New AI Pipeline Enables Automated Segmentation of Remote Sensing Imagery with 99% Accuracy

Researchers have developed a pipeline that integrates zero-shot AI detection and segmentation tools to achieve robust, automated segmentation of remote sensing images. By leveraging a sliding window hyper-inference approach and an outlier rejection step, the pipeline enhances the identification of features such as buildings, trees, and vehicles in aerial and satellite imagery. This solution is implemented as a user-friendly Python package, LangRS, making advanced remote sensing segmentation accessible to a wide range of users.

The amount of aerial and satellite imagery captured worldwide continues to grow at an incredible pace, yet efficiently identifying and labeling features in these images remains challenging. The new pipeline developed by researchers at Politecnico di Milano and the National Technical University of Athens tackles this issue by combining advanced AI models with smart data-handling strategies. Professor Maria Antonia Brovelli from Politecnico di Milano explains that general-purpose AI models often struggle when asked to locate unfamiliar objects without explicit training, but their approach using sliding window hyper inference and outlier-rejection steps significantly reduces computational burden and improves accuracy.

The pipeline leverages open-source foundation models like Segment Anything Model (SAM) and Grounding DINO in a strategic two-step process. First, it intentionally over-detects objects to ensure even the smallest details are captured through a sliding window approach that applies the detection model to smaller image patches. This method not only reduces computational burden but also enhances detection accuracy. Next, the system refines results by filtering out irrelevant bounding boxes using statistical and data-driven techniques, with the remaining high-quality bounding boxes passed to SAM for precise segmentation mask generation.

Operating in a zero-shot manner, the models were used in an off-the-shelf fashion without additional fine-tuning or retraining on external data. In aerial images with spatial resolution of less than 1 meter, the pipeline achieved outstanding segmentation results reaching up to 99% accuracy. The researchers hope this pipeline will make automated remote sensing imagery analysis more accessible, speeding up applications from environmental surveys to urban planning. The research findings are documented in the journal Artificial Intelligence in Geosciences with DOI 10.1016/j.aiig.2025.100105.