Sonar Classifier Research

For my senior thesis, I did a study on the effectiveness using a Machine Learning algorithm to classify acoustic sonar data. For the classifier's training data, I generated synthetic imagery from MATLAB using a large database of handwritten digits. However, test data would have tocome from physical models.

I needed to scan these physical models using an aboveground synthetic sonar system. In order to get the proper test models, I took the original data set of handwritten digits and converted them into 3D surface using MATLAB. Then, I converted these plots into STL files which could then be 3D printed and scanned in the sonar system. Below are some images of letter models used in the setup and their corresponding sonar responses, each using a different filtering method.

The full paper can be accessed here.