An artificial intelligence (AI) framework capable of both quickly detecting damaged buildings in crisis zones and accurately estimating the size of bird flocks has been developed by a team of computer scientists at the University of Massachusetts Amherst.

The framework, dubbed DISCount, reportedly combines the speed and data-analyzing power of AI with the reliability of human analysis to immediately deliver reliable estimates that can quickly assess and count features from very large collections of images.

Source: University of Massachusetts AmherstSource: University of Massachusetts Amherst

To enable the computer vision tool to accurately count buildings damaged during events like earthquakes or wars, while simultaneously helping ornithologists to get accurate estimates of the size of bird flocks, DISCount uses AI to analyze very large data sets — for example, all the images taken of a particular region in a decade — to determine what specific smaller set of data a human researcher should examine.

This smaller data set might include, for instance, all the images from a handful of days that the computer vision model determined best capture the extent of building damage in that region. Meanwhile, a researcher can subsequently hand-count the damaged buildings from the smaller data set of images and the algorithm will use that data to determine the number of buildings affected across that region. Furthermore, DISCount will reportedly be able to estimate the accuracy of the human-derived estimate.

"DISCount works significantly better than random sampling for the tasks we considered," the researchers concluded. "And part of the beauty of our framework is that it is compatible with any computer-vision model, which lets the researcher select the best AI approach for their needs. Because it also gives a confidence interval, it gives researchers the ability to make informed judgments about how good their estimates are."

An article detailing the research,DISCount: Counting in Large Image Collections with Detector-Based Importance Sampling,” appears in the journal Proceedings of the AAAI Conference on Artificial Intelligence.

To contact the author of this article, email