Dynamic infrared scene projection is a technology for converting infrared (IR) digital image sequences into IR radiating image sequences. One of the most promising technologies is light down-conversion technology based on photoinduced opaque (PIO) effect. A parametric end-to-end steady-state model was proposed to describe the PIO mechanism. It consisted of three submodels such as the silicon parameter model, the photoinduced free carrier model, and the constitutive relation model. Furthermore, the parametric full-link from pump photons’ power density, photoinduced free carrier concentration, complex dielectric constant, and complex refractive index to the emissivity was constructed and mathematically analyzed by the above model. In order to verify the model, an intrinsic silicon wafer was pumped by a continuous-wave running Nd:YAG laser when the wafer was heated up to 321, 423, 459, and 498 K, respectively. Correspondingly, the emissivity integrated from 3 to 5 μm, with the increase of the pump power density measured by an IR camera. The measurement result agreed well with the theoretical result computed by the parametric end-to-end steady-state model. The maximum apparent temperature of the region illuminated by the laser with a pump power density of 407 W cm − 2 is up to 453 K when the silicon wafer was kept at 498 K. At the same time, the background temperature was elevated to 330 K owing to the initial free-carrier absorption enhancement.
A novel infrared small target detection algorithm based on potential regions proposal is proposed in this paper. Potential regions mean subsets (size are 16 by 16 in this paper) with small targets of an infrared image. A convolution neural network (CNN) classifier has been trained by using constructed datasets to discriminate potential regions of an input image. Traditional methods such as tophat transform, max-mean and max-median filter are used to suppress the background and noise of potential regions. Some experiments are carried out to verify the algorithm performance, and the results show that the gains of signal noise ratio and contrast ratio have better performance than traditional methods.