Prev-Talk | Next-Talk | All-Talks | Talks-Sorted
Talk-16: Using the MOOS-IvP with Machine Learning and AUV Behaviors for Target Classification Based on Acoustic Scattered Fields
Erin Fischell, Laboratory for Autonomous Marine Sensing Systems, Massachusetts Institute of Technology (MIT)
One challenge for AUV autonomy is the onboard classification of underwater targets in mine countermeasures type applications. A method for machine learning classification of underwater target geometry using the bistatic acoustic scattered fields has been developed and demonstrated using simulation. This method consists of two parts: model training/analysis and target classification. While the Support Vector Machine (SVM) model training is conducted offline, target classification must carried out in real time. This classification process includes signal processing, machine learning, path planning and vehicle behavior elements that have been implemented in the LAMSS MOOS-IvP simulation environment in preparation for use in the field. Once a target has been localized using the target tracker, a path planner charts a path through the critical waypoints determined by model analysis. The vehicle behavior utilizes that path and a broadside behavior to guide the vehicle through the important parts of the scattered field. The target amplitude is calculated and fed into a classifier that maps it to the feature space with existing data and returns the classification and confidence. Once sufficient confidence is achieved the vehicle returns to a search mode or moves onto classification of the next target.
Categories:
- Autonomous Underwater Vehicles (AUVs)
- MOOS-IvP
- Machine Learning
- Mine Countermeasures