Harnessing AI for ecosystem surveillance: A case study on wetland mapping with transformer models

Wetlands play a crucial role in maintaining ecological balance, making their accurate classification and monitoring imperative. This paper presents a novel approach to wetland classification using satellite imagery, harnessing the power of state-of-the-art deep learning architectures. We introduce the application of the advanced Transformer and Convolutional Neural Network models, to the domain of wetland classification, demonstrating its potential in handling intricate spatial data. To further enhance the model's performance, we employ a transfer learning strategy, fine-tuning the transformer on the comprehensive BigEarthNet-S2 satellite image dataset. This approach capitalizes on the model's pre-trained knowledge while tailoring it to the specific characteristics of wetland imagery. Our research also introduces a modularized workflow, designed to be both robust for the current classification task and adaptable for future advancements in the field. The results showcase a significant improvement in classification accuracy. This research not only sets a new standard in the satellite-based monitoring of wetlands but also contributes to the broader dialogue on resource and energy conservation.