An Improved Forest Fire Detection Method Based on the Detectron2 Model and a Deep Learning Approach

Sensors, Journal Year: 2023, Volume and Issue: 23(3), P. 1512 - 1512

Published: Jan. 29, 2023

With an increase in both global warming and the human population, forest fires have become a major concern. This can lead to climatic shifts greenhouse effect, among other adverse outcomes. Surprisingly, activities caused disproportionate number of fires. Fast detection with high accuracy is key controlling this unexpected event. To address this, we proposed improved fire method classify based on new version Detectron2 platform (a ground-up rewrite Detectron library) using deep learning approaches. Furthermore, custom dataset was created labeled for training model, it achieved higher precision than models. robust result by improving model various experimental scenarios 5200 images. The detect small over long distances during day night. advantage algorithm its long-distance object interest. results proved that successfully detected 99.3%.

Language: Английский

---

A YOLOv6-Based Improved Fire Detection Approach for Smart City Environments

Sensors, Journal Year: 2023, Volume and Issue: 23(6), P. 3161 - 3161

Published: March 16, 2023

Authorities and policymakers in Korea have recently prioritized improving fire prevention emergency response. Governments seek to enhance community safety for residents by constructing automated detection identification systems. This study examined the efficacy of YOLOv6, a system object running on an NVIDIA GPU platform, identify fire-related items. Using metrics such as speed, accuracy research, time-sensitive real-world applications, we analyzed influence YOLOv6 efforts Korea. We conducted trials using dataset comprising 4000 photos collected through Google, YouTube, other resources evaluate viability recognition tasks. According findings, YOLOv6's performance was 0.98, with typical recall 0.96 precision 0.83. The achieved MAE 0.302%. These findings suggest that is effective technique detecting identifying items Multi-class random forests, k-nearest neighbors, support vector, logistic regression, naive Bayes, XGBoost performed SFSC data system's capacity objects. results demonstrate objects, highest accuracy, values 0.717 0.767. followed forest, 0.468 0.510. Finally, tested simulated evacuation scenario gauge its practicality emergencies. show can accurately real time within response 0.66 s. Therefore, viable option classifier provides when attempting achieving remarkable results. Furthermore, identifies objects while they are being detected real-time. makes tool use initiatives.

Language: Английский

---

Forest Fire Detection and Notification Method Based on AI and IoT Approaches

Future Internet, Journal Year: 2023, Volume and Issue: 15(2), P. 61 - 61

Published: Jan. 31, 2023

There is a high risk of bushfire in spring and autumn, when the air dry. Do not bring any flammable substances, such as matches or cigarettes. Cooking wood fires are permitted only designated areas. These some regulations that enforced hiking going to vegetated forest. However, humans tend disobey disregard guidelines law. Therefore, preemptively stop people from accidentally starting fire, we created technique will allow early fire detection classification ensure utmost safety living things Some relevant studies on forest have been conducted past few years. there still insufficient notification systems for monitoring disasters real time using advanced approaches. came up with solution convergence Internet Things (IoT) You Only Look Once Version 5 (YOLOv5). The experimental results show IoT devices were able validate falsely detected undetected YOLOv5 reported. This report recorded sent department further verification validation. Finally, compared performance our method those recently reported approaches employing widely used matrices test achieved results.

Language: Английский

---

Improved Face Detection Method via Learning Small Faces on Hard Images Based on a Deep Learning Approach

Sensors, Journal Year: 2023, Volume and Issue: 23(1), P. 502 - 502

Published: Jan. 2, 2023

Most facial recognition and face analysis systems start with detection. Early techniques, such as Haar cascades histograms of directed gradients, mainly rely on features that had been manually developed from particular images. However, these techniques are unable to correctly synthesize images taken in untamed situations. deep learning's quick development computer vision has also sped up the a number learning-based detection frameworks, many which have significantly improved accuracy recent years. When detecting faces software, difficulty small, scale, position, occlusion, blurring, partially occluded uncontrolled conditions is one problems identification explored for years but not yet entirely resolved. In this paper, we propose Retina net baseline, single-stage detector, handle challenging problem. We made network improvements boosted speed accuracy. Experiments, used two popular datasets, WIDER FACE FDDB. Specifically, benchmark, our proposed method achieves AP 41.0 at 11.8 FPS single-scale inference strategy 44.2 multi-scale strategy, results among one-stage detectors. Then, trained model during implementation using PyTorch framework, provided an 95.6% faces, successfully detected. Visible experimental show outperforms seamless achieved performance evaluation matrices.

Language: Английский

---

Real-Time Deep Learning-Based Drowsiness Detection: Leveraging Computer-Vision and Eye-Blink Analyses for Enhanced Road Safety

Sensors, Journal Year: 2023, Volume and Issue: 23(14), P. 6459 - 6459

Published: July 17, 2023

Drowsy driving can significantly affect performance and overall road safety. Statistically, the main causes are decreased alertness attention of drivers. The combination deep learning computer-vision algorithm applications has been proven to be one most effective approaches for detection drowsiness. Robust accurate drowsiness systems developed by leveraging learn complex coordinate patterns using visual data. Deep algorithms have emerged as powerful techniques because their ability automatically from given inputs feature extractions raw Eye-blinking-based was applied in this study, which utilized analysis eye-blink patterns. In we used custom data model training experimental results were obtained different candidates. blinking eye mouth region coordinates applying landmarks. rate eye-blinking changes shape analyzed measuring landmarks with real-time fluctuation representations. An performed real time proved existence a correlation between yawning closed eyes, classified drowsy. 95.8% accuracy drowsy-eye detection, 97% open-eye 0.84% 0.98% right-sided falling, 100% left-sided falling. Furthermore, proposed method allowed analysis, where threshold served separator into two classes, “Open” “Closed” states.

