Wildfire Smoke Classification Based on Synthetic Images and Pixel- and Feature-Level Domain Adaptation.

Jun Mao,Jiyan Yin,Ye Tian,Wenbin Cui,Change Zheng

doi:10.3390/s21237785

Abstract

Training a deep learning-based classification model for early wildfire smoke images requires a large amount of rich data. However, due to the episodic nature of fire events, it is difficult to obtain wildfire smoke image data, and most of the samples in public datasets suffer from a lack of diversity. To address these issues, a method using synthetic images to train a deep learning classification model for real wildfire smoke was proposed in this paper. Firstly, we constructed a synthetic dataset by simulating a large amount of morphologically rich smoke in 3D modeling software and rendering the virtual smoke against many virtual wildland background images with rich environmental diversity. Secondly, to better use the synthetic data to train a wildfire smoke image classifier, we applied both pixel-level domain adaptation and feature-level domain adaptation. The CycleGAN-based pixel-level domain adaptation method for image translation was employed. On top of this, the feature-level domain adaptation method incorporated ADDA with DeepCORAL was adopted to further reduce the domain shift between the synthetic and real data. The proposed method was evaluated and compared on a test set of real wildfire smoke and achieved an accuracy of 97.39%. The method is applicable to wildfire smoke classification tasks based on RGB single-frame images and would also contribute to training image classification models without sufficient data.

Highlights

The frequency and size of wildfires have increased dramatically worldwide over the past few decades [1], posing a significant threat to natural resources while damaging the lives and property of individuals
We describe the building process of the synthetic smoke dataset, the CycleGAN-based pixel-level domain adaptation process, and the feature-level domain adaptation process based on ADDA with DeepCORAL in SSeennssoorrss 22002211, 2211, xx FFOORR PPEEEERR RREEVVIIEEWW
The 2000 photorealistic smoke images were subjected to a series of data augmentations such as horizontal flip, gamma correction, color dithering, and contrast enhancement, and a total of 5000 images were selected as smoke samples in the source domain for featurelevel domain adaptation (FDA)

Summary

Introduction

The frequency and size of wildfires have increased dramatically worldwide over the past few decades [1], posing a significant threat to natural resources while damaging the lives and property of individuals. In deep learning-based smoke recognition, obtaining diverse wildfire smoke data as a positive sample is challenging due to the episodic nature of fire events, while it is relatively easy to collect forest environments as a negative sample. Under such conditions, the trained models are prone to false positive smoke detection, making it difficult to obtain satisfactory results. Most of the visible images of wildfire smoke are acquired from RGB cameras carried by drones or webcams set up on lookouts. Acquiring a rich and diverse set of high-quality data is essential for deep learning-based smoke recognition tasks

Objectives

Results

Discussion

Conclusion