Residual-based Neural Network for Unmodeled Distortions in Coordinate Transformation

This dataset contains real and simulated coordinate data used to evaluate the performance of neural networks in modeling unmodeled distortions during coordinate transformations. The data support the study entitled "Residual-based Neural Network for Unmodeled Distortions in Coordinate Transformation", which investigates how observation residuals derived from least-squares estimation can be used to improve transformation accuracy through machine learning techniques. The files include: Input coordinates: image-space or distorted coordinates used as predictors in transformation models. Reference coordinates: ground-truth or undistorted coordinates serving as transformation targets. Residuals: computed from initial geometric transformations (e.g., 2D projective) using least-squares estimation. Neural network predictions: coordinate outputs derived from models trained on residuals or direct coordinate mappings. Scripts and metadata: MATLAB files for running repeated leave-one-out validation, neural network training, and statistical error analysis. The dataset is particularly relevant for researchers working with image rectification, photogrammetric calibration, and geodetic transformation, especially where conventional models fail to fully capture local nonlinear distortions.