Active Fire Detection in Landsat-8 Imagery: a Large-Scale Dataset and a Deep-Learning Study (2101.03409v2)

Abstract: Active fire detection in satellite imagery is of critical importance to the management of environmental conservation policies, supporting decision-making and law enforcement. This is a well established field, with many techniques being proposed over the years, usually based on pixel or region-level comparisons involving sensor-specific thresholds and neighborhood statistics. In this paper, we address the problem of active fire detection using deep learning techniques. In recent years, deep learning techniques have been enjoying an enormous success in many fields, but their use for active fire detection is relatively new, with open questions and demand for datasets and architectures for evaluation. This paper addresses these issues by introducing a new large-scale dataset for active fire detection, with over 150,000 image patches (more than 200 GB of data) extracted from Landsat-8 images captured around the world in August and September 2020, containing wildfires in several locations. The dataset was split in two parts, and contains 10-band spectral images with associated outputs, produced by three well known handcrafted algorithms for active fire detection in the first part, and manually annotated masks in the second part. We also present a study on how different convolutional neural network architectures can be used to approximate these handcrafted algorithms, and how models trained on automatically segmented patches can be combined to achieve better performance than the original algorithms - with the best combination having 87.2% precision and 92.4% recall on our manually annotated dataset. The proposed dataset, source codes and trained models are available on Github (https://github.com/pereira-gha/activefire), creating opportunities for further advances in the field

Active Fire Detection in Landsat-8 Imagery Using Deep Learning Techniques

The paper "Active Fire Detection in Landsat-8 Imagery: a Large-Scale Dataset and a Deep-Learning Study" presents a significant advancement in the application of deep learning for the detection of active fires using satellite imagery. The research addresses a critical aspect of environmental management by enhancing the capability of identifying active fires through satellite data, specifically utilizing data from Landsat-8.

Overview and Dataset

The primary contribution of this paper is the introduction of a substantial dataset tailored for active fire detection. The dataset comprises over 150,000 image patches, amounting to more than 200GB, extracted from Landsat-8 images captured during wildfires across various global locations in August and September 2020. The images include a range of spectral data with 10-band spectral imagery, which were used to create outputs using three established handcrafted algorithms for active fire detection. Moreover, a subset of the dataset is manually annotated to provide ground-truth data for model evaluation.

Methodology

The paper investigates the use of Convolutional Neural Networks (CNNs) for the task of active fire detection, demonstrating how these networks can approximate and in some cases surpass traditional handcrafted algorithms. Three variations of the U-Net architecture were employed:

• A 10-channel U-Net using all available bands.
• A 3-channel U-Net focusing on the channels with the strongest fire signals.
• A light version of the 3-channel U-Net with fewer parameters for efficiency.

These networks are evaluated not only for their capability to replicate traditional algorithm outputs but also for their performance when trained on different target scenarios, including the intersection and majority voting results of the traditional methods.

Results

The results highlight the proficiency of CNNs in active fire detection, with the best model achieving 87.2% precision and 92.4% recall against the manually annotated dataset. This precision-recall balance underlines the effectiveness of deep learning in providing robust detection capabilities. The experiments underscore that a reduced set of spectral channels (3-channel input) is often sufficient, leading to resource savings in bandwidth and storage without significantly impacting performance.

Implications and Future Directions

The paper has several implications for the field of remote sensing and environmental monitoring. The introduced dataset, alongside the demonstrated efficacy of CNN architectures, sets a precedent for future research and development of machine learning-based approaches in satellite image analysis. The findings suggest that machine learning techniques can successfully mitigate the limitations of threshold-based detection inherent in handcrafted algorithms.

Looking forward, there are numerous avenues for research enhancement. These include exploring more sophisticated neural network architectures, incorporating temporal dynamics into fire detection, and expanding the approach to other types of satellite data, such as from Sentinel-2. Additionally, integrating these models into real-time monitoring systems could vastly improve response times and resource allocation during wildfire outbreaks.

Conclusion

This paper exemplifies the integration of deep learning with environmental management, focusing on improving the detection of active fires through satellite imagery. By introducing a large-scale dataset and providing a detailed paper on CNN-based fire detection, it opens the door for further exploration and application of AI in remote sensing, promising enhanced capabilities in environmental monitoring and response strategies.

Overall, the research presented provides a comprehensive analysis and demonstration of the utility of deep learning methods in active fire detection, showcasing both practical applicability and theoretical advancements in the field.