Integrated Algorithm Based on Bidirectional Characteristics and Feature Selection for Fire Image Classification DOI Open Access

Zuoxin Wang,

Xiaohu Zhao, Yuning Tao

et al.

Electronics, Journal Year: 2023, Volume and Issue: 12(22), P. 4566 - 4566

Published: Nov. 8, 2023

In some fire classification task samples, it is especially important to learn and select limited features. Therefore, enhancing shallow characteristic learning accurately reserving deep characteristics play a decisive role in the final task. this paper, we propose an integrated algorithm based on bidirectional feature selection for image called BCFS-Net. This from two modules, module module; hence, algorithm. The main process of as follows: First, construct convolution obtain multiple sets traditional convolutions dilated mining Then, improve Inception V3 module. By utilizing attention mechanism Euclidean distance, points with greater correlation between maps generated by are selected. Next, comprehensively consider integrate richer semantic information dimensions. Finally, use further features complete We validated feasibility our proposed three public datasets, overall accuracy value BoWFire dataset reached 88.9%. outdoor 96.96%. Fire Smoke 81.66%.

Language: Английский

Efficient fire and smoke detection in complex environments via adaptive spatial feature fusion and dual attention mechanism DOI
Jie Hu, Lei Wang, Bo Peng

et al.

Digital Signal Processing, Journal Year: 2025, Volume and Issue: 159, P. 104982 - 104982

Published: Jan. 8, 2025

Language: Английский

Citations

3
An Improved Forest Fire and Smoke Detection Model Based on YOLOv5 DOI Open Access
Junhui Li, Renjie Xu, Yunfei Liu

et al.

Forests, Journal Year: 2023, Volume and Issue: 14(4), P. 833 - 833

Published: April 18, 2023

Forest fires are destructive and rapidly spreading, causing great harm to forest ecosystems humans. Deep learning techniques can adaptively learn extract features of smoke. However, the complex backgrounds different fire smoke in captured images make detection difficult. Facing background smoke, it is difficult for traditional machine methods design a general feature extraction module extraction. effective many fields, so this paper improves on You Only Look Once v5 (YOLOv5s) model, improved model has better performance First, coordinate attention (CA) integrated into YOLOv5 highlight targets improve identifiability features. Second, we replaced YOLOv5s original spatial pyramidal ensemble fast (SPPF) with receptive field block (RFB) enable focus global information fires. Third, path aggregation network (PANet) neck structure bi-directional pyramid (Bi-FPN). Compared our at [email protected] by 5.1%.

Language: Английский

Citations

24
CNTCB-YOLOv7: An Effective Forest Fire Detection Model Based on ConvNeXtV2 and CBAM DOI Creative Commons
Yiqing Xu, Jiaming Li, Long Zhang

et al.

Fire, Journal Year: 2024, Volume and Issue: 7(2), P. 54 - 54

Published: Feb. 12, 2024

In the context of large-scale fire areas and complex forest environments, task identifying subtle features aspects can pose a significant challenge for deep learning model. As result, to enhance model’s ability represent its precision in detection, this study initially introduces ConvNeXtV2 Conv2Former You Only Look Once version 7 (YOLOv7) algorithm, separately, then compares results with original YOLOv7 algorithm through experiments. After comprehensive comparison, proposed ConvNeXtV2-YOLOv7 based on exhibits superior performance detecting fires. Additionally, order further focus network crucial information fires minimize irrelevant background interference, efficient layer aggregation (ELAN) structure backbone is enhanced by adding four attention mechanisms: normalization-based module (NAM), simple mechanism (SimAM), global (GAM), convolutional block (CBAM). The experimental results, which demonstrate suitability ELAN combined CBAM lead proposal new method detection called CNTCB-YOLOv7. CNTCB-YOLOv7 outperforms an increase accuracy 2.39%, recall rate 0.73%, average (AP) 1.14%.

Language: Английский

Citations

10
Precision-Boosted Forest Fire Target Detection via Enhanced YOLOv8 Model DOI Creative Commons

Zhaoxu Yang,

Yifan Shao,

Wei Ye

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(6), P. 2413 - 2413

Published: March 13, 2024

Forest fires present a significant challenge to ecosystems, particularly due factors like tree cover that complicate fire detection tasks. While technologies, YOLO, are widely used in forest protection, capturing diverse and complex flame features remains challenging. Therefore, we propose an enhanced YOLOv8 multiscale method. This involves adjusting the network structure integrating Deformable Convolution SCConv modules better adapt complexities. Additionally, introduce Coordinate Attention mechanism Detection module more effectively capture feature information enhance model accuracy. We adopt WIoU v3 loss function implement dynamically non-monotonic optimize gradient allocation strategies. Our experimental results demonstrate our achieves mAP of 90.02%, approximately 5.9% higher than baseline network. method significantly improves accuracy, reduces False Positive rates, demonstrates excellent applicability real scenarios.

Language: Английский

Citations

7
Review of Modern Forest Fire Detection Techniques: Innovations in Image Processing and Deep Learning DOI Creative Commons

Berk Özel,

Muhammad Shahab Alam, Muhammad Umer Khan

et al.

