This paper aims to develop an automatic feature extraction system for detecting icebergs in Antarctica. Extracting suitable features to discriminate an iceberg from sea ice and land melting based on its content is tedious. Especially in Synthetic Aperture Radar data, high image content is highly affected by speckle noise. Establishing the appropriate spatial relationship between pixels is not producing much accuracy with the standard low-level features. The proposed method introduces the two-level iceberg detection and tracking algorithm. The available samples were used to train the first-level convolution neural network-based features. False-positive predictions have been removed using the multiscale contourlet-based Haralick texture features in the second level. The final detected iceberg movement has been tracked using the temporal image data. The distance moved in both temporal images is computed with the help of latitude and longitude information. The proposed methodology exhibited the best performance over state-of-the-art methods and acquired 79.1% precision and 83.8 F1 score.
2
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
To improve the recognition rate in different conditions, a multiscale face recognition method based on nonsubsampled contourlet transform and support vector machine is proposed in this paper. Firstly, all face images are decomposed by using nonsubsampled contourlet transform. The contourlet coefficients of low frequency and high frequency in different scales and various angles will be obtained. Most significant information of faces is contained in coefficients, which is important for face recognition. Then, the combinations of coefficients are applied as study samples to the support vector machine classifiers. Finally, the decomposed coefficients of testing face image are used to test classifiers, then face recognition results are obtained. The experiments are performed on the YaleB database and the Cambridge University ORL database. The results indicate that the method proposed has performs better than the wavelet-based method. Compared with the wavelet-based method, the proposed method can make the best recognition rates increase by 2.85% for YaleB database and 1.87% for ORL database, respectively. Our method is also suitable for other face databases and appears to work well.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.