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

---

TFNet: Transformer-Based Multi-Scale Feature Fusion Forest Fire Image Detection Network

Fire, Journal Year: 2025, Volume and Issue: 8(2), P. 59 - 59

Published: Jan. 30, 2025

Forest fires pose a severe threat to ecological environments and the safety of human lives property, making real-time forest fire monitoring crucial. This study addresses challenges in image object detection, including small targets, sparse smoke, difficulties feature extraction, by proposing TFNet, Transformer-based multi-scale fusion detection network. TFNet integrates several components: SRModule, CG-MSFF Encoder, Decoder Head, WIOU Loss. The SRModule employs multi-branch structure learn diverse representations images, utilizing 1 × convolutions generate redundant maps enhance diversity. Encoder introduces context-guided attention mechanism combined with adaptive (AFF), enabling effective reweighting features across layers extracting both local global representations. Head refine output iteratively optimizing target queries using self- cross-attention, improving accuracy. Additionally, Loss assigns varying weights IoU metric for predicted versus ground truth boxes, thereby balancing positive negative samples localization Experimental results on two publicly available datasets, D-Fire M4SFWD, demonstrate that outperforms comparative models terms precision, recall, F1-Score, mAP50, mAP50–95. Specifically, dataset, achieved metrics 81.6% 74.8% an F1-Score 78.1%, mAP50 81.2%, mAP50–95 46.8%. On M4SFWD these improved 86.6% 83.3% 84.9%, 89.2%, 52.2%. proposed offers technical support developing efficient practical systems.

Language: Английский

---

Semantic Segmentation of Aerial Imagery: A Novel Approach Leveraging Hierarchical Multi-scale Features and Channel-based Attention for Drone Applications

International journal of engineering. Transactions B: Applications, Journal Year: 2024, Volume and Issue: 37(5), P. 1022 - 1035

Published: Jan. 1, 2024

Drone semantic segmentation is a challenging task in computer vision, mainly due to inherent complexities associated with aerial imagery. This paper presents comprehensive methodology for drone and evaluates its performance using the ICG dataset. The proposed method leverages hierarchical multi-scale feature extraction efficient channel-based attention Atrous Spatial Pyramid Pooling (ASPP) address unique challenges encountered this domain. In study, of compared several state-of-the-art models. findings research highlight effectiveness tackling segmentation. outcomes demonstrate superiority over models, showcasing potential accurate results contribute advancement drone-based applications, such as surveillance, object tracking, environmental monitoring, where precise crucial. obtained experimental that outperforms these existing approaches regarding Dice, mIOU, accuracy metrics. Specifically, achieves an impressive scores 86.51%, 76.23%, 91.74%, respectively.

Language: Английский

---

A deep learning based approach for semantic segmentation of small fires from UAV imagery

Remote Sensing Letters, Journal Year: 2025, Volume and Issue: 16(3), P. 277 - 289

Published: Jan. 17, 2025

In this paper, we propose novel techniques for fire segmentation from unmanned aerial vehicle (UAV) images. (1) We the ObjectDetection+CIELAB thresholding technique, which leverages a pre-trained object detector such as YOLO (``you only look once'') to generate bounding boxes. then apply in CIELAB colour space within these regions detect pixels. This approach significantly improves speed by streamlining task into more focused detection and classification task. (2) further introduce SEG-4CHANNEL generates pixel mask using method. is integrated fourth channel various networks, allowing models concentrate on while minimising background interference. (3) Finally, explore AttentionSeg incorporates an attention module framework (e.g., SegFormer-B5) that utilises all four channels. It combines advantages of colour-space model, convolution neural network (CNN) transformer. large design networks backbones FLAME (Fire Luminosity Airborne-based Machine learning Evaluation) dataset. Our best AttentionSeg-B5, segments with intersection-over-union (IoU) score 84.15% 91.39% F1-score. The code has been released at https://github.com/CandleLabAI/FireSegmentation.

