Land Classification Using SAR Imagery********

该项目的重点是利用MDA公司提供的合成孔径雷达图像进行土地分类。雷达成像系统基于回波原理：天线发射微波辐射的相干波束，该波束被地球表面吸收、反射和反向散射。信号的后向散射部分由天线接收并处理以形成图像。所产生的图像比光学图像噪声更大，但可以提供光学图像无法容易获得的附加信息。迄今为止，MDA公司使用的是其卫星产生的SAR图像，其模式识别方法是基于低级图像处理的简单方法。现在，考虑到深度学习在许多不同应用上的突破，MDA公司有兴趣评估深度学习模型在SAR图像上的性能。因此，该项目的目的是开发、调整和评估最有前途的深度学习模型，用于SAR图像上的光学图像（例如卷积神经网络），以进行土地分类。该项目旨在为MDA提供专业知识和工程解决方案，以评估卷积神经网络的识别性能。MDA将提供有关SAR图像的知识和技术以及培训和验证数据。MDA提供的数据和他们在SAR图像上的专业知识与我的视觉识别小组的专业知识相结合，将为MDA提供极好的条件，开始一个新的和有前途的SAR图像高性能识别研究。********

This project focuses on land classification using Synthetic Aperture Radar imagery provided by MDA Corporation. Radar imaging systems are based on the echo principle: an antenna emits a coherent beam of microwave radiation, which is absorbed, reflected, and backscattered by the Earth's surface. The backscattered portion of the signal is received by the antenna and processed to form an image. The produced image is noisier than optical images but can provide additional information that cannot be easily obtained by optical images. So far MDA Corporation has used the SAR images produced by its satellites with simple pattern recognition methods based on low level image processing. Now, considering the breaking through of deep learning on many different applications, MDA Corporation is interested in evaluating the performance of deep learning models on SAR images. Thus, the aim of this project is to develop, adapt and evaluate the most promising deep learning models for optical images (e.g. Convolutional Neural Networks) on SAR images for land classification.**This project aims to provide MDA with expertise and engineering solutions in order to evaluate the recognition performance of a convolutional neural network. MDA will provide knowledge and know how on SAR images as well as training and validation data. The combination of the data provided by MDA and their expertise on SAR images together with the expertise of my group on visual recognition will provide MDA with excellent conditions to start a new and promising line of research on high-performance recognition on SAR images. ********