Flame Images Research Articles

Application of Multivariate Reinforcement Learning Engine in Optimizing the Power Generation Process of Domestic Waste Incineration

Garbage incineration is an ideal method for the harmless and resource-oriented treatment of urban domestic waste. However, current domestic waste incineration power plants often face challenges related to maintaining consistent steam production and high operational costs. This article capitalizes on the technical advantages of big data artificial intelligence, optimizing the power generation process of domestic waste incineration as the entry point, and adopts four main engine modules of Alibaba Cloud reinforcement learning algorithm engine, operating parameter prediction engine, anomaly recognition engine, and video visual recognition algorithm engine. The reinforcement learning algorithm extracts the operational parameters of each incinerator to obtain a control benchmark. Through the operating parameter prediction algorithm, prediction models for drum pressure, primary steam flow, NOx, SO2, and HCl are constructed to achieve short-term prediction of operational parameters, ultimately improving control performance. The anomaly recognition algorithm develops a thickness identification model for the material layer in the drying section, allowing for rapid and effective assessment of feed material thickness to ensure uniformity control. Meanwhile, the visual recognition algorithm identifies flame images and assesses the combustion status and location of the combustion fire line within the furnace. This real-time understanding of furnace flame combustion conditions guides adjustments to the grate and air volume. Integrating AI technology into the waste incineration sector empowers the environmental protection industry with the potential to leverage big data. This development holds practical significance in optimizing the harmless and resource-oriented treatment of urban domestic waste, reducing operational costs, and increasing efficiency.

Assessment of flame stability through a convolutional denoising autoencoder and statistical analysis

Flame stability assessment is essential for optimizing combustion operation and improving combustion quality. However, an accurate and reliable assessment of stability is difficult, heavily relying on prior expert knowledge and massive labeled data. This study proposes a novel method for flame stability assessment through flame images and deep learning techniques. In this method, the deep image features are extracted by an unsupervised convolutional denoising autoencoder (CDAE), and then quantitatively analyzed by a stability index. In particular, the CDAE introduces a new loss function composed of denoising coding constraints and reconstruction similarity to improve its training efficiency. The stability index is established based on clustering analysis and statistical analysis of the deep image features, with a numerical interval of [0, 1]. The effectiveness of the proposed method is verified by the flame images obtained from ethylene-air diffusion combustion conditions. Results show that the proposed method extracts representative flame features accurately and quantifies the flame stability with strong robustness and generalization ability.

Combustion state identification of MSWI processes using ViT-IDFC

Municipal solid waste (MSW) incineration processes often experience fluctuations in the physical and chemical properties of waste materials, leading to unstable combustion and increased flue-gas emissions. This hinders the development of greener waste-treatment methods. Currently, experts rely on their experience in observing flames during field operations to determine the combustion state. However, this process is energy consuming, subjective, and not optimal for controlling MSW incineration (MSWI). To address this issue, this study proposes a dataset of MSWI flame images for identifying combustion states using a vision-transformer-improved deep forest classification (ViT-IDFC) algorithm. The ViT-IDFC algorithm extracts multilayer visual transformation features from flame images based on a pre-trained ViT model’s transformer coding layer. The experience of domain experts was used to select the depth features. The algorithm then constructed an IDFC model by combining the selected ViT visual transformation features and the original flame image as inputs for the cascade forest. The proposed method was evaluated using a dataset of flame images, and the results demonstrated accuracies of 98.15% and 96.43% in identifying the left and right grate flame images, respectively, which are acceptable in the industry.

Flame Detection Based on Improved YOLOv7

A flame detection algorithm with improved YOLOv7 is proposed to address the problems that small targets in flame detection are easy to be missed and misdetected, and are easily disturbed by bright objects during detection. In the improved algorithm, the SE attention module is introduced to enhance the perception of the model on the channel; ConvNeXt is used to retain the features of small target flames for small targets among flame images; Employing K-means ++ clustering algorithm to gain the anchor points that match the flame size more to enhance the detection accuracy.The results testify that the improved algorithm we proposed in this paper is higher than the YOLOv7 in terms of average accuracy and precision.

