Satellite Imagery Change Detection Model

I have a time-series collection of remote sensing data—specifically multi-spectral satellite imagery—and I need an AI model that can reliably pinpoint and quantify changes that occur from one acquisition date to the next. The end goal is an automated workflow that flags where and when significant alterations appear, whether those are shifts in vegetation, newly built structures, or other surface transformations. To give you an idea of scope, the raw scenes are already orthorectified and radiometrically corrected. What I’m missing is the machine-learning layer that will take two (or more) aligned images, learn the patterns of normal variability, and then output clear, georeferenced change masks and summary statistics. Key expectations • Model architecture, training pipeline, and inference script packaged in Python (TensorFlow, PyTorch, or another proven deep-learning framework). • Clear instructions for reproducing results on my own machine, including environment file and command-line steps. • Evaluation report showing accuracy metrics on a held-out test set I will provide after initial proof of concept. If you have experience with satellite imagery, convolutional networks, and change detection techniques such as Siamese or UNet-based approaches, I’d like to see examples of similar work in your bid.