PyroFocus: A Deep Learning Approach to Real-Time Wildfire Detection in Multispectral Remote Sensing Imagery

TL;DR

This paper tackles real-time onboard wildfire detection in multispectral imagery by proposing PyroFocus, a two-stage cascade that first classifies fire-containing patches and then performs FRP regression or segmentation only on those patches. Through extensive evaluation on MASTER/FIREX-AQ-like data, the approach demonstrates substantial reductions in inference time (up to ~73% for segmentation and ~50% for FRP) while preserving high accuracy, highlighting its potential for edge deployment on airborne and spaceborne platforms. The work also provides a comparative analysis of CNN and Transformer architectures, identifying ResNet and SSRN as strong candidates for onboard missions, and outlines practical considerations and future hardware-focused optimizations. Overall, PyroFocus offers a scalable, resource-conscious framework for fast, fine-grained wildfire monitoring that can be adapted to other time-sensitive remote sensing tasks.

Abstract

Rapid and accurate wildfire detection is crucial for emergency response and environmental management. In airborne and spaceborne missions, real-time algorithms must distinguish between no fire, active fire, and post-fire conditions, and estimate fire intensity. Multispectral and hyperspectral thermal imagers provide rich spectral information, but high data dimensionality and limited onboard resources make real-time processing challenging. As wildfires increase in frequency and severity, the need for low-latency and computationally efficient onboard detection methods is critical. We present a systematic evaluation of multiple deep learning architectures, including custom Convolutional Neural Networks (CNNs) and Transformer-based models, for multi-class fire classification. We also introduce PyroFocus, a two-stage pipeline that performs fire classification followed by fire radiative power (FRP) regression or segmentation to reduce inference time and computational cost for onboard deployment. Using data from NASA's MODIS/ASTER Airborne Simulator (MASTER), which is similar to a next-generation fire detection sensor, we compare accuracy, inference latency, and resource efficiency. Experimental results show that the proposed two-stage pipeline achieves strong trade-offs between speed and accuracy, demonstrating significant potential for real-time edge deployment in future wildfire monitoring missions.

Figures (7)

• Figure 1: The distribution of pixels in our classification mask for our whole dataset, before data augmentation. The No Fire class contains orders of magnitude more pixels than any other class in the distribution (note the logarithmic scale on the y-axis).
• Figure 2: Bar charts comparing accuracy, precision, recall, and F1 score across all models.
• Figure 3: Normalized confusion matrices for each model. Each cell shows the fraction of predictions for a given class (row) classified into each predicted category (column).
• Figure 4: Comparison of an original RGB image of an active wildfire (left), the model’s predicted FRP heatmap (center), and the overlay of the predicted FRP on the RGB image (right). FRP is in MW.
• Figure 5: Actual vs. predicted values of pixels for FRP prediction.