| Forest-fires are sudden,destructive,and difficult to deal with.They cause serious harm and loss to forests,forest ecosystems,and human beings.The only effective and efficient way to reduce the Forest-fire incidences is by using accurate fire detection systems.Most of the current methods suffer from a high false alarm rate due to(ⅰ)fire-color objects and moving fire-like objects,and(ⅱ)overdependence only on color for detection of fire or using inappropriate color space and techniques.The main purpose of this research is to address the problem of the high false alarm rate in existing Forest-fire detection methods by increasing the true detection rate and reducing false alarms rate.We developed Forest-fires detection methods using rule-based and flame multi-feature fusion to avoid the drawbacks of the current fire detection methods.The main research content is as follows:(1)Research on Forest-fire detection and identification using image processing and support vector machine(SVM);by adopting color and motion features with special wavelet analysis.The brightness contrast of real fire regions is higher than that of fire-like objects,therefore we used spatial wavelet analysis because,it has an excellent distinguishing ability in this aspect.The new method consists of moving region extraction by applying the background subtraction model.CIE L*a*b*color spaces was adopted to determine candidate fire regions.And the next step is computing special wavelet analysis for the candidate fire regions.SVM is then used to classify the region to either real fire or non-fire,as it provides high performance and accurate classification results with limited training data set.The main idea of the SVM is to create an optimal hyper plane to divide the input data set into two classes with maximum margin.The data in this study is nonlinearly separable,no hyper plane may exist to separate the input data into two parts;therefore,non-linear radial basis function(RBF)kernel is used.This method is accurately distinguishing between genuine fire and fire-like objects,witch making the approach applicable in Forest-fire alarm systems.(2)Research on Forest-fire detection using a Rule-Based image processing algorithm and temporal variation by adopting color and motion features with a temporal variation.The shape and size of the flame are totally changeable as a result of burning materials and airflow;they produce higher temporal-variation.In contrast,rigid and non-fire objects produce lower temporal variation thus,temporal variation was adopted to eliminate spurious fire regions.This method consists of the "Movement Containing Region Detection" algorithm and we developed fire-color algorithm using YCbCr color space due to its effectiveness and less complexity.The approach is applicable to many areas and environment due to its good stable performance in different conditions,as shown in the experimental result.(3)Research on the Forest-fire detection method based on the Multi-features fashion of the flam.The method adopted motion and color features in the pre-processing stage,(RGB)and(HSI)color spaces are used to get the potential fire region.Flame features such as boundary complexity and area variability are used.Finally,a support vector machine was used for classification.We added the data acquired during the fire pre-processing stage to reduce computational redundancy which will enhance the response time.Experimental results have shown that this method can detect fires in complex environments with accuracy up to 98%and 0.4 average processing times per frame in seconds thus,it can be applied to incorporate with a fully automatic surveillance system observing forest motion zone for early fire detecting system.All of the proposed methods were tested using dataset videos,10 of them are offered at(http://www.ultimatechase.com/),and 13 at(http://signalee.bilkent.edu.tr/visifire),10 forest fire videos are offered at(https://www.storyblocks.com/video/search/forest+fire)and we downloaded two fire-like moving object video clips from the public video-sharing site,("youtube.com").The experimental results showed that the proposed approaches improve the accuracy and reduce the false alarm rate compared with a state of the art technique.Our proposed methods correct detection rate is close to 97%,which implies that our proposed new methods have higher stability and accuracy,this indicates that the proposed approaches are correct and can be efficiently applied in different automatic Forest-fire detection systems,complex environment,observing forest motion zone and other areas. |
