Research On Retrieving Algorithm Of Sea Wave Height Based On 3D Convolutional Network

The ocean monitoring system is mainly responsible for visual monitoring and parameter detection of ocean conditions.With the rise of artificial intelligence technology,the application of deep learning technology to ocean wave detection has become the development trend of ocean monitoring systems.At present,most of the researches on applying deep learning to the detection of ocean wave parameters only focus on the extraction of two-dimensional features of ocean wave images by two-dimensional convolutional neural networks,and the network model can only learn the spatial information of ocean wave images.In order to make use of the time information of ocean wave motion,this paper proposes a neural network model(Wave-3DCNN)based on the three-dimensional convolution kernel by increasing the time dimension of the two-dimensional convolution kernel into a three-dimensional convolution kernel,and uses this model The algorithm for inversion and detection of sea wave height is realized.The main work of this paper is as follows:First of all,according to the requirements of wave height detection,the wave height level is divided more finely,and the comparison standard for the wave level and wave height value of the Yunxiao data set is made.Based on the traditional extraction of spatial features of ocean waves,the two-dimensional convolution kernel is expanded into a three-bit convolution kernel in the time dimension to construct a feature extraction block(3D-Block)that can simultaneously extract the spatial-temporal information of ocean waves.On the basis of the feature extraction block,a lightweight wave high-level classification network model(Wave-3DCNN)is constructed.Secondly,perform data preprocessing and truth labeling on the collected ocean wave video to create a ocean wave video data set.On this data set,Wave-3Dcnn is trained and model optimized,so that the network classification accuracy rate reaches 88%.Then compare and analyze the model in this paper with two-dimensional convolutional neural network and other three-dimensional convolutional neural network models.Next,test the model in this paper on the test set and invert the grade result into wave height results.The average relative error of the tested wave height results is counted and compared with other models,and the reliability of the test results is evaluated and analyzed.By comparing the evaluation index of the classification model,the comparison of the wave height detection error and the complexity analysis of the model,it proves that the model in this paper has greater superiority and reliability in the classification of the wave level and the detection of the wave height.Finally,the algorithm is actually applied to sea wave video detection,which can directly process offline or real-time video of sea waves,classify and detect different levels of waves,and display the detection results on the sea wave video in real time.