Design And Implementation Of Fire Monitoring Terminal Based On Image Recognition

The use of varied electrical devices and a significant number of flammable materials have led to an increase in the frequency of fires year after year as urbanization progresses.Rapid and accurate fire detection in the early stages of a fire is critical for minimizing casualties and property losses.For identification,most image fire detectors send images to cloud servers or local servers.The identifying effect is constrained by network bandwidth and local server performance,and it cannot be performed when the network is down.Therefore,a fire monitoring terminal based on image recognition is proposed in this thesis.The improved infrared fire detection model is accelerated by using the characteristics of FPGA(Field-Programmable Gate Array)parallel computing to achieve all-weather rapid fire detection of the terminal.Local execution of fire detection tasks minimizes the quantity of data that terminals post to the server,reduces the stress on the cloud server,and enhances task execution performance.The following are the main points of this thesis:1.Combining with the research status of fire detection technology at home and abroad,this thesis designs a fire monitoring terminal based on image recognition,and puts forward the overall design scheme of software and hardware of the fire monitoring terminal according to the analysis of terminal functions and requirements.2.An improved YOLO(You Only Look Once)network infrared fire detection model is proposed to improve the problem of low recognition accuracy caused by few pixels and few accessible characteristics when infrared images are employed for fire detection.To extract infrared tiny target characteristics and increase the accuracy of fire detection,a residual network and Spatial Pyramid Pooling(SPP)are used to fuse deep and shallow features.Experimental results show that the improved model can improve the accuracy of fire identification.3.Since the improved fire detection model requires a substantial amount of computation,it is implemented on an ARM+FPGA heterogeneous architecture to accelerate the model by utilizing FPGA parallel processing capabilities.To begin,the model parameters are quantified to ensure that the model consumes as little terminal storage and processing resources as possible.The parallel computing section of the redesigned model structure is next analyzed,and the calculation is accelerated on FPGA.Multi-channel transmission is employed between the ARM and the FPGA to reduce data transmission delays,and finally realize the rapid fire detection of the terminal.Experimental results show that the terminal can run the improved infrared fire detection model locally and realize all-weather fire detection with almost constant accuracy.4.According to the function and requirements analysis,the hardware and software design of the fire monitoring terminal are completed,and the terminal function is tested.The test results show that all functions of the terminal are normal.When a fire is detected,the terminal can timely control the action of the local equipment,and the local execution of the detection task can effectively reduce network data transmission and improve task execution efficiency.