This paper proposes a convolutional neural network (CNN) method to estimate subsurface temper-ature (ST) in the Pacific Ocean from a suite of satellite remote sensing measurements. These include sea surface temperature(SST), sea surface height (SSH), and sea surface salinity (SSS). We propose using the multisource sea surface parameters to establish a monthly CNN model to reconstruct the ocean subsurface temperature (ST) and use Argo data for accurate validation. The results show that the CNN can accurately estimate the ST of the Pacific Ocean by using the model. We trained the model for 12 months. The most prominent months are January, April, July, and October with average mean square error (MSE) values of 0.2659, 0.3129, 0.5318, and 0.5160, and the average coefficients of determination (R2) were 0.968, 0.971, 0.949, and 0.967, respectively. This study improves the accuracy of ST estimation and the good results based on reanalysis indicate that the model is promising to be applied to satellite observations.

Underwater acoustic sensor networks (UWA-SNs) are envisioned to perform monitoring tasks over the large portion of the world covered by oceans. Due to economics and the large area of the ocean, UWA-SNs are mainly sparsely deployed networks nowadays. The limited battery resources is a big challenge for the deployment of such long-term sensor networks. Unbalanced battery energy consumption will lead to early energy depletion of nodes, which partitions the whole networks and impairs the integrity of the monitoring datasets or even results in the collapse of the entire networks. On the contrary, balanced energy dissipation of nodes can prolong the lifetime of such networks. In this paper, we focus on the energy balance dissipation problem of two types of sparsely deployed UWA-SNs: underwater moored monitoring systems and sparsely deployed two-dimensional UWA-SNs. We first analyze the reasons of unbalanced energy consumption in such networks, then we propose two energy balanced strategies to maximize the lifetime of networks both in shallow and deep water. Finally, we evaluate our methods by simulations and the results show that the two strategies can achieve balanced energy consumption per node while at the same time prolong the networks lifetime.

With the rapid development of marine IoT (Internet of Things), ocean MDTN (Mobile Delay Tolerant Network) has become a research hot spot. Long-term trajectory prediction is a key issue in MDTN. There are no long-term fine-grained trajectory prediction methods proposed for ocean vessels because a vessel’s mobility pattern lacks map topology support and can be easily influenced by the fish moratorium, sunshine duration, etc. A traditional on-land trajectory prediction algorithm cannot be directly utilized in this field because trajectory characteristics of ocean vessels are far different from that on land. To address the problem above, we propose a novel long-term trajectory prediction algorithm for ocean vessels, called L-VTP, by utilizing multiple sailing related parameters and K-order multivariate Markov Chain. L-VTP utilizes multiple sailing related parameters to build multiple state-transition matrices for trajectory prediction based on quantitative uncertainty analysis of trajectories. Trajectories’ sparsity of ocean vessels results in a critical state missing problem of a high-order state-transition matrix. L-VTP automatically traverses other matrices in a specific sequence in terms of quantitative uncertainty results to overcome this problem. Furthermore, the different mobility models of the same vessel during the day and the night are also exploited to improve the prediction accuracy. Privacy issues have been taken into consideration in this paper. A quantitative model considering Markov order, training metadata and privacy leak degree is proposed to help the participant make the trade-off based on their customized requirements. We have performed extensive experiments on two years of real-world trajectory data that include more than two thousand vessels. The experiment results demonstrate that L-VTP can realize fine-grained long-term trajectory prediction with the consideration of privacy issues. The average error of 4.5-hour fine-grained prediction is less than 500 m. In addition, the proposed method can be extended to 10-hour prediction with an average error of 2.16 km, which is also far less than the communication range of ocean vessel communication devices.

Indian Ocean Dipole (IOD) is a large-scale physical ocean phenomenon in the Indian Ocean that plays an important role in predicting the El Nino Southern Oscillation in the tropical Pacific. Predicting the occurrence of IOD is of great significance to the study of climate change and other marine phenomena. Generally, the IOD index is calculated to judge whether the IOD occurs. In this paper, a convolutional LSTM (convLSTM) neural network is used to build the deep learning model to predict the sea surface temperature in the next seven months and calculate the IOD index. Through the analysis of marine atmospheric data with complex temporal and spatial relationships, the wind field signal knowledge of the physical ocean is introduced to predict IOD phenomenon by combining the prior knowledge of the physical ocean and deep learning. The experimental results show that the average correlation of IOD index time series to the true IOD index time series is 82.87% from 2015 to 2018, seven months ahead for IOD prediction. IOD manifests as sea surface temperature (SST) anomaly changes, and this thesis verifies that the wind field signal information has a positive impact on the prediction of IOD changes. Moreover, the convLSTM can predict this anomaly better. The IOD index line graph can generally fit the real IOD index variation trend, which has a profound impact on the study of the IOD phenomenon.

Impacted by global warming, the global sea surface temperature (SST) has increased, exerting profound effects on local climate and marine ecosystems. So far, investigators have focused on the short-term forecast of a small or medium-sized area of the ocean. It is still an important challenge to obtain accurate large-scale and long-term SST predictions. In this study, we used the reanalysis data sets provided by the National Centers for Environmental Prediction based on the Internet of Things technology and temporal convolutional network (TCN) to predict the monthly SSTs of the Indian Ocean from 2014 to 2018. The results yielded two points: Firstly, the TCN model can accurately predict long-term SSTs. In this paper, we used the Pearson correlation coefficient (hereafter this will be abbreviated as “correlation”) to measure TCN model performance. The correlation coefficient between the predicted and true values was 88.23%. Secondly, compared with the CFSv2 model of the American National Oceanic and Atmospheric Administration (NOAA), the TCN model had a longer prediction time and produced better results. In short, TCN can accurately predict the long-term SST and provide a basis for studying large oceanic physical phenomena.

In this paper, we propose a deep learning-based method for classifying abnormal judgments of music song interpretation problems, using a computer to record the singer’s voice and then analyze and judge it with a trained method model, pointing out the main problems that exist in the song singing process, which is of practical significance for the singer’s self-learning and related teaching guidance. We collected more than 300 singers’ audio, and the data was calibrated and classified by professional researchers in the field of music, mainly divided into seven categories: vocal pallor, inability to keep up with the rhythm, running out of tune, lack of breath, unclear spitting, narrow range (the highs can’t go up and the lows can’t come down), and true and false voice problems. The audio feature extraction process applies Short Time Fourier Transform (STFT), Mel Frequency Cepstrum Coefficient (MFCC), and spectral prime to extract the features of song audio, and a randomly selected part of the audio is used as the training and validation dataset. This dataset is trained using a residual neural network and efficient neural network. The experimental results of our model show that the data training accuracy is about 90.1%, which achieves good results.