Monitoring Camera Research Articles

Aircraft engine danger areas incursion detection using keypoint detection and IoT

Operational safety at airports has long been a subject of active research, with a growing focus on the safety of aircraft stands in recent years. The phase during which an aircraft taxi into its designated stand with engines still running poses safety risks to surrounding vehicles. However, there is currently a lack of effective methods for detecting potential conflicts and hazards during this taxiing phase. To address this issue, this study utilizes apron activity videos as a monitoring data source and introduces an integrated Internet of Things (IoT) conflict detection model. This system combines object detection, engine keypoint detection, coordinate conversion, and conflict warning system to provide timely alerts when vehicles incursion into the engine danger areas. Firstly, the paper streamlines the fourth-level branch and network of HRNet, resulting in the HRNet-3stage network. This network is then compared with Lite-HRNet to determine the optimal choice. Secondly, Recognizing the limitations of traditional video conflict detection based on pixel distance, a fixed monitoring camera coordinate conversion algorithm is designed to convert pixel coordinates into actual coordinates on the aircraft stand, thereby improving the accuracy of conflict detection based on distance measurement. Thirdly, considering the risks associated with engine inlet and exhaust, engine design parameters, static spacing standards, and the dynamic anti-collision process within the aircraft stand, the study proposes a method for classifying four types of aircraft engine danger areas. Corresponding conflict detection models are designed for potential scraping incidents when vehicles incursion into these danger areas. Upon detecting a vehicle entering the aircraft engine danger areas, the IoT system sends warning messages through the airport control tower monitoring system. Finally, we construct an apron sandbox to validate the conflict detection model. This validation results in an impressive F1-score exceeding 90% and a detection delay of less than 100 ms. Our innovative approach, supported by keypoint detection networks and IoT, effectively addresses the detection of critical aircraft components within stands. It comprehensively analyzes incursion issues into engine danger areas, offering a novel perspective for understanding and mitigating potential conflicts arising from aircraft taxiing into stands.

Research on the Detection of Traffic Flow based on Video Images

Based on the current level of social development, everyone's demand for cars has increased rapidly. At present, the total number of motor vehicles and drivers in China ranks first in the world. With the rapid development of deep learning, the method of vehicle flow statistics based on video can directly use the existing traffic monitoring camera to realize the detection of vehicles, and some traffic flow detection based on YOLOv1, YOLOv2, YOLOv3, YOLOv4 and other algorithms have problems such as insufficient accuracy and low efficiency. Therefore, this paper proposes to use YOLOv5 to replace the original algorithm to achieve object detection, tracking, and processing. I improve the efficiency of the statistics of the traffic flow.

Surface Heat Monitoring with High-Resolution UAV Thermal Imaging: Assessing Accuracy and Applications in Urban Environments

The urban heat island (UHI) effect, where urban areas experience higher temperatures than surrounding rural regions, necessitates effective monitoring to estimate and address its diverse impacts. Many existing studies on urban heat dynamics rely on satellite data with coarse resolutions, posing challenges in analyzing heterogeneous urban surfaces. Unmanned aerial vehicles (UAVs) offer a solution by providing thermal imagery at a resolution finer than 1 m. Despite UAV thermal imaging being extensively explored in agriculture, its application in urban environments, specifically for surface temperatures, remains underexplored. A pilot project conducted in Athens, Georgia, utilized a UAV with a FLIR Vue Pro R 640 thermal camera to collect thermal data from two neighborhoods. Ground data, obtained using a handheld FLIR E6-XT infrared imaging camera, were compared with UAV thermal imagery. The study aimed to assess the accuracy of the UAV camera and the handheld camera for urban monitoring. Initial testing revealed the handheld’s accuracy but tendency to underpredict, while UAV camera testing highlighted considerations for altitude in both the rjpg and tiff image pixel conversion models. Despite challenges, the study demonstrates the potential of UAV-derived thermal data for monitoring urban surface temperatures, emphasizing the need for careful model considerations in data interpretation.

