遙測影像的應用一直以來是熱門的研究議題，例如從紅外線衛星影像中進行土地分類便是一項重要的工作。然而，由於IR影像對影像處理來說是一種高維影像，如何從中擷取對土地分類有用的資訊便是一項重要的工作。因此，對高維的衛星影像進行降維，並且在降維空間中擷取重要特徵便直接關係著後續分類結果的成效。 假設有了一個好的降維工具可以得到好的降維空間，那麼如何進一步於降維空間中擷取重要特徵呢?近年來由於電腦硬體的進步使得深度學習成為影像處理中最熱門的一項技術。深度學習其實是一種倒傳遞類神經網路，其主要的架構稱之為Convolutional Neural Network(CNN)，CNN是一種特別針對影像處理而設計的類神經網路架構，它的捲積層(Convolution layer)其實就是一種影像特徵，然而受限於過去電腦運算處理速度，CNN並沒有特別受到矚目，直到近來電腦運算速度的長足進步，CNN可以去計算大量的捲積層，這意味它可以得到大量的特徵，而且是非線性且透過監督式學習得來的特徵，因此對特定目標的分類與學習特別有效。所以，將CNN用於上述降維空間中擷取重要特徵將是一個很好的選擇。有鑑於此，我們期望將深度學習也應用到遙測影像的領域之中以提高土地分類的正確率。 ;The application of remote sensing images has always been a hot research topic. For example, land classification from infrared satellite images is an important task. However, since IR images are high-dimensional images for image processing, how to extract useful information for land classification is an important issue. Therefore, the dimension reduction of high-dimensional remote sensing images and the acquisition of important features in the dimension reduction space are directly related to the effectiveness of subsequent classification results. Assuming that we have a good dimension reduction tool obtain a good feature space, how to further extract important features in the feature space? Recently, due to the advancement of computer hardware, deep learning has become the most popular image processing technology. Deep learning is actually a back propagation neural network. Its main architecture is termed Convolutional Neural Network (CNN). CNN is a neural network architecture specially designed for image processing. In fact, the Convolution layer is an image feature. However, due to the speed of computer computing in the past, CNN has not been particularly noticeable. Until the recent advancement in computer computing speed, CNN can calculate a large number of convolution layers, which means that it can get a lot of features. Moreover, it is a non-linear and supervised learning feature, so it is particularly effective for classification and learning of specific goals. Therefore, it is a good choice to apply CNN for extracting salient features in the above dimension reduction space. Therefore we expect to apply deep learning to the field of remote sensing images to improve the accuracy of land classification.
