Classification of Coastal Altimetric Waveforms using Machine Learning Technology

Waveforms are the primary observations of the satellite radar altimeters, from which several crucial oceanography parameters, such as sea surface height, significant wave height and sea surface wind speed, are retrieved. Aiming at the problem of the complex and irregular shape of the radar altimeter waveforms in the coastal zone, an approach for radar altimeter coastal waveform classification using machine learning technology is proposed. This approach can fulfill the requirements of intelligent recognition of the radar altimeter waveforms. This paper processed and analyzed of a large number of altimeter waveforms offshore Hong Kong coast. High-quality waveform samples were singled out from the original waveform datasets, and eight waveform features were extracted as the basis for classification. After dimensionality reduction with principal component analysis, five features were retained. The features after dimensionality reduction were input into a Support Vector Machine (SVM) anomaly detection module, and the waveforms identified as abnormal were input to the BP (Back Propagation) neural network classifier, where the abnormal waveforms are divided into three categories. Different types of waveforms are suitable for different retracking algorithms. The performance of anomaly detection and neural network classifier were evaluated. The anomaly detection accuracy rate is 94.60%, and the neural network classification accuracy rate is about 95%. Experimental results show that the use of SVM anomaly detection and BP neural network classifier for radar altimeter coastal waveform classification is satisfying and the classification specification can be achieved.