Vishal Monga, Penn State professor of electrical engineering, received a four-year, $1.6 million grant from the Strategic Environmental Research and Development Program for his work on imaging and classifying underwater military munitions. Credit: Kelby Hochreither/Penn State
Detecting and remediating underwater munitions with AI: A Q&A with Vishal Monga
March 13, 2024
By Sarah Small
UNIVERSITY PARK, Pa. — A team of researchers led by Vishal Monga, Penn State professor of electrical engineering, received a four-year, $1.6 million grant from the Strategic Environmental Research and Development Program for their work on imaging and classifying underwater military munitions. The team’s research focuses on physics-inspired artificial intelligence models for analysis of 3D underwater sonar imagery to remove harmful munitions and other contaminants from lakes, oceans and similar waterbodies, according to Monga.
Penn State News spoke with Monga about the research and its potential applications.
Q: What types of munitions are commonly found in waterbodies? How are they harmful?
A: Underwater sites may contain a variety of munitions types, including bombs, projectiles, mortars, grenades and rockets. Munitions ranging in size from 20-millimeter projectiles to 2,000-pound bombs can be distributed on the surface or buried at these sites, showing no visible evidence of their presence.
Such munitions in the sea represent a worldwide threat for humans and marine inhabitants because they pose two kinds of danger. The first is that explosive ordnance can still detonate — for example, when mines are disturbed by bottom-fishing nets or when construction work for wind farms is initiated without prior examination of the sea floor for old munitions. The explosion of a sea mine would trigger a shock wave that would kill all of the marine creatures in the close vicinity and shred the blood vessels and alveoli of whales, seals and other marine mammals over a greater distance.
The second danger to the environment is chemical in nature. Both explosive and chemical weapons contain various components that are increasingly leaking into the water due to the ongoing decay of the metal munition coverings, eventually dissolving in the water and being distributed throughout the ocean by marine currents. Humans that come in contact with such waterbodies are at continual risk.
Q: What is currently done to detect underwater munitions? What do you hope to achieve with this grant?
A: Diver intervention and investigation — using a highly experienced trained dive team with diver-operated metal detection apparatus, camera and video facility for recognition and identification purposes — has been the traditional approach for underwater munition detection and removal. The above, however, is expensive and poses significant danger to the divers engaged.
Our overarching goal is to automatically detect, classify and remediate unexploded ordinance (UXO) and other contaminants in the underwater environment. We will accomplish this by the analysis of underwater 3D sonar imagery with sensors especially designed for munition detection. New Artificial Intelligence (AI) methods will be developed in the work that will incorporate underwater and acoustic phenomenology in the analysis of this sonar imagery to automatically detect and identify munition or contaminant type. Such high accuracy automated approaches should cut down cost substantially for munition remediation and crucially reduce the involvement of human participants.
Q: What are some of the potential “real-world” applications currently for this research?
A: One of the most obvious applications — with commercialization efforts in motion — is the development of automated technology tools for munition detection, recognition and remediation in underwater environments. The economic footprint of such potential technology could be massive. For instance, the U.S. Department of Defense estimates that 15 million acres at approximately 1,500 sites in the United States are UXO contaminated. Countries worldwide face a similar challenge. Substantial money, time and resources – both physical and human – are spent in munition detection with grave dangers to involved maritime workers, such as divers. There is potential to deploy AI based munition detection and recognition tools worldwide paving the way for significant cost savings and a safer world.
While many munitions contaminated sites are a legacy of major historical wars — World War II, for example, is notorious in this regard — this problem is not going away but getting worse. Military munition training and testing areas have deliberately been situated in water environments for over a hundred years and wars such as the Russia-Ukraine and Israel-Hamas conflicts only exacerbate the problem.
Q: How will the research be conducted? What does this project aim to do that has not been done before?
A: The research will be conducted by a team of researchers from multiple institutions across the US and involves a unique collaboration between sonar sensor developers situated at the University of Washington (UW), led by consultant with senior expertise Tim Marston, and the Penn State Applied Research Lab (ARL), including co-principal investigator Dave Williams and consultant with senior expertise Dan Brown. Image data processing and analysis experts will be led by myself at Penn State but also involve co-principal investigator Suren Jayasuriya of Arizona State University (ASU). My research team along with colleagues at ASU will analyze sensor data provided by UW and Penn State-ARL to both form high quality underwater sonar imagery and subsequently analyze it to answer the questions of whether munitions are present in the said imagery and recognize its type for remediation purposes.
While the use of sonar sensors has become popular for this problem in the past decade, an Achilles heel has been the difficulty in high quality image formation and analysis, especially in demanding environmental conditions where munitions may be obscured, and/or easily confused with natural rocks and other materials. For the first time, we are developing phenomenology enriched AI approaches that should help overcome this long standing technical challenge. In a departure from black-box machine learning, which shows little awareness of problem domain, we aim for a comprehensive integration of the physical nature of the underwater domain with the modeling capacity of modern AI methods to develop AI-based munition detection and recognition that can be reliably deployed.
Q: Is there anything unique to Penn State that enables your research or helps you pursue this project?
A: Penn State ARL researchers were amongst the earliest to develop sophisticated underwater sonar sensors, and an interaction with them drew me to the problem. My group at Penn State has a decade-long history of computational image processing and analysis. I saw the opportunity to merge these complementary strengths and worked to identify other key players in the area — at UW and ASU namely — to form a team that led to a winning proposal.
Share this story:
