Underwater Target Recognition Based On Deep Learning

With the increasing importance of ocean development in recent years,as well as the great benefits of military confrontation,coastal defense,civil energy development and seabed detection,underwater target recognition system is becoming a research hotspot.Therefore the research of underwater target recognition based on deep learning is of great significance to practical application.Underwater target recognition is a image recognition process in which underwater acoustic images are generated after the subsequent treatment of signals collected based on sonar.Now underwater target recognition has a wide range of application.Therefore many countries in the world have researched a lot in this field to control the commanding point.Deep learning is a kind of special neural network,which has an input layer,an output layer and multiple hidden layers,belonging to a sub-field of machine learning.The main principle is that,the training data through the input layer could be processed by multiple hidden layers and some features could be extracted by a lot of training,then we could use the model to classify or predict in practice.Traditional methods have some problems in underwater target recognition,such as slow processing speed and low recognition rate.To solve the above problems,a new convolutional neural network model called DCNN is proposed.In DCNN,the previous model is improved.The activation function is set to maxout,and the pooling method is set to stochastic pooling.Three methods are used to reduce overfitting.Those methods are local response normalization,DropConnect design and extended data set.In order to solve the vague problem of underwater images,we do preprocessing to underwater images.Several denoising methods are compared in preprocessing,and a new threshold denoising method is proposed,which has higher signal-to-noise ratio.In the experiment,in order to improve the recognition accuracy,the parameters in DCNN,such as learning rate,batch and momentum impulse,are optimized.The experimental results show that DCNN has higher accuracy in underwater target recognition.