Taking aerial footage of zones of interest was once a challenge: resources were limited, and the quality of footage often left much to be desired. Now that full-motion video has become commonplace, a new problem has arisen: there is so much data that geospatial intelligence analysts cannot meet the demand with traditional methods. CACI is addressing the need for quick-turn, reliable analysis by using an advanced artificial intelligence software technique known as deep learning.
Deep learning involves training computers to recognize patterns in data via trial and error. Since the number of full-motion video images is extraordinarily high, it is difficult to label them all in a way that helps analysts quickly locate and track objects of interest. By using clustering techniques to score initial outputs in vast sets of images, our artificial intelligence experts have found a way to jump-start the process of labeling and greatly speed the intelligence cycle. Analysts are presented with a scored output that they can further refine, rather than having to sift through the initial sets of thousands of images. As analysts review and add their own input, the networks are continually retrained, thus providing more accurate output over time.
As the neural networks take over mundane labeling, human analysts have more time to fully address complex analysis. The CACI system process is data-type agnostic, so there is no need for a one-off workflow when a new type of data is introduced. The result is high quality analysis that gives our customers the results they need, when they need them.