Efficient Target Detection and Classification of SAR Images Using Z-Buffer Convolutional Neural Networks
Autor: | S. Mohamed Mansoor Roomi, P. Vasuki, G. Maragatham, A. Shakin Banu |
---|---|
Rok vydání: | 2020 |
Předmět: |
Computer science
business.industry Deep learning ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Wavelet transform Pattern recognition Convolutional neural network Image (mathematics) symbols.namesake ComputingMethodologies_PATTERNRECOGNITION Fourier transform Region of interest symbols Noise (video) Artificial intelligence business Smoothing |
Zdroj: | Smart Trends in Computing and Communications: Proceedings of SmartCom 2020 ISBN: 9789811552236 |
Popis: | Target detection and classification for military, geographical, and other scientific research areas are the demanding requirement. This paper aims to detect the targets effectively as well as enhances the quality of detected image before classification by deep learning techniques. A Z-buffer convolutional neural network (Z-BCNN) model consisting of two phases is more suitable for target detection, and classification of SAR image is proposed in this paper. In the first phase, the region of interest (ROI) is extracted from the background of the SAR image by means of cavity detection algorithm and elliptical Fourier descriptors are computed to describe the characteristics of target outline, whereas in the second phase, the noise present in the detected SAR image is reduced by using wavelet transform-based brute force thresholding algorithm with directional smoothing. Afterward the detection of hidden targets, classification of military vehicles from SAR images is done by using Z-buffer convolutional neural network. The proposed methodology gives significant results in terms of classification accuracy compared with the other existing algorithms. |
Databáze: | OpenAIRE |
Externí odkaz: |