A Comparative Analysis of AI-Driven Marine Debris Detection Methods for Indian Ocean Plastic Monitoring

Abstract

Plastic pollution continues to accumulate in marine environments at an alarming rate, with millions of tonnes entering the oceans annually. Given the length of India\'s coastline, early detection of floating debris is essential, as undetected objects may disperse, fragment, or sink before any cleanup effort can be initiated. Satellite remote sensing paired with artificial intelligence has become the go-to tool for this problem, yet a close reading of the published work reveals something surprising: researchers rarely explain why they chose one detection method over another. Most studies simply use whatever technique fits the sensor already in hand, without weighing it against alternatives suited to the deployment at hand — a gap even more pronounced for Indian coastal waters, which remain strikingly under-studied relative to the wider global literature. Motivated by this, the present paper reviews twenty-five published studies and compares seven AI-driven detection-method families — spectral-index screening, classical machine learning, CNN and U-Net segmentation, YOLO-based detection, hyperspectral classification, SAR-based screening, and drift-forecasting models — not merely on accuracy, but on sensor cost, robustness to confounders such as algae and biofouling, and real-world deployment readiness. The review finds that spectral-index and classical machine-learning approaches remain the most field-tested and affordable, reporting 86–98% accuracy on Sentinel-2 imagery, while drift forecasting is still reported largely qualitatively, with little validation against real drift tracks, and Indian coastal research remains scattered, single-site, and hard to compare across studies. These findings point to a pressing need for low-cost, transparently reported detection frameworks built for Indian shores, alongside a coordinated regional benchmark, drift-forecasting pipelines validated against in-situ tracks, and multi-sensor fusion combining optical, hyperspectral, and SAR data.

Introduction

Marine plastic pollution is a growing environmental challenge, with millions of tonnes of debris entering the oceans each year. Satellite remote sensing offers the only practical way to monitor debris continuously over large coastal regions, but detecting marine plastic depends on the choice of detection method. Existing approaches include spectral indices, classical machine learning, CNN/U-Net segmentation, YOLO object detection, hyperspectral imaging, SAR-based detection, and drift forecasting models, each with different strengths, costs, and limitations. However, most studies select methods based on available sensors rather than systematically comparing their suitability for operational coastal monitoring.

This paper reviews and compares seven major marine debris detection method families, evaluating them in terms of accuracy, sensor requirements, operational readiness, robustness, and deployment potential. It highlights important research gaps, including limited validation of drift prediction models, insufficient studies focused on Indian coastlines, and poor reporting on whether proposed systems use live or simulated data. Based on this review, the paper justifies a decision-support architecture that combines Floating Debris Index (FDI) screening, CNN-based classification, hotspot clustering, and drift forecasting, while clearly distinguishing implemented components from simulated ones.

The literature shows that spectral-index and classical machine-learning methods are inexpensive and interpretable but sensitive to environmental conditions. CNN and U-Net segmentation provide better detection of diffuse and small debris but require large, well-annotated datasets. YOLO detectors are suitable for real-time UAV and USV applications but struggle with very small or occluded objects. Hyperspectral imaging achieves the highest material discrimination but is costly and mainly validated under controlled conditions. SAR-based methods provide all-weather monitoring but cannot reliably distinguish plastic from other objects without optical data, while drift forecasting models help predict future debris movement but remain insufficiently validated with real-world observations.

The review also highlights increasing research on Indian coastal waters, where studies using FDI, Naïve Bayes, Random Forest, SVM, and drone-based segmentation have produced promising results. Nevertheless, Indian marine debris monitoring still lacks coordinated datasets, comprehensive validation, and operational deployment strategies.

Overall, the study concludes that no single detection method is universally optimal. Instead, integrated pipelines that combine complementary techniques are better suited for practical, agency-level marine debris monitoring. The proposed architecture serves as a literature-supported example of such a pipeline while emphasizing the need for improved validation, transparent reporting, and greater focus on operational deployment and Indian coastal environments.