Language: Английский

---

FireSeg: A weakly supervised fire segmentation framework via pre-trained latent diffusion models

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 126852 - 126852

Published: Feb. 1, 2025

Language: Английский

---

Research and application of deep learning object detection methods for forest fire smoke recognition

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 10, 2025

Language: Английский

---

FlameTransNet: Advancing Forest Flame Segmentation with Fusion and Augmentation Techniques

Forests, Journal Year: 2023, Volume and Issue: 14(9), P. 1887 - 1887

Published: Sept. 17, 2023

Forest fires pose severe risks, including habitat loss and air pollution. Accurate forest flame segmentation is vital for effective fire management protection of ecosystems. It improves detection, response, understanding behavior. Due to the easy accessibility rich information content remote sensing images, techniques are frequently applied in segmentation. With advancement deep learning, convolutional neural network (CNN) have been widely adopted achieved remarkable results. However, images often high resolutions, relative entire image, regions relatively small, resulting class imbalance issues. Additionally, mainstream semantic methods limited by receptive field CNNs, making it challenging effectively extract global features from leading poor performance when relying solely on labeled datasets. To address these issues, we propose a method based deeplabV3+ model, incorporating following design strategies: (1) an adaptive Copy-Paste data augmentation introduced learn samples (Images that cannot be adequately learned due other factors) effectively, (2) transformer modules concatenated parallelly integrated into encoder, while CBAM attention mechanism added decoder fully image features, (3) dice mitigate problem. By conducting validation our self-constructed dataset, approach has demonstrated superior across multiple metrics compared current state-of-the-art methods. Specifically, terms IoU (Intersection over Union), Precision, Recall category, exhibited notable enhancements 4.09%, 3.48%, 1.49%, respectively, best-performing UNet model. Moreover, advancements 11.03%, 9.10%, 4.77% same aforementioned as baseline

Language: Английский

---

Wildfire Impact Analysis and Spread Dynamics Estimation on Satellite Images Using Deep Learning

Journal of the Indian Society of Remote Sensing, Journal Year: 2024, Volume and Issue: 52(6), P. 1385 - 1403

Published: May 25, 2024

Language: Английский

---

A multi-branch dual attention segmentation network for epiphyte drone images

Image and Vision Computing, Journal Year: 2024, Volume and Issue: 148, P. 105099 - 105099

Published: May 31, 2024

Language: Английский

---

A Multi-Branch Dual Attention Segmentation Network for Epiphyte Drone Images

Published: Jan. 1, 2024

Sampling and monitoring of epiphytes growing in trees inside canopy using Unmanned Aerial Vehicles(UAV’s) provides better approach for Botanist. However, the images captured by UAV’s usually contain complex background, uneven lighting small targets. Apart from this, obtaining large number diverse, high-quality target is difficult due to accessibility issues with canopy. This poses a significant challenge existing advanced segmentation networks. In recent years, Deep Learning (DL) has witnessed widespread adoption image methodologies, including popular approaches like U-shaped architectures, vision transformer-based models, hybrid Nevertheless, their reliance on substantial quantities data effective training limitation when applied smaller datasets exhibiting heterogeneous quality. Furthermore, these networks often incorporate deep encoders, an increased convolution filters, heightened emphasis local features rather than global features, aspects crucial achieving accurate segmentation. Appropriate attention while segmenting target/s quality ensures predictions boundary regions correct mapping pixels respective classes. Therefore, we propose multi-branch dual network segment epiphyte drone images. The proposed consist dedicated parallel branches extracting during encoding stage. encoder passed spatial channel modules relevant focus given important target. two fused summation decoder crossed fusion technique effectively combine complement multiple branches. validated dataset 132 training. output compared state art transformer used previous study [40]. more predicting class labels. Specifically, taken close target, under low light are zoomed cropped. We calculated Intersection over Union (IoU) score conduct quantitative analysis trained model's performance across various qualities test cases. qualitative assessment predicted mask presented through falsecolor images, highlighting accurately regions, areas where failed, instances false predictions. produced exhibited 5% improvement average IoU 48% increase or shadow conditions, remarkable 68% that were cropped, as model.

Language: Английский

---