Comparative Analysis of the Effectiveness of Petroleum-Product Contamination Detection in Aquatic Systems Using Pre-Trained YOLO-Family Neural Network Models and the SMDNetV4 and PSPNet Architectures

Introduction. The PSPNet, SMD-Net, and YOLO26-Seg semantic segmentation architectures are considered for the task of detecting oil spills in satellite imagery. Particular attention is paid to comparing segmentation accuracy under identical training conditions. Materials and Methods. PSPNet uses a ResNet101 backbone with a pyramid pooling module for semantic segmentation. SMD-Net is a lightweight shared-backbone architecture with two decoders and CTGM, CoordAtt, and ASPP modules. YOLO26-Seg combines a YOLO26 encoder backbone with fixed weights and a lightweight decoder. The following metrics are compared: IoU (Intersection over Union), defined as the ratio of the intersection of the predicted and ground-truth regions to their union; Dice (Dice coefficient), a measure of segment overlap based on the intersection area; accuracy, the proportion of correct predictions among all predictions; inference time, the time required by the model to process one input; and the number of trainable parameters. Results. After 50 epochs of training on the existing dataset masks without manual annotation, PSPNet significantly outperformed SMD-Net in binary oil-spill segmentation (IoU = 0.704, Dice = 0.770, Acc = 0.896 versus IoU = 0.467, Dice = 0.575, Acc = 0.671 for SMD-Net). The additionally fine-tuned YOLO26l-seg model demonstrated moderate pixellevel metrics (IoU = 0.320) at a reduced detection threshold, which limits its applicability to this task. Discussion. The choice of architecture for oil-spill segmentation is governed by a trade-off between accuracy and speed: PSPNet provides the highest segmentation quality, SMD-Net performs worse across all metrics, and YOLO26-SemSeg cannot reliably detect oil spills without task-specific training. Conclusion. PSPNet significantly outperforms SMD-Net and YOLO26-Seg in binary oil-spill segmentation under identical training conditions (IoU = 0.704 versus 0.467 for SMD-Net), confirming the effectiveness of the ResNet101 backbone combined with pyramid pooling.

1. Sukhinov A.I., Nikitina A.V., Chistyakov A.E. Using multichannel satellite images for predictive modelling the “bloom” phytoplankton processes in shallow waters on supercomputer. Computational Mathematics and Information Technologies. 2017;1(2):128–140. (In Russ.) https://doi.org/10.23947/2587-8999-2017-2-128-140

2. Dunaeva A.V., Kornilov F.A. Building detection in remote sensing images using a digital surface model. Computational Mathematics and Information Technologies. 2017;1(2):185–193. (In Russ.) https://doi.org/10.23947/2587-8999-2017-2-185-193

3. Shchetinin E.Yu., Shevchuk A.A. Semantic segmentation with uncertainty estimation based on the Dirichlet model and anisotropic regularization. Computational Mathematics and Information Technologies. 2026;10(1):7–20. (In Russ.) https://doi.org/10.23947/2587-8999-2026-10-1-7-20

4. Sukhinov A.I., Solomakha D.A., Sidoryakina V.V. Oil spill detection based on advanced LBP neuroalgorithms for noisy satellite images. Computational Mathematics and Information Technologies. 2026;10(1):58–71. (In Russ.) https://doi.org/10.23947/2587-8999-2026-10-1-58-71

5. Sukhinov A.I. An improved method for recognizing objects of marine and coastal systems based on a combination of local binary patterns and neural network methods. Computational methods and programming. 2025;26(3):366–379. (In Russ.) https://doi.org/10.26089/NumMet.v26r324

6. Vasiliev M.E., Shalimov A.S., Savina O.A. Review of YOLO versions: Single-stage convolutional neural network model. Universum: Technical Sciences. 2025;1(6): 36–46.

7. Stetsiuk V.O. Application of a modification of the grid-characteristic method using superimposed grids to explicitly distinguish the interface of media in modeling the relief of the oceanic shelf. Computational Mathematics and Information Technologies. 2023;7(3):20–27. (In Russ.) https://doi.org/10.23947/2587-8999-2023-7-3-20-27

8. Documentation for the yolov8l-world model. URL: https://docs.ultralytics.com/ru/models/yolo-world/ (accessed: 16.04.2026).

9. Documentation for the yolov8l-obb model. URL: https://docs.ultralytics.com/ru/tasks/obb/ (accessed: 16.04.2026).

10. Burmistrov S.А. Working with YOLOV8. Detection, segmentation, tracking of objects, as well as preparation of your own dataset and training. URL: https://habr.com/ru/articles/821971/ (accessed: 16.04.2026).

11. Gallagher J., Skalski P. How to Detect Objects with YOLO-World. URL: https://blog.roboflow.com/how-todetect-objects-with-yolo-world/ (accessed: 16.04.2026).

12. Sukhinov A.A., Ostrobrod G.B. Effective face detection on an Epiphany multicore processor. Computational Mathematics and Information Technologies. 2017;1(1):1–13. (In Russ.)

13. Zakharova A.A., Nebaba S.G. Application of the background subtraction method and the Viola-Jones algorithm to increase the speed of face tracking in the personality recognition system. Computational Mathematics and Information Technologies. 2018;2(1):25–32. (In Russ.) https://doi.org/10.23947/2587-8999-2018-2-1-25-32

14. Sukhinov A.I., Panasenko N.D. Comparative investigation of neural and locally binary algorithms for image identification of plankton populations. Computational Mathematics and Information Technologies. 2022;6(2):70–80. (In Russ.) https://doi.org/10.23947/2587-8999-2022-1-2-70-80

15. Khusnullin Sh.R., Koledina K.F., Alimbekova S.R., Ishmuratov F.G. Machine learning in the analysis of the influence of an electromagnetic field on the rate of corrosion and salt deposition of oilfield equipment. Computational Mathematics and Information Technologies. 2023;7(1):70–76. (In Russ.) https://doi.org/10.23947/2587-8999-2023-6-1-70-76

16. Panasenko N.D., Sukhinov A.I. A program for recognizing heterogeneous objects on the surface of marine and coastal systems based on remote sensing data. Certificate of state registration of the computer program No. 2021617654 Russian Federation. 2021.

17. Arzumanyan R.V., Sukhinov A.I. Investigation of the effect of the global GPU memory access pattern on performance. Bulletin of the Don State Technical University. 2016;2(85):105–110. (In Russ.) https://doi.org/10.12737/19687