Generation of Synthetic Data Based on Real Images for the Detection of Plastic Waste in an Urban Canal

• Ana Sofia da Silva Barbosa UFPE
• Antonio Celso Dantas Antonino UFPE
• Manoel Henriques de Sá Campos Filho UPE
• Tsang Ing Ren UFPE
• Jean Firmino Cardoso UFPE

DOI: https://doi.org/10.5753/eniac.2026.31534

Resumo

Monitoring plastic waste in urban waterways demands continuous image analysis, yet building annotated datasets for this task is costly, requiring manual inspection of thousands of recorded frames to find usable examples. This paper investigates whether synthetic data can replace or reduce that annotation effort in a Brazilian urban canal under permanent camera surveillance. We contribute a real annotated dataset (S1) and three synthetic variants: images generated by DALL·E conditioned on real canal photographs (S2), copy-paste composites using SAM-extracted waste instances (S3), and a hybrid combining all three sources equally (S4). All datasets are intended to be made publicly available to support future research. Training YOLOv11s on each configuration and evaluating all four models on a fixed held-out set of 1,000 real canal images, the real dataset (S1) achieved the best generalisation (Precision = 0.8723, Recall = 0.9244), while the copy-paste strategy (S3) demonstrated the best balance among synthetic configurations (Precision = 0.3516, Recall = 0.5790). The T2I model (S2) generalised poorly to real conditions despite its strong internal validation score (mAP@0.5 = 0.9866). These results highlight a critical gap between synthetic validation performance and real-world generalisation, and suggest that visual consistency within the training set matters more than data source diversity.

Referências

Barbosa, A. S. d. S. (2025). Prompts for DALL·E synthetic image generation — Cavouco Canal plastic waste detection.

Devi, B. S., Gupta, D., and Singh, R. P. (2024). A deep learning approach to plastic bottle waste detection on the water surface using YOLOv6 and YOLOv7. Engineering, Technology & Applied Science Research.

Ghiasi, G., Cui, Y., Srinivas, A., Qian, R., Lin, T.-Y., Cubuk, E. D., Le, Q. V., and Zoph, B. (2021). Simple copy-paste is a strong data augmentation method for instance segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2918–2928.

Jocher, G. and Qiu, J. (2024). Ultralytics YOLO11. [link]. Version 11.0.0.

Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., White-head, S., Berg, A. C., Lo, W.-Y., Dollár, P., and Girshick, R. (2023). Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4015–4026.

Peng, C., He, B., Xi, W., and Lin, G. (2024). Improved YOLOv7 algorithm for floating waste detection based on GFPN and long-range attention mechanism. Wuhan University Journal of Natural Sciences, 29(4):338–348.

Ramesh, A., Dhariwal, P., Nichol, A., Chu, C., and Chen, M. (2022). Hierarchical text-conditional image generation with CLIP latents. arXiv preprint arXiv:2204.06125.

Ramsurrun, N., Suddul, G., Armoogum, S., and Foogooa, R. (2024). Plastic water bottle detection model using computer vision in aquatic environments. Scientific Reports, 14.

Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10684–10695.

Sio, G. A., Guantero, D., and Villaverde, J. (2022). Plastic waste detection on rivers using YOLOv5 algorithm. In Proceedings of the International Conference on Computing Communication and Networking Technologies (ICCCNT), pages 1–6. IEEE Press.

Tomas, J. P. Q., Tupas, J. E. E., Soniel, M. T., Caruz, C. H. M. E., and Babar, D. B. (2024). Real-time detection of floating debris in waterways using YOLOv8. In Proceedings of the 14th International Workshop on Computer Science and Engineering (WCSE), pages 19–21, Phuket Island, Thailand.

ttylu (2024). Floating plastic waste river dataset. [link]. Available at Roboflow Universe. Accessed: May 2026. Contains 319 images.

UNEP (2021). From pollution to solution: A global assessment of marine litter and plastic pollution. Technical report, United Nations Environment Programme.