Conclusion

This paper argued that detection methods should be chosen not for how well they perform in isolation, but for how well they fit a given deployment. Toward that end, seven AI-driven detection-method families — spectral-index screening, classical machine learning, CNN/U-Net segmentation, YOLO-family detection, hyperspectral classification, SAR-based screening, and Lagrangian/learned drift forecasting — were compared on the terms that matter for continuous, agency-facing coastal monitoring rather than benchmark performance alone. This comparison surfaced a consistent pattern: drift forecasts go largely unvalidated against real tracks, Indian coastal studies remain fragmented and difficult to benchmark against one another, and deployment status — what is actually live versus simulated — is rarely disclosed with any precision. Building on this synthesis, an implemented case-study pipeline was presented: a low-cost architecture combining FDI screening, CNN classification, DBSCAN clustering, and drift forecasting, purpose-built for Indian coastal waters. Its value lies not in a new detection algorithm, but in demonstrating what a literature-justified, transparently scoped architecture looks like when every stage is honestly labelled as live or fallback — a level of disclosure this review found conspicuously absent elsewhere. This conclusion remains deliberately bounded. No externally validated field accuracy, drift accuracy, or latency figures are claimed for any stage of the pipeline. Its contribution is architectural and operational — assembling reviewed techniques into a defensible, transparently reported design — not algorithmic novelty. What this work does establish is a template: a structured basis for choosing between detection families, and a working example of the deployment honesty the field needs more of. The path forward is concrete. It includes training the CNN stage on MARIDA/MADOS rather than synthetic spectra — potentially via self-supervised, cross-sensor pre-training approaches such as CSM-Debris that reduce dependence on site-specific labelled data [24] — wiring in live CMEMS/ERA5 drift fields, and reporting same-conditions field validation against Indian coastal ground truth. Beyond this specific pipeline, the wider field still needs what this review found missing throughout: a coordinated regional benchmark and genuine multi-sensor fusion, so that the next detection system built for India\'s coastline does not have to start, once again, from a single sentence justifying its sensor choice.

