Department of Computer Science and Electrical Engineering
The aim of the thesis was to assess the performance of a mathematical technique known as Principal Components Analysis (PCA) for classifying low resolution radar images. The study presents the development and final results of a classification system based on the PCA mathematical method. The system is used to identify low-resolution radar ship images as one of four possible ship classes that include Container Ship, Bulk Carrier, Tugboat and Speedboat.
PCA is a technique used to reduce the dimensionality of a data set in which there are a large number of interrelated variables, while retaining as much as possible of the variation present in the data set. This is achieved by representing a group of data by a low dimensional feature space based on the eigenvectors corresponding to the eigenvalues, of the covariance matrix representative of the data. Only those eigenvectors corresponding to the largest eigenvalues (that is, those that contain most of the information about the data set) are retained.
To investigate PCA as a classifying technique three different types of test were examined. The first test analyzed the system’s performance when using large data sets. Smaller data sets were used in the second test type. The final test type examined the outcome when test images were closely related to the training data.
Results from all test types proved that the classification system accurately identifies test images with related training sets. This is a reflection of PCA ability to effectively summarize and differentiate data by the feature space.
It is concluded that the PCA technique employed in the system is a good
and efficient tool for classificying low resolution radar images.
Complete conference paper:
conference_paper.pdf 
Complete thesis:
thesis.pdf
Additional materials:
Training
Sets Grouping Matlab Program
Classification
System Matlab Programs
Please e-mail me any
comments or queries you have. 