Implementation of a Convolutional Neural Network for Object Classification

Cristian Cuevas-Mora* and Marjorie Darrah
Department of Mathematics, West Virginia University, Morgantown, WV 26506

Presentation No.: 54

Assigned Category (Presentation Format): Engineering (Poster Presentations)

Student’s Major: Mathematics

A Light Detection and Ranging (LiDAR) sensor is capable of collecting 3D data that can be used for object classification. The LiDAR data analysis process starts with detecting objects in the 3D set and then uses a Convolutional Neural Network (CNN) to try to recognize (or classify) these objects. Our study describes the implementation of the VoxNet CNN algorithm and how it classifies target objects in a 3D set. The data collection was accomplished through the use of an unmanned aquatic vehicle with a 4D Tech Solutions RedTail LiDAR sensor, which was used to scan the seafloor to detect cables and other objects. To start the process we train the CNN on the seafloor data so it can classify several different objects. We want to be able to take a portion of the 3D LiDAR data and give a user the classification of the object within the data. The goal is to optimize this process to see how few data points are needed before a successful classification can be made.

Funding: Louis Stokes Alliances for Minority Participation, National Science Foundation

Program/mechanism supporting research/creative efforts: Other