This thesis investigates the use of different classifiers for integrating remotely sensed data with other spatial data derived from digital elevation models to produce maps showing areas affected by salinity in the south west agricultural region of Western Australia. A method is developed for accurately mapping and monitoring salinity on a broad-scale using costeffective data. In addition, a cost-effective method is presented for predicting areas at risk from salinity over broads regions. Maximum likelihood classification using single-date Landsat Thematic Mapper image is used as a benchmark to determine whether the integration of multi-temporal Landsat data and landform data produces more accurate salinity maps. Decision trees and neural networks are used to map saline areas in the Ryan's Brook catchment, located approximately 50 kilometres southwest of Kojonup, using two Landsat images and two landform attributes (water accumulation and downhill slope). A conditional probabilistic network is then used to impose a known relationship between input attributes and salinity status. In this way, changes in salinity through time can be modelled using all of the available Landsat data. The results show a large improvement on the maximum likelihood, decision tree and neural network classifiers. The network is used to produce a time-series of salinity maps for the upper Blackwood and Frankland-Gordon catchments. These maps are used as inputs to the predictions of salinity risk areas. The prediction of salinity risk areas is approached using decision tree classifiers, so that the derived models can be easily interpreted by end-users. A simple decision tree for predicting salinity risk is developed. Rules are extracted from the decision tree, and refined to form some straightforw...

Abstract — In this paper, we propose the use of Support Vector Machines (SVM) for automatic hyperspectral data classification and knowledge discovery. In the first stage of the study, we use SVMs for crop classification and analyze their performance in terms of efficiency and robustness, as compared to extensively used neural and fuzzy methods. Efficiency is assessed by evaluating accuracy and statistical differences in several scenes. Robustness is analyzed in terms of (a) suitability to working conditions when a feature selection stage is not possible, and (b) performance when different levels of Gaussian noise are introduced at their inputs. In the second stage of this work, we analyze the distribution of the support vectors (SV) and perform sensitivity analysis on the best classifier in order to analyze the significance of the input spectral bands. For classification purposes, six hyperspectral images acquired with the 128-band HyMAP spectrometer during the DAISEX-1999 campaign are used. Six crop classes were labelled for each image. A reduced set of labelled samples is used to train the models and the entire images are used to assess their performance. Several conclusions are drawn: (1) SVMs yield better outcomes than neural networks regarding accuracy, simplicity and robustness; (2) training neural and neurofuzzy models is unfeasible when working with high dimensional input spaces and great amounts of training data; (3) SVMs perform similarly for different training subsets with varying input dimension, which indicates that noisy bands are successfully detected; and (4) a valuable ranking of bands through sensitivity analysis is achieved. Index Terms — Hyperspectral imagery, crop classification, knowledge discovery, Support Vector Machines, neural networks.

This paper presents a method for Data Mining and Knowledge Discovery in Image Data. This method is based on the Self-Organizing Map (SOM) which is an unsupervised artificial neural network algorithm. The SOM possesses unique properties of clustering, classification, modelling and visualization and is used here as a Data Mining tool. This enables us to get informative yet simpler pictures of the image data space whose dimension, complexity and amount are large for human observation alone. For this purpose, a Landsat-5 TM satellite Remotely Sensed Image of the Lower Thames Valley area in the vicinity of Heathrow Airport (London, UK) was used in this study for evaluating the proposed SOM Neural network and comparing its results with conventional classification algorithms which are used in the Remote Sensing field.

Abstract—In this study, satellite-based remote sensing techniques were developed for identifying smoke from forest fires. Both artificial neural networks (NN) and multithreshold techniques were explored for application with imagery from the Advanced Very High Resolution Radiometer (AVHRR) aboard NOAA satellites. The NN was designed such that it does not only classify a scene into smoke, cloud, or clear background, but also generates continuous outputs representing the mixture portions of these objects. While the NN approach offers many advantages, it is time consuming for application over large areas. A multithreshold algorithm was thus developed as well. The two approaches may be employed separately or in combination depending on the size of an image and smoke conditions. The methods were evaluated in terms of Euclidean distance between the outputs of the NN classification, using error matrices, visual inspection, and comparisons of classified smoke images with fire hot spots. They were applied to process daily AVHRR images acquired across Canada. The results obtained in the 1998 fire season were analyzed and compared with fire hot spots and TOMS-based aerosol index data. Reasonable correspondence was found, but the signals of smoke detected by TOMS and AVHRR are quite different but complementary to each other. In general, AVHRR is most sensitive to low, dense smoke plumes located near fires, whereas smoke detected by TOMS is dispersed, thin, elevated, and further away from fires. Index Terms—AVHRR, classification, fire, neural networks, smoke.

