The role of data augmentation and synthetic imagery in improving marine debris detection models

Marine litter can damage marine organisms and it can destroy their habitats. It can pollute the seawater, so we need to detect and monitor marine litter to maintain the health of the marine environment. Deep learning can detect marine litter because it is accurate and efficient, therefore it requires a large amount of high-quality data to achieve an accurate detection. Obtaining such data is difficult and expensive, thus this paper presents the use of larger datasets and synthesised images to overcome this limitation. The augmentation of existing data and the generation of synthetic images that resemble realistic scenes improve the performance of the deep learning system in detecting litter, and the deep learning system is capable of detecting litter in various scenes. This demonstrates the effectiveness of augmentation techniques and synthesised images for deep learning, which can be extended to litter detection in the marine environment, supporting marine conservation, because it can help maintain the health of the marine environment.