Research On The Method For Predicting Firelines Continuously Based On Spatial And Temporal Characteristics Of Forest Fire Spreading

Forest fires are very hazardous to the ecological environment,and they occurrence and spread is a complex process of physical and chemical changes,whose influencing factors include fuels,terrain and climate.The mechanism of forest fire spreading is complex and relatively difficult to study.In this paper,by exploring the spatial and temporal characteristics of forest fire spreading,on the basis of developing an UAV forest fire detection platform,a continuously predicting model of firelines based on spatial and temporal characteristics is designed.The main research work includes:The multi-sensor forest fire detection platform based on the UAV is designed.Calibration methods between infrared cameras,LIDAR and integrated navigation were explored to achieve the data fusion between multiple sensors,which can achieve the collecting of fire datas and the locating of firelines.The set calibration points were used to apply the perspective transformation to the images recorded by the UAV,which let the fire has the same coordinates.Simulation software was used to obtain simulation data for large scale fire spread processes,and the correlation between influence factors and the grid spread time was analysed to extract the key factors for the spread of forest fires modelling and analysis.A single-step prediction model of fireline combining spatial features of the forest fire was designed,and a encoder-decoder fireline location prediction model was proposed based on the fireline.A convolutional neural network was designed to extract the features of the local influence factors of the fire spreading process.The attention mechanism was used to calculate the different degrees of influence of the fireline fire point at the current moment on the fire point at the next moment.The average validation loss was obtained through cross-validation,and the performance of the model is analysed and compared.A multi-step fireline prediction model based on temporal characteristics was designed.Enhancing the temporal dependence of continuous prediction models on fires by modelling and analysing the temporal dimension of fires.In response to the lack of a clear correspondence between fire points between two adjacent moments in the fire spread process,the attention mechanism was used to extract features from fire lines segments,in this way it reduced the encoding length and improves the degree of information memory.The optimal model structure is obtained by hyperparameter tuning.Outdoor spot burn data and simulated data were used for validation analysis of the prediction models.For outdoor spot burn data the time series based multi-step fireline prediction model had an F-value above 0.75.For simulated data the predicted F-value was above 0.79.While the predicted fireline was globally located by a multi-sensor detection platform with an F-value of 0.71.This paper focuses on designing a continuous fire spread prediction model based on a UAV multi-sensor detection platform for mining the spatial and temporal characteristics of the fire spreading,establishing temporal and spatial dependencies through algorithms to enhance the prediction accuracy of the model,achieving the lightweight end-to-end prediction of the fire spreading process and using the UAV forest fire detection platform to achieve the global positioning of the fire line location,which is a guide to fire fighting.