Non-fire Pixels Research Articles

Early fire detection using wavelet based features

In recent years, millions of hectares of vegetation worldwide have been destroyed by wildfires and forest fires. Computer vision-based fire classification, which separates pixels from image or video datasets into fire and non-fire categories, has gained more attention recently due to technological innovations. Fire pixels in an image or video can be classified using either a deep learning strategy or a traditional machine learning approach. Deep learning algorithms could process enormous volumes of data, but their training model performance is constrained because they fail to consider the differences in complexity between training samples. Moreover, it is not obvious to train a deep learning model without a dedicated GPU unit.Similarly, deep learning techniques that have a scarcity of training data and insufficient features exhibit poor performance in intricate real-world fire situations. Consequently, to categorize fire and non-fire pixels from the processed photos from the publicly accessible datasets, Corsican and FLAME, as well as the aerial private dataset Firefront_Gestosa, the current study uses a lightweight technique based on SVM and a refined set of features.The present research implemented a novel framework for fire detection and classification from a variety of RGB and Infra-red images acquired during real missions, addressing the significant requirement for swift and accurate recognition of diverse types of flames, ranging from wildfires to industrial and domestic fires. The framework employs wavelet decomposition-based features, including wavelet length, standard deviation, variance, energy, and Shannon’s entropy, extracted through a sliding window sampling method within a machine learning approach. It should be noted that managing multidimensional data to train a model is difficult in machine learning applications. This issue is solved adopting a feature selection approach, which eliminates redundant or unnecessary data that affects the functionality of the model. Thus, to enhance model performance, feature selection using ranking algorithms based on theoretical mutual information is applied in combination to the Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel.Extensive experiments demonstrate exceptional results, notably with Haar wavelets, achieving an impressive overall accuracy of 99.43% and remarkable performance across specificity, precision, recall, F-measure, and G-mean metrics. Thus, the present study showcases the potential of advanced image processing techniques to significantly advance fire detection and classification, thereby contributing to fire prevention, management, and research in various contexts.

Fire images classification based on a handcraft approach

In recent years, wildfires and forest fires have ravaged millions of hectares of the forest all over the world. Recent technological breakthroughs have increased interest in computer vision-based fire classification that classifies fire and non-fire pixels from image or video datasets. Fire pixels from an image or video can be classified using either a traditional machine learning approach or a deep learning approach. Presently, the deep learning approach is the mainstream in forest fire detection studies. Although deep learning algorithms can handle vast amounts of data, they ignore the variation in complexity among training samples and as a result, their training model performance is limited. Furthermore, deep learning approaches with little data and features perform poorly in real-world challenging fire scenarios. As a result, the current study adopts a machine learning technique to extract higher-order features from the processed images from the publicly available datasets: Corsican dataset and FLAME, and a private dataset: Firefront_Gestosa, for classifying fire and non-fire pixels. It should be emphasized that in machine learning applications, handling multidimensional data to train a model is challenging. Feature selection is used to overcome this problem by removing redundant or irrelevant data that has an impact on the model's performance. In this paper, information-theoretic feature selection approaches are used to choose the most important features for classification while minimizing the computational cost. The traditional machine classifier, Support Vector Machine (SVM) is adopted in the present work, that works on the discriminative features input selected from the feature selection technique. The SVM performs the classification of fire and non-fire pixels with a Radial Basis Function (RBF) kernel, and the model's performance is measured using assessment measures such as overall accuracy, sensitivity, specificity, precision, recall, F-measure, and G-mean. The model draws an overall accuracy of 96,21%, a sensitivity of 94,42%, a specificity of 97,99%, a precision of 97,91%, a recall of 94,42%, an f-measure and g-mean values of 96,13% and 96,19% respectively.

A New Color Model for Fire Pixels Detection in PJF Color Space

Since the number of fires in the world is rising rapidly, automatic fire detection is getting more and more interest in computer vision community. Instead of the usual inefficient sensors, captured videos by video surveillance cameras can be analyzed to quickly detect fires and prevent damages. This paper presents an early fire-alarm raising method based on image processing. The developed work is able to discriminate fire and non-fire pixels. Fire pixels are identified thanks to a rule-based color model built in the PJF color space. PJF is a newly designed color space that enables to better reflect the structure of the colors. The rules of the model are established through examining the color nature of fire. The proposed fire color model is assessed over the largest dataset in the literature collected by the authors and composed of diverse fire images and videos. While considering color information only, the experimental findings of detecting flame pixels candidates are promising. The suggested method achieves up to 99.8% fire detection rate and 8.59% error rate. A comparison with the state-of-the-art color models in different color spaces is also carried out to prove the performance of the model. Based on the color descriptor, the developed approach can accurately detect fire areas in the scenes and accomplish the best compromise between true and false detection rates.

