Component Design And Implementation Of Automatic Road Extraction System For Remote Sensing Image

Road extraction based on high-resolution remote sensing imagery has always been an important topic in the field of remote sensing.In the past 30 years,many researchers have proposed a number of methods based on different principles,but due to the complexity of remote sensing image features and the limitations of the method.An algorithm with sufficient adaptability has not yet appeared.With the improvement of computer computing power,deep learning has been continuously developed in various fields,especially image processing,and it has become a hot direction for researching road extraction of high-resolution remote sensing images.Because the extraction technology of road network information is of great significance in the field of military field and disaster rescue command,there is still no software system that can carry various algorithms and quickly extract road information.Therefore,this paper focuses on deep learning methods.Research on high-resolution remote sensing imagery road extraction and result optimization filtering,and developed a set of remote sensing image road extraction system with practical application value based on component technology.The research work of this paper is as follows:(1)The structure of full convolutional neural network is studied,and a road extraction model of high resolution remote sensing image based on similarity mapping is proposed.The model has road extraction capability for remote sensing images,and has no special requirements on the input data size,and does not require excessive human participation.And research on small sample training and application in the training of the model.The accuracy of the road extraction obtained by the model after the experimental data training in this paper is 86.5%.(2)The noise filtering model based on deep learning is studied to process the road extraction result image with excessive noise.The road extraction result filtering model is established by using the anti-learning thought and the loop learning idea respectively.The Generative Adversarial Networks model has a certain image enhancement capability in the process of filtering,that is,it complements a part of the missing road.After the generative confrontation model is processed,the road extraction result rejects the noise by 7.3%.After the cyclic learning model is processed,the road extraction result rejects noise by 6.4%.It can be seen that the filtering model is helpful for the accurate improvement of the road extraction results.(3)The componentization idea is studied.The above-mentioned deep learning model algorithm is componentized by dynamic link library technology,and the algorithm component is embedded into the designed road extraction software system.Developed a user-friendly,logically structured software system that adds any algorithm plug-ins whenever required.