Abstract:Wildfire detection using satellite images is a widely studied task in remote sensing with many applications to fire delineation and mapping. Recently, deep learning methods have become a scalable solution to automate this task, especially in the field of unsupervised learning where no training data is available. This is particularly important in the context of emergency risk monitoring where fast and effective detection is needed, generally based on high-resolution satellite data. Among various approaches, Anomaly Detection (AD) appears to be highly potential thanks to its broad applications in computer vision, medical imaging, as well as remote sensing. In this work, we build upon the framework of Vector Quantized Variational Autoencoder (VQ-VAE), a popular reconstruction-based AD method with discrete latent spaces, to perform unsupervised burnt area extraction. We integrate VQ-VAE into an end-to-end framework with an intensive post-processing step using dedicated vegetation, water and brightness indexes. Our experiments conducted on high-resolution SPOT-6/7 images provide promising results of the proposed technique, showing its high potential in future research on unsupervised burnt area extraction.

What problem does this paper attempt to address?

The paper attempts to address the problem of automatic extraction of fire-burned areas using high-resolution satellite images. Specifically, the authors focus on how to effectively identify burned areas after a fire through unsupervised anomaly detection methods in the absence of labeled data. This issue is particularly important in emergency risk monitoring, as rapid and accurate fire detection is crucial for fire management and disaster reduction. ### Background and Motivation - **Hazards of Wildfires**: Wildfires cause severe damage to ecosystems, infrastructure, and even human lives. - **Importance of Early Detection and Mapping**: Early detection and fire mapping play a key role in effective fire management and disaster reduction. - **Application of Remote Sensing Technology**: Satellite remote sensing technology has become an important tool for monitoring and analyzing wildfire events, providing high-resolution multispectral images. - **Advantages of Deep Learning**: In recent years, deep learning has achieved significant success in computer vision tasks and has also become a powerful tool for remote sensing data analysis. - **Challenges of Labeled Data**: The main challenge faced by most supervised learning methods is the availability of labeled data, especially in anomalous events like fires. - **Potential of Unsupervised Anomaly Detection**: Unsupervised anomaly detection (AD) has emerged as a promising approach because it does not require labeled data, using only normal image data for training. ### Methods and Contributions - **Using VQ-VAE Framework**: The authors constructed an unsupervised burned area extraction framework based on the Vector Quantized Variational Autoencoder (VQ-VAE). - **End-to-End Framework**: This framework includes preprocessing, the VQ-VAE model, and post-processing steps. - **Post-Processing Steps**: Vegetation, water, and brightness indices are used in post-processing to reduce false positives. - **Experimental Validation**: Experiments were conducted on high-resolution SPOT-6/7 images in two different scenarios, demonstrating the effectiveness of the method. ### Experimental Results - **Quantitative Evaluation**: In the two scenarios, the post-processed F1 scores were 0.588 and 0.675, respectively, indicating that the post-processing steps significantly improved accuracy. - **Qualitative Evaluation**: The effectiveness of the method was further validated through visualizing the model's reconstruction outputs and anomaly maps. ### Conclusion - **Potential of Unsupervised Anomaly Detection**: The study shows that unsupervised anomaly detection has great potential in extracting burned areas from high-resolution satellite images. - **Future Work Directions**: Further research is needed on the application of anomaly detection metrics in latent space, exploring other VAE-based models, and considering the impact of seasonal and sensor variations. In summary, this paper proposes a method for extracting burned areas from high-resolution satellite images based on unsupervised anomaly detection, providing new ideas and technical support for future fire monitoring and management.

Burnt area extraction from high-resolution satellite images based on anomaly detection

Wildfire Detection Through Deep Learning Based on Himawari-8 Satellites Platform

Deep Learning Approaches for Wildland Fires Using Satellite Remote Sensing Data: Detection, Mapping, and Prediction

Unsupervised burned areas detection using multitemporal synthetic aperture radar data

Analyzing Multispectral Satellite Imagery of South American Wildfires Using Deep Learning

Vegetation change detection and recovery assessment based on post-fire satellite imagery using deep learning

An Adaptive and Extensible System for Satellite-Based, Large Scale Burnt Area Monitoring in Near-Real Time

Integration of Multiple Spectral Indices and a Neural Network for Burned Area Mapping Based on MODIS Data

A Practical Method for High-Resolution Burned Area Monitoring Using Sentinel-2 and VIIRS

Deep-learning-based Burned Area Mapping Using the Synergy of Sentinel-1&2 Data

Two Scalable Approaches for Burned-Area Mapping Using U-Net and Landsat Imagery

Automated Extraction of Forest Burn Severity Based on Light and Small UAV Visible Remote Sensing Images

Rapid Deforestation and Burned Area Detection using Deep Multimodal Learning on Satellite Imagery

Active fire detection in Landsat-8 imagery: A large-scale dataset and a deep-learning study

Deep learning approach for uav-based wildfire detection and segmentation

Burned area detection from a single satellite image using an adaptive thresholds algorithm

Wildfire Impact Analysis and Spread Dynamics Estimation on Satellite Images Using Deep Learning

Deep Learning Approach to Improve Spatial Resolution of GOES-17 Wildfire Boundaries Using VIIRS Satellite Data

Unsupervised Wildfire Change Detection based on Contrastive Learning

Wildland Fire Detection and Monitoring Using a Drone-Collected RGB/IR Image Dataset

BASNet: Burned Area Segmentation Network for Real-Time Detection of Damage Maps in Remote Sensing Images