Research On Flame Detection Method Based On Video

As one of the major disasters that seriously threaten social development and the safety of human life and property.The hazards of fire mainly include: endangering life safety,causing economic losses,destroying civilization achievements,affecting social stability,and destroying the ecological environment.However,real-time flame detection is a challenging task due to the diversity of colors,textures,and shapes of flames,as well as the presence of a large number of flammable objects similar to flames.At present,the accuracy of flame detection in related research is still restricted by bottleneck problems such as lack of flame data sets,insufficient model generalization ability,and dependence on high-performance machines.In order to effectively solve the above key problems,this paper constructs a new large-scale flame data set,and starts from the direction of multi-feature fusion and model lightweight,research and develop a video-based real-time flame detection method.The key research content and innovation points of this paper are as follows.In view of the relatively small number of publicly available flame datasets and the inaccurate labeling of some data,we have spent a lot of time and energy on expanding,cleaning and re-labeling the dataset to build a new large-scale high-quality flame dataset.dataset,providing a new benchmark for future flame research.In view of the fact that existing flame detection methods are easily affected by factors such as weather conditions,light intensity,and background interference,in order to achieve real-time and accurate flame detection in complex scenes.Based on the YOLOv5 algorithm,this paper combines multiple loss functions and multi-feature fusion to propose a real-time and efficient flame detection method.In order to alleviate the imbalance problem of positive and negative samples and make full use of the information of difficult samples,a focal loss function is introduced.At the same time,a multi-feature fusion method is designed by combining the dynamic and static features of flames to achieve the purpose of eliminating false flames.The method in this paper has significantly improved the accuracy,speed,precision and generalization ability,and can effectively reduce the false positive rate.Existing flame detection methods often rely on high-performance machines,and the speed at the embedded and mobile terminals is not satisfactory.Moreover,most of the existing flame detection has slow detection speed and high false detection rate,especially the problem of small-scale flame detection that cannot be solved.This paper proposes a YOLO-based flame detection method,which uses depthwise separable convolution to improve flame detection.Detect the network structure of the model and improve accuracy using various data augmentation techniques and a bounding box-based loss function.Experiments show that the proposed lightweight flame detection model improves both accuracy and speed.A video-based real-time flame detection system is designed and implemented.The main functions include locking the flame target,dynamically displaying the flame area,and real-time flame warning.Multi-channel video is used to test the flame detection system,and the results show that the system can rapidly and accurately identity flames,to satisfy the requirements of a high detection rate and a low false alarm rate.