海洋廢棄物(Marine Debris)一直是海洋與海岸區域長期的環境污染問題，常有公司或民間團體主動發起淨灘活動以維護環境，若能在清除海廢的同時記錄各種廢棄物的類型、數量以及位置等資訊，透過掌握這些海廢的特性與來源，讓相關學者與地方政府能夠調整廢棄物的管理與監測工作，藉此擬定正確防範或減緩策略。近年來隨著人工智慧(Artificial Intelligence, AI)技術在物件識別任務上的進步，預期可利用AI協助海洋廢棄物的調查工作。本研究目的是透過使用辨識效率高且準確性高的YOLO模型來建置海洋廢棄物的辨識模型，使其可以快速且正確地辨識其位置與類別，協助淨灘志願者及相關研究人員的統計結果工作，成為後續對於海廢統計、分析、定位與建立資料庫等應用的基礎。
海洋廢棄物具有場景變化大、多種類別與樣態及通常為較小物件且不完整等特性，這些特性都會影響模型學習與辨識時的效果。因此，本研究應用多種影像擴增方法來提升模型訓練效果，加大輸入模型的影像尺寸以提升模型對細部特徵的辨識，同時加深模型網路的結構來強化模型對複雜特徵的提取能力，最後調整了訓練超參數以加強模型訓練效果。
本研究發現實際拍攝的海洋廢棄物影像中，有幾種類別的海廢因較難在海岸上蒐集到，使訓練資料中的數量較少，導致模型對於此類的辨識效果有待加強。因此，本研究利用了SAM將海洋廢棄物從網上蒐集到的影像中分割出來，並合成到額外拍攝的海灘背景上，提出一種利用影像合成產生新影像的方法來增加訓練資料量。本研究發展之辨識模型驗證結果顯示模型辨識效果良好，對於牙刷有著最高辨識率達99.5%，對於常出現在海岸上的寶特瓶廢棄物辨識率達96.5%，顯示本研究經最佳化且使用影像合成擴增訓練資料的海洋廢棄物辨識模型具有良好的辨識準確性與可靠性。

Marine debris has long been a significant environmental issue for oceans and coastal areas. Companies and volunteer groups often organize coastal clean-up activities to solve this problem. Recording data on the types, quantities, and locations of collected marine debris can help researchers and local governments improve waste management and monitoring, leading to more effective prevention and mitigation strategies. This study aims to develop a marine debris recognition model using the highly efficient and accurate YOLO model. The model can quickly and accurately identify the location and type of debris, aiding volunteers and researchers in statistical analysis. Marine debris often varies in scene, type, and appearance, and is usually small and fragmented, posing challenges for model training and recognition. To address this, we applied various image augmentation techniques, increased input image size, deepened the model structure to enhance feature extraction, and adjusted training hyperparameters. Our findings indicate that some types of debris are underrepresented in training data due to their rarity on beaches, affecting recognition accuracy. To counter this, we used SAM to segment debris from online images and composite them onto additional beach backgrounds that we collected, creating new training images. The optimized model demonstrated excellent recognition performance, with a 99.5% accuracy for toothbrushes and 96.5% for PET bottles. These results highlight the model's high accuracy and reliability for marine debris identification, forming a solid foundation for future debris statistics, analysis, and database creation.

[1] 洪昱翔、陳威成、董東璟(2023)，海洋廢棄物識別AI模型之研發與測試。第45屆海洋工程研討會, A113-A114。
[2] 洪昱翔、陳威成、董東璟(2024)，開發深度學習模型應用於海灘垃圾之辨識。第25屆海洋與水下技術研討會, 第139-141頁。
[3] 荒野保護協會 (2023)，荒野2023年成果報告。
[4] 黑潮海洋文教基金會 (2023)，黑潮2023年成果報告。
[5] 董東璟、莊卉婕 (2012) ，海洋垃圾調查與分析。海洋及水下科技季刊，57-65。
[6] Chen, L. C., Zhu, Y., Papandreou, G., Schroff, F., & Adam, H. (2018). Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European conference on computer vision (ECCV) (pp. 801-818).
