A comprehensive survey of research towards AI-enabled unmanned aerial systems in pre-, active-, and post-wildfire management

Information Fusion, Journal Year: 2024, Volume and Issue: 108, P. 102369 - 102369

Published: March 22, 2024

Wildfires have emerged as one of the most destructive natural disasters worldwide, causing catastrophic losses. These losses underscored urgent need to improve public knowledge and advance existing techniques in wildfire management. Recently, use Artificial Intelligence (AI) wildfires, propelled by integration Unmanned Aerial Vehicles (UAVs) deep learning models, has created an unprecedented momentum implement develop more effective Although survey papers explored learning-based approaches wildfire, drone disaster management, risk assessment, a comprehensive review emphasizing application AI-enabled UAV systems investigating role methods throughout overall workflow multi-stage including pre-fire (e.g., vision-based vegetation fuel measurement), active-fire fire growth modeling), post-fire tasks evacuation planning) is notably lacking. This synthesizes integrates state-of-the-science reviews research at nexus observations modeling, AI, UAVs - topics forefront advances elucidating AI performing monitoring actuation from pre-fire, through stage, To this aim, we provide extensive analysis remote sensing with particular focus on advancements, device specifications, sensor technologies relevant We also examine management approaches, monitoring, prevention strategies, well planning, damage operation strategies. Additionally, summarize wide range computer vision emphasis Machine Learning (ML), Reinforcement (RL), Deep (DL) algorithms for classification, segmentation, detection, tasks. Ultimately, underscore substantial advancement modeling cutting-edge UAV-based data, providing novel insights enhanced predictive capabilities understand dynamic behavior.

Language: Английский

---

Detection of forest fire using deep convolutional neural networks with transfer learning approach

Applied Soft Computing, Journal Year: 2023, Volume and Issue: 143, P. 110362 - 110362

Published: May 9, 2023

Language: Английский

---

Advancing Maritime Safety: Early Detection of Ship Fires through Computer Vision, Deep Learning Approaches, and Histogram Equalization Techniques

Fire, Journal Year: 2024, Volume and Issue: 7(3), P. 84 - 84

Published: March 8, 2024

The maritime sector confronts an escalating challenge with the emergence of onboard fires aboard in ships, evidenced by a pronounced uptick incidents recent years. ramifications such transcend immediate safety apprehensions, precipitating repercussions that resonate on global scale. This study underscores paramount importance ship fire detection as proactive measure to mitigate risks and fortify comprehensively. Initially, we created labeled custom dataset. collected images are varied their size, like having high- low-resolution Then, leveraging YOLO (You Only Look Once) object algorithm developed efficacious accurate model for discerning presence vessels navigating marine routes. was trained 50 epochs more than 25,000 images. histogram equalization (HE) technique also applied avoid destruction from water vapor increase detection. After training, ships were input into inference after HE, be categorized two classes. Empirical findings gleaned proposed methodology attest model’s exceptional efficacy, highest accuracy attaining noteworthy 0.99% across both fire-afflicted non-fire scenarios.

Language: Английский

---

Deep artificial intelligence applications for natural disaster management systems: A methodological review

Ecological Indicators, Journal Year: 2024, Volume and Issue: 163, P. 112067 - 112067

Published: May 6, 2024

Deep learning techniques through semantic segmentation networks have been widely used for natural disaster analysis and response. The underlying base of these implementations relies on convolutional neural (CNNs) that can accurately precisely identify locate the respective areas interest within satellite imagery or other forms remote sensing data, thereby assisting in evaluation, rescue planning, restoration endeavours. Most CNN-based deep-learning models encounter challenges related to loss spatial information insufficient feature representation. This issue be attributed their suboptimal design layers capture multiscale-context failure include optimal during pooling procedures. In early CNNs, network encodes elementary representations, such as edges corners, whereas, progresses toward later layers, it more intricate characteristics, complicated geometric shapes. theory, is advantageous a extract features from several levels because generally yield improved results when both simple maps are employed together. study comprehensively reviews current developments deep methodologies segment images associated with disasters. Several popular models, SegNet U-Net, FCNs, FCDenseNet, PSPNet, HRNet, DeepLab, exhibited notable achievements various applications, including forest fire delineation, flood mapping, earthquake damage assessment. These demonstrate high level efficacy distinguishing between different land cover types, detecting infrastructure has compromised damaged, identifying regions fire-susceptible further dangers.

Language: Английский

---

Deep Learning Approaches for Wildland Fires Remote Sensing: Classification, Detection, and Segmentation

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(7), P. 1821 - 1821

Published: March 29, 2023

The world has seen an increase in the number of wildland fires recent years due to various factors. Experts warn that will continue coming years, mainly because climate change. Numerous safety mechanisms such as remote fire detection systems based on deep learning models and vision transformers have been developed recently, showing promising solutions for these tasks. To best our knowledge, there are a limited published studies literature, which address implementation classification, detection, segmentation As such, this paper, we present up-to-date comprehensive review analysis methods their performances. First, previous works related including reviewed. Then, most popular public datasets used tasks presented. Finally, discusses challenges existing works. Our shows how approaches outperform traditional machine can significantly improve performance detecting, segmenting, classifying wildfires. In addition, main research gaps future directions researchers develop more accurate fields.