Combustion Characteristics of a Swirl-Radial-Injection Composite Fuel Grain with Applications in Hybrid Rockets

The combustion characteristics of a swirl-radial-injection composite fuel grain were experimentally and numerically investigated. This composite grain permits swirl-radial oxidizer injection based on three hollow helical blades, each having a constant hollow space allowing uniform oxidizer injection into the main chamber along the axial direction. The oxidizer enters from channel inlets located along a hollow outer wall. This wall, together with the three blades, is fabricated as one piece from acrylonitrile-butadiene-styrene using three-dimensional printing. Paraffin-based fuel is embedded in the spaces between adjacent blades. Firing tests were conducted with gaseous oxygen as the oxidizer, using oxidizer mass flow rates ranging from 7.45 to 30.68 g/s. Paraffin-based fuel grains using conventional fore-end injection were used for comparison. Regression rate boundaries were determined taking into account the erosion of the oxidizer channels. The data show that the regression rate was significantly increased even at the lower limit. Images of the combustion chamber flame and of the exhaust plume were also acquired. The flame was found to be concentrated in the main chamber and a smoky plume was observed, consistent with the high regression rate. A three-dimensional simulation was employed. The present design was found to improve fuel/oxidizer mixing and combustion efficiency compared with a fuel grain using fore-end injection. Both the experimental results and numerical simulations confirmed the potential of this swirl-radial-injection fuel grain.

Research on influence of oxygen on flow/combustion characteristics in a scramjet combustor equipped with an O2-pilot strut

With the help of an O2-pilot strut, the stable supersonic combustion was achieved in a scramjet combustor fueled by liquid kerosene. Objective of this work is to investigate the effect of different mass flow rate of oxygen on the cold flow and combustion characteristics to explain the role of the supplied oxygen in supersonic combustion. A series of numerical simulations and experiments were carried out under the high inflow enthalpy condition (Ma = 2.8). High-speed photography and pressure measurement were used to record the flame images and pressure of combustor. A skeleton kerosene mechanism was used for combustion simulation for more details of combustion. Simulation results indicated that by adjusting the mass flow rate of oxygen, the severe rich-kerosene zone around the ignitor could be improved. Under the low mass flow rate of oxygen condition, limited by over-high equivalence ratio the ignition delay time was too long and heat release was too small to establish the global flame. On the contrary, under the high mass flow rate of oxygen condition, limited by over-low equivalence ratio, only little kerosene could be ignited with most of the kerosene carried downstream by high-speed flow resulting in a sharp increase in flame lift distance. The boundary of mass flow rate of supplied oxygen for combustion stabilization was obtained. The results in this paper would be valuable for reveling the ignition and flame stabilization mechanism in supersonic airflow.

A simpler tractable contour technique to model thermal radiation from buoyant diffusion flames

This work presents an original technique based on parametric interpolation to estimate the view factor between unsteady flames and a small target. The technique was tested on a buoyant diffusion flame generated in an ethylene-fueled porous burner that emulates a pool fire. Flame images were processed to determine the flame contours in terms of parametric polynomials. A mathematical formulation based on the Stokes’ theorem was developed to estimate the view factor between the flame and a radiometer, whose contour integrals were solved numerically. These results were compared to ones obtained with the point-source method and the solid flame model, assuming the flame as a cylinder and a cone. The proposed technique captures the influence of geometry on the radiant flux field more robustly than these methods, providing estimations of thermal radiation between the point and solid flame models.

Explosive characteristics of non-uniform methane-air mixtures in half-open vertical channels with ignition at the open end

In this study, the explosive characteristics of non-uniform methane-air mixtures in a 1.5 m half-opened vertical channel with ignition at the open end were experimentally studied. The results revealed significant oscillation characteristics in the flame image, velocity, and pressure variations. Notably, a strong similarity was observed between the flame velocity and pressure profiles. Furthermore, it was experimentally demonstrated that the explosion characteristics of methane-air mixtures in vertical channels with the same average volume fraction differed. These variations in characteristics were attributed to the degree of non-uniformity, where both the direction and steepness of the methane concentration gradient influenced the explosion flame behavior. When the diffusion time was very short, resulting in a small concentration above the channel and a large gradient, an obvious cusp appeared at the flame front, which was absent in the case of longer diffusion times. In addition, as the methane diffusion time increased, the methane in the vertical channel gradually accumulated upward. Consequently, the time required to reach maximum velocity and pressure after ignition decreased, leading to a larger maximum amplitude. In summary, our study highlights the importance of considering non-uniformity and diffusion time in understanding the explosion characteristics of methane-air mixtures in vertical channels.

