PROJECT TITLE :

Supervised Polarimetric SAR Image Classification Using Tensor Local Discriminant Embedding - 2018

ABSTRACT:

Feature extraction may be a terribly vital step for polarimetric artificial aperture radar (PolSAR) image classification. Several dimensionality reduction (DR) strategies have been employed to extract options for supervised PolSAR image classification. But, these DR-primarily based feature extraction strategies only consider every single pixel independently and thus fail to require into account the spatial relationship of the neighboring pixels, therefore their performance might not be satisfactory. To deal with this issue, we introduce a novel tensor local discriminant embedding (TLDE) methodology for feature extraction for supervised PolSAR image classification. The proposed technique combines the spatial and polarimetric info of each pixel by characterizing the pixel with the patch targeted at this pixel. Then each pixel is represented as a third-order tensor of which the primary 2 modes indicate the spatial info of the patch (i.e., the row and also the column of the patch) and the third mode denotes the polarimetric info of the patch. Primarily based on the label info of samples and therefore the redundance of the spatial and polarimetric info, a supervised tensor-primarily based DR technique, known as TLDE, is introduced to find three projections which project every pixel, that is, the third-order tensor into the low-dimensional feature. Finally, classification is completed based on the extracted options using the closest neighbor classifier and also the support vector machine classifier. The proposed methodology is evaluated on two real PolSAR data sets and therefore the simulated PolSAR knowledge sets with varied number of looks. The experimental results demonstrate that the proposed methodology not only improves the classification accuracy greatly however additionally alleviates the influence of speckle noise on classification.