The code for the "Anomaly-Detection-in-Satellite-Videos-using-Diffusion-Models " will be available soon here
Timely identification of extreme events such as wildfires, cyclones, or floods using satellite data is paramount for effective disaster management. While various earth-observing satellites furnish data on disasters, those in geostationary orbit offer information at intervals as frequent as every minute, essentially crafting a space-based video feed. Our central focus is the detection of anomalies, specifically wildfires and smoke, in these satellite videos. In contrast to prior endeavors in anomaly detection within surveillance videos, this study introduces a system tailored for high-frequency satellite videos, placing particular emphasis on two anomalies. Unlike the majority of existing CNN-based methods for wildfire detection that rely on labeled images or videos, our unsupervised approach addresses the challenges posed by high-frequency satellite videos with a high intensity of clouds. These CNN-based methods can only identify fires once they have reached a certain size and are susceptible to false positives. We frame the challenge of wildfire detection as a general anomaly detection problem. Introducing an inno- vative unsupervised approach involving diffusion models, which are state-of-the-art generative models for anomaly detection in satellite videos, we adopt a ”generating-to-detecting” strategy. The diffusion model is trained to generate normal videos without anomalies, and when tested on abnormal videos containing anomalies like fire or smoke, it encounters difficulty generat- ing corresponding frames. Performance evaluation, measured through AUC-ROC, underscores the superior efficacy of the diffusion model over CNN and GAN-based methods in detecting anomalies in these high-frequency satellite videos characterized by a high intensity of clouds.
The dataset utilized in our paper can be accessed through the link below. It comprises extracted normal clips from day scenes, and we have generated 500 normal clips, each containing 14 frames, using the provided clips.
https://drive.google.com/drive/folders/14q9irXkDozr-HMUtRj2NZVQOl2Q9sdNT?usp=share_link
We also offer the extracted clips specifically captured during night scenes with low light conditions.
https://drive.google.com/drive/folders/1-WsjM-1OCY_kxBT5Ehv-3ljTer2BhSta?usp=share_link