Camera trapping and telemetry for detecting and quantifying animal interactions: Not anything goes

Telemetry using Global Positioning Systems (GPS) and camera trapping are two widely used approaches to detect animal interactions involving wildlife. However, while the former follows an individual-based approach, the latter follows a population-based approach, which may lead to different outcomes when describing interaction networks. In this study, we compared the performance of GPS telemetry, regular grid camera trapping (CT-RG), and resource monitoring camera trapping (in which cameras are placed monitoring a valuable resource for animals, such as water or food; CT-RM) in detecting and quantifying animal interactions. We used spatially explicit simulated trajectories of 150 individuals of 3 imaginary species. We simulated 9 different scenarios of resource visiting frequency and spatial resource aggregation in the environment to assess the potential influence of animal behavior and landscape structure. The simulated trajectories were sampled using the three approaches with six different sampling efforts each, and the direct and indirect interactions between the three species were quantified. Additionally, CT-RG and CT-RM were applied in Doñana National Park to evaluate their differences and limitations in field application. In terms of simulation, CT-RM offered the best performance in detecting direct interactions, and GPS achieved good performance when using a high number of fixes per hour. Both CT-RM and GPS also achieved good performance in detecting indirect interactions. CT-RG, on the other hand, missed most direct and indirect interactions, but correctly identified the pairs of species that interacted indirectly at some point. In the field trial, CT-RM and CT-RG detected almost the same number of species pairs interacting indirectly. However, CT-RG missed at least half of the species pairs that interacted directly. The spatial aggregation of the resource favors the performance of GPS over CT-RM, while the opposite occurs when the frequency of animal visits to the resource increases. To summarize, CT-RM is the most reliable method for detecting interactions when resource monitoring is feasible (e.g., small ponds in arid areas). Otherwise, GPS may show better performance, although it may miss direct interactions if a high fix rate is not used. Finally, the use of CT-RG should be limited to identify which species could potentially interact indirectly, but not to quantify these interactions.

Early Fire Detection System by Using Automatic Synthetic Dataset Generation Model Based on Digital Twins

Fire is amorphous and occurs differently depending on the space, environment, and material of the fire. In particular, the early detection of fires is a very important task in preventing large-scale accidents; however, there are currently almost no learnable early fire datasets for machine learning. This paper proposes an early fire detection system optimized for certain spaces using a digital-twin-based automatic fire learning data generation model for each space. The proposed method first automatically generates realistic particle-simulation-based synthetic fire data on an RGB-D image matched to the view angle of a monitoring camera to build a digital twin environment of the real space. In other words, our method generates synthetic fire data according to various fire situations in each specific space and then performs transfer learning using a state-of-the-art detection model with these datasets and distributes them to AIoT devices in the real space. Synthetic fire data generation optimized for a space can increase the accuracy and reduce the false detection rate of existing fire detection models that are not adaptive to space.

Improvement of crack detectivity for noisy concrete surface by machine learning methods and infrared images

In order to maintain serviceability and reliability of concrete structures, it is essential to assess their condition as concrete structures deteriorate in time. Cracks develop in concrete due to several reasons such as severe loading, environmental effects, chemical effects etc. and cause durability loss in the structure which may also lead to loss of stability. In this research, crack detection is realized by machine learning and an infrared image. The effects of infrared images on crack detection are confirmed by random forest algorithm to select useful explanatory features. Selected features are applied to random forest algorithm and neural network algorithm. Effective filters are selected as a feature selection technique to improve the accuracy. Crack detection is also conducted by U-Net with RGB and infrared images, and the detection characteristics are compared to conventional methods. The performance of two conventional machine learning methods, random forest and neural network, are evaluated based on F1 score and false positives. Applying selected features improves the accuracy of the crack detection from an infrared image. False positives decreased due to monitoring conditions and camera specifications in the infrared image. The most effective image processing filter is the blur filter for each algorithm. Comparing algorithms for crack detection using selected features, different accuracy values are obtained. U-Net enables more accurate crack detection compared to conventional methods. The number of false positives is reduced compare to conventional method. In the detection results by three algorithms, infrared image affects the balance of false negatives and false positives.

