid learning procedure with an RBF two-stage learning technique exploiting labeled data to adapt hidden unit parameters.

Fuzzy Neural Networks for Pattern Recognition

The objective of this paper is to discuss a state-of-the-art of methodology and algorithms for integrating fuzzy sets and neural networks in a unique framework for dealing with pattern recognition problems, in particular classification procedures. Methods of pattern recognition are studied in two main streams, namely supervised and unsupervised learning. We propose our own definition of fuzzy neural integrated networks. This criterion is proposed as a unifying framework for comparison of algorithms. In the first part of the this paper, classification methods based on rule sets or numerical data are reviewed, together with specific methods for handling classification in image processing. In the second part of this paper, several fuzzy neural clustering models are reviewed and compared. These models are: i) Self-Organizing Map (SOM); ii) Fuzzy Learning Vector Quantization (FLVQ); iii) Carpenter-Grossberg- Rosen Fuzzy Adaptive Resonance Theory (CGR Fuzzy ART); iv) Growing Neural Gas (GNG); and v) Fully self-Organizing Simplified Adaptive Resonance Theory (FOSART).

Parallel genetic algorithm for the design of neural networks: an application to the classification of remotely sensed data

We consider the problem of classification of remote sensed data from LANDSAT Thematic Mapper images. The data have been acquired in July 1986 on an area locate din South Italy. We compare the performance obtained by feed-forward neural networks designed by a parallel genetic algorithm to determine their topology with the ones obtained by means of a multi-layer perceptron trained with Back Propagation learning rule. The parallel genetic algorithm, implemented on the APE100/Quadrics platform, is based on the coding scheme recently proposed by Sternieri and Anelli and exploits a recently proposed environment for genetic algorithms on Quadrics, called AGAPE. The SASIMD architecture of Quadrics forces the chromosome representation. The coding scheme provides that the connections weights of the neural network are organized as a floating point string. The parallelization scheme adopted is the elitistic coarse grained stepping stone model, with migration occurring only towards neighboring processors. The fitness function depends on the mean square error.After fixing the total number of individuals and running the algorithm on Quadrics architectures with different number of processors, the proposed parallel genetic algorithm displayed a superlinear speedup. We report results obtained on a data set made of 1400 patterns.

8-Band Image Data Processing of the Worldview-2 Satellite in a Wide Area of Applications

Recent years have seen advances in remote sensing in many fields with applications at a spatial scale which range from global to local. As a consequence, the need to observe the Earth with more specialized and sophisticated sensors and data analysis techniques to obtain more accurate information has increased. On the 8th October 2009 a new second nextgeneration Worldview-2 satellite was launched by DigitalGlobe: it represents the latest innovation among sensors for the acquisition of remote sensed imagery. It has an advanced agility due to control moment gyros (like Worldview-1) and combines an average revisiting time of 1.1 days around the globe with a large scale collection capacity. Moreover, it is also the first commercial satellite able to provide panchromatic imagery at 46 cm of spatial resolution and 8-band multispectral imagery at 1.84 m spatial resolution. In addition to the standard panchromatic and multispectral BLUE, GREEN, RED and NEAR INFRARED (NIR1) bands the Worldview-2 sensor has:

Fuzzy neural-network-based segmentation of multispectral magnetic-resonance brain images

This study investigates the applicability of a multimodular neuro-fuzzy system in the multispectral analysis of magnetic resonance (MR) images of the human brain. The system consists of two components: an unsupervised neural module for image segmentation in tissue regions and a supervised module for tissue labeling. The former is the fuzzy Kohonen clustering network (FKCN). The latter is a feed-forward network based on the back-propagation learning rule. The results obtained with the FKCN have been compared with those extracted by a self organizing map (SOM). The system has been used to analyze the multispectral MR brain images of a healthy volunteer. The data set included the proton density (PD), T2, T1 weighted spin-echo (SE) bands and a new T1- weighted three dimensional sequence, i.e. the magnetization- prepared rapid gradient echo (MP-RAGE). One of the main objectives of this study has been to evaluate the usefulness of brain imaging with the MP-RAGE sequence in view of automatic tissue classification. To this purpose, a quantitative evaluation has been provided on the base of some labeled areas selected interactively by a neuro- radiologist from the input raw images. Quantitative results seem to indicate that the MP-RAGE sequence may provide higher tissue separability than the T1-weighted SE sequence.

Feature extraction and pattern classification for remotely sensed data analysis by a modular neural system

In this paper a modular neural network architecture is proposed for classification of Remote Sensed data. The neural network learning task of the supervised Multi Layer Perceptron (MLP) Classifier has been made more efficient by pre-processing the input with an unsupervised feature discovery neural module. Two classification experiments have been carried for coping with two different situations, very usual in real remote sensing applications: the availability of complex data, such as high dimensional and multisourced data, and on the contrary, the case of imperfect low dimensional data set, with a limited number of samples. In the first experiment on a multitemporal data set, the Linear Propagation Network (LPN) has been introduced to evaluate the effectiveness of neural data compression stage before classification. In the second experiment on a poor data set, the Kohonen Self Organising Feature Map (SOM) Network has been introduced for clustering data before labelling. In the paper is also illustrated the criterion for the selection of an optimal number of cluster centres to be used as node number of the output SOM layer. The results of the two experiments have confirmed that modular learning performs better than the non-modular one in learning quality and speed.

Spectral rules and geostatistic features for characterizing olive groves in QuickBird images

The variogram image extracted from the panchromatic band of a QuickBird image has been analyzed to discriminate between similar spectral vegetated land use classes, such as forest and dense olive groves, once spectral features have been exploited. Such images seem to reproduce the direction and inter-row spacing of oriented texton-based textures present in the original input image. The variogram based texture algorithm (VBTA), recently proposed in literature for analyzing aerial images, tries to determine if texture in a region is not oriented, as in the case of forested and grassland areas or is oriented in one or two main directions, as in the case of vineyards and olive groves, respectively. If texture is oriented, the number of orientations, the orientation angle and periods (i.e. the distance between textons in each direction) are extracted, the latter depending on agricultural practices. This paper concentrates on: 1) the experimental selection of some variables useful for variogram analysis: i) the size of the input image moving window for variogram calculation; ii) some threshold values to be used for variogram normalization and for testing the presence/absence of any directionality in the variogram image; 2) The methodology for orientation and period estimation from the cumulative variance of the variogram image. The results obtained for a Southern Italy area characterized by intensive olive trees cultivation reveals the effectiveness of VBTA for QuickBird images.