Research On Campus Safety Fire Detection Method Based On Deep Learning

In recent years,using electrical security in student quarters,fire warning is a serious difficulty in school safety management work,which has become a research hotspot in the area of Campus Intelligence at the moment.With the progress of science and technology,the variety of electrical appliances has gradually become wide,and students’ behaviors such as using malignantly loaded offending electrical appliances in the quarters,or smoking cigarettes in the quarters,and using bright fire are very easy to cause fires,which will bring great threats to the students’ life and property safety once fire occurs.In this paper,using deep learning technology as the core technology combined with edge computing cutting edge technology,we propose a deep learning based method for campus safety fire detection,which can conduct real-time monitoring of fireworks information,prompt warning of fire emission,and at the same time can accurately detect the violation of quarters using malignant load appliances,which is of great significance for the safety of campus.Major research efforts in this paper:（1）In response to the campus fire warning problem,a fire detection strategy based on a improvement of the FAM-YOLOv4 algorithm is proposed.Aiming at the original algorithmic parametrical computation is too large to reach the inference speed problem of real-time detection in edge devices,selecting the YOLOv4-small model,and Mobilenetv2 after synchronization shrinkage is used instead of the original backbone network,which is used to improve the inference speed of the algorithm in edge devices.Aiming at the poor effect of pyrotechnic target detection,proposing a feature aggregation method.At the same time,CBAM attention mechanism is introduced into the network output part to improve the ability of the algorithm model to capture global information.Mosaic and Mixup data augmentation methods are used in the training stage to improve the generalization ability of the model.The final improved network compared to the original algorithm’s m AP@0.5 and m AP@.5 :.95 improves by 1.6%and 0.5%,respectively,while boosting 19 fps in speed.（2）Aiming at the illegal appliances with malignant load in the dormitory,proposing a malignant load detection method based on the reconstructed Res Net model.The method collects the power time series data of air conditioning and malignant load appliances through the intelligent socket,and uses the reconstructed Res Net model to classify the two time series data,so as to realize the effective monitoring of malignant load behavior,Further reduce the potential safety hazards of dormitory buildings,and can also be used as an auxiliary system of fire early warning system.（3）Design and develop a campus safety fire detection system model based on edge computing.The system model uses cameras to capture images in the campus,including real-time monitoring images of roads,dormitory buildings,teaching buildings and laboratories.Jetson AGX Xavier edge equipment realizes real-time monitoring of fire disasters in the campus by calling fire and smoke detection algorithm and malignant load detection algorithm.Once the fireworks target is identified or the occurrence of malignant load behavior is found,the alarm information will be sent through the edge end to warn the fire in time.Figure[36] table[6] reference[78]...