传统法与深度学习级联结合的近实时火点监测

Real-Time Fire Detection by Cascading Traditional Approaches with Deep Learning

针对山火监测精度和时效性不足的问题,提出一种级联传统法的多通道卷积神经网络(Multichannel Convolutional Neural Network,MCNN)近实时火点监测算法。首先,结合最大类间方差和空间上下文法,利用背景亮温空间信息差异筛选出潜在火点。然后,采用集成学习思想构建三个卷积神经网络通道,各通道分别输入光谱信息、空间上下文信息及时间地理信息特征的不同组合,同时,通过粒子群优化算法搜索各通道的最佳权重,获取三个通道的火点联合预测概率,实现火点准确识别。结果表明,相比于单通道卷积神经网络(Convolutional Neural Network,CNN)模型,MCNN精确度达到0.88,漏报率降低0.16,并且较日本气象厅官方产品漏报率降低0.06,且实验中模型运行平均耗时172 s。因此,文章提出的MCNN模型可实现较高精度的近实时火点监测,为火灾应急处理提供科学依据。

Aiming at the problem of insufficient accuracy and timeliness of wildfire monitoring,a near realtime fire monitoring algorithm using a Multichannel Convolutional Neural Network(MCNN)with a cascaded traditional method is proposed.Firstly,by combining the OTSU method and the spatial context method,potential fire points are identified by exploiting the differences in background brightness temperature spatial information.Secondly,using the idea of ensemble learning,three convolutional neural network channels are constructed.Each channel takes different combinations of spectral information,spatial context information,and temporalgeographical information features as input.The optimal weights for each channel are obtained by using the particle swarm optimization algorithm to search for the best weights,and the joint prediction probabilities of fire points from the three channels are obtained,achieving accurate fire point recognition.The results show that compared to a single-channel Convolutional Neural Network(CNN)model,the MCNN achieves a precision of 0.88 and reduces the omission rate by 0.16.Furthermore,compared to the Japan Meteorological Agency’s official product,the omission rate is reduced by 0.06.In addition,the highest runtime of the model in the experiment is 268 seconds.Therefore,the MCNN model proposed in this paper can achieve high-precision near real-time fire point detection,providing a scientific basis for emergency fire response.