Model research for monitoring forest fires based on UAV multispectral remote sensing
[Objective]Forest fire monitoring often uses thermal infrared cameras to identify forest fires,which have high monitoring accuracy but depend on hardware performance and have limited applications.Less research is conducted on the application of multi-spectral remote sensing images collected by unmanned aerial vehicles(UAVs)at low altitudes in forest fire monitoring.[Method]A forest with lush canopy was selected as the experimental area,and according to the flaming and smoldering status of the fire point,combined with the shading type of the tree crown,it was divided into four working condition experimental groups,flaming and sheltered fire,flaming and shelterless fire,smoldering and sheltered fire,and smoldering and shelterless fire,and the no-fire scenario was used as the control group to carry out the forest fire simulation experiments.The corresponding image data were collected by UAV equipped with thermal infrared,multispectral and visible image sensors,and three machine learning algorithms,Random forest(RF),Support vector machine(SVM),and Back propagation neural network(BP),were used to establish a binary classification model for forest fire monitoring,and accuracy,precision,recall and F1-score for model performance evaluation.[Result]Combining the four working conditions,the thermal infrared camera and visible light camera applied the support vector machine(SVM)model with the highest accuracy,and the multispectral camera applied the random forest(RF)model with the highest accuracy.The thermal infrared camera monitored the highest overall accuracy of 100%,the multispectral camera was close to 100%,and the visible camera reached 85%under all conditions.The thermal infrared camera had the highest accuracy,followed by the multispectral camera,and the visible light camera had the worst performance among the three cameras.The multispectral camera could be a better substitute for the thermal infrared camera in three working conditions,and the visible light camera failed to show a better substitution effect in all four working conditions.[Conclusion]By secondary processing of images using machine learning,multispectral cameras can replace thermal infrared cameras in forest fire monitoring applications,and visible light cameras can partially replace thermal infrared cameras in open fire situations.Using multispectral or visible light cameras to replace thermal infrared cameras for forest fire monitoring can significantly reduce monitoring costs.
forest fire detectionmultispectralUAVmachine learning
万方数据
维普