Spray combustion of fast-pyrolysis bio-oils under engine-like conditions

In the EU funded SmartCHP project, biomass derived fast-pyrolysis bio-oil (FPBO) is used to fuel modified stationary diesel engines for combined heat and power applications. In this study, the spray combustion characteristics of fast-pyrolysis bio-oil and FPBO/ethanol blends are experimentally investigated in a combustion research unit. Due to the special physicochemical properties of fast-pyrolysis bio-oil, major modifications are made to the combustion research unit to facilitate testing this fuel, including the installation of a high-temperature chamber and a heavy-fuel oil injection system. Experimental tests are carried out at a chamber wall temperature of 750 °C and an ambient pressure of 50 bar. The pressure-based heat release analysis is used to evaluate fuel ignition and combustion characteristics, while a high-speed imaging technique is adopted to visualize natural luminosity of spray flames through a borescope.Results show that fast-pyrolysis bio-oil has lower sooting tendency than diesel, and it significantly alleviates the fuel dribble phenomenon. At injection pressure of 300 bar, the poor atomization of fast-pyrolysis bio-oil leads to rather grainy-looking spray flame images, and nozzle orifices becoming clogged within couple of injections. The increase of injection pressure to 900 bar results in an obvious improvement of the atomization quality and nozzle durability for fast-pyrolysis bio-oil. Adding ethanol into fast-pyrolysis bio-oil could improve fuel atomization and further reduce the natural luminosity. The ignition delay of fast-pyrolysis bio-oil is slightly retarded by 10% ethanol addition, while it is advanced by 30% ethanol addition.

Assessment of burner structure for methane-oxygen inverse diffusion flame based on laser detection: Reaction characteristics and soot analysis

The structure of a burner is a key issue of boilers and gasifiers in several industries. It significantly influences the flame formation and combustion reaction determining the mixing degree of fuel and oxidant. This study aims to explore the characteristics of flames at different angles and evaluate their advantages in specific applications. The present work explores the correlation between laser-induced breakdown threshold energy and flame temperature based on laser-induced breakdown spectroscopy (LIBS). Subsequently, the local structure and reaction characteristics of the flame at different burner angles (45, 60 and 90°) were investigated with simulation. The local equivalence ratio of the laminar flame was used to determine the distribution of reactants and components in the flame. The results indicated that the breakdown threshold energy was positively correlated with the temperature of the flame, and it was dependent on the gas concentration. With increased laser energy, the O/H ratio could be used as a function of the local equivalence ratio of the flame. In addition, the study of flame images and the spontaneous radiation of soot revealed that the C/H ratio in the flame reached the maximum value when the burner incident angle was 45°. Through simulation and experimental research, we found that the flame at 60° angle is the most stable, produces the least soot, has the highest combustion efficiency, and has better thermal radiation intensity.

Fiber-based high-speed 3D schlieren imaging.

This Letter reports the first demonstration of a high-speed three-dimensional (3D) schlieren technique based on the combination of fiber imaging, Toepler's lens-type schlieren, and computed tomography (CT). The technique uses a single high-speed camera, two xenon lamps, and a series of fiber bundles to simultaneously capture the schlieren images of turbulent flames from seven orientations with a framerate beyond tens of kHz. The presented method complements the existing technique with advantages of being flexible, high speed, and low cost. The 3D schlieren technique is first demonstrated and validated on the turbulent premixed flame and stable laminar premixed flame, respectively. Then, the 3D schlieren technique is used to measure the transient, dynamic ignition process. The results show that time-resolved 3D fundamental properties of ignition kernels (i.e., structure and edge speed) can be obtained by the technique.