A Contextual and Multitemporal Active-Fire Detection Algorithm Based on FengYun-2G S-VISSR Data

Wildfires are one of the most destructive disasters on the planet. They also significantly impact the land surface. Satellite data have been widely used to detect the outbreak and monitor the expansion of fire incidents for damage assessment and disaster management. Polar-orbiting satellite data have been used for several decades but data from geostationary satellites, which can provide observations with a high temporal resolution, have received much less attention. This paper utilizes data from FengYun-2G, a Chinese geostationary satellite, to detect wildfires in two selected research regions in January 2016. The detection algorithm systemizes image-based analysis to filter out obvious nonfire pixels and temporal analysis to confirm the true detections. Fire detection is based on comparisons between predicted and observed values. The results show that the proposed method has some advantages compared with the use of polar-orbiting satellite data, including early detection and continuous observation. The validation work is conducted based on the collection 6.1 Global Monthly Fire Location Product generated from fire detections by Moderate Resolution Imaging Spectroradiometer (MODIS) sensors. The average accuracy within the target time is 56%, while the omission error rate is over 78%. In detail, the algorithm has a lower omission error rate in Australia while it fails in detecting most of the fire pixels in India. The dominance of small fire incidents, as well as low spatial resolution greatly limit the detection ability. Many small fires were beyond the ability of Stretched Visible and Infrared Spin Scan Radiometer (S-VISSR) data when no significant fire characteristics could be captured. Future development of the algorithm will focus on improving the results by enhancing the adaption to different regions, as well as, including multisource data sets.

Automatic fire pixel detection using image processing: a comparative analysis of rule-based and machine learning-based methods

This paper presents a comparative analysis of state-of-the art image processing-based fire color detection rules and methods in the context of geometrical characteristics measurement of wildland fires. Two new rules and two new detection methods using an intelligent combination of the rules are presented, and their performances are compared with their counterparts. The benchmark is performed on approximately two hundred million fire pixels and seven hundred million non-fire pixels extracted from five hundred wildland images under diverse imaging conditions. The fire pixels are categorized according to fire color and existence of smoke; meanwhile, non-fire pixels are categorized according to the average intensity of the corresponding image. This characterization allows to analyze the performance of each rule by category. It is shown that the performances of the existing rules and methods from the literature are category dependent, and none of them is able to perform equally well on all categories. Meanwhile, a new proposed method based on machine learning techniques and using all the rules as features outperforms existing state-of-the-art techniques in the literature by performing almost equally well on different categories. Thus, this method, promises very interesting developments for the future of metrologic tools for fire detection in unstructured environments.

An approach of surface coal fire detection from ASTER and Landsat-8 thermal data: Jharia coal field, India

Radiant temperature images from thermal remote sensing sensors are used to delineate surface coal fires, by deriving a cut-off temperature to separate coal-fire from non-fire pixels. Temperature contrast of coal fire and background elements (rocks and vegetation etc.) controls this cut-off temperature. This contrast varies across the coal field, as it is influenced by variability of associated rock types, proportion of vegetation cover and intensity of coal fires etc. We have delineated coal fires from background, based on separation in data clusters in maximum v/s mean radiant temperature (13th band of ASTER and 10th band of Landsat-8) scatter-plot, derived using randomly distributed homogeneous pixel-blocks (9×9 pixels for ASTER and 27×27 pixels for Landsat-8), covering the entire coal bearing geological formation. It is seen that, for both the datasets, overall temperature variability of background and fires can be addressed using this regional cut-off. However, the summer time ASTER data could not delineate fire pixels for one specific mine (Bhulanbararee) as opposed to the winter time Landsat-8 data. The contrast of radiant temperature of fire and background terrain elements, specific to this mine, is different from the regional contrast of fire and background, during summer. This is due to the higher solar heating of background rocky outcrops, thus, reducing their temperature contrast with fire. The specific cut-off temperature determined for this mine, to extract this fire, differs from the regional cut-off. This is derived by reducing the pixel-block size of the temperature data. It is seen that, summer-time ASTER image is useful for fire detection but required additional processing to determine a local threshold, along with the regional threshold to capture all the fires. However, the winter Landsat-8 data was better for fire detection with a regional threshold.

