Research on Large-scale Ocean Date Analysis based on Machine Learning

The rapid development of electronic information and communication technology has led to the widespread use of a variety of marine monitoring technologies, including aerospace remote sensors, automatic buoys, multi-beam echo-sounders, and underwater detection equipment. This has resulted in a significant increase in the volume of marine big data. This work puts forth an innovative model that is based on the most recent machine learning techniques. In consideration of the distinctive attributes of high-dimensional, nonlinear, and strongly spatiotemporally correlated marine data, the model employs an enhanced deep neural network architecture. Initially, the Variational Autoencoder (VAE) was employed for the purpose of reducing the dimensionality of the data set and extracting the most pertinent features, thus enabling the effective handling of the noise and redundant information inherent to marine data. Subsequently, the spatiotemporal dependencies in the data were mined using a spatiotemporal convolutional neural network (ST-CNN) with a self-attention mechanism, enabling high-precision prediction and classification of ocean parameters. The model has been optimised to account for the distinctive characteristics of marine data, thereby enhancing its capacity to adapt to complex marine environments.