xy1377660586,Yan,github

bootstrap

The most popular HTML, CSS, and JavaScript framework for developing responsive, mobile first projects on the web.

fine-tuning-a-pre-trained-cnn-for-first-year-sea-ice-and-multi-year-sea-ice-cp-imagery-classificatio

Mapping first-year sea ice and multi-year sea ice in the oceans is significant for many applications. For example, ship navigation and weather forecast. Accurate and robust classification methods of multi-year ice and first-year ice are in demand [2]. Hybrid-polarity SAR architecture will be included in future SAR missions such as the Canadian RADARSAT Constellation Mission (RCM). These sensors will enable the use of compact polarimetry (CP) data in wide swath imagery [1]. Convolutional neural networks (CNNs) are becoming increasingly popular in many research communities due to availability of large image datasets and high-performance computing systems. As Convolutional networks (ConvNets) have achieved great success on many image classification tasks, I pursue this method for the classification of image patches from compact polarimety (CP) imagery into first-year ice and multi-year ice is applicable. In this course project, my work is kind of like the first practice of the CP imagery classification by fine-tuning a pre-trained convolutional neural network (CNN). Specifically, fine-tuning the last fully-connected layer of a pre-trained convolutional networks, I extract patches from simulated CP images as my dataset, the classification accuracy of the test set achieved 91.3% by fine-tuning a pre-trained CNN, compared to 49.4% classification accuracy by training from scratch.