Research Of Nearshore Ocean Wave Grade Classification Based On Deep Learning

Ocean is the cradle of mankind and an important environment for human survival and development.The area of the ocean is about 2/3 of the earth's area.The marine resources are abundant and carries many human life,such as maritime transport or marine fishing and aquaculture.However,marine environment is extremely complex.The development and exploitation of marine resources,marine disaster prevention and reduction,and marine scientific research all depend on the accurate grasp of basic ocean conditions and the internal laws of the tides and ocean waves.Ocean waves bring many disasters and loss to human every year.According to the data of the State Oceanic Administration,only in 2016,there were 36 times of disaster wave processes with effective wave height above 4 meters in China's coastal areas.Due to the disaster,the direct economic loss can be 37 million yuan.So it is of great significance to research as well as monitor the waves and forecast the waves.The traditional ocean wave monitoring methods such as the ocean current meter monitoring is difficult to monitor,with low real-time and convenience;the wave meter monitoring has the disadvantage of low automation;the radar monitoring and photophotographic method have the disadvantage of high cost.In order to master the ocean wave rule more accurately,it is necessary to monitor the waves continuously for a long time.Video surveillance has the characteristics of real-time and non-contact,which can make up for the shortcomings of traditional monitoring methods.However,the ocean wave grade video data is large and complex and it is difficult to extract features artificially.Convolutional nerual network(CNN)is one of the successful models of deep learning.CNN improves the deficiency of low-level features that are difficult to extract and select.Thus,it has been widely used in the field of video and image processing.The periodicity of ocean waves is fast,the temporal spatial correlation of the time series image is high and the feature is complex.The traditional machine learning methods need to extract the prior knowledge artificially,making the classification and recognition of the ocean wave image difficult.In view of this,the paper attempts to classify the nearshore ocean waves based on the live video surveillance data and convolution neural network(CNN).According to the ocean wave grade image data,under the relation of international general ocean wave grade standard and wave meter synchronously measure ocean wave grade data,the label of ocean wave grade data set is obtained.The ocean wave grade dataset suitable for deep learning is constructed by using data augmentation technology for small sample data.Upon the traditional CNN structure,in the process of Back Propagation(BP),we introduce an elasticity factor based on video correlation to revise and optimize the sensitivity.Finally we proposed an ocean wave classification model structure(Wave-CNNs)based on deep learning.We carry out comparsion and analysis of the classification results between Wave-CNNs and traditional machine learning methods.In addition,combining with ocean wave grade dataset and Wave-CNNs model,based on the correction of BP error sensitivity,further experiments were conducted on the input frame size,data set ratio,network depth,and hyper-parameter optimization.After this,we obtain the optimized parameters and strategies suitable for classification of ocean wave grade.Firstly,we carried out the research of classification of offshore ocean wave grade.One is,based on the ocean wave video monitoring data of Dayang mountain deep-water port ecological base,a series of processing was used to construct a three-grade ocean wave dataset.We contructed the three-grade ocean wave data set suitable for deep learning,including wave grade training sets,validation sets,and test sets.The other is,on the basis of high temporal spatial correlation of ocean wave video,we introduced an error function as an elasticity factor.The sensitivity is improved by optimizing the internal model bias(i.e.the error is the derivative of all input to a node)to decrease overfitting and improve model generalization ability.Then we design the parameters of the convolution layer and the sampling layer,such as the size and number of convolution kernel,sampling radius,putting forward the architecture of the ocean wave grade deep learning model(Wave-CNNs).Wave-CNNs are compared and verified with traditional machine learning models(support vector machine,Bayesian networks)on the constructed ocean wave-grade dataset.The result showed that the classification accuracy of Wave-CNNs is superior to that of support vector machines and Bayesian networks classification.Afterwards,in view of how to improve the classification accuracy of the Wave-CNNs model to the ocean wave grade dataset,a global and local optimization strategy is proposed for the nearshore ocean wave grade dataset.One is,before and after global optimization strategy,the Wave-CNNs experiments showed that the number and size of convolution kernels,the introduction of regularization and Dropout can effectively improve the model classification accuracy and improve the recognition accuracy.Second is,using Wave-CNNs for local optimization,Wave-CNNs were used to conduct compared experiments on frame size,data set ratio,network depth,and hyper-parameter optimization.In terms of classification of ocean wave grade,the optimal frame size,the best ratio of wave grade training set and validation set,the best network depth,the best learning rate,and the optimal batch size are obtained.These parameters make Wave-CNNs' classification accuracy increased from 70.33% to 92.33%.In summary,this paper deeply analyzes the oceanic long-range wave-grade video data of Dayang mountain,constructs a nearshore wave garde data set for deep learning,and proposes a nearshore ocean wave-grade classification model(Wave-CNNs)based on deep learning.The research results obtained in this dissertation can be applied to the ocean wave video monitoring system,and provide theoretical support for early warning and forecasting of nearshore ocean wave grade.