The optical images quality in dark underwater background is usually degraded, influencing the results of accurate identification, terrain mapping and seabed exploration. Thus, the super-resolution methods for dark underwater optical images attract extensive research interest. In this work, the SwinIR model is provided for dark underwater optical images super-resolution. Here, the EUVP DARK dataset with dark underwater background is employed, which clusters 5500 paired images. The SwinIR networks has high speed of process with large-size image capability and plentiful image detail. Compared with the traditional SRGAN method, the super-resolution reconstruction speed increases about 18.3%, and the peak pignal-to-noise ratio (PSNR) and structural similarity (SSIM) of SwinIR test results on the EUVP DARK dataset increase by 14.4% and 15.8% respectively, the results illustrate that our method improving the accuracy and quality of reconstruction. In conclusion, the hierarchical structure of SwinIR and self-attention mechanism adaptive attention weights in generated to the image, enabling precise adjustment and control of detail and texture. This method provides an efficient approach to dark underwater image quality enhancement.
Infrared and visible image fusion technology aims to integrate the radiation information of infrared images and the details of visible images into one image, which is widely used in video surveillance, target tracking etc.. Thereinto, maritime surveillance often uses infrared and visible images to monitor ships. With the rapid development of deep learning technology, numerous image fusion algorithms have sprung up. However, maritime images can be affected by extreme illumination conditions during fused. Note that the quality of maritime images is extremely significant for identifying ship targets. Therefore, the quality of fused image is necessary for further investigation. Here, a progressive image fusion network with a lighting awareness module, named PIAFusion, is provided for maritime images fusion. Contributed by the light sensing subnetwork, PIAFusion can adaptively fuse the common information and complementary information of images. Evaluated with five operators in daytime and night cases, the results demonstrate that this method significantly improve the fused image quality, compared to common fusion methods (i.e., GTF and DenseFuse). In addition, this method preserves more details of the source image. The progressive fusion network illustrates better fusion performance for maritime images in different lighting cases.
KEYWORDS: Modulation, Digital signal processing, Optical communications, Telecommunications, Tolerancing, Signal to noise ratio, Fiber lasers, Chromium, Optical networks, Optical amplifiers
A modulation format identification (MFI) method based on the nonlinear power transformation via logistics regression is adopted for coherent optical receives systems. The amplitude variance, fourth power transformation and fast Fourier transform of input signals are utilized for special features extraction in our work. Five typical optical modulation formats (i.e.,16/32/64QAM and Q/8PSK) with the transmission rate of 28 GBaud are numerically simulated to demonstrate the feasibility. The simulation results show that our method has great performance even under low optical signal noise ratio (OSNR). Compared with the MFI algorithm based on Stokes space and asynchronous delay tapped sampling, our MFI algorithm requires less time to achieve similar performance of optical receive systems. Especially, this method exhibits tolerances to the laser linewidth and nonlinearity.
Curve fitting algorithm is traditionally utilized for Brillouin optical time domain analyzer (BOTDA) temperature extraction, with high time cost. To improve the extraction speed, general regression neural network (GRNN) is introduced into BOTDA temperature extraction. As a feed-forward network, only one parameter is needed to be learned in GRNN and its structure is self-adaptive for various samples with different scale, making it easier to train. The performance of GRNN is investigated in simulation and experiment different signal-to-noise ratios, pump pulse widths, and frequency scanning steps. The results show that, with the similar or better accuracy, GRNN achieves faster processing speed which is 7 times over curve fitting methods. The fast processing speed, and high extraction accuracy make GRNN approach a potential way of real-time BOTDA temperature extraction.
An improved CenterNet is proposed for signal recognition with time-frequency image input. The signal is transformed into time-frequency image by short-time Fourier transform, hence, the signal recognition is transformed into investigating the object detection problem in the field of image detection. Then, the advanced achievements of image detection are adopted to enhance the performance of signal recognition. Here, an improved CenterNet-based object detection network, which demonstrates great advantages in detection speed, is proposed. The results show that the proposed method identifies the signal modulation format with high speed. After training and testing on the self-collected data set with 6 types and 7800 samples, the mean average precision achieves 98.38% and the frame per second reaches 21.4. Compared with the original CenterNet, the detection speed increases more than 4 times while only reducing recognition accuracy by 0.3%, where this algorithm gives a promising way for applications of real-time signal recognition.
Image segmentation is the most fundamental part of computer vision, which is the foundation of all other methods of image processing. The quality of image segmentation technology will affect the subsequent processing considerably. Comparing with traditional image segmentation algorithms, image segmentation algorithm based on deep learning is constantly proposed, with high performance and efficiency. But there is also a lot of room for improvement. For example, key parts such as fastening bolt are usually small in size, polluted and covered, and do not have enough characteristic information, so it is difficult to obtain satisfactory results. These factors affect the accuracy of the test, which is easy to cause serious accidents. As traditional methods sometimes cannot meet the requirement of high-accuracy result, deep learning play a particularly important role in facing those problems. To solve the problem that traditional object recognition methods are not robust enough to extract image features, parts recognition accuracy is low, and segmentation is not possible, we have made some modifications based on Mask R-CNN. In this method, convolutional neural network is used to extract features from part images. Then we use some annotated images from dataset to fine-tuned Mask R-CNN network to guarantee the accuracy. At the same time, data enhancement and k-folding cross-validation are carried out to improve the robustness of the model. Finally, the result of part recognition and segmentation by building the experimental platform proves the significance of the method.
