Research On Fire Recognition Algorithm Based On Video Image

Fire is a common disaster.Once a fire breaks out,people's lives and property will be greatly threatened.Fire identification and timely warning in the early stage of fire will have positive significance for protecting people's lives and property.For a long time,people have used various temperature-sensitive,smoke-sensitive and photosensitive detectors to collect data to determine whether a fire has occurred.However,due to the installation position of the sensor and the effective distance of detection,the detection range is restricted,and the sensor has a single judgment information,which is susceptible to interference from ambient light,temperature and humidity,and airflow,thereby generating false alarms,false negatives,etc.,reliability and stability is difficult to guarantee.In addition,the propagation of temperature,smoke,radiation and other parameters caused by fire takes time,and the speed of fire spread is extremely fast.The traditional fire detection system can not meet the fire prevention requirements at all.For these issues,based on video images,three methods based on fire video images are designed.Based on the traditional fire image processing method,a fire identification method based on multi-feature fusion is designed.Firstly,combining the RGB criterion and HIS criterion,setting the appropriate threshold conditions,based on the OpenCV open source library,the corresponding pixel area of the flame in the video is recognized on the VS2013 platform,and the original image is binarized.The flame area is marked with a rectangular frame;then,based on this,the gray image of the original image is proposed to be clustered by the K-Means algorithm,and multiplied by the suspected flame region to obtain the contour of the suspected flame region,and the circularity and eccentricity are further added to complete the identification of the fire.The experimental results show that the method can extract most of the fire image contours and the recognition effect is better.Based on the classification recognition method of deep learning,a classification network of fire video images based on convolutional neural network is designed.The convolutional neural network is applied to the field of fire identification.The convolutional neural network fire classification and recognition architecture with both accuracy and processing complexity is proposed.A convolutional neural network based on deep separable convolution is designed and built.The fire classification recognition model is obtained,which shows considerable accuracy on the test set.Compared with other classification networks,the network has less computational complexity.For hardware devices with the same processing power,the network takes less time.This provides the possibility for the application of the network on the mobile and embedded devices.Based on the deep learning detection method,a convolutional neural network method for identifying and locating fire video images is proposed.The model is inspired by the SSD algorithm.The model is fine-tuned by the idea of migration learning,and the fire video image detection model is obtained through a large number of trainings.Different fire scenes and non-fire scenes were selected for detection,which have good recognition results,indicating that the detection model has a good generalization ability.In order to further improve the accuracy of model detection and reduce the interference of fire-like scenes,based on the positioning frame,the position coordinate changes of the two frames before and after the video are proposed,and the interference caused by the static fire scene to the fire video detection is eliminated.