A very high-resolution scene classification model using transfer deep CNNs based on saliency features
- PDF / 994,567 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 18 Downloads / 173 Views
ORIGINAL PAPER
A very high-resolution scene classification model using transfer deep CNNs based on saliency features Osama A. Shawky1 · Ahmed Hagag2
· El-Sayed A. El-Dahshan1,3 · Manal A. Ismail4
Received: 21 January 2020 / Revised: 14 July 2020 / Accepted: 6 October 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Developing remote sensing technology enables the production of very high-resolution (VHR) images. Classification of the VHR imagery scene has become a challenging problem. In this paper, we propose a model for VHR scene classification. First, convolutional neural networks (CNNs) with pre-trained weights are used as a deep feature extractor to extract the global and local CNNs features from the original VHR images. Second, the spectral residual-based saliency detection algorithm is used to extract the saliency map. Then, saliency features from the saliency map are extracted using CNNs in order to extract robust features for the VHR imagery, especially for the image with salience object. Third, we use the feature fusion technique rather than the raw deep features to represent the final shape of the VHR image scenes. In feature fusion, discriminant correlation analysis (DCA) is used to fuse both the global and local CNNs features and saliency features. DCA is a more suitable and cost-effective fusion method than the traditional fusion techniques. Finally, we propose an enhanced multilayer perceptron to classify the image. Experiments are performed on four widely used datasets: UC-Merced, WHU-RS, Aerial Image, and NWPU-RESISC45. Results confirm that the proposed model performs better than state-of-the-art scene classification models. Keywords VHR image · Saliency detection · Features fusion · Transfer learning · Scene classification · Deep CNNs
1 Introduction With the growth of Earth observation technology, many different types of high-resolution images are available for the Earth’s surface. The very high-resolution remote sensing image (VHR) is one of the best surface observation data sources for mapping the urban landscape which are obtained from optical sensor and characterized by less than 50 cm resolution. The speed development of remote sensing technology facilitates the acquisition of remote sensing images with high and very high spatial resolution. Therefore, robust
B
Ahmed Hagag [email protected]; [email protected]
1
Faculty of Computers and Information Technology, Egyptian E-Learning University, Dokki, Giza 12611, Egypt
2
Department of Scientific Computing, Faculty of Computers and Artificial Intelligence, Benha University, Benha 13518, Egypt
3
Department of Physics, Faculty of Science, Ain Shams University, Abbasia, Cairo 11566, Egypt
4
Faculty of Engineering, Helwan University, Helwan, Cairo 11731, Egypt
remote sensing image classification techniques are needed to understand the surface of the Earth’s VHR image in order to allocate a label for each image based on its contents. Extracting salient features from satellite images is the main challenge in c
Data Loading...