Development of a multi-temporal Kalman filter approach to geostationary active fire detection & fire radiative power (FRP) estimation

Most active fire detection algorithms applied to data from geostationary Earth Observation (EO) satellites are adjustments of those originally developed for polar-orbiting systems, and thus the high temporal imaging frequencies offered from geostationary systems are often not fully utilized within such detection approaches. Here we present a new active fire detection algorithm that fully exploits geostationary data's temporal dimension, including both for detecting actively burning fires and quantifying their fire radiative power (FRP). The approach uses a robust matching algorithm to model each pixels diurnal temperature cycle (DTC) in the middle infra-red (MIR) spectral band, the most important band for active fire detection. For each pixel, a Kalman filter (KF) is used to blend the set of basis DTCs with the actual geostationary observations during times of confirmed cloud- and fire-free measurements, allowing for estimates of the pixels true non-fire ‘background’ signal to be provided throughout the day, even when a fire maybe present. This is different to the standard ‘spatial contextual’ approach, where the non-fire background signal is estimated from nearby non-fire pixels. A series of spectral thresholds are then applied to the analyzed pixel in order to identify whether the actual observation departs sufficiently from the estimated non-fire signal to confidently suggest that the pixel contains an active fire. If it does, the difference between the fire pixel and non-fire background signal estimate in the MIR is used to estimate the fire radiative power (FRP) output. We apply this new ‘Kalman Filter Algorithm’ (KFA) to one month of African imagery acquired by the Spinning Enhanced Visible and Infrared Imager (SEVIRI), which is carried onboard the geostationary Meteosat satellite. We compare the resulting active fire detections and FRPs to those produced using the prototype (offline) version of the ‘spatial contextual’ based ‘Fire Thermal Anomalies’ (FTA) algorithm, here termed the pFTA, now used to generate the operational SEVIRI FRP-PIXEL product in the EUMETSAT Land Surface Analysis Satellite Applications Facility (http://landsaf.meteo.pt/), and also compare results to simultaneous detections from the MODIS MOD14/MYD14 active fire products. The KFA shows some advantages over the pFTA algorithm, detecting a greater number of fire pixels, up to ~80% more at the peak of the diurnal fire cycle. These additional fire pixels are primarily low FRP fires (<30MW), which results in the overall fire radiative energy (FRE) increasing by only 20% compared to the pFTA algorithm, though this is still a substantial difference. Comparison against simultaneous MODIS active fire observations confirms that the KFA detects more of the MODIS-detected active fires than does the pFTA (60% more), but at the expense of doubling the false alarm rate. Analysis of the ability of the KFA to aid in the estimation of FRP by providing a more certain estimate of the fire pixels background signal indicates a small 0.2K reduction (rmsd) between the ‘estimate’ of the background temperature and the ‘truth’. One limitation of the KFA approach currently is that it is computationally costly, and also requires the full day diurnal variation to be measured prior to the multi-temporal fire detection process. In its present form it is therefore unsuited to the generation of real-time products. We recommend future work concentrate on extracting maximum performance and utility of geostationary active fire observations by blending both ‘spatial contextual’ and ‘multi-temporal’ approaches.

Detection of Active Wildland Fires Using Multitemporal MODIS Images

The fire fractional area and radiances at 4 and 11 ¿m of active fires in the boreal forests in Siberia and Mongolia were estimated using Advanced Spaceborne Thermal Emission and Reflection Radiometer images. In addition, a stochastic fire model was constructed. The efficiency of a biband threshold method used for detecting active fires on the basis of the thermal anomaly obtained from multitemporal Moderate Resolution Imaging Spectroradiometer (MODIS) images was evaluated using the stochastic fire model. The simulation results of fire detection indicated that the aforementioned method accurately detected 80% of the fire pixels with false alarms that were less than 2 x 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-5</sup> when the radiances of the nonfire pixels at 4 and 11 ¿m were estimated using past MODIS images with a standard deviation (SD) of 1 K. The analyses also indicated that, in order to obtain fewer errors than those obtained using the conventional contextual algorithm (MOD14) from the objective area (omission errors that are less than 62% and commission errors that are less than 2 x 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-5</sup> ), it is essential to carry out the nonfire radiance estimation using past images with an SD of less than 3 K. During an actual application of the multitemporal method to several fire cases, fewer errors occurred than those that occurred using MOD14.

Fire detection based on vision sensor and support vector machines

This paper proposes a new vision sensor-based fire-detection method for an early-warning fire-monitoring system. First, candidate fire regions are detected using modified versions of previous related methods, such as the detection of moving regions and fire-colored pixels. Next, since fire regions generally have a higher luminance contrast than neighboring regions, a luminance map is made and used to remove non-fire pixels. Thereafter, a temporal fire model with wavelet coefficients is created and applied to a two-class support vector machines (SVM) classifier with a radial basis function (RBF) kernel. The SVM classifier is then used for the final fire-pixel verification. Experimental results showed that the proposed approach was more robust to noise, such as smoke, and subtle differences between consecutive frames when compared with the other method.