Evaluation of Spray and Combustion Models for Simulating Dilute Combustion in a Direct-Injection Spark-Ignition Engine

Abstract Dilute combustion in spark-ignition engines has the potential to improve thermal efficiency by mitigating knock and by reducing throttling and wall heat losses. However, ignition and combustion processes can become unstable for dilute operation due to a lowered laminar flame speed, resulting in excessive cycle-to-cycle variability (CCV) of the combustion process. To compensate for the slower combustion in less reactive mixtures, a modified intake port geometry can be employed to generate a strong tumble flow in the cylinder and elevate turbulence levels around the spark plug, thereby promoting a faster transition to turbulent deflagration. Consequently, optimizing combustion chamber geometry and operating strategy is crucial to maximizing the benefits of using dilute combustion with enhanced in-cylinder turbulence across a wide range of operating conditions. Computational fluid dynamics (CFD) simulations can be utilized for virtual engine optimization tasks, but this would require the models to be truly predictive regarding the impact of changes to the engine design and operational parameters. In this study, multicycle large-eddy simulations (LES) are performed for a direct-injection spark-ignition engine to investigate the model performance in predicting engine combustion characteristics with respect to changes in the intake configuration. A tumble plate that blocks the lower part of the intake port inlet is used to vary the tumble. A set of CFD models that have been recently developed are employed, which takes into account the drag of nonspherical droplets, flash-boiling behavior of liquid sprays, spray-wall interaction, surrogate formulation of a research-grade E10 gasoline, and fast chemical kinetic solvers. Simulation results are compared to experimental engine data in terms of cylinder pressure, apparent heat release rate, mass fraction burned timing, and flame images. It is found that LES employing the state-of-the-art CFD models are capable of properly predicting the spray processes and reproducing the measured mean cylinder pressure for the case with the tumble plate. On the other hand, the LES over-predicts the combustion rate during the early combustion stage and under-estimates the CCV, and these discrepancies become larger when the tumble plate is removed.

Experimental research on the venting characteristics of methane and coal dust two-phase system mixture explosions

This research focused on the venting characteristics and influence factor of methane and coal dust mixture explosions. The corresponding relationship between the venting parameters was analyzed and the effect of methane concentration was obtained. Results indicate that the venting flame propagation and pressure rise of vessel interior and exterior presented the clear corresponding relationship. The venting flame underwent the evolution process of “mushroom”, “spherical” and “strip” structures. At 6% methane concentration, the venting flame was longest and its propagation velocity was fastest. During the extinguishing stage, the flame exhibited an obvious oscillation. The venting flame appeared a secondary venting phenomenon at 6% and 8% methane concentrations, and the phenomenon was more evidently at 8% methane concentration. The pressure at the venting pipe was the greatest and its change was also most significant, even appeared a larger negative value. Compared with the vessel interior, the pressure of vessel exterior was continuously decreased as the venting distance increased. The venting pressure was the greatest at 6% methane concentration, followed by 8% methane concentration. The pressures of four points at 4% methane concentration were respectively reduced by 25.0%, 34.9%, 43.8% and 81.8%. The corresponding flame temperature also presented the same change trend. Besides, the venting flame image showed a good correspondence with the temperature field distribution and the highest temperature appeared in the center of spherical flame. The secondary venting resulted in the recurrence of high temperature flame at the venting port (500℃-650℃) and was affected by the methane concentration inside the vessel.

Fundamental Study of Premixed Methane Air Combustion in Extreme Turbulent Conditions Using PIV and C-X CH PLIF

This paper presents the flow and flame characteristics of a highly turbulent reactive flow over a backward-facing step inside a windowed combustor. Flow and combustion experiments were performed at Re = 15,000 and Re = 30,000 using high-resolution 10 kHz PIV and 10 kHz PLIF diagnostic techniques. Grid turbulators (Grid) with two different hole diameters (HD of 1.5 mm and 3 mm) and blockage ratios (BR of 46%, 48%, 62%, and 63%) were considered for the turbulence study. Grids introduced different turbulent length scales (LT) in the flow, causing the small eddies and turbulence intensity to increase downstream. The backward-facing step increased the turbulence level in the recirculation zone. This helped to anchor the flame in that zone. The small HD grids (Grids 1 and 3) produced continuous fluid structures (small-scale), whereas the larger HD grids (Grids 2 and 4) produced large-scale fluid structures. Consequently, the velocity fluctuation was lower (~25.6 m/s) under small HD grids and higher (~27.7 m/s) under large HD grids. The flame study was performed at ɸ = 0.8, 1.0, and 1.2 using C-X CH PLIF. An Adaptive MATLAB-based flame imaging scheme has been developed for turbulent reacting flows. Grids 1 and 3 induced more wrinkles in the flame due to higher thermal instabilities, pressure fluctuation, and diffusion under those grids. The flamelet breakdown and burnout events were higher under Grids 2 and 4 due to higher thermal diffusivity and a slower diffusion rate. It was observed that the flame wrinkling and flame stretching are higher at Re = 30,000 compared to Re = 15,000. The Borghi–Peters diagram showed that the flames were within the thin reaction zone except for Grid 1 at Re =15,000, where flames fell in the corrugated zone. It was observed from PIV and PLIF analyses that Re and LT mostly controlled the flame and flow characteristics.

