AN INTELLIGENT HYBRID DEEP LEARNING FRAMEWORK FOR ROBUST OIL SPILL DETECTION AND MULTI-CLASS CLASSIFICATION IN SYNTHETIC APERTURE RADAR IMAGERY

Marine oil spills constitute a persistent and devastating environmental crisis, demanding rapid and accurate detection for effective mitigation. While Synthetic Aperture Radar (SAR) satellite imagery is the premier tool for all-weather monitoring, distinguishing oil spills from spectrally similar look-alike phenomena remains a significant challenge for automated systems. This paper introduces a novel, end-to-end hybrid deep-learning framework designed to address this critical problem. Our methodology integrates a semantic segmentation stage using a DeepLabV3+ architecture with an Xception backbone for precise pixel-wise mapping of marine features, followed by a sophisticated classification stage. This stage employs a modified Xception network, enhanced with Swish activation functions for superior feature extraction, whose outputs are fed into a Multiclass Support Vector Machine (MSVM) for robust classification into "Oil Spill," "Look-alike," and "Others." A dedicated pre-processing module using Wavelet Coefficient Shrinkage (WCS) filtering is implemented for effective speckle noise reduction. The system is rigorously evaluated on a public benchmark dataset from the European Space Agency. The proposed model achieves a state-of-the-art overall classification accuracy of 98.20%, with precision, recall, and F1-score of 99.63%, 98.33%, and 98.96%, respectively. Area under the Curve (AUC) values for the three classes were 0.992 (Oil Spill), 0.95 (Look Alike), and 1.00 (Others), demonstrating exceptional class separability. A comprehensive comparative analysis confirms that our hybrid approach outperforms contemporary state-of-the-art methods, including Deep CNNs (VGG-16, ResNet-50), GAN-augmented models, and Vision Transformers, particularly in terms of recall for oil spills (100%) and computational efficiency. The results robustly validate the framework's efficacy, offering a powerful, accurate, and deployable solution for real-time marine surveillance and environmental protection agencies.