Marine Debris Detection Model with Custom Dataset Using Instance Segmentation

Object detection and instance segmentation are fundamental task in computer vision with applications ranging from autonomous driving to robotics. In this study, presented a custom dataset implementation on Mask R-CNN model, for object detection and instance segmentation. Our dataset images acquired from Motobu beach, Okinawa, Japan. Used to CV AT (Computer Vision Annotation Tool) for annotation. Containing a diverse range of objects of interest. Coastal debris affects marine ecosystem. The standard responses set in SGDs projects involve cutting down plastic waste at the source, beach cleanups, use of the circular economy, recycling, education, and a reduction in packaging, among other solutions. We fine-tuned the Mask R-CNN architecture using transfer learning techniques, initializing the model with weights pretrained on a large-scale dataset. Our proposed model is based on instance segmentation Mask R-CNN and SIFT (Scale invariant features transform) matching deep learning neural network. The results showed successfully recognize seven categories of marine debris, using a specific custom dataset. Marine debris global environmental issue and debris detection challenging, because of weather condition like (cloud, shine, and sunset), occulated and small size object, long distance from fix camera. Debris issues not only beaches but also under water and any localization. when the flood affects any area in the world, garbage is accumulated into deep sea. We developed a deep learning-based Mask R-CNN project for the marine environment auto-clean and multi-class classification process of marine debris detection from (Kakaxi fixed camera) images.