Kömür Yakıcılarında Yanma Verimi Tahmini için Gömülü Platformda Çalışabilen Evrişimsel Sinir Ağının Parametre Analizi

Accurately and effectively calculating combustion efficiency in coal burners is crucial for industrial boiler manufacturers. Two main approaches can be used to calculate boiler efficiency: 1) Analyzing the gas emitted from the flue; 2) Visualizing the combustion chamber in the boiler. Flue gas analyzers, which are not user-friendly, come with high costs. Additionally, the physical distance between the flue and the combustion chamber causes the measurement to be delayed. Methods based on visualizing the combustion chamber do not have these disadvantages. This study proposes a system based on visualizing the combustion chamber and has two contributions to the literature: 1) for the first time, the modern Convolutional Neural Networks (CNN) approach is used to estimate combustion efficiency; 2) the CNN architecture with optimal parameters can work on an embedded platform. When classical classification techniques and a CPU-supported processor card are used, efficiency can be calculated from one flame image in 1.7 seconds, while this number increases to approximately 20 frames per second (34 times faster) when the proposed CNN architecture and GPU-supported processor card are used. The results obtained demonstrate the superiority of the proposed CNN architecture and hardware over classical approaches in estimating coal boiler combustion efficiency.

Ammonia/methane dual-fuel injection and Co-firing strategy in a swirl flame combustor for pollutant emissions control

The co-firing of ammonia with hydrocarbon fuel is a promising way to reduce the emissions of carbon dioxides. However, conventional way of co-firing under premixed lean swirl flame results in high NOx pollutant emissions, thus requiring an innovative mitigation strategy to solve the CO2-NOx conundrum. In the present study, the combustion characteristics and emissions performance of a partially-mix ammonia-methane dual flames are examined and compared with fully premixed swirl flame. Flame imaging shows a distinctive luminous central jet flame surrounded by compact swirl flame for the dual flame, as opposed to the lean bluish single-stage premixed flame. The flow field of the dual flames shows the absence of the typical recirculation flow of premixed flame, yet stabilisation of the flame is barely affected. The dual-flame shows reduced NO and CO emissions by at least a factor of 2 compared with the fully premixed flame. Low emissions of NO and CO can be attained for NH3–CH4 dual-fuel flames at φ ∼0.65. The reduced NO emission via the dual-fuel injection technique can lead to improved economics for the subsequent flue gas treatment to meet the emission regulation targets.

Role of Altitude in Influencing the Spray Combustion Characteristics of a Heavy-Duty Diesel Engine in a Constant Volume Combustion Chamber. Part I: Free Diesel Jet

Heavy-duty diesel engines operating in plateau regions experience deteriorated combustion. However, the lack of up-to-date information on the spray-combustion process limits the fundamental understanding of the role of altitude. In this work, the in-cylinder thermodynamic conditions of a real diesel engine operating under different altitudes were reproduced in a constant-volume combustion chamber (CVCC). The liquid spray, ignition, and combustion processes were visualized in detail using different optical diagnostics. Apart from predictable results, some interesting new findings were obtained to improve the understanding of free spray-combustion processes with different altitudes. The spatial distributions of ignition kernels provided direct evidence of higher peak pressure rise rates for high-altitude diesel engines. The percent of stoichiometric air was calculated to confirm that the net effect of altitude was an increase in the amount of air-entrained upstream of the lifted flame; therefore, the soot levels deduced from flame images were inconsistent with those from real engines, revealing that accelerating the soot oxidation process could effectively reduce engine soot emissions in plateau regions. Finally, a novel schematic diagram of the spray flame structure was proposed to phenomenologically describe the role of altitude in influencing the spray-combustion process of a free jet.

