| Statistical Programs |
| College of Agriculture | University of Idaho |
| Seminar Announcement |
| "Applied Statistics in Agriculture" |
|
Determination of Container Fill Types by Statistical
Classification Analysis of Acoustic Resonance Signatures
Presented By |
| Dr. Larry G. Blackwood |
|
Idaho National Engineering and Environmental Laboratory |
| Tuesday, December 2 3:30 P. M. Ag. Science 62 |
|
Researchers at the Idaho National Engineering and Environmental Laboratory
have developed a noncontacting, nondestructive inspection system for
identifying fill characteristics for various types of containers. The
inspection system, based on using a laser vibrometer to measure the
response of containers to acoustic excitation, is capable of detecting
subtle changes in container vibration characteristics due to variations in
the physical properties of different fill materials. Potential areas of
application of the technology range from the food industry (e.g.,
identification of spoiled contents of food containers) to the military
(e.g., verification of contents of unexploded artillery shells).
Determination of fill types based on the laser-acoustic system data is accomplished by employing statistical classification analysis. . We have investigated various statistical classification methods for suitability to the these data. Since each measurement with the laser vibrometer produces an acoustical spectra comprised of several thousand data points, "feature selection" (i.e., the choice of variables for the analysis) is an important step in developing a classification procedure. This talk describes a simple, graphics-based, method of feature selection developed for the acoustic spectrum data. The method is demonstrated on a set of test data in which acoustical signatures were obtained for empty, liquid, water, and juice filled drinking containers. A k-nearest neighbor classification tree was developed for discriminating between the four fill types. The feature selection process was used to determine appropriate variables for analysis at each branch in the classification tree. Performance of the classification method was tested over time and compared to other competing classification methods. All interested faculty, staff, and graduate students are invited to attend. |
| Return to Seminar Listing. |