Spectral-Spatial Deep Learning model for seaweed cultivation mapping using PlanetScope imagery in Pangkajene and Islands Regency

The efficient mapping of seaweed cultivation over large areas is essential for supporting sustainable management of coastal resources. This study introduces a novel Spectral-Spatial Deep Learning model that integrates spectral and spatial data from high-resolution remote sensing imagery to automate and improve the accuracy of seaweed cultivation mapping. Based on a Convolutional Neural Network architecture, UNet, enhanced with a Spectral-Spatial Attention Module, the model effectively captures the complex relationships between seaweed and its environment. PlanetScope imagery, known for its high spectral and spatial resolution, serves as the primary input data. The model’s performance was evaluated using evaluation metrics, achieving an accuracy of 94.71 %, loss of 13.09 %, precision of 80.93 %, recall of 73.63 %, and Intersection over Union (IoU) of 48.51 % on the training data. For the validation data, the model attained an accuracy of 93.64 %, loss of 16.75 %, precision of 84.34 %, recall of 57.57 %, and IoU of 42.98 %. These results demonstrate the model’s ability to rapidly and accurately map seaweed cultivation areas, making it a valuable tool for environmental monitoring. Highlights Efficient mapping of seaweed cultivation over large areas is crucial for sustainable coastal resource management. Remote sensing technology, integrated with Artificial Intelligence (AI) methods like Deep Learning, enhances the efficiency of seaweed cultivation mapping. The integration of remote sensing’s spatial, spectral, and temporal advantages with Deep Learning-based image segmentation forms the foundation for automating seaweed cultivation mapping.