Forest Burned Area Detection Based on Improved PSPNet
In order to improve the detection accuracy of forest burned area,this paper uses Sentinel-2 satellite images after the fire to propose a forest burned area detection model based on improved PSPNet.This model employs ResNet34 with dilated convolution as the backbone network and fuses the RFB module and ULSAM module inside the backbone network to enhance its feature extraction capability.Finally,skip connection is used to make the decoder part of the model make full use of the four-level feature maps output by the backbone network.The experimental results show that the MIoU and overall accuracy of the improved PSPNet model is 91.86%and 96.89%,respectively,which is 1.52%and 0.67%higher than PSPNet.Compared with other semantic segmentation models,segmentation outcomes achieved by the improved model exhibit richer details and have better generalization performance.
国家科技期刊平台
NSTL
万方数据
