Norwegian People’s Aid Cuts Ukraine Landmine Survey Times by 90% Using AWS and Safe Pro AI Image Analysis

With the scale of the humanitarian crisis in Ukraine, NPA needed to act immediately. It established its first office in the country in mid-2022 to support the clearance of areas contaminated by mines and ERW. Its primary mission in Ukraine is to protect civilians from the threat of explosive devices and minimize the loss of life. Because of the ongoing situation, Ukraine—one of the world’s leading producers of grain, including wheat, corn, and sunflower meal—has seen its agricultural production fall, threatening its economy and impacting global food prices. Agricultural production has been impeded as a result of the contamination of farmland by mines and ERW.

Safe Pro AI estimates that it would take over 120 years and billions of dollars to clear all visible mines and ERW from Ukraine using traditional methods. By speeding up the NTS process with AI and drone-based image analysis, NPA could potentially reduce clearance time to a few years at a fraction of the cost. This is the essential first step in landmine clearance, where potentially mined areas are examined to identify, assess, and collect information to determine if landmines or ERW are present or not.

The NTS process involves traveling to locations to interview residents and government officials to understand whether areas could have been mined or contain ERW. The suspected areas are then surveyed from a distance using binoculars or drones. Using the SpotlightAI platform, the organization can create high resolution, detailed maps and use GPS satellite navigation data to pinpoint where landmines and ERW have been detected.

The huge volume of drone imagery required to survey areas for suspected landmines and ERW in Ukraine makes employing trained human analysts to manually process each captured image unfeasible. The cloud-based SpotlightAI platform, built on AWS, uses machine learning, deep learning, and applied computer vision to rapidly process images and generate reports. Safe Pro AI was able to accelerate development of its platform after working closely with AWS technical support to reduce image-processing times while improving scalability, system performance, reliability, and security.

When users upload a survey or a series of drone images, the platform uses Amazon Elastic Container Service (Amazon ECS) to orchestrate image and data processing tasks in Amazon Elastic Compute Cloud (Amazon EC2) and AWS Fargate. This reduces the need to manage servers and allows for updates to be caried out without incurring downtime issues. To manage workflow and to control costs, the process is integrated with AWS Batch, which provides batch processing for machine learning (ML) model inference. By using a just-in-time compute services for user image uploads, Safe Pro AI can avoid the need to pay for always-on instances and can easily start, stop, and manage multiple tasks.

To store all image and map data, SpotlightAI uses Amazon Simple Storage Service (Amazon S3) as the platform’s common storage repository. When a user uploads image or survey data, a batch process is initiated with all outputs automatically stored in the same repository. Because access to the Amazon S3 bucket is fast—and with all images stored in one central place—storage costs and process times can be kept to a minimum.

NPA can conduct each NTS survey with SpotlightAI, scan multiple hectares of land, capture a series of drone images, then upload them to AWS for processing by the Safe Pro AI software platform. The resulting analysis produces sub-centimeter, high-resolution maps with GPS-tagged geospatial information showing landmine and ERW locations. These are then incorporated into NPA databases for use in its geographic information system (GIS).

Using SpotlightAI and AWS, NPA has cut the time it takes to conduct the surveys by almost 90 percent compared with manual image analysis because of a much faster, autonomous image review and analysis process. NPA can then use the reports generated to inform land release operations.

The SpotlightAI platform can also create large, high-resolution orthomosaic maps, created by stitching together thousands of individual images into a single depiction of a survey area. “We can now understand the situation on the ground much more quickly,” says Le Xuan Hoang, GIS manager at NPA. “While we are still at a very early stage, we estimate that non-technical survey analysis costs can be cut by more than 80 percent.” The data generated by SpotlightAI can also be easily shared and used by other stakeholders including government and local officials, to help increase landmine situational awareness.

Safe Pro AI uses several AWS services, including Amazon Inspector and Amazon GuardDuty, to ensure that the underlying infrastructure for SpotlightAI is secure and protected against any vulnerabilities. To track any unintended configuration changes that also might compromise the infrastructure, SpotlightAI uses a combination of Amazon CloudWatch to monitor resources and applications and AWS CloudTrail to track user activity and API usage.

The company uses Amazon Inspector to scan for vulnerabilities throughout the infrastructure, especially on Amazon EC2 instances. More importantly, it is helpful in automatically scanning the Amazon Elastic Container Registry (Amazon ECR) container images for vulnerabilities, outdated dependency packages and other problematic packages. Alerts setup in CloudTrail and CloudWatch track configuration changes that may be unintended allows Safe Pro AI to keep on top of anything that could compromise the infrastructure.

SpotlightAI has also helped NPA cut analysis time from 3–4 minutes per image to one-fifth of a second by using Amazon EC2 G4 Instances. “The image analysis process is intensive and requires highly trained human analysts to review tens of thousands of drone images,” says Hoang. “We’ve had great, on-the-ground support from Safe Pro AI, as well as training for our operatives to make sure everything works the way we need it to.”