Modified Hopfield Neural Network Classification Algorithm For Satellite Images
Loading...
Date
2016-05
Authors
Ahmed Asal Kzar
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Water is an essential material for living creatures. Human activities and natural influences have an effecting on water quality, and this is considered one of the largest problems facing living forms. Several techniques have been used to address this problem. Remote sensing images have proven their efficiency for water quality mapping. Artificial neural networks have been proven their efficiency as mathematical models with remote sensing techniques for water quality mapping. Hopfield neural network type is a simple, common, fast and efficient network, but, this network is unable to deal with colour images. The researcher in this work addressed this problem through a new modification of this network type. Named MHNNA, which is suitable for classification of colour images. The modification is based on three factors: a zero diagonal weight matrix, HNN as a feedforward network, and weights of bits in the energy function equation. It solves the local minimum problem, time consuming, and it deals with the numbers of image bands deeply. The efficiency of the new algorithm with the sampled color satellite images for water quality mapping has been proven by data validation. MHNNA has been tested with noisy images, proving steadfast until 66% of 30 quantities of noise were added. Two images of THEOS type and two of ALOS type with different dates have been considered for remote sensing images. In addition, a new algorithm is developed to estimate pollutant concentrations by adding new steps to MHNNA for estimation. The efficiency of the algorithm with the colour satellite images for water quality mapping has been proven via validation data that showed higher values of the correlation coefficient (R), reaching up to 0.996 for TSS and lower values of the root mean square error (RMSE), decreasing to 0.564 mg/m^3 for chlorophyll, when compared with the standard method, which is the minimum distance classifier for classification and multi-regression for estimation.
Description
Keywords
Remote sensing images have proven their efficiency , for water quality mapping.