الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 129 |  |  |
| | | from | 129 | | |
Abstract
Nowadays, the analysis and classification of the remote sensing images is very important in many practical applications, such as natural hazards and geospatial object detection, large forest fires mapping, precision agriculture, urban planning, vegetation mapping, and military monitoring. For those reasons, many researchers carried out their experiments on the satellite images for the task of scene classification. Hyperspectral imagery has an important feature which aids in differentiating materials of interest. Indeed, it has detailed spectral information which raises substantially the power of discrimination. from a classification perspective, a common paradigm to analyze hyperspectral data is the pixel-based approach, in which the single image pixels are classified by means of the spectral information they convey. Recently, convolutional neural networks (CNNs) have shown particularly effective for several analysis tasks including segmentation, classification, and object detection. CNNs either require a large amount of training data or have to be fine-tuned on the specific dataset and thus classification task. In this work, we aim to exploit the full potential of the spectral information conveyed by each image pixel by merging the convolutional support vector machine (CSVM), which is an alternative supervised learning strategy based on support vector machines (SVMs), and (1D-CNN) approach in one network. We call the proposed architecture one dimensional convolutional support vector machine (1D-CSVM). Basically, it processes and analyzes the spectral signature of each pixel through a cascade of multiple convolutional and reduction layers and ends by a classification layer. Each convolutional layer in 1D-CSVM uses the linear SVMs as filter banks to generate a set of feature vectors. Indeed, spectral information is very important to decide the nature of each point on the ground. In this work, we aim at exploiting as much as possible the high potential of this rich information source.