Efficient deep feature selection for remote sensing image recognition with fused deep learning architectures
- PDF / 3,682,589 Bytes
- 19 Pages / 439.37 x 666.142 pts Page_size
- 24 Downloads / 199 Views
Efficient deep feature selection for remote sensing image recognition with fused deep learning architectures Fatih Özyurt1
© Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract Convolutional neural networks (CNNs) have recently emerged as a popular topic for machine learning in various academic and industrial fields. It is often an important problem to obtain a dataset with an appropriate size for CNN training. However, the lack of training data in the case of remote image research leads to poor performance due to the overfitting problem. In addition, the back-propagation algorithm used in CNN training is usually very slow and thus requires tuning different hyper-parameters. In order to overcome these drawbacks, a new approach fully based on machine learning algorithm to learn useful CNN features from Alexnet, VGG16, VGG19, GoogleNet, ResNet and SqueezeNet CNN architectures is proposed in the present study. This method performs a fast and accurate classification suitable for recognition systems. Alexnet, VGG16, VGG19, GoogleNet, ResNet and SqueezeNet pretrained architectures were used as feature extractors. The proposed method obtains features from the last fully connected layers of each architecture and applies the ReliefF feature selection algorithm to obtain efficient features. Then, selected features are given to the support vector machine classifier with the CNN-learned features instead of the FC layers of CNN to obtain excellent results. The effectiveness of the proposed method was tested on the UC-Merced dataset. Experimental results demonstrate that the proposed classification method achieved an accuracy rate of 98.76% and 99.29% in 50% and 80% training experiment, respectively. Keywords CNN · Feature reduction · ReliefF algorithm · Image recognition · UC-Merced dataset
* Fatih Özyurt [email protected] 1
Department of Informatics, Firat University, Elazig, Turkey
13
Vol.:(0123456789)
F. Özyurt
1 Introduction Over the past few decades, remote sensing has undergone dramatic changes in the spatial resolution of the image and increases in the acquisition rate. Innovations in the field of computer technology and increasing spatial resolution offer new opportunities to improve image detection and remote sensing, enabling the development of new approaches. The classification of land use from remote sensing images is vital for the monitoring and management of human development in terms of scope and scale. Moreover, automatic classification through machine learning is crucial in order to understand the Earth’s ever-changing surface, including anthropogenic changes. The automatic land cover classification of high-resolution remote sensing images is highly desirable but poses many difficulties due to the wide variety of volumes and various daily collected satellite and weather images. This is further complicated by a wide range of features and scales related to geographically diverse human areas on earth. Conventional image feature extraction and image analysis techniques were
Data Loading...
