MIT Researchers Develop Deep Learning Method to Recuperate Objects in Low Light

Artificial Intelligence News

MIT_Researchers_Develop_Deep_Learning_Method_to_Recover_Objects_in_Low_Light MIT Researchers Develop Deep Learning Method to Recuperate Objects in Low LightA team of engineers at Massachusetts Institute of Technology (MIT) has developed a new imaging technique that illustrates Deep Neural Networks (DNNs), can be utilized to irradiate transparent features like biological tissues and cells in images taken with very diminutive light. The researchers used a DNN to recreate transparent objects from images of the objects held in near complete darkness.

To commence, the researchers discussed a database of 10,000 integrated circuits, each etched with a different pattern. Rather than etching each of the 10,000 patterns over as many glass slides, the researchers utilized a state spatial light modulator to show the pattern on a glass slide, recreating the same optical effect that an actual etched slide would have. For this, the researchers pointed a camera at an aluminum frame carrying the light modulator. Then they used the device to reproduce each of the 10,000 patterns from the database. The researchers wrapped the complete experiment so it was protected from light, and practiced the light modulator to quickly spin through each pattern, akin to a slide carousel. After that, they took images of each transparent pattern, in very low lighting conditions that produced salt-and-pepper images that resembled little more than static on a television screen.

After training the neural network on the 10,000 images of a variety of interracial circuit patterns, the team built a totally new pattern. They took an image of this pattern, again in darkness, filled it into the neural network, and analyzed the patterns that the neural network reconstructed, both with and without the physical law embedded in the network. The researchers found that both methods reconstructed the original transparent pattern fairly well, but the physics-informed reconstruction produced a sharper, more precise image.