Nearshore wave grade video monitoring based on deep learning

Deep learning is an important research field of machine learning. As an effective tool for Big Data processing and analysis, it has been paid more attention. Based on multi-source long time sequence nearshore wave video data labeled by wave synchronous measure in situ, the ocean-specific wave grade deep learning model architecture (Wave-CNNs) was proposed. We constructed the wave train and test datasets which were suitable for deep learning in marine environment. The video monitoring images were preprocessed by using the data augmentation technology to improve the generalization ability of model. According to the correlation of video, error function was introduced to optimize model sensitivity. Finally, the proposed improved deep learning model was applied to 3000-sample wave image training set, and the results were verified by 300-sample wave image test set. The results showed that Wave-CNNs achieved 66.6% recognition accuracy, which was superior to traditional Bayes and SVM methods.