Abstract
Remote sensing scene classification has numerous applications on land cover land use. However, classifying the scene images into their correct categories is a challenging task. This challenge is attributable to the diverse semantics of remote sensing images. This nature of remote sensing images makes the task of effective feature extraction and learning complex. Effective image feature representation is essential in image analysis and interpretation for accurate scene image classification with machine learning algorithms. The recent literature shows that convolutional neural networks are mighty in feature extraction for remote sensing scene classification. Additionally, recent literature shows that classifier-fusion attains superior results than individual classifiers. This article proposes the adaptive deep co-accordance feature learning (ADCFL). The ADCFL method utilizes a convolutional neural network to extract spatial feature information from an image in a co-occurrence manner with filters, and then this information is fed to the multigrain forest for feature learning and classification through majority votes with ensemble classifiers. An evaluation of the effectiveness of ADCFL is conducted on the public datasets Resisc45 and Ucmerced. The classification accuracy results attained by the ADCFL demonstrate that the proposed method achieves improved results.
Highlights
T HE key problem in computer vision is to develop algorithms for effective image feature processing to detect and group objects into categories independent of scale, illumination, clutter, and pose positions
The review is classed into five aspects, that is; the developments and challenges in remote sensing, conventional feature representation methods, deep learning and convolutional neural networks, feature learning through Convolution neural networks (CNNs) weights, and deep forests
CNNs are a type of deep learning strategy for image feature learning which applies in task classification
Summary
T HE key problem in computer vision is to develop algorithms for effective image feature processing to detect and group objects into categories independent of scale, illumination, clutter, and pose positions. Convolutional neural networks(CNNs) extract high capacity image feature parameters through convolution and pooling processes to yield image-featurerepresentations In this regard, there are various deep learning architectures in literature [3] [8] which have been adopted for remote sensing image classifications [9] [10]. The effectiveness of deep feature extraction is evident in recent literature [9] [10] [11] on remote sensing scene classification In machine learning, it is common for standard feature learning algorithms to exhibit performance variations on different datasets. In their work [17], they utilize CNNLeNet-5 to extract deep features from digital handwritten images They apply multiple classifiers to learn the features of digit for multi-class classification problem.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