Language: Английский

---

Deep Learning-Based Weed Detection Using UAV Images: A Comparative Study

Drones, Journal Year: 2023, Volume and Issue: 7(10), P. 624 - 624

Published: Oct. 7, 2023

Semantic segmentation has been widely used in precision agriculture, such as weed detection, which is pivotal to increasing crop yields. Various well-established and swiftly evolved AI models have developed of late for semantic detection; nevertheless, there insufficient information about their comparative study optimal model selection terms performance this field. Identifying a helps the agricultural community make best use technology. As such, we perform cutting-edge deep learning-based detection using an RGB image dataset acquired with UAV, called CoFly-WeedDB. For this, leverage models, ranging from SegNet DeepLabV3+, combined five backbone convolutional neural networks (VGG16, ResNet50, DenseNet121, EfficientNetB0 MobileNetV2). The results show that UNet CNN best-performing compared other candidate on CoFly-WeedDB dataset, imparting Precision (88.20%), Recall (88.97%), F1-score (88.24%) mean Intersection Union (56.21%). From study, suppose could potentially be by concerned stakeholders (e.g., farmers, industry) detect weeds more accurately field, thereby removing them at earliest point

Language: Английский

---

Fire Detection and Notification Method in Ship Areas Using Deep Learning and Computer Vision Approaches

Sensors, Journal Year: 2023, Volume and Issue: 23(16), P. 7078 - 7078

Published: Aug. 10, 2023

Fire incidents occurring onboard ships cause significant consequences that result in substantial effects. Fires on can have extensive and severe wide-ranging impacts matters such as the safety of crew, cargo, environment, finances, reputation, etc. Therefore, timely detection fires is essential for quick responses powerful mitigation. The study this research paper presents a fire technique based YOLOv7 (You Only Look Once version 7), incorporating improved deep learning algorithms. architecture, with an E-ELAN (extended efficient layer aggregation network) its backbone, serves basis our system. Its enhanced feature fusion makes it superior to all predecessors. To train model, we collected 4622 images various ship scenarios performed data augmentation techniques rotation, horizontal vertical flips, scaling. Our through rigorous evaluation, showcases capabilities recognition improve maritime safety. proposed strategy successfully achieves accuracy 93% detecting minimize catastrophic incidents. Objects having visual similarities may lead false prediction by but be controlled expanding dataset. However, model utilized real-time detector challenging environments small-object detection. Advancements models hold potential enhance measures, exhibits potential. Experimental results proved method used protection monitoring port areas. Finally, compared performance those recently reported fire-detection approaches employing widely matrices test classification achieved.

Language: Английский

---

Deep learning in cropland field identification: A review

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 222, P. 109042 - 109042

Published: May 17, 2024

Language: Английский

---

A Systematic Literature Review and Bibliometric Analysis of Semantic Segmentation Models in Land Cover Mapping

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(12), P. 2222 - 2222

Published: June 19, 2024

Recent advancements in deep learning have spurred the development of numerous novel semantic segmentation models for land cover mapping, showcasing exceptional performance delineating precise boundaries and producing highly accurate maps. However, to date, no systematic literature review has comprehensively examined context mapping. This paper addresses this gap by synthesizing recent mapping from 2017 2023, drawing insights on trends, data sources, model structures, metrics based a 106 articles. Our analysis identifies top journals field, including MDPI Remote Sensing, IEEE Journal Selected Topics Earth Science, Transactions Geoscience Sensing Letters, ISPRS Of Photogrammetry And Sensing. We find that research predominantly focuses cover, urban areas, precision agriculture, environment, coastal forests. Geographically, 35.29% study areas are located China, followed USA (11.76%), France (5.88%), Spain (4%), others. Sentinel-2, Sentinel-1, Landsat satellites emerge as most used sources. Benchmark datasets such Vaihingen Potsdam, LandCover.ai, DeepGlobe, GID frequently employed. Model architectures utilize encoder–decoder hybrid convolutional neural network-based structures because their impressive performances, with limited adoption transformer-based due its computational complexity issue slow convergence speed. Lastly, highlights existing key gaps field guide future directions.

Language: Английский

---

Explainable Lightweight Block Attention Module Framework for Network-Based IoT Attack Detection

Future Internet, Journal Year: 2023, Volume and Issue: 15(9), P. 297 - 297

Published: Sept. 1, 2023

In the rapidly evolving landscape of internet usage, ensuring robust cybersecurity measures has become a paramount concern across diverse fields. Among numerous cyber threats, denial service (DoS) and distributed (DDoS) attacks pose significant risks, as they can render websites servers inaccessible to their intended users. Conventional intrusion detection methods encounter substantial challenges in effectively identifying mitigating these due widespread nature, intricate patterns, computational complexities. However, by harnessing power deep learning-based techniques, our proposed dense channel-spatial attention model exhibits exceptional accuracy detecting classifying DoS DDoS attacks. The successful implementation framework addresses posed imbalanced data its potential for real-world applications. By leveraging mechanism, precisely identify classify attacks, bolstering defenses servers. high rates achieved different datasets reinforce robustness approach, underscoring efficacy enhancing capabilities. As result, holds promise scenarios, contributing ongoing efforts safeguard against threats an increasingly interconnected digital landscape. Comparative analysis with current reveals superior performance model. We 99.38%, 99.26%, 99.43% Bot-IoT, CICIDS2017, UNSW_NB15 datasets, respectively. These remarkable results demonstrate capability approach accurately detect various types assaults. inherent strengths learning, such pattern recognition feature extraction, overcomes limitations traditional methods, efficiency systems.

Language: Английский

---