researchsquare.comSource publication:
Loading content…
Infrared light is invisible to the eye, and conventional IR cameras rely on narrow-bandgap detector arrays plus cooling, making them costly and bulky. SIVIS sends IR and visible beams through one shared path, modulates both with a digital micromirror device, and uses the IR intensity read by a single-pixel single-photon detector as feedback so a genetic algorithm evolves the illumination pattern toward the target shape, while a matching visible pattern is projected onto the target or a screen. Static and moving targets are visualized in real time in both transmission and reflection, remaining discernible down to about 0.11 photons per pixel per detection (about 0.1 pW/cm²), with demonstrations in IR anti-counterfeiting and imaging through roughly 2 mm of biological tissue.
An infrared scene can be seen directly as a visible pattern by the naked eye or an ordinary silicon camera, without first reconstructing an infrared image. Earlier infrared-to-visible upconversion relied either on nonlinear crystals with high-power pulsed lasers, on molecular systems needing deoxygenated solvents, or on rare-earth nanoparticles with low quantum yields, and generally required higher excitation power densities. Static targets and targets translating at 0.1 mm/s were visualized in real time in both transmission and reflection, with the cost function and contrast-to-noise ratio rising across generations and peaking in the dynamic case when the target fully occupied the field of view.
The system uses the single-pixel detector reading to drive the evolution of the illumination pattern, which converges onto the target outline. Single-pixel imaging normally uses pre-designed illumination patterns and then computationally reconstructs the image from the measured intensity sequence. The digital micromirror device runs at a 10 kHz refresh rate, giving about 5 ms per generation with 10 patterns, or roughly 200 generations per second; the genetic algorithm uses a population of 10, a cost function equal to the ratio of detector intensity to the pattern pixel sum, and a mutation rate decaying exponentially from 0.01 to 0.0025.
Infrared and visible light share one optical path and are modulated by the same device, so the projected visible pattern stays geometrically synchronized with the infrared target. Conventional upconversion schemes separate detection from display, forming an image first and projecting it afterward. An 852 nm infrared beam and a 532 nm visible beam are coaxially combined by a long-pass dichroic mirror and co-modulated by the digital micromirror device; the visible light is diverted to a display screen by a beam splitter, while the infrared signal passes an 850/50 nm band-pass filter into the single-photon detector. Projection magnification is about 23×, and the illuminated field of view is about 10×10 mm², corresponding to a 32×32 pixel pattern.
The demonstration covers transmission and reflection configurations, static targets and targets translating at 0.1 mm/s, an illuminated field of view of about 10×10 mm² with 32×32 pixel patterns, and about 23× magnification. On the application side it shows positive and negative dual-mode visualization of infrared anti-counterfeiting ink patterns, vascular structures in a hand phantom under about 1.5 mm of artificial skin, and a target under about 2 mm of chicken tissue. The system currently uses dual-wavelength active structured illumination at 852 nm and 532 nm, and the authors note it can also be adapted to a passive configuration.
The authors state that the current limitation lies mainly in the optimization efficiency of the conventional genetic algorithm, which takes about 0.5 s to go from random speckle to a recognizable shape; future work could use faster optimization algorithms such as gradient descent or reinforcement learning and raise the data transfer bandwidth between the computer and the digital micromirror device. In the tissue penetration demonstration, captured images show some blurring compared with non-scattering targets, though the vascular structures remain distinguishable to the naked eye.