Case Studies

Deep Learning For Image Recognition

Customer Challenges

The mission of our customer was hampered by a common and growing challenge across the federal government community. The rate of growth in the volume and diversity of potentially valuable data is severely outpacing the capacity of the human analyst to harness, distill and make sense of it to inform decision making. In tandem, a significant time lapse persists between data collection and arriving at actionable intelligence, introducing further risks to the mission from an over-saturation of just-in-time information. These problems are particularly pronounced in the field of digital imagery. Our customer had a pressing need for advanced computer vision techniques capable of streamlining the analysis of large volumes of digital imagery to identify actionable intelligence in lock-step with the pace of the mission.

Solution Features

Our artificial intelligence specialists at ATG developed a neural networks model for automating the analysis of image and video data sets. We applied advanced deep learning techniques to the automation of facial recognition, object detection, anomaly detection, natural language processing, audio processing, and a wider range of image and video-related applications. We used open source tools and API’s, such as Google’s TensorFlow, carefully tailoring the configuration to our customer’s specific needs. The model was developed in a modular manner without any proprietary software, using industry best practice open standards, such as XML, and J2EE. This made integration within existing systems simpler, with the flexibility to fit within a build-low/deploy-high construct. Our engineers trained the system by introducing a baseline set of pre-vetted, unique data-sets provided by our customer, refining the system’s ability to work well with new data sets as they are introduced. Our team provided a maintenance plan and full system documentation to our customer, and enabled a continuous level of human interaction with the system to assess its performance, address anomalies, sustain its ability to flex with the mission and, ultimately, give confidence to the results it was producing.

Benefits to the Customer Mission

Our analytics approach to image recognition advanced our customer’s mission in a number of ways, including:

 

  • Shortening the time-to-value between data ingestion and actionable intelligence for very large data sets.
  • Enabling more efficient human interaction with the system to validate its results.

 

 

An Acacia Group Company