Research On Semantic Segmentation Algorithm Of Remote Sensing Image Based On Cascade Feature Fusion Network

With the rapid development of remote sensing technology,remote sensing images have played an important role in human production and life.Remote sensing images contain increasingly rich information data.How to effectively extract useful information from them has become an important issue in remote sensing image processing in recent years.Semantic segmentation can classify and label geographic information in remote sensing images,which provides a basis for further recognition,measurement and analysis of remote sensing images.SegNet,a commonly used semantic segmentation model,performs well in semantic segmentation tasks.However,due to the complex background,high resolution,different target scales and complex edge details of remote sensing images,the accuracy of SegNet in remote sensing image segmentation needs to be further improved,especially in edge and texture details and multi-scale object segmentation.To solve this problem,the SegNet model needs to be further optimized and improved to make it more suitable for remote sensing image segmentation tasks.The main research work in this paper is as follows.P-SegNet model based on the SegNet model is proposed.The improvement is as follows.Remove the pooling of the encoder’s last layer and use only convolution to reduce the loss of spatial information and enhance the model’s perception of details.Build a Bottleneck layer to deepen the network while reducing the number of parameters.Introduce a pyramid pooling module(PPM)to improve the network’s awareness of global information.Secondly,UEP-SegNet model is proposed based on the above model.This model introduces a skip connection structure to fuse the feature maps of the encoder and decoder at the same scale to further improve the segmentation effect.At the same time,the activation function of the model is replaced by ELU to speed up the convergence of the model,avoid neuron death,and improve network stability.Finally,the two proposed models P-SegNet and UEP-SegNet are validated on an open CCF remote sensing image data set.The experimental results show that the segmentation accuracy of the model proposed in this paper is better than that of the original SegNet model and has achieved a good segmentation effect.