Page 49 - AJWEP-v22i3

SpillNet CNN model for oil spill detection

Table 2 presents a comparative analysis of the SAR images, and it offers a time- and cost-effective
developed model with existing CNN models employed approach to oil spill management. This study contributes
for oil spill detection. The results of the comparative to advancing automated oil spill detection by offering a
analysis showed that the proposed model (integrating reliable and efficient solution for early oil spill detection
multiple DSCL, BN, and RC) aligns well with the and environmental monitoring in general. However,
outcomes of existing studies, with some improvements. there is a need for larger, more diverse annotated datasets
This indicates that the proposed model can train on to further improve the model’s robustness. Future work
oil spill datasets and accurately detect oil spills using should focus on enhancing segmentation precision and
SAR image analysis. The proposed model can also processing speed to optimize the model for real-world
autonomously differentiate between oil spills and look- applications.
alike regions with similar features.
Table 3 presents a further comparative analysis of Acknowledgments
the developed model with existing CNN models used
for oil spill detection, demonstrating that the proposed None.
SpillNet model offers improvements over some existing
CNN models deployed for similar applications. Funding

5. Conclusions, recommendations, and future None.
works
Conflict of interest
This study developed and evaluated a novel CNN model
for oil spill detection using SAR images. Specifically, The authors declare no conflicts of interest.
an integrated multiple DSCL, BN, and RC were
proposed to enhance feature extraction and learning Author contributions
capabilities. The dataset employed consisted of SAR
images obtained from the Sentinel-1 satellites, part of Conceptualization: All authors
the European Space Agency’s Copernicus program, and Investigation: All authors
the model was trained on an NVIDIA Tesla T4 GPU Methodology: All authors
available on Google Colab, with programming done in Writing-original draft: All authors
the Python 3.7 environment. The proposed algorithm Writing-review & editing: All authors
leveraged the unique characteristics of SAR data and
the robust feature extraction capabilities of CNNs Availability of data
to enhance detection accuracy. The SpillNet model
achieved a mean precision of 0.623542, mean recall of The data supporting the findings of this study are
0.571559, accuracy of 0.946947, mean IoU of 0.58124, available from the corresponding author upon reasonable
mean pixel accuracy of 0.598060, and mean specificity request.
of 0.944469, with a negligible error rate of 0.061053.
These results outperform those obtained by existing References
deep-learning CNN segmentation models, including
UNet, LinkNet, FPN, and PSPNet. This demonstrates 1. Del Frate F, Petrocchi A, Lichtenegger J, Calabresi G.
that the proposed model surpasses other models in the oil Neural networks for oil spill detection using ERS-
spill classification and segmentation tasks. Furthermore, SAR data. IEEE Trans Geosci Remote Sens.
the model effectively differentiated between actual oil 2000;38(5):2282-2287.
spill areas and their look-alikes, thereby improving doi: 10.1109/36.868885
detection accuracy. This indicates that the proposed 2. Hou L, Samaras D, Kurç T, Gao Y, Davis J, Saltz J. Patch-
based Convolutional Neural Network for whole Slide
model is both robust and efficient, making it a valuable Tissue Image Classification. In: 2016 IEEE Conference
tool for oil spill detection in real-world applications. By Computer Vision Pattern Recognition CVPR; 2016.
maintaining a balance between precision and recall, the doi: 10.1109/cvpr.2016.266
model minimized false alarms and missed detections. 3. Iqbal J, Vogt M, Bajorath J. Activity landscape image
Thus, the developed “SpillNet” model is recommended analysis using convolutional neural networks. J
for implementation in the detection of oil spills using Cheminform. 2020;12(1):34.

Volume 22 Issue 3 (2025) 43 doi: 10.36922/ajwep.8282