. This paper describes some preliminary results of an innovative methodology for classifying agricultural crops using multi-temporal, multi-spectral and multi-source remotely-sensed data. The procedure firstly characterises the individual training class data by considering these data values as a function of time of imaging and waveband. An analytical surface is fitted to these data points, which may be irregularly spaced. Initially a polynomial surface was used, and the coefficients of the surface were input to: an artificial neural network; maximum likelihood algorithm and minimum distance rule procedure to make comparisons. Results show that classification accuracy is significantly improved in comparison with the use of a single date image. However, the high-dimensional bivariate polynomial functions used to characterise the time/waveband distribution of the training data for each class show significant edge effects, and ongoing work is concerned with the comparison of alter...

. Artificial neural networks (ANNs) have been widely used for many classification purposes and generally proved to be more powerful than conventional statistical techniques. However, the use of ANNs requires decisions on the part of the user which may affect the accuracy of the resulting classification. One of these decisions concerns the determination of the optimum network structure for a particular problem. In fact, network structure has a direct effect on the generalisation capabilities of the network. Pruning techniques can be used to reduce network size and thus improve generalisation capabilities. In this study, a feed-forward artificial neural network learning a classification task by backpropagation algorithm was used to classify agricultural crops from microwave SAR and optical SPOT images. Three major pruning algorithms (magnitude based pruning, optimum brain damage, and optimal brain surgeon) were then analysed to find out their performance and visualised to understand thei...

In 1999, the NASA Jet Propulsion Lab presented a proposal for a six satellite navigation and communication network for Mars called the Mars Network. This thesis investigates the performance of the Mars Network both theoretically, using figures of merit commonly applied to satellite navigation systems on Earth, and in the position domain using simulated observations.

Errors associated with visual inspection and interpretation of radargrams often inhibits the intensive surveying of widespread areas using ground-penetrating radar (GPR). To automate the interpretive process, this paper presents an application of a fuzzy-neural network (F-NN) classifier for unsupervised clustering and classification of soil profile using GPR imagery. The classifier clusters and classifies soil profiles strips along a traverse based on common pattern similarities that can relate to physical features of the soil (e.g., number of horizons; depth, texture and structure of the horizons; and relative arrangement of the horizons, etc). This paper illustrates this classification procedure by its application on GPR data, both simulated and actual real-world. Results show that the procedure is able to classify the profile into zones that corresponded with those obtained by visual inspection and interpretation of radargrams. Results indicate that an F-NN model can supply real-time soil profile clustering and classification during field surveys.

Advances in software process technology have rendered many existing methods of size assessment and effort estimation inapplicable. The use of automation in the software process, however, provides an opportunity for the development of more appropriate software size-based effort estimation models. A specification-based size assessment method has therefore been developed and tested in relation to process effort on a preliminary set of systems. The results of the analysis confirm the assertion that, within the automated environment class, specification size indicators (that may be automatically and objectively derived) are strongly related to process effort requirements.

The expansion of urban areas has a negative impact on the environment. The increase of impervious or sealed surfaces is directly proportional to this expansion. The estimation of sealed surfaces has often been executed using remote sensing imagery, although only on a local to regional scale, using medium and recently available high resolution images like LANDSAT TM and IKONOS. In order to develop a global policy and strategy on urban expansion matters, consistent time series of area statistics on urban land use on a national and global level will become indispensable. This research explores the possibilities of SPOT VEGETATION imagery, with a spatial resolution of 1 km, for urban monitoring in order to generate statistics of sealed surfaces over larger zones. While low resolution imagery offers the advantage of covering a large area in small temporal intervals, its spatial resolution is too coarse to monitor most urban objects. In order to tackle this problem, a sub-pixel classification was applied and unmixed sealed surface area statistics were produced. Endmember selection is a key element in the sub-pixel classification process in which the spectrally complex sealed surface class should be distinguished from other general classes. To find the most favourable temporal interval or period for endmember selection, several datasets were developed and explored. SPOT-VEGETATION images were acquired in summer and winter for Flanders (Belgium). This region is characterised by a highly fragmented urban land-cover and large availability of reference data. Spectral unmixing of the multitemporal datasets illustrates that the endmember spectra differs for three different endmember selection techniques, affecting the quality of the final sub-pixel classification. The paper argues that the