An experimental study on marine debris location and recognition using object detection

The large amount of debris in our oceans is a global problem that dramatically impacts marine fauna and flora. While number human-based campaigns have been proposed to tackle this issue, these efforts deemed insufficient due the insurmountable existing litter. In response that, there exists high interest use autonomous underwater vehicles (AUV) may locate, identify, collect garbage automatically. To perform such task, AUVs consider state-of-the-art object detection techniques based on deep neural networks their reported performance. Nevertheless, generally require amounts data with fine-grained annotations. work, we explore capabilities reference detector Mask Region-based Convolutional Neural Networks for automatic location classification context limited availability. Considering recent CleanSea corpus, pose several scenarios regarding available train study possibility mitigating adverse effects scarcity synthetic scenes. Our results achieve new state art establishing future research. addition, it shown task still has room improvement lack can be somehow alleviated, yet extent.