| abstract |
The present invention relates to a SAR image sea ice classification method based on a deep convolutional neural network, comprising the following steps: S01: Segment existing SAR images of sea ice; S02: Perform data preprocessing; S03: Perform model training and establish Model; S04: Process the sea ice SAR images to be classified; S05: Merge the classification results. The advantage is that the convolutional neural network model constructed by this method can realize the automatic extraction of image-based features without too much manual intervention; it is an end-to-end classification method for SAR image sea ice, which can achieve the business of sea ice monitoring. The level of automation meets the real-time requirements of offshore operators; the model relies on a large number of labeled samples to automatically extract image features without relying on expert knowledge; the stochastic gradient descent method is used to accelerate convergence, and the quality of model training is judged according to the loss function and accuracy ; use normalization to solve the problem of gradient disappearance or gradient diffusion in network parameter optimization backpropagation. |