Adversarial Robustness for Deep Learning-based Wildfire Prediction Models

TL;DR

This work tackles the robustness of DNN-based wildfire smoke detection under data scarcity by introducing WARP, a model-agnostic adversarial robustness procedure that uses global Gaussian noise overlays and local cloud-like patches to test CNN and transformer detectors. By comparing YOLOv8n (CNN) and RT-DETR-l (transformer) on NEMO and HPWREN data, the study reveals that transformers are more susceptible to global perturbations while CNNs show localization vulnerabilities under local perturbations. The authors report concrete metrics for robustness, such as L, α_i, β_i, and γ_i, and propose wildfire-specific data augmentation strategies to mitigate identified weaknesses. Overall, WARP provides a practical, architecture-agnostic pathway toward more reliable, early wildfire warning systems and informed data augmentation to improve robustness in real-world surveillance settings.

Abstract

Rapidly growing wildfires have recently devastated societal assets, exposing a critical need for early warning systems to expedite relief efforts. Smoke detection using camera-based Deep Neural Networks (DNNs) offers a promising solution for wildfire prediction. However, the rarity of smoke across time and space limits training data, raising model overfitting and bias concerns. Current DNNs, primarily Convolutional Neural Networks (CNNs) and transformers, complicate robustness evaluation due to architectural differences. To address these challenges, we introduce WARP (Wildfire Adversarial Robustness Procedure), the first model-agnostic framework for evaluating wildfire detection models' adversarial robustness. WARP addresses inherent limitations in data diversity by generating adversarial examples through image-global and -local perturbations. Global and local attacks superimpose Gaussian noise and PNG patches onto image inputs, respectively; this suits both CNNs and transformers while generating realistic adversarial scenarios. Using WARP, we assessed real-time CNNs and Transformers, uncovering key vulnerabilities. At times, transformers exhibited over 70% precision degradation under global attacks, while both models generally struggled to differentiate cloud-like PNG patches from real smoke during local attacks. To enhance model robustness, we proposed four wildfire-oriented data augmentation techniques based on WARP's methodology and results, which diversify smoke image data and improve model precision and robustness. These advancements represent a substantial step toward developing a reliable early wildfire warning system, which may be our first safeguard against wildfire destruction.

Figures (14)

• Figure 1: Smoke object detection from surveillance video images. A DNN object detection model creates bounding boxes (see green box) to locate smoke as the target object. Image adapted from ALERTWildfire.
• Figure 2: WARP workflow.
• Figure 3: Local noise injection. (a) An example image with injected cloud-like noise at a grid location, highlighted by the red circle. The green bounding box indicates the ground-truth location of the smoke. (b) The cloud-like PNG patch used as local noise, with a background added for visibility. Adapted from the internet.
• Figure 4: Illustration of the proposed localization deception test.
• Figure 5: mAP Percentage Loss plotted across all iterations.