We present a high-resolution terahertz (THz) reflective imaging system, operating at 2.52 THz, that employs a continuous-wave THz gas laser and a pyroelectric detector. The spatial resolution was evaluated from the system's modulation transfer function and tested by scanning a series of resolution targets. To further improve the image quality, Lucy-Richardson method was adopted to restore the scanning result. With the scanning spot profile measured using knife edge method, a satisfying restoration result can be obtained. Finally, the system's performance was observed by imaging some different test objects.

High-resolution terahertz reflective imaging and image restoration

We present a high-resolution (HR) reconstruction model and algorithms for terahertz imaging, taking advantage of super-resolution methodology and algorithms The algorithms used include projection onto a convex sets approach, iterative backprojection approach, Lucy–Richardson iteration, and 2D wavelet decomposition reconstruction Using the first two HR reconstruction methods, we successfully obtain HR terahertz images with improved definition and lower noise from four low-resolution (LR) 22×24 terahertz images taken from our homemade THz–TDS system at the same experimental conditions with 10 mm pixel Using the last two HR reconstruction methods, we transform one relatively LR terahertz image to a HR terahertz image with decreased noise This indicates potential application of HR reconstruction methods in terahertz imaging with pulsed and continuous wave terahertz sources

High-resolution reconstruction for terahertz imaging

Rapid 3D object detection and modeling using range data from 3D range imaging camera for heavy equipment operation

During the real-aperture-scanning imaging process, terahertz (THz) images are often plagued with the problem of low spatial resolution. Therefore, an accommodative super-resolution framework for THz images is proposed. Specifically, the 3D degradation model for the imaging system is firstly proposed by incorporating the focused THz beam distribution, which determines the relationship between the imaging range and the corresponding image restoration level. Secondly, an adjustable CNN is introduced to cope with this range dependent super-resolution problem. By simply tuning an interpolation parameter, the network can be adjusted to produce arbitrary restoration levels between the trained fixed levels without extra training. Finally, by selecting the appropriate interpolation coefficient according to the measured imaging range, each THz image can be coped with its matched network and reach the outstanding super-resolution effect. Both the simulated and real tested data, acquired by a 160 ∼ 220 GHz imager, have been used to demonstrate the superiority of our method.

Adaptive terahertz image super-resolution with adjustable convolutional neural network.

A new method of fiber-optic based beam shaping is investigated both numerically and experimentally. A cylindrically symmetric method of lines (MoLs) is developed to simulate the device. The device is fabricated by fusion splicing a predetermined length of multimode fiber (MMF) to a single-mode fiber. The multimode interference (MMI) effects create ring-shaped field profiles at certain positions inside the MMF. The shaped beam can be used in medical applications requiring particular irradiation patterns.

Fiber-optic beam shaper based on multimode interference.

Laser radar is able to produce both intensity images and range images. These two images provide the full 3-D information of the target. The quality of the range image is negatively impacted by noise arising from dropouts and range anomalies. This paper proposes an algorithm for range image background suppression, utilizing the intensity image, a mathematical morphology (MM) filter and a mean filter. An order-mean-filter algorithm is applied to suppress the range anomaly. A comparison of the filter effects between various templates and templates in weighted mode was made, resulting in the selection of the order-mean-filter algorithm of the best multiple template. It was applied to the range image formed by the target in the actual radar.

Range image noise suppression in laser imaging system

To convert a given single-mode He–Ne laser beam into a ring-shaped intensity distribution, a diffractive optical element with 16 levels was designed by YG amplitude-phase retrieval and iteration algorithm. This beam shaper is investigated experimentally and compared with the results of the computer design. The results show good conformity, and the measured diffraction efficiency is 87.2%. In addition, the diffractive effect is observed when the distance between diffractive optical element and CCD is changed.

Investigation of diffractive optical element for shaping a Gaussian beam into a ring-shaped pattern

Resolution of continuous-wave CW terahertz scanning im- age is limited by many factors among which the aperture effect of finite focus diameter is very important. We have investigated the factors that affect terahertz THz image resolution in details through theory analysis and simulation. On the other hand, in order to enhance THz image res- olution, Richardson-Lucy algorithm has been introduced as a promising approach to improve image details. By analyzing the imaging theory, it is proposed that intensity distribution function of actual THz laser focal spot can be approximatively used as point spread function PSF in the res- toration algorithm. The focal spot image could be obtained by applying the pyroelectric camera, and mean filtering result of the focal spot image is used as the PSF. Simulation and experiment show that the algorithm implemented is comparatively effective. © 2010 Society of Photo-Optical Instru-

Continuous-wave terahertz scanning image resolution analysis and restoration

The range image has received considerable attention in the automatic target recognition field; however, a mass of range images are generally difficult to be collected in a real application. Therefore, with small samples of laser radar (ladar) range images, classifier ensembles-support vector machine (SVM) ensembles and back propagation neural networks (BPNN) ensembles-are applied to improve the performance of target recognition in this paper. There are three aspects of experiments. First, the performances of SVM and BPNN are compared with different numbers of training sets. Secondly, the SVM ensembles and the BPNN ensembles are applied to improve the performance. Thirdly, the performances of the SVM ensembles and the BPNN ensembles are analyzed with the view angle of the tested samples changing while the view angle of the trained samples is invariant. The experimental results demonstrate that the recognition rate is effectively improved by classifier ensembles. The SVM is superior to the BPNN when the number of the training sets is not less than the feature dimensions; however, the BPNN has a better approximating ability when the number of training sets is small and lower than feature dimensions.

Qi Li

Papers

Range image noise suppression in laser imaging system

Investigation of diffractive optical element for shaping a Gaussian beam into a ring-shaped pattern

Continuous-wave terahertz scanning image resolution analysis and restoration

Target recognition for small samples of ladar range image using classifier ensembles