Language: Английский

---

Remote Sensing Object Detection Meets Deep Learning: A metareview of challenges and advances

IEEE Geoscience and Remote Sensing Magazine, Journal Year: 2023, Volume and Issue: 11(4), P. 8 - 44

Published: Oct. 24, 2023

Remote sensing object detection (RSOD), one of the most fundamental and challenging tasks in remote field, has received long-standing attention. In recent years, deep learning techniques have demonstrated robust feature representation capabilities led to a big leap development RSOD techniques. this era rapid technical evolution, article aims present comprehensive review achievements learning-based methods. More than 300 papers are covered review. We identify five main challenges RSOD, including multiscale detection, rotated weak tiny with limited supervision, systematically corresponding methods developed hierarchical division manner. also widely used benchmark datasets evaluation metrics within field as well application scenarios for RSOD. Future research directions provided further promoting

Language: Английский

---

YOLO-SF: YOLO for Fire Segmentation Detection

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 111079 - 111092

Published: Jan. 1, 2023

Owing to the problems of missed detection, false and low accuracy current fire detection algorithm, a segmentation YOLO-SF, is proposed. This algorithm combines instance technology with YOLOv7-Tiny object improve its accuracy. We gather images that include both non-fire elements create dataset (FSD). The head YOLOR adopted model enhance ability express details. MobileViTv2 module introduced build backbone network, which effectively reduces parameters while ensuring network's extract features. Efficient Layer Aggregation Network (ELAN) neck network augmented Convolutional Block Attention Module (CBAM) broaden receptive field attention image channel spatial information. Additionally, Varifocal Loss used address problem inaccurate positioning in edge areas images. Compared for Box Mask, precision increases by 5.9% 6.2%, recall 2.5% 3.3%, mAP 4% 6%. In addition, FPS reaches 55.64, satisfying requirements real-time detection. improved exhibits good generalization performance robustness.

Language: Английский

---

MS-FRCNN: A Multi-Scale Faster RCNN Model for Small Target Forest Fire Detection

Forests, Journal Year: 2023, Volume and Issue: 14(3), P. 616 - 616

Published: March 19, 2023

Unmanned aerial vehicles (UAVs) are widely used for small target detection of forest fires due to its low-risk rate, low cost and high ground coverage. However, the accuracy is still not ideal irregular shape, different scale how easy it can be blocked by obstacles. This paper proposes a multi-scale feature extraction model (MS-FRCNN) fire improving classic Faster RCNN model. In MS-FRCNN model, ResNet50 replace VGG-16 as backbone network alleviate gradient explosion or dispersion phenomenon when extracting features. Then, map output input into Feature Pyramid Network (FPN). The advantage FPN will help improve ability obtain detailed information. At same time, uses new attention module PAM in Regional Proposal (RPN), which reduce influence complex backgrounds images through parallel operation channel space attention, so that RPN pay more semantic location information fires. addition, soft-NMS algorithm instead an NMS error deletion detected frames. experimental results show that, compared baseline proposed this achieved better performance fires, was 5.7% higher than models. It shows strategy image mechanism suppress interference adopted really detection.

Language: Английский

---

A Semi-Supervised Method for Real-Time Forest Fire Detection Algorithm Based on Adaptively Spatial Feature Fusion

Forests, Journal Year: 2023, Volume and Issue: 14(2), P. 361 - 361

Published: Feb. 11, 2023

Forest fires occur frequently around the world, causing serious economic losses and human casualties. Deep learning techniques based on convolutional neural networks (CNN) are widely used in intelligent detection of forest fires. However, CNN-based fire target models lack global modeling capabilities cannot fully extract contextual information about targets. CNNs also pay insufficient attention to vulnerable interference invalid features similar fires, resulting low accuracy detection. In addition, require a large number labeled datasets. Manual annotation is often annotate huge amount datasets; however, this takes lot time. To address these problems, paper proposes model, TCA-YOLO, with YOLOv5 as basic framework. Firstly, we combine Transformer encoder its powerful capability self-attention mechanism CNN feature extraction network enhance Secondly, order model’s focus targets, integrate Coordinate Attention (CA) mechanism. CA not only acquires inter-channel but considers direction-related location information, which helps model better locate identify Integrated adaptively spatial fusion (ASFF) technology allows automatically filter out useless from other layers efficiently fuse suppress complex backgrounds area for Finally, semi-supervised save manual labeling effort. The experimental results show that average TCA-YOLO improves by 5.3 compared unimproved YOLOv5. outperformed detecting targets different scenarios. ability was much improved. Additionally, it could more accurately. misses fewer less likely be interfered fire-like focused at small-target FPS reaches 53.7, means speed meets requirements real-time

Language: Английский

---

Video Fire Detection Methods Based on Deep Learning: Datasets, Methods, and Future Directions

Fire, Journal Year: 2023, Volume and Issue: 6(8), P. 315 - 315

Published: Aug. 14, 2023

Among various calamities, conflagrations stand out as one of the most-prevalent and -menacing adversities, posing significant perils to public safety societal progress. Traditional fire-detection systems primarily rely on sensor-based detection techniques, which have inherent limitations in accurately promptly detecting fires, especially complex environments. In recent years, with advancement computer vision technology, video-oriented fire owing their non-contact sensing, adaptability diverse environments, comprehensive information acquisition, progressively emerged a novel solution. However, approaches based handcrafted feature extraction struggle cope variations smoke or flame caused by different combustibles, lighting conditions, other factors. As powerful flexible machine learning framework, deep has demonstrated advantages video detection. This paper summarizes deep-learning-based video-fire-detection methods, focusing advances commonly used datasets for recognition, object detection, segmentation. Furthermore, this provides review outlook development prospects field.

Language: Английский

---