Convolutional Neural Networks-Based Sea Ice Detection From TDS-1 Data

This paper presents a convolutional neural networks (CNNs)-based scheme for sea ice detection from TechDemoSat1 (TDS-1) delay-Doppler maps (DDMs). DDM images were employed as input and SIC data from passive microwave sensors were modified as targeted output. Two different CNNs were designed and tested with the full-size DDM data (128-by-20 pixels) and the further processed ones (40-by-20 pixels, and with fixed position in each image). The performance was also compared with the existing neural networks (NNs)-based method. It was found that when DDM data are adequately preprocessed CNNs and NNs share similar accuracy, otherwise the former outperforms the latter. Further conclusion was thus drawn that CNNs were more tolerant to the data format changes than NNs.