SpillNet: A modified convolutional neural network model for oil spill detection

Rapid and accurate detection of oil spills is crucial for initiating timely response measures to mitigate environmental impacts. This study proposes an oil spill detection method based on a modified convolutional neural network, termed “SpillNet.” The architecture integrates multiple depthwise separable convolutional layers, batch normalization, and residual connections to enhance feature extraction and learning capabilities. The dataset consists of synthetic aperture radar images obtained from Sentinel-1 satellites, part of the European Space Agency’s Copernicus program. Model training was conducted on an NVIDIA Tesla T4 GPU available on Google Colab, with up to 12GB of random access memory. Programming was carried out in the Python environment using Python 3.7, and all required libraries were installed through pip. The results indicate that the proposed model achieves an accuracy of 0.946947, a mean Intersection over Union of 0.58124, and a mean specificity of 0.944469. These results demonstrate that the proposed model outperforms existing models in the oil spill segmentation task. This study contributes to advancing automated oil spill detection by offering a reliable and efficient solution for early oil spill detection and environmental monitoring.