[7] Chang, M., Xing, Y. Y., Zhang, Q. Y., Han, S. J., & Kim, M. (2020). A CNN Image Classification Analysis for 'Clean-Coast Detector' as Tourism Service Distribution. Journal of Distribution Science, 18(1), 15-26.
[8] Chiu, C. C., Liao, C. P., Kuo, T. C., & Huang, H. W. (2020). Using citizen science to investigate the spatial-temporal distribution of floating marine litter in the waters around Taiwan. Marine Pollution Bulletin, 157, 111301.
[9] Dave, B., Mori, M., Bathani, A., & Goel, P. (2023). Wild Animal Detection using YOLOv8. Procedia Computer Science, 230, 100-111.
[10] Fallati, L., Polidori, A., Salvatore, C., Saponari, L., Savini, A., & Galli, P. (2019). Anthropogenic Marine Debris assessment with Unmanned Aerial Vehicle imagery and deep learning: A case study along the beaches of the Republic of Maldives. Science of The Total Environment, 693, 133581.
[11] Geyer, R., Jambeck, J. R., & Law, K. L. (2017). Production, use, and fate of all plastics ever made. Science advances, 3(7), e1700782.
[12] Gonçalves, G., Andriolo, U., Gonçalves, L., Sobral, P., & Bessa, F. (2020). Quantifying marine macro litter abundance on a sandy beach using unmanned aerial systems and object-oriented machine learning methods. Remote Sensing, 12(16), 2599.
[13] He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 2961-2969).
[14] Huang, B., Chen, G., Zhang, H., Hou, G., & Radenkovic, M. (2023). Instant deep sea debris detection for maneuverable underwater machines to build sustainable ocean using deep neural network. Science of The Total Environment, 878, 162826.
[15] Jambeck, J. R., Geyer, R., Wilcox, C., Siegler, T. R., Perryman, M., Andrady, A., ... & Law, K. L. (2015). Plastic waste inputs from land into the ocean. Science, 347(6223), 768-771.
[16] Kuo, F. J., & Huang, H. W. (2014). Strategy for mitigation of marine debris: analysis of sources and composition of marine debris in northern Taiwan. Marine Pollution Bulletin, 83(1), 70-78.
[17] Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., ... & Girshick, R. (2023). Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 4015-4026).
[18] LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989). Backpropagation applied to handwritten zip code recognition. Neural computation, 1(4), 541-551.
[19] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C. Y., & Berg, A. C. (2016). Ssd: Single shot multibox detector. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14 (pp. 21-37). Springer International Publishing.
[20] PlasticsEurope, E. P. R. O. (2017). Plastics-the facts 2017. An analysis of European latest plastics production, demand and waste data.
[21] Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28.
[22] Ronneberger, O., Fischer, P., & Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 (pp. 234-241).
[23] Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 779-788).
[24] Song, K., Jung, J. Y., Lee, S. H., & Park, S. (2021). A comparative study of deep learning-based network model and conventional method to assess beach debris standing-stock. Marine Pollution Bulletin, 168, 112466.
[25] Thompson, R. C., Moore, C. J., Vom Saal, F. S., & Swan, S. H. (2009). Plastics, the environment and human health: current consensus and future trends. Philosophical transactions of the royal society B: biological sciences, 364(1526), 2153-2166.
[26] Teng, C., Kylili, K., & Hadjistassou, C. (2022). Deploying deep learning to estimate the abundance of marine debris from video footage. Marine Pollution Bulletin, 183, 114049.
[27] Walther, B. A., Kunz, A., & Hu, C. S. (2018). Type and quantity of coastal debris pollution in Taiwan: A 12-year nationwide assessment using citizen science data. Marine pollution bulletin, 135, 862-872.
[28] Wang, L. C., Chun-Te Lin, J., Dong, C. D., Chen, C. W., & Liu, T. K. (2021). The sorption of persistent organic pollutants in microplastics from the coastal environment. Journal of Hazardous Materials, 420, 126658.
[29] Wang, C. Y., Bochkovskiy, A., & Liao, H. Y. M. (2023). YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 7464-7475).
[30] Xuan, K., Deng, L., Xiao, Y., Wang, P., & Li, J. (2023). SO-YOLOv5: Small object recognition algorithm for sea cucumber in complex seabed environment. Fisheries Research, 264, 106710.