Information, Journal Year: 2024, Volume and Issue: 15(9), P. 538 - 538

Published: Sept. 3, 2024

Fire detection and extinguishing systems are critical for safeguarding lives minimizing property damage. These especially vital in combating forest fires. In recent years, several fires have set records their size, duration, level of destruction. Traditional fire methods, such as smoke heat sensors, limitations, prompting the development innovative approaches using advanced technologies. Utilizing image processing, computer vision, deep learning algorithms, we can now detect with exceptional accuracy respond promptly to mitigate impact. this article, conduct a comprehensive review articles from 2013 2023, exploring how these technologies applied extinguishing. We delve into modern techniques enabling real-time analysis visual data captured by cameras or satellites, facilitating smoke, flames, other fire-related cues. Furthermore, explore utilization machine training intelligent algorithms recognize patterns features. Through examination current research development, aims provide insights potential future directions learning.

Language: Английский

Citations

5
An efficient model for real-time wildfire detection in complex scenarios based on multi-head attention mechanism DOI
Xiaotian Wang, Zhongjie Pan, Huajian Gao

et al.

Journal of Real-Time Image Processing, Journal Year: 2023, Volume and Issue: 20(4)

Published: June 2, 2023

Language: Английский

Citations

13
An Efficient Forest Fire Target Detection Model Based on Improved YOLOv5 DOI Creative Commons
Long Zhang, Jiaming Li, Fuquan Zhang

et al.

Fire, Journal Year: 2023, Volume and Issue: 6(8), P. 291 - 291

Published: July 31, 2023

To tackle the problem of missed detections in long-range detection scenarios caused by small size forest fire targets, initiatives have been undertaken to enhance feature extraction and precision models designed for imagery. In this study, two algorithms, DenseM-YOLOv5 SimAM-YOLOv5, were proposed modifying backbone network You Only Look Once version 5 (YOLOv5). From perspective lightweight models, compared YOLOv5, SimAM-YOLOv5 reduced parameter 28.57%. Additionally, although showed a slight decrease recall rate, it achieved improvements average (AP) varying degrees. The algorithm 2.24% increase precision, as well 1.2% rate 1.52% AP YOLOv5 algorithm. Despite having higher size, outperformed terms detection.

Language: Английский

Citations

12
Semi-Supervised Method for Underwater Object Detection Algorithm Based on Improved YOLOv8 DOI Creative Commons
Siyi Xu, Jian Wang, Qingbing Sang

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(3), P. 1065 - 1065

Published: Jan. 22, 2025

Deep learning-based object detection technology is rapidly developing, and underwater detection, an important subcategory, plays a crucial role in various fields such as structure repair maintenance, well marine scientific research. Some of the major challenges are relatively limited availability image video datasets high cost acquiring high-quality, diverse training data. To address this, we propose novel method, SUD-YOLO, based on Mean Teacher semi-supervised learning strategy. More specifically, it combines small number labeled samples with large unlabeled samples, using teacher model to guide generation pseudo-labels. In addition, multi-scale pseudo-label enhancement module developed specifically issue low-quality overcome model’s difficulty feature extraction, integrate receptive-field attention mechanism local spatial features then design lightweight head task alignment concept further improve extraction capability. Experimental results DUO dataset show that, by only 10% data, proposed method achieves average precision 50.8, which improvement 11.0% over fully supervised YOLOv8 algorithm, 11.3% YOLOv11 9.3% Efficient 3.4% Unbiased while 20% computational required.

Language: Английский

Citations

0
A Small-Sample Target Detection Method for Transmission Line Hill Fires Based on Meta-Learning YOLOv11 DOI Creative Commons

Yaoran Huo,

Shuicheng Yan,

Jian Xu

et al.

Energies, Journal Year: 2025, Volume and Issue: 18(6), P. 1511 - 1511

Published: March 19, 2025

China has a large number of transmission lines laid in the mountains and forests other regions, these enable national strategic projects such as west-east power project. However, occurrence mountain fires corresponding areas will seriously affect projects. At same time, yield fewer image samples complex backgrounds. Based on this, this paper proposes line hill fire detection model with YOLOv11 basic framework, named meta-learning attention YOLO (MA-YOLO). Firstly, feature extraction module it is replaced meta-feature module, scale head adjusted to detect smaller-sized targets. After re-weighting learns class-specific vectors from support set uses them recalibrate mapping meta-features. To enhance model’s ability learn target features backgrounds, adaptive fusion (AFF) integrated into process improve capabilities, filter out useless information features, reduce interference backgrounds detection. The experimental results show that accuracy MA-YOLO improved by 10.8% few-shot scenarios. misses targets different scenarios less likely be affected

Language: Английский

Citations

0
Dynamic fire and smoke detection module with enhanced feature integration and attention mechanisms DOI Creative Commons
Ammar Amjad,

Aamer Mohamed Huroon,

Hsien-Tsung Chang

et al.

Pattern Analysis and Applications, Journal Year: 2025, Volume and Issue: 28(2)

Published: April 5, 2025

Language: Английский

Citations

0