A New Optimized GA-RBF Neural Network Algorithm for Oil Spill Detection in SAR Images

Abstract

Marine ecosystem is affected seriously by the illegal discharge of oil spills. Marine oil spills can be detected using SAR image processing. Most effective indicators are detection accuracy and efficiency. In marine environment, oil spills are detected using images of Synthetic Aperture Radar (SAR). Cloudiness and conditions of weather cannot affect the images of SAR. Very calm sea area’s backscatter value will most probably equals oil spill’s backscatter value. Oil spill causes short-gravity waves and dampens capillary. Various techniques are used for the detection of oil spills. Dark areas are detected by these techniques. These areas are having a high probability of being an oil spill. These methods involve a lot of non-linearity, which makes the process a complex one. In the input space of multi-dimension, non-linear data can be handled effectively by neural network. The use of NN is getting increased in remote sensing. Well organized explicit relation between output and input are not required by NN. Own relationship is computed in NN. Genetic Algorithm based, a new optimized Radial Basis Function (RBF) neural network algorithm is proposed in this work. It is termed as GA-RBF algorithm. RBF neural network’s structure and weights are optimized by using this genetic algorithm. Hybrid optimizing encoding is done simultaneously. Detection of the Oil spill is done by using various SAR image samples for training. High value of efficiency and accuracy is produced by this proposed technique as shown by experimentation.