Effect of the primary oxidizer stream jet velocities on the ignition and combustion characteristics of pulverized coal under MILD oxy-coal combustion conditions

MILD oxy-coal combustion technology is a prospective technology with the potential for CO2 capture and storage, and it is characterized by the intense flue gas recirculation to preheat the fuel and to dilute the oxidizer due to the oxidizer or fuel jet with high velocity. The effect of the primary oxidizer stream jet velocities ranging from 1.7 m/s to 11.2 m/s on the coal ignition and combustion characteristics were experimentally investigated on a flat flame pulverized coal burner at various oxygen concentrations of 5%-21% and gas temperatures of 1473–1873 K under O2/CO2 and O2/N2 atmospheres. The coal flame images were recorded by the digital camera, and the flame structure evolution and ignition delay times of coal particles were analyzed at different primary oxidizer stream jet velocities and hot coflow conditions. The coal combustion radiation spectrum was monitored by an optical fiber spectrometer and the coal particle temperatures were derived with the two-color pyrometry. The results showed that the bright yellow soot flame gradually disappeared with the decrease of O2 concentration when CO2 replaced N2. The difference of ignition delay times between O2/CO2 atmosphere and O2/N2 atmosphere decreased from 2.8 ms to 0.2 ms when the primary air jet velocity increased from 1.7 m/s to 11.2 m/s at 1473 K and 5% O2. This meant the elevated jet velocity was beneficial to offset the delayed effect of the physicochemical properties of CO2 on coal ignition. The peak coal particle temperature increased by 68 K at 1873 K and 5% O2 under O2/CO2 atmosphere with the highest jet velocity owing to the accelerated diffusion of O2 into coal particle stream. Moreover, a dimensionless coal particle temperature parameter η was proposed to characterize the variation of the peak coal particle temperature caused by the improved heat transfer and mass transport between the coflow and coal particles under high jet velocity. The maximum value of η was 1.053 at 1873 K and 21% O2 under O2/CO2 atmosphere with the jet velocity of 11.2 m/s, indicating the increased jet velocity had the predominant effect on the increase of peak particle temperature at high gas temperature and O2 concentration.

Reconstructing the flame in supersonic combustion flow by using the distribution of the wall pressure based on deep learning

Two models based on the deep learning-based convolutional neural network (CNN) and the re-parameterized convolutional neural network (RepCNN) were designed to reconstruct the flame in the combustor. Experiments were performed on a ground-pulse combustion wind tunnel at a fixed inlet Mach number of 2.5 and different pressures to inject hydrogen to obtain the relevant datasets. The results showed that both models could reconstruct the image of the flame in the combustor based on pressures of the upper and lower walls as well as the pressure at which hydrogen was injected. The average structural similarity index between the reconstructed image of the flame and its actual/original image was 0.9553, the average peak signal-to-noise ratio was 34.201, and the average correlation coefficient was 0.9819. The speed of reconstruction of the image using the RepCNN model improved by 40.7% at the cost of a slightly lower accuracy compared with the CNN model, and it took only 2.85 ms to reconstruct the image of a single flame. The lightweight feature of the RepCNN provides an important foundation for monitoring the model to reconstruct the image of the flame in real time. The work here simplifies requirements on the hardware for ground wind tunnel tests and provides a new idea for examining the characteristics of the flame in small combustors.

Fully Synthetic Videos and the Random-Background-Pasting Method for Flame Segmentation

Video-based flame detection (VFD) aims to recognize fire events by using image features. Flame segmentation is an essential task in VFD, providing suspected regions for feature analysis and object recognition. However, the lack of positive flame samples makes it difficult to train deep-learning-based VFD models effectively. In this paper, we propose the assumption that we can train a segmentation model with virtual flame images and design experiments to prove it. We collected many virtual flame videos to extend existing flame datasets, which provide adequate flame samples for deep-learning-based VFD methods. We also apply a random-background-pasting method to distribute the flame images among different scenarios. The proposed method trains a flame segmentation model with zero real flame images. Moreover, we perform segmentation testing using real flame images, which the model has never used, to see if the model trained using ‘fake’ images can segment real objects. We trained four segmentation models based on FCN, U-Net, Deeplabv3, and Mask-RCNN using synthetic flame video frames and obtained the highest mPA of 0.783 and mIoU of 0.515. The experimental results on the FIRE-SMOKE-DATASET and the Fire-Detection-Image-Dataset demonstrate that the ‘fake’ flame samples generated by the proposed random-background-pasting method can obviously improve the performance of existing state-of-the-art flame segmentation methods using cross-dataset evaluation settings.