Research And Implementation Of Remote Sensing Image Segmentation Algorithm Based On Attention Mechanism

The purpose of semantic segmentation of remote sensing images is to use classification labels to mark each pixel of remote sensing images.Traditional semantic segmentation of remote sensing images mainly interprets remote sensing feature information through texture,boundary,spectrum,length and other information.In recent years,with the rapid development of deep learning models,deep learning algorithms have been applied to remote sensing image processing,which has greatly improved the performance of traditional methods.However,in practical applications,the low signal-to-noise ratio and high resolution of high-resolution remote sensing images have led to a sharp increase in the demand for high-resolution remote sensing image storage.At the same time,due to the close distribution of buildings,streets,parks and other public facilities in urban scenes,the traditional remote sensing image segmentation method extracts buildings from high-resolution remote sensing images,which has low segmentation accuracy and incomplete edge extraction.Therefore,this paper will focus on designing more efficient algorithms to solve these problems.In addition,taking the large demand for remote sensing image storage space as the starting point,a set of remote sensing image algorithm management system is designed and implemented.The service platform architecture design in the cloud environment is studied,and the algorithm is used to segment and identify the remote sensing image integrated in the system.The work of this paper is as follows:(1)Combined with the characteristics of urban scene images,aiming at the problem that the method based on feature pyramid network FPN in deep learning loses semantic information due to channel reduction when fusing features,and the segmentation of small-sized targets and target boundaries is inaccurate when interpreting remote sensing images,a model SGE-FPNet is proposed.The model integrates spatial group enhanced attention SGE and sub-pixel fusion feature module SFF.By grouping channels,sub-pixel convolution is used to directly sample high-order semantic information,extract more semantic features,and enhance the model ’s ability to extract contextual semantic information.At the same time,the network structure is simple,the computational complexity is low,the number of parameters is small,and the performance is significantly improved with low computational cost,which has strong practicability.(2)Aiming at the problems of misclassification,omission,low segmentation accuracy and slow speed of the current Deeplab v3 +model in the segmentation of multi-class targets of remote sensing images in urban scenes,a lightweight semantic segmentation model Deeplab_v3S+based on attention mechanism is proposed.Firstly,in the backbone network,the SE attention module is introduced to perform feature selection on important channels to enhance the feature extraction ability.Secondly,the depthwise separable convolution is used to replace the dilated convolution in ASPP to reduce the overall parameters of the model and speed up the segmentation.Finally,in the decoding stage,the deep semantic information is up-sampled by sub-pixel convolution,and the elements are summed with the shallow semantic information to enrich the fused feature semantic information.(3)A set of remote sensing image algorithm management system is designed and implemented.Based on Kubernetes,containerized application services in cloud environment are realized.Visualization and distributed scheduling schemes are designed to achieve load balancing.At the same time,hierarchical architecture is used to design each functional module.Semantic segmentation algorithm is used to segment remote sensing images and display the results,and the results are uploaded to the cloud,which is easy to upgrade and maintain the system.