We’ve got almost full-color night vision to work with

This site may earn authorized commissions from links on this page. Terms of use.

(Photo: Brown Lab, UC Irvine Ophthalmology Department)
Current night vision technology has its drawbacks: it is useful, but it is essentially monochromatic, which makes it difficult to accurately identify things and people. Fortunately, through deep learning night vision seems to be undergoing a change with full-color visibility.

Scientists at the University of California, Irvine have experimented with reconstructing colorful night vision using deep learning algorithms. The algorithm uses invisible infrared images with the naked eye; Humans can only see light waves from 400 nanometers (which we see as purple) to 700 nanometers (red), while infrared devices can see up to one millimeter. Infrared is therefore an essential element of night vision technology, as it allows people to “see” what we normally perceive as complete darkness.

Although used to color scenes captured on infrared before thermal imaging, it is not perfect. Thermal imaging uses a technique called pseudocolor to “map” each shade of color from a monochromatic scale, resulting in a helpful but highly unrealistic image. It does not solve the problem of identifying objects and persons in low or light conditions.

Paratroopers are conducting an operation in Iraq, as seen through a traditional night vision device. (Photo: Speci Lee Davis, US Army / Wikimedia Commons)

On the other hand, scientists at UC Irwin wanted to create a solution that would create an image similar to what a human being can see in visible spectral light. They used a monochromatic camera sensitive to visible and near-infrared light to capture color palettes and facial images. They then train a evolutionary neural network to predict visible spectral images using only near-infrared images provided. The training process produced three architectures: a baseline linear regression, a U-Net inspired CNN (UNet), and an augmented U-Net (UNet-GAN), each capable of producing about three images per second.

Once Neural Network created color images, a team of engineers, opticians, surgeons, computer scientists, and doctoral students – provided images to graders who chose which outputs most closely matched the ground truth image. This response has helped the team choose which neural network architecture is most effective, surpassing UNet-GAN without zoom-in status.

The team revealed them at UC Irvine The result In the journal PLOS ONE on Wednesday, they hoped that their technology could be applied to security, military operations and animal surveillance, although their skills tell them that it could be applied to reduce vision loss during eye surgery.

Read now:

Leave a Reply

Your email address will not be published.