SegNet: A Segmented Deep Learning based Convolutional Neural Network Approach for Drones Wildfire Detection

TL;DR

The paper tackles real-time wildfire detection from UAV imagery under data and hardware constraints. It introduces SegNet, a segmented CNN that partitions high-resolution frames into 12 segments to reduce feature maps and accelerate inference while preserving amorphous-fire features. The approach achieves about 98.2% test accuracy on a drone-derived dataset with a per-image processing time of roughly 240 ms, outperforming AlexNet and GoogleNet in both speed and accuracy. The method is demonstrated as viable for onboard deployment, requiring modest memory (~44 MB) and simple workflow, which supports rapid, automated early-warning in wildfire surveillance.

Abstract

This research addresses the pressing challenge of enhancing processing times and detection capabilities in Unmanned Aerial Vehicle (UAV)/drone imagery for global wildfire detection, despite limited datasets. Proposing a Segmented Neural Network (SegNet) selection approach, we focus on reducing feature maps to boost both time resolution and accuracy significantly advancing processing speeds and accuracy in real-time wildfire detection. This paper contributes to increased processing speeds enabling real-time detection capabilities for wildfire, increased detection accuracy of wildfire, and improved detection capabilities of early wildfire, through proposing a new direction for image classification of amorphous objects like fire, water, smoke, etc. Employing Convolutional Neural Networks (CNNs) for image classification, emphasizing on the reduction of irrelevant features vital for deep learning processes, especially in live feed data for fire detection. Amidst the complexity of live feed data in fire detection, our study emphasizes on image feed, highlighting the urgency to enhance real-time processing. Our proposed algorithm combats feature overload through segmentation, addressing challenges arising from diverse features like objects, colors, and textures. Notably, a delicate balance of feature map size and dataset adequacy is pivotal. Several research papers use smaller image sizes, compromising feature richness which necessitating a new approach. We illuminate the critical role of pixel density in retaining essential details, especially for early wildfire detection. By carefully selecting number of filters during training, we underscore the significance of higher pixel density for proper feature selection. The proposed SegNet approach is rigorously evaluated using real-world dataset obtained by a drone flight and compared to state-of-the-art literature.

Figures (9)

• Figure 1: Wildfires is a global and growing threat: (a) Global tree cover loss due to forest fires, 2001-2022 (courtesy to Ref32). (b) Global distribution of forest land Ref44. (c) Number of fires and suppression cost with area burnt along with linear trend lines from 1991-2015 (National Interagency Fire Center Ref42). (d) Wildfire metrics of Canada, USA, Australia, and Brazil Ref27.
• Figure 2: Environmental challenges: (a) Images with fog can be misclassified as smoke caused by wildfire, and (b) Red trees in the forest could yield undesired false results by model due to misclassification of the red spots (trees) in the image Ref50.
• Figure 3: Dataset and preparation: (a) flames in a complete image, (b) flames in a segmented image, (c) flames in normal image (courtesy to WXChasing), and (d) flames in an augmented image, (e) no scaling (courtesy to Bloomberg), (f) Loss of information when scaling an image’s resolution.
• Figure 4: Segmentation of a complete image into twelve segmented images avoids loss of critical information and enables fire spots to be the prominent feature in the segmented image for better feature detection and selection. This approach can only be implemented on objects, under detection, that are amorphous in nature (do not exhibit a consistent shape or form).
• Figure 5: The proposed SegNet architecture specifically designed to be light on computations. This model can be deployed in drones with weak processing systems. The input image to this architecture is a segmented image achieved after segmentation of a complete $1280\times720$ image into 12 smaller $320\times240$ segments.