Taking the bait: Environmental factors affecting bait retention and hook disposition in the Gulf of Maine Bottom Longline Survey

Longline fishery surveys are major contributors to groundfish species abundance estimates. Catchability is often assumed constant, but bait is lost or consumed at inconsistent rates during and across gear sets. Accounting for hook availability and gear saturation can improve abundance estimates, but these data are challenging to collect. Our objectives were to determine whether video camera systems can be used to collect these data from the Northeast Fisheries Science Center’s Gulf of Maine Bottom Longline Survey and to explore the associations between hook disposition percentages and environmental characteristics. Electronic monitoring camera systems were deployed on partner vessels during two fall and two spring seasons to capture video of gear retrieval. During post-survey review, individual hooks were recorded as one of the following hook dispositions: bait, empty hook, fish, and invertebrate. Percentages of these hook dispositions were characterized across categorical (season, day or night, and bottom substrata) and continuous (bottom temperature, bottom depth, and substrate rugosity) survey station variables. Nonmetric multidimensional scaling and logistic regression analyses revealed correlations between retained bait and low bottom temperatures, the spring season, shallow depths, rough bottom substrata, and high terrain rugosity. These characteristics were typical of shallow, inshore, and topographically variable survey strata, suggesting that catchability differs among habitat types. This information will be incorporated into catch rate models to develop standardized indices of abundance for groundfish stock assessments. This study is an example of the utility of electronic technologies to supplement on board data collection during cooperative fishery surveys.

Smart Dairy Farming-The Potential of the Automatic Monitoring of Dairy Cows' Behaviour Using a 360-Degree Camera.

The aim of this study is to show the potential of a vision-based system using a single 360° camera to describe the dairy cows' behaviour in a free-stall barn with an automatic milking system. A total of 2299 snapshots were manually evaluated, counting the number of animals that were lying, standing and eating. The average capture rate of animals in the picture is 93.1% (counted animals/actual numbers of animals). In addition to determining the daily lying, standing and eating times, it is also possible to allocate animals to the individual functional areas so that anomalies such as prolonged standing in the cubicle or lying in the walkway can be detected at an early stage. When establishing a camera monitoring system in the future, attention should be paid to sufficient resolution of the camera during the night as well as the reduction of the concealment problem by animals and barn equipment. The automatic monitoring of animal behaviour with the help of 360° cameras can be a promising innovation in the dairy barn.

Intelligent and secure real-time auto-stop car system using deep-learning models

In this study, we introduce an innovative auto-stop car system empowered by deep learning technology, specifically employing two Convolutional Neural Networks (CNNs) for face recognition and travel drowsiness detection. Implemented on a Raspberry Pi 4, our system is designed to cater exclusively to certified drivers, ensuring enhanced safety through intelligent features. The face recognition CNN model accurately identifies authorized drivers, employing deep learning techniques to verify their identity before granting access to vehicle functions. This first model demonstrates a remarkable accuracy rate of 99.1%, surpassing existing solutions in secure driver authentication. Simultaneously, our second CNN focuses on real-time detecting+ of driver drowsiness, monitoring eye movements, and utilizing a touch sensor on the steering wheel. Upon detecting signs of drowsiness, the system issues an immediate alert through a speaker, initiating an emergency park and sending a distress message via Global Positioning System (GPS). The successful implementation of our proposed system on the Raspberry Pi 4, integrated with a real-time monitoring camera, attains an impressive accuracy of 99.1% for both deep learning models. This performance surpasses current industry benchmarks, showcasing the efficacy and reliability of our solution. Our auto-stop car system advances user convenience and establishes unparalleled safety standards, marking a significant stride in autonomous vehicle technology.

Habitat suitability and crop damage risk caused by wild boar in Guizhou Plateau, China

AbstractCrop damage caused by wild boar (Sus scrofa) in many agroforestry regions of the world has affected agricultural development and farmers' livelihoods. Understanding the patterns of wild boar crop damage is important for properly managing these conflicts, and can improve the effectiveness of existing prevention strategies. We collected a comprehensive dataset from Guizhou Province, China, from 2017 to 2020, obtained from motion‐sensitive camera monitoring, transect surveys, interview surveys, and wildlife management administrations. These data include 3,156 records of wild boar activity in natural areas and 576 records of crop damage. We applied a multi‐scale MaxEnt model to map the regional distribution of wild boar habitat suitability and crop damage risk in Guizhou Province and focused on the role of habitat suitability in wild boar crop damage from a regional perspective. Wild boar habitat suitability exhibits a divergent pattern, expanding from the central area of Guizhou Province to the eastern, northern, southern, and northwestern regions, with a gradually increasing trend. Suitability of habitat for wild boar was associated with composite topographic index, mean coldest season temperature, isothermality, Shannon diversity index, edge density, and distance to village roads. Similarly, crop damage caused by wild boars in agroforestry areas is mainly concentrated in the northeast, with the risk increasing from west to east. The risk of damage was associated with temperature seasonality, roughness, distance to village roads, precipitation in the warmest quarter, edge density, and contagion index. Furthermore, in agroforestry areas, there is a positive correlation between wild boar habitat suitability and crop damage risk, with Spearman rank correlation coefficients ranging from 0.5759 to 0.5932. This correlation is not perfectly synchronized on a regional scale. This study indicates that the relationship between habitat suitability and crop damage caused by wild boar in Guizhou Province is primarily influenced by changes in population size rather than the habitat. Consequently, we emphasize the importance of controlling wild boar populations through hunting as a way to effectively reduce the risk of crop damage in agroforestry areas, without neglecting the need for proper habitat management.

