ABSTRACT
Estimating how much macro-plastic is present in the world’s oceans is one of the most critical environmental challenges of our day. The most popular techniques for counting the amount of floating plastic trash take a long time and only work in a relatively narrow region. To overcome the shortcomings, an automatic identification method based on a deep learning framework was developed to find and quantify the marine trash. As a result, it is suggested to utilize the FMA-YOLOv5 approach, an improved version of the YOLOv5 technique that uses a feature map attention (FMA) layer at the end of the backbone for detection and classification. The deep learning-based segmentation model that was used by the image classifier allowed it to quantify the marine debris in the sample. The quantity of coastal debris objects predicted by the suggested strategy was compared with the estimation of the manual monitoring method in order to demonstrate the differences in predicting the marine debris standing-stock. Apparently, the proposed method verified on seven categories of items yielded a mean average precision of 0.90. The outcomes offer vital data for formulating efficient marine trash control programs and regulations.
