|Institution:||Wright State University|
|Keywords:||Electrical Engineering; vibrometry; classification; accelerometers; target recognition|
|Full text PDF:||http://rave.ohiolink.edu/etdc/view?acc_num=wright1421104791|
Laser vibrometry provides a method to identify running vehicles’ unique signatures using non-contact measurements. A vehicle’s engine, size, materials, shape, and other attributes affect its vibration signature. To develop the capability to classify and identify these signatures, a robust aided target recognition (AiTR) end-to-end process is evaluated and expanded. The main challenge in classifying a vehicle’s vibration signatures is presented by the operating conditions and parameters that vary as a function of sensor, environment, and collection locations on the target, among others. Some of the parameters affecting the vibration signatures include weather, terrain, sensor location, sensor type, and engine speed. Another challenge in vehicle classification is the determination of signal features that can overcome the differences created by these varying operating conditions. The end-to-end process consists of signal preprocessing, feature extraction, feature selection, classification, and identification. A total of 11 features from automatic speech recognition, seismology, and structural analysis and previously utilized in vibration exploration were used in this end-to-end process. Features were selected by two feature selection methods to determine the best feature set for vehicle classification. Finally, four classifiers were used to identify the vehicles’ signatures. Confusion matrices were used as metrics to evaluate the effectiveness of the end-to-end process. The entire process was tested on two sets of data: a military vehicle collection using accelerometers and a civilian vehicle collection using a laser vibrometer and accelerometers.