一种基于卷积神经网络的林火蔓延模型

A Forest Fire Spread Model Based on Convolutional Neural Network

基于速度公式来预测某时刻森林火灾火线位置的传统林火蔓延模型,其精度相对较低。针对上述问题,本文提出了一种卷积神经网络(CNN),其包含一个输入通道和一个输出通道,输入通道输入影响林火蔓延各种因素的二维图像和t时刻的初始燃烧图,输出通道输出t+1时刻的燃烧概率图,然后将燃烧概率图经过F-Measure的阈值处理得到预测的燃烧图。随机使用两场火的数据来检验CNN模型的鲁棒性,其Kappa系数分别为0.89和0.87,预测结果与真实结果高度一致。CNN模型对火线位置预测准确率方面高于传统林火蔓延模型,这有利于消防人员快速找到火线位置并进行处理。

The accuracy of traditional forest fire spread models based on velocity formula to predict the location of a forest fire line at a certain time is relatively low.To solve the above problems,in this paper,a convolutional neural network(CNN)involving an input channel and an output channel was proposed.In the input channel,the 2d images of various factors affecting the spread of a forest fire and the initial burning diagram at time t were input,and in the output channel,the burning probability diagram at time t+1 was output.Then the combustion probability graph was processed by the threshold value of F-Measure to obtain the predicted combustion graph.The data of two fires were randomly used to test the robustness of the CNN model,with its Kappa coefficients being 0.89 and 0.87,respectively,and the predicted results being highly consistent with the real results.The prediction accuracy of CNN model is higher than that of traditional forest fire spread models,which is beneficial for firefighters to find and deal with the location of a fire line.