Oil Spill Detection Using Convolutional Neural Network

Abstract

Oil spills from large tankers, vessels, and pipeline cracks lead to significant harm and damage to the marine ecosystem. The persistent threat of oil spills necessitates advanced detection methods to safeguard ecosystems and economies. In this study, we proposed a classification technique using a convolutional neural network (CNN) framework for the identification and classification of oil spill images. Traditional approaches grapple with complex patterns and varying conditions, prompting us to harness CNN's proficiency in image recognition. The model, feature extraction, and classification were rigorously evaluated using accuracy, precision, recall, and F1-score metrics. Employing a dataset containing labeled instances of oil and non-oil spills, the classification technique using CNN achieved an accuracy of 94.30%, precision of 83.01%, recall of 88.70%, and an F1-score of 85.70%. These results underscore the proposed potential for accurate oil spill detection. The comparative analysis revealed that while the proposed technique had a slightly lower accuracy than one existing model of 96% and 92%, it excelled in the precision of 79% and 76%, recall of 80% and 84%, and F1-score of both 80%. This highlights its potential as a valuable tool for oil spill detection, offering a more balanced approach to minimizing false positives and false negatives.