Forest Fire Video Recognition Technology Based On Edge Computing

According to the scientific report of global forest resources assessment in 2020,China's forest coverage rate is about 22.96%,with the most annual increase amout in the world.However,the distribution of forest land in China is obviously regionalized.The main forest areas are the southern collective forest area,the northeast forest area,the northern forest area and the southwest forest area.The forest ecological environment is relatively fragile and prone to serious and catastrophic forest fires.Therefore,it is of great significance to use the existing efficient technologies and methods to realize the monitoring and early warning of forest fire and strengthen the research of forest fire monitoring system.In this paper,the traditional forest fire smoke video recognition algorithm is optimized,and the image processing,artificial intelligence,pattern recognition and cloud computing technology are comprehensively used to establish a fire data monitoring and early warning cloud platform system based on edge computing,which improves the timeliness and effectiveness of forest fire video monitoring.1)The video data should be processed before extracting the key frames of forest fire smoke video.We proposed a video shot boundary detection algorithm based on SURF and SIFT features in this paper.The accuracy rate of shot boundary detection is 93%(Cut shots)and 76%(Gradual shots),the recall rate is 95%(Cut shots)and 90%(Gradual shots),and the fl score is 93.9%(Cut shots)and 82.4%(Gradual shots).The experimental results indicate that this algorithm is better than the pixel frame difference algorithms,and has better recall and precision for video features.2)The traditional clustering method needs to adjust the clustering center and the number of clusters manually.Here,an improved k-means based key frame extraction algorithm for forest fire smoke video is proposed to solve this problem.The algorithm is based on the traditional K-means algorithm and introduces the concept of hierarchical clustering and k-nearest neighbor algorithm,the recognition accuracy of key frames is 94%,the recall rate is 87%,and the fl score is 90.3%.The experimental results showed that this algorithm could fully express video content while ensuring accuracy and reducing redundancy.3)The video proximity of cloud,fog and smoke is very high.In order to reduce the false alarm rate of forest fire monitoring,a forest smoke recognition model is constructed by combining HOG features with support vector machine,namely the HOG_SVM model,the accuracy rate of forest smoke recognition reached 87.7%.This algorithm could facilitate network model processing,and it could not only meet the requirements of edge computing terminal equipment but reduce the calculation amount and improve the training recognition effect.4)A forest fire smoke video target recognition model based on improved Faster R-CNN is proposed.This model takes the advantages of feature extraction between CNN networks to extract smoke features.To ensure that the built model meetsthe needs of edge computing terminal equipment,the Soft-NMS algorithm is used to improve the Faster R-CNN model,and the Impro FRCNN model is established to ensure fast and accurate calculation.The accuracy of forest fire smoke targetrecognition reaches 93.1%.5)By analyzing the characteristics of forest fire monitoring data and the functional requirements of the device connection module,the overall design scheme of the edge computing smart gateway software system based on Edge X Foundry is proposed;we design the software architecture of the device connection module based on the CoAP device communication protocol specification;we establish a registration configuration service based on the Consul service discovery framework;we build a command control service component that accesses the unified interface of the device,and deeply mined the data through Impro FRCNN to achieve unified management of devices and data;we design and implement edge smart gateway equipment;we have implemented the edge platform training model and issued the module to enhance the compatibility of the system.In summary,we use the edge computing technology to improve the architecture of traditional cloud center for the forest fire video recognition technology in this paper,and put forward the overall design scheme of edge computing platform based on edge x foundry framework.Then according to the plan combined with embedded technology and recognition algorithm,we have built a terminal-edge-cloud structure forest fire recognition edge computing system.Through innovatively combining forest firerecognition algorithms with edge computing,we fully solve the problems of high failure rate and poor timeliness caused by the intensive growth of traditional cloud terminal equipment andwewill provide a new method for the automation and informatizationof forest fire prevention for our country.