References

[1] L. Biermann, D. Clewley, V. Martinez-Vicente, and K. Topouzelis, \"Finding plastic patches in coastal waters using optical satellite data,\" Scientific Reports vol. 10, p. 5364, 2020. [2] K. Themistocleous, C. Papoutsa, S. Michaelides, and D. Hadjimitsis, \"Investigating detection of floating plastic litter from space using Sentinel-2 imagery,\" Remote Sensing, vol. 12, no. 16, p. 2648, 2020. [3] K. Kikaki, I. Kakogeorgiou, P. Mikeli, D. E. Raitsos, and K. Karantzalos, \"MARIDA: A benchmark for marine debris detection from Sentinel-2 remote sensing data,\" PLOS ONE, vol. 17, no. 1, p. e0262247, 2022. [4] J. Mifdal, N. Longépé, and M. Rußwurm, \"Towards detecting floating objects on a global scale with learned spatial features using Sentinel-2,\" ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. 3, pp. 285–293, 2021. [5] M. Rußwurm, S. J. Venkatesa, and D. Tuia, \"Large-scale detection of marine debris in coastal areas with Sentinel-2,\" iScience, vol. 26, p. 108402, 2023. [6] S. Sannigrahi, B. Basu, A. S. Basu, and F. Pilla, \"Development of automated marine floating plastic detection system using Sentinel-2 imagery and machine learning models,\" Marine Pollution Bulletin, vol. 178, p. 113527, 2022. [7] M. Kremezi, V. Kristollari, V. Karathanassi, K. Topouzelis, P. Kolokoussis, N. Taggio, A. Aiello, G. Ceriola, E. Barbone, and P. Corradi, \"Increasing the Sentinel-2 potential for marine plastic litter monitoring through image fusion techniques,\" Marine Pollution Bulletin, vol. 182, p. 113974, 2022. [8] B. Badams, U. U. Sheikh, N. A. Wahab, S. A. R. Abu Bakar, M. I. Masud, M. Khouj, U. Waheed, Z. A. Arfeen, and K. M. Goh, \"A2ANet: Real-time detection of floating marine debris using atrous convolution and channel attention,\" Ecological Informatics, vol. 94, p. 103620, 2026. [9] A. Cernian and M.-E. Iliuta, \"Empowering sustainability through AI-driven monitoring: The DEEP-PLAST approach to marine plastic detection and trajectory prediction for the Black Sea,\" Water, vol. 17, no. 22, p. 3318, 2025. [10] A. El Bergui, A. Porebski, and N. Vandenbroucke, \"A lightweight spatial and spectral CNN model for classifying floating marine plastic debris using hyperspectral images,\" Marine Pollution Bulletin, 216, p. 117965, 2025. [11] J. Pretorius, S. Haupt, and B. Sibolla, \"Exploring the potential of remote sensing to detect marine plastic debris in the South African Ocean region,\" ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. X-G-2025, pp. 665–671, 2025. [12] W. Zhou, F. Zheng, G. Yin, Y. Pang, and J. Yi, \"YOLOTrashCan: A deep learning marine debris detection network,\" IEEE Transactions on Instrumentation and Measurement, vol. 72, pp. 1–12, 2023. [13] H. Booth, W. Ma, and O. Karaku?, \"High-precision density mapping of marine debris and floating plastics via satellite imagery,\" Scientific Reports, vol. 13, p. 6822, 2023. [14] A. Alboody, N. Vandenbroucke, A. Porebski, R. Sawan, F. Viudes, P. Doyen, and R. Amara, \"A new remote hyperspectral imaging system embedded on an unmanned aquatic drone for the detection and identification of floating plastic litter using machine learning,\" Remote. 15, p. 3455, 2023. [15] O. Botvynko, C. Granero-Belinchon, S. G. Roux, N. B. Garnier, and A. Benzinou, \"Deep learning for Lagrangian drift simulation at the sea surface,\" arXiv preprint arXiv:2211.09818, 2022; extended as \"Neural prediction of Lagrangian drift trajectories on the sea surface,\" Artificial Intelligence for the Earth Systems, vol. 4, no. 3, 2025. [16] M. P. Raju, S. Veerasingam, V. Suneel, F. S. Asim, H. A. Khalil, M. Chatting, P. Suneetha, and P. Vethamony, \"A machine learning-based detection, classification, and quantification of marine litter along the central east coast of India,\" Frontiers in Marine Science, vol. 12, 2025. [17] V. Nivedita, S. Sabarunisha Begum, G. Aldehim, A. M. Alashjaee, M. A. Arasi, M. Y. Sikkandar, T. Jayasankar, and S. Vivek, \"Plastic debris detection along coastal waters using Sentinel-2 satellite data and machine learning techniques,\" Marine Pollution Bulletin, vol. 209, part A, p. 117106, 2024. [18] S. Sabarunisha Begum, K. Nithya, M. Sujatha, T. Jayasankar, N. B. Prakash, S. Srinivasan, and S. Vivek, \"Automated system for identifying marine floating plastics to enhance sustainability in coastal environments through Sentinel-2 imagery and machine learning models,\" Ocean Science Journal, 2024. [19] S. Krishnakumar, S. Anbalagan, K. Kasilingam, P. Smrithi, S. Anbazhagi, and S. Srinivasalu, \"Assessment of plastic debris in remote islands of the Andaman and Nicobar Archipelago, India,\" Marine Pollution Bulletin, vol. 151, p. 110841, 2020. [20] R. Kiruba-Sankar, K. Saravanan, S. Adamala, K. Selvam, K. L. Kumar, and J. Praveenraj, \"First report of marine debris in Car Nicobar, a remote oceanic Island in the Nicobar archipelago, Bay of Bengal,\" Regional Studies in Marine Science, vol. 61, p. 102845, 2023. [21] S. K. Sivadas, P. Mishra, T. Kaviarasan, M. Sambandam, K. Dhineka, M. V. R. Murthy, S. Nayak, D. Sivyer, and D. Hoehn, \"Litter and plastic monitoring in the Indian marine environment: A review of current research, policies, waste management, and a roadmap for multidisciplinary action,\" Marine Pollution Bulletin, vol. 176, p. 113424, 2022. [22] P. Thanabalan, K. Gayathrri, P. Mishra, T. Usha, S. K. Dash, and S. R. Marigoudar, \"Monitoring of marine floating debris and beach litter using satellite and drones: A synergistic approach on policy and decision making,\" Journal of the Indian Society of Remote Sensing, 2025. [23] C. Pattiaratchi, M. van der Mheen, C. Schlundt, B. E. Narayanaswamy, A. Sura, S. Hajbane, R. White, N. Kumar, M. Fernandes, and S. Wijeratne, \"Plastics in the Indian Ocean – sources, transport, distribution, and impacts,\" Ocean Science, vol. 18, pp. 1–28, 2022. [24] E. Dalsasso, M. Rußwurm, C. Donner, S. Darmon, R. de Vries, M. Volpi, and D. Tuia, \"Self-supervised pre-training enables marine debris detection across sensors,\" Remote Sensing of Environment, vol. 339, 2026. [25] J. Cui, S. Zhou, G. Xu, X. Liu, and X. Gao, \"Marine debris detection in real time: A lightweight UTNet model,\" Journal of Marine Science and Engineering, vol. 13, p. 1560, 2025.

Copyright

Copyright © 2026 Nimisha Nair, Darshan Gadekar, Nalaksh Randhawa, Roshan Kotkondawar, Krushna Taiwade. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.