Seawater Temperature Profile Reconstruction Based on Transfer Learning

Seawater temperature plays a key role in ocean thermal structure, water mass division, and acoustic field distribution. To address the problem of constructing full-depth temperature profiles with limited data, this paper establishes a transfer learning framework based on the Back Propagation (BP) neural network. By fine-tuning the network with few target domain data on the basis of pre-training a large amount of source domain data, the full-depth temperature profile is constructed with small samples. Two thermistor chain data from the South China Sea are used to analyze the accuracy of temperature profile construction under two scenarios of whether there are internal solitary waves. The results show that, compared with the traditional BP neural network, the BP pre-training and fine-tuning model can achieve high construction accuracy in both scenarios.