Three-Stage Deep Learning Framework for Video Surveillance

The escalating use of security cameras has resulted in a surge in images requiring analysis, a task hindered by the inefficiency and error-prone nature of manual monitoring. In response, this study delves into the domain of anomaly detection in CCTV security footage, addressing challenges previously encountered in analyzing videos with complex or dynamic backgrounds and long sequences. We introduce a three-stage deep learning architecture designed to detect abnormalities in security camera videos. The first stage employs a pre-trained convolutional neural network to extract features from individual video frames. Subsequently, these features are transformed into time series data in the second stage, utilizing a blend of bidirectional long short-term memory and multi-head attention to analyze short-term frame relationships. The final stage leverages relative positional embeddings and a custom Transformer encoder to interpret long-range frame relationships and identify anomalies. Tested on various open datasets, particularly those with complex backgrounds and extended sequences, our method demonstrates enhanced accuracy and efficiency in video analysis. This approach not only improves current security camera analysis but also shows potential for diverse application settings, signifying a significant advancement in the evolution of security camera monitoring and analysis technologies.

Design of Supply Chain Logistics Intelligent Management Information System Based on GIS Optimization Model

With the improvement of productivity and the development of internet and information technology, many e-commerce systems and supply chain logistics systems have emerged, allowing for online purchases of food such as vegetables and fruits, as well as clothing and shoes. Nowadays, online shopping platforms such as Duoduo Maicai, Taobao, JD, and Meituan Youxuan have been integrated into people's lives, and various supermarkets and shopping malls are also everywhere. Therefore, the safety, efficiency, and real-time location of goods delivery need to be taken into account. Traditional supply chain logistics has problems such as requiring a large amount of manpower, inaccurate system positioning, inaccurate designed delivery routes, low efficiency, and inability to understand the status of drivers and vehicles. GIS (Geographic Information System) can assist systems in designing delivery routes, obtaining real-time locations of drivers or goods, and navigating. Therefore, in order to save manpower and ensure the safety and efficiency of goods delivery, this article used genetic algorithms to optimize and design a supply chain logistics intelligent management information system for GIS. This system had three modules, namely the client, driver, and terminal. The client mainly had functions such as publishing orders, checking the estimated arrival time of goods, and learning the real-time location of goods. The driver end mainly had functions such as real-time positioning, navigation, communication, viewing delivered orders, and monitoring cameras. The terminal mainly had functions such as viewing driver monitoring cameras and managing background information. Experiments showed that optimizing GIS could more accurately locate drivers and goods, assist drivers in navigation, assist systems in arranging delivery routes, and calculate the estimated arrival time of goods, resulting in an efficiency improvement of nearly 12%.

Research on Image Detection Algorithm of Rail Traffic Congestion Degree Based on Convolutional Neural Networks

With the sustainable development of the social economy and the continuous maturity of science and technology, urban rail transit has developed rapidly. It solved the problems of urban road load and people’s travel and brought about the problem of rail transit passenger congestion. The image detection algorithm for rail transit congestion is established based on the convolutional neural networks (CNN) structure to realize intelligent video image monitoring. The CNN structure is optimized through the backpropagation (BP) algorithm so that the model can detect and analyze the riding environment through the monitoring camera and extract the relevant motion characteristics of passengers from the image. Furthermore, the crowding situation of the riding environment is analyzed to warn the rail transit operators. In practical application, the detection accuracy of the algorithm reached 91.73%, and the image processing speed met the second-level processing. In the performance test, the proposed algorithm had the lowest mean absolute error (MAE) and mean square error (MSE). In Part B, the MAE and MSE values of the model were 16.3 and 24.9, respectively. The error values were small, so the performance was excellent. The purpose of this study is to reduce the possibility of abnormal crowd accidents at stations and provide new ideas for intelligent management of rail transit.

