Research On Sonar Image Quality Evaluation Method And Construction Of Evaluation Platform

In recent years,with the rapid development of science and technology,the value of sonar technology in underwater rescue,seabed resources development and marine military field is becoming more and more important,which puts forward higher requirements for the quality of sonar image.Due to the complexity of underwater acoustic channel transmission environment,sonar image is easy to degrade in the acquisition process,which seriously affects the subsequent feature extraction and target recognition.At present,the research on sonar image quality evaluation is relatively less,and the performance of the research results is not satisfactory.Therefore,to put forward and improve the sonar image quality evaluation model is helpful to the collaborative development of the current civil field,marine science field and marine military field.In view of the current research status of sonar image quality evaluation,this dissertation designs full reference and no reference quality evaluation methods for sonar image from the perspective of traditional image analysis method and convolution neural network.The main tasks are as follows:（1）The dissertation analyzes and summarizes some mainstream image quality evaluation algorithms and the characteristics of different types of images at home and abroad,and expounds some traditional image analysis techniques and convolutional neural network technologies.On this basis,the research status of sonar image quality evaluation algorithm is analyzed,and the significance of this research is clarified.（2）Research on image quality evaluation algorithm of full reference sonar based on multi-scale structure fusion.Considering the unique structural feature information of sonar image,this dissertation first extracts the microstructure（texture information such as brightness,sharpness and contrast）and macro structure（contour information such as saliency）of sonar image.Then the similarity measure is used to measure the similarity between distorted image and lossless image.Finally,the objective quality fraction of sonar image is predicted by pixel level weighted fusion of the two structures.The experimental results on SIQD database show that the proposed algorithm has a significant improvement in prediction accuracy,monotonicity and consistency,and the execution efficiency is also improved by 6 times.（3）Research on image quality evaluation algorithm of no reference sonar based on deep neural network.In order to pursue more efficient performance,this dissertation designs a dual path deep neural network.In this method,batch normalization and skip operation are used to reduce training time and speed up feature extraction.Then,the global average pooling layer and the fully connected layer are used to fuse the structural features of sonar images extracted from the two paths.Finally,the experimental results on the newly established SIQD database show that the prediction performance and execution efficiency of the designed network model are significantly improved.（4）Design and implementation of sonar image quality evaluation platform.In order to apply the two algorithms in practice,this dissertation uses QT designer and pychar software to design an interactive platform for sonar image quality evaluation,which is convenient for staff to use.The main functions include: user login module,user registration module,password retrieval module and sonar image quality evaluation main interface display module corresponding to the two algorithms.It can realize the screening of high-quality sonar images and save human resources greatly.