Oil Spill Segmentation Using Enhanced U-NET and Fuzzy System Hybrid

The suggested method integrates an advanced U-Net architecture with a fuzzy logic system to improve segmentation outcomes for accuracy and interpretability. The Enhanced U-Net incorporates additional convolutional layers, enhanced dropout rates, batch normalization, and improved skip connections, facilitating superior detection of complex features in oil spill photos while reducing overfitting. Significant advances comprise Leaky ReLU activation, dynamic learning rate scheduling, and a bespoke weighted loss function that together address the class imbalance problem and enhance the model's training efficiency. Data supplementation and improved preprocessing enhance the model's generalization capability. Fuzzy logic introduces subtleties into the segmentation process, significantly diminishing false positives and negatives while yielding more dependable and interpretable outcomes. The hybrid approach enhances oil spill detection accuracy, hence significantly augmenting its utility for environmental monitoring.