Efficacy of compliance with ventilator-associated pneumonia care bundle: A 24-month longitudinal study at Bach Mai Hospital, Vietnam.

To decrease the risk of complications from ventilator-associated pneumonia, it is essential to implement preventative measures in all ICU patients. Since 2018, with the help of Japanese experts, we have applied a ventilator-associated pneumonia care bundle with 10 basic standards in patient care and monitoring. Therefore, we conducted a study to evaluate the results of applying 10 solutions to prevent ventilator-associated pneumonia over 24 months. A cross-sectional descriptive study with longitudinal follow-up for 24 months on 170 mechanically ventilated patients at the Center for Critical Care Medicine, Bach Mai Hospital. According to the Centers for Disease Control (CDC, 2021), the diagnosis of ventilator-associated pneumonia is when pneumonia appears 48 h after intubation by confirmation by at least two doctors. Evaluate compliance with each solution in the care bundle through camera monitoring, medical records, and directly on patients daily. The rate of ventilator-associated pneumonia is 12.9%, the frequency of occurrence is 16.54 of 1000 days. The compliance rate for complete compliance with a 10-item ventilator-associated pneumonia was only 1.8%, while the average value was 84.1%. Average values of compliance with each solution for hand hygiene, head elevation 30-45 degrees, oral hygiene, stopping sedation, breathing circuit management, cuff pressure management, hypoplastic suction, Spontaneous breathing trial (SBT) daily and assessed extubation, mobilization and early leaving bed, ulcer and thrombosis prevention were 96.9%, 97.3%, 99.4%, 81.5%, 99.9%, 99.9%, 86.3%, 83.5%, 49.3%, and 46.4%, respectively. The time to appear ventilator-associated pneumonia in the high compliance group was 46.7 ± 5.0 days, higher than in the low compliance group, 10.3 ± 0.7 days, p < 0.001. A 10-item ventilator-associated pneumonia care bundle has helped reduce the incidence of ventilator-associated pneumonia. To reduce the risk of ventilator-associated pneumonia and shorten ICU and hospital stays, it is essential to fully adhere to subglottic secretion suction, daily SBT, and early mobilization and leaving the bed.

Smart railways: the design and construction of an autonomous inspection and maintenance vehicle

The Railway Inspection and Maintenance Vehicle (RIMV) is a state-of-the-art autonomous ground robot for the railway environment. The robot has specialised hardware tailored for specific functionalities, ensuring optimal performance in its designated tasks. It incorporates a GPS, IMU, and encoder for robust localisation, and it integrates a LiDAR system and a depth camera for collision avoidance. Prioritising safety, the RIMV features an IP camera for real-time environmental monitoring and is equipped with emergency stop mechanisms. An Ubuntu Linux machine manages the robot’s data processing, complemented by a 5G Wi-Fi router for remote monitoring and operation when necessary. ROS2 is the primary platform for robot operations, standing out for its open-source solution and enhanced cybersecurity features. We validated the RIMV in a heritage railway setting with an ultrasonic inspection kit to detect rail-track defects. Beyond this specific application, the RIMV is versatile and designed to serve as a platform for other autonomous inspection and maintenance tasks within the railway sector. Its adaptability highlights its potential to enhance efficiency and safety in railway maintenance, offering invaluable assistance to human operators across varied scenarios.

IoT-Based Strawberry Disease Detection With Wall-Mounted Monitoring Cameras

This paper proposes StrawberryTalk, an Internet of Things (IoT) platform for image-based strawberry disease detection. StrawberryTalk reuses the wall-mounted monitoring cameras without extra hardware cost. The contributions of StrawberryTalk are the utilization of IoT for automatic photo shoot and the wind detection mechanism to eliminate the obscure photos due to the wind effects. Also, the data preprocessing and the infection detection models are manipulated as IoT devices to simplify the implementation of the multi-cascade Artificial Intelligence (AI) models. We derive the relationship between the camera zoom factor and the distance between the camera and the strawberries for optimal disease detection. Accuracy of detection may be affected by obscure photos. In terms of eliminating obscure photos due to wind effect, we analytically derive the relationship between the wind alert delay and the number of obscure photos that must be retaken. For the greenhouses in the Bao Mountain, we only need to retake one photo. Based on the experiments, the mean average precision (mAP) of StrawberryTalk (to detect exact spots in a leaf) can be up to 92.37%, which is better than the previous approaches. To detect if a pot has infected leaves, the accuracy of StrawberryTalk can be up to 97.92% at the zoom factor 30X. In commercial operation, it is important to detect all infected strawberry pots. StrawberryTalk is able to detect all infected pots (i.e., recall is 100%) with a camera of zoom factor of 12X. The accuracy is 96.88%.

