ABSTRACT

Early detection of oil spills and quick interventions are key elements in reducing the menace caused by oil spills in our environment. In this research work, oil spills classification system based on laser fluorosensor spectra data was modeled and simulated. This is to distinguish oil spills from various backgrounds and to classify oil spills into different products. Artificial Neural Network (ANN) toolbox in Matlab/Simulink with MLP (multi-layer perceptron) based supervised architecture was used for the simulation. Using the data in form of 90-channel spectra as inputs, the ANN presents the analysis and estimation results of oil products and various backgrounds as outputs. The network was trained to understand numerous spectra data of laser fluorosensor for different oil spill products (light oil, medium oil, and heavy oil) and various backgrounds (water, sand and stone). The trained network was tested using data set to the network. A back propagation learning algorithm with an optimizer based on gradient descent method was used during the training of the network. The effect of different number of hidden layers (2, 3, and 4) and the different  number of neurons (7,7), (7,7,7), (7,7,7,7), (7,8), (9,7),  (8,10) and (10,10) in hidden layers was checked on the overall performance and accuracy of the MLP artificial neural network and the results were compared. It was found that the ANN with MLP based supervised architecture performed well when the number of neurons in hidden layers is the same ((7,7), (7,7,7), (7,7,7,7), (10,10)) and an average of 100% classification result was achieved. The network however performed badly when it was trained with different number of neurons ((7,8), (9,7), (8,10), in the hidden layers and an average of 18.3% classification result was achieved. It was also found that laser fluorosensors must be operated at wavelengths between 308 nm and 340 nm to produce well distinguished fluorescence spectra. Finally, it has been noted from this research work that laser fluorosensors are currently the most effective remote sensors as they can detect oil spills in almost all backgrounds (land, water, ice, snow etc.).