Marine litter, particularly plastic debris, poses a significant environmental challenge globally. Detecting floating debris in the marine environment using satellite remote sensing remains a complex task due to the limited availability of high-resolution data and the coarseness of existing datasets. This study explores the potential of active few-shot meta-learning for improving the detection of marine debris. The results demonstrate that active learning methods incorporating uncertainty-based sampling, such as entropy and query by committee, outperform other strategies in terms of recall and average precision. Yet, diversity-based methods are found to be limited by the poor representativeness of the feature space used for clustering samples. Additionally, the study highlights the influence of regional characteristics on detection performance and the impact of class imbalance on active learning strategies. To further enhance marine debris detection, future research directions are identified, including training meta-models specifically on marine debris data and tuning decision thresholds. The suggested methodology shows promise for enhancing the efficiency of remote sensing-based monitoring of marine debris, thereby assisting environmental management and conservation efforts.