Real-time grid detection in sheet metal fiber laser cutting through coaxial monitoring

Real-time defects’ estimation and control of the cut quality through the coaxial camera monitoring of the kerf are amongst the most promising developments for laser cutting. In industrial systems, sheets are positioned on a metallic grid creating discontinuities in the cutting process due to unpredictable thickness, blown and resolidified material and the time varying position of the grid. The interaction between laser radiation escaping from the kerf and the grid, along with the difficult outflow of molten material, causes changes in the process emission images. These changes influence the defect estimation approach, and, consequently, the control action. In this work, a real-time grid identification and classification-based machine learning algorithm was developed and tested during the fusion cutting of 6 mm thick Al5754, exploiting a NIR coaxial monitoring system. Real-time control experiments with dross estimates were performed, demonstrating a correct identification of the grid and highlighting a feasible industrial application.

User perception and ergonomic display layout design of truck camera monitor system

A camera monitor system (CMS) provides side/rear views via in-vehicle displays by replacing the existing mirrors with miniature cameras. Two surveys were conducted to determine the overall perception of truck drivers toward CMS compared to the conventional mirrors and to identify the ergonomic CMS display layout. Thirty experienced truck drivers evaluated CMS compared to the existing mirror system using six evaluation measures (intention to use, safety, convenience, rapidity and accuracy, preference, and superiority). Each respondent also evaluated eight different layout design alternatives using seven ergonomic evaluation measures (rapidity, accuracy, safety, convenience, learnability, intuitiveness, and preference). The design alternatives were developed by combining the locations of two truck CMS displays – one is side display replacing the main and wide-angle exterior mirrors and the other one is front display replacing the front mirror. The study results revealed that truck drivers rated the CMS highly compared to the existing mirror system across all six measures, suggesting the need for further academic and industrial research/development and thereby wider introduction of truck CMS. Also, the layout wherein the side and front displays were positioned at the A-pillar and middle-top-of-windshield locations, respectively, was found to be the best and its design characteristics were: 1) spatial correspondence between the provided information and the in-vehicle location for both the side and front displays, 2) spatial compatibility with driver expectation for the side display, 3) reduction in the eye movement required for the front display, and 4) spatial compatibility by relative position between the side and front displays.

Wildlife Camera Monitoring Revealed the Northern Goshawk as a Predator on Gyrfalcon Nestlings

Wildlife Camera Monitoring Revealed the Northern Goshawk as a Predator on Gyrfalcon Nestlings

An Image-Based Fire Monitoring Algorithm Resistant to Fire-like Objects

Due to its wide monitoring range and low cost, visual-based fire detection technology is commonly used for fire detection in open spaces. However, traditional fire detection algorithms have limitations in terms of accuracy and speed, making it challenging to detect fires in real time. These algorithms have poor anti-interference ability against fire-like objects, such as emissions from factory chimneys, clouds, etc. In this study, we developed a fire detection approach based on an improved YOLOv5 algorithm and a fire detection dataset with fire-like objects. We added three Convolutional Block Attention Modules (CBAMs) to the head network of YOLOv5 to improve its feature extraction ability. Meanwhile, we used the C2f module to replace the original C2 module to capture rich gradient flow information. Our experimental results show that the proposed algorithm achieved a mAP@50 of 82.36% for fire detection. In addition, we also conducted a comparison test between datasets with and without labeling information for fire-like objects. Our results show that labeling information significantly reduced the false-positive detection proportion of fire-like objects incorrectly detected as fire objects. Our experimental results show that the CBAM and C2f modules enhanced the network’s feature extraction ability to differentiate fire objects from fire-like objects. Hence, our approach has the potential to improve fire detection accuracy, reduce false alarms, and be more cost-effective than traditional fire detection methods. This method can be applied to camera monitoring systems for automatic fire detection with resistance to fire-like objects.