Recently, convolution neural network (CNN) has been widely used in single image super-resolution (SR). However, the traditional network structure has the problems of fewer convolution layers and slow convergence speed. In this paper, an image super-resolution method based on deep residual network is proposed. Through the deepening of the network structure, more receptive fields are obtained. Thus, more pixel information is utilized to improve the reconstruction accuracy of the model. The feature extraction process is carried out directly in low resolution space, and the images are sampled by shuffling the pixels at the end of the network. The learning method combining global residual and local residual is used to improve the convergence speed of the network while recovering the high-frequency details of the images. In order to make full use of image feature information, feature maps extracted from different residual blocks are fused. In addition, parametric rectified linear unit (PReLU) is used as the activation function, and the Adam optimization method is used to further improve the reconstruction effect. The experimental results of benchmark datasets show that the proposed method is superior to other methods in subjective visual effects and objective evaluation indicators.
The detection of rail surface defects is of great significance for railway safety. To detect the rail surface defect, the laserinduced ultrasonic rail propagation model is established by the finite element method. The intrinsic relationship between the defect depth, of the defect on rail surface and the acoustic surface wave is investigated by discussing the variation of the reflected wave and the transmitted wave both in the time and frequency domain, respectively. Quantitative evaluation of defect depth is given based on the energy of the reflected and transmitted wave, which providing a promising theoretical way for the estimation of the rail surface defect feature.
A two-pump fiber optical parametric amplifier (FOPA) based on the photonic crystal fiber (PCF) with As2S3 background in the mid-infrared (MIR) region is investigated numerically. The genetic algorithms are used to optimize the gain of the FOPA, and the amplifier peak gain, bandwidth and flatness are investigated in detail for the variety of the fiber length, the input signal and pump power. In addition, the comparison of the gain spectra between considering and neglecting the loss of the PCF is given. The results show that the wideband gain spectra with high peak gain can be obtained by using a short length of the PCF with the relatively low pump power, which show the great potential of the FOPA at wavelengths in the MIR region.
Atmospheric anisoplanatic effect is an important problem to be solved in telescope observation of space target imaging. Numerical simulation of atmospheric anisoplanatic imaging is the basis for studying the restoration of anisoplanatic images. Based on the propagation theory of light waves on the inhomogeneous turbulent path and multilayer phase screens distribution model, this paper establishes a theoretical model of atmospheric imaging for space targets under anisoplanatic conditions. The near-surface atmosphere can be divided into several stratifications of atmosphere at different altitudes. Find out the best phase screen distribution location for each atmospheric stratification, and use the multilayer phase screens at different altitudes to represent the atmospheric anisoplanatic effect. The phase change of the light wave emitted by each point on the space object through the atmosphere is represented by a phase screen, and the final phase size is the superposition of the phase of the light wave passing through the phase screens of each layer. A series of spatial target images are simulated by different layers of phase screens for anisoplanatic imaging, and combined with theoretical analysis to find the best phase screen position and the number of layers. The experimental results show that the three-layer phase screen can accurately simulate the atmospheric anisoplanatic imaging while maintaining the computational efficiency, and effectively reflect the changes of the point spread function (PSF) when the spatial position changes. The imaging results have no ringing and edge effects, and can accurately represent the influence of atmospheric anisoplanatic effect on atmospheric imaging.
During the long-time working circle, the wheels will be damaged to a certain degree caused by the wearing, the impact, the loads, the climate and so on. In order to evaluate the health of the wheels and reduce the potential losses, many effective methods are used in railway health monitoring, such as laser method or ultrasonic method. But few of them can reach the demand of the real-time online detection, and integrate more comprehensive inspection function at the same time. A composite detection scheme for wheel-tread defects based on FBG sensing technique has been investigated in this paper. By collecting and analyzing the data from the sensors which are distributed on tracks and rails, we can precisely evaluate the Wheel-flats and also measure some other parameters used in rail health monitoring scheme such as speed, loads and axle counting measurement.
Bandwidth gain characteristics of two-pump fiber optical parametric amplifier (FOPA) with photonic crystal fiber (PCF) are discussed. The pump depletion is considered in. Broader 62dB gain can be obtained in the range from 1370nm to 1790nm.The effects of dispersion characteristics of the PCF on the gain bandwidth are analyzed. And the influences of signal and pump powers are also taken into account. From the analysis that by properly selecting the PCF and setting the pump and signal powers, wider gain bandwidth can be achieved, which is useful for the dense wavelength diversion multiplexing (DWDM)system.
The gain optimization characteristics of a one-pump fiber-optical parametric amplifier with an optical band-pass filter (OBPF) reducing the idler wave is investigated numerically. The phase mismatching is compensated, thus both the maximum gain and the gain bandwidth are optimized with the OBPF inserted in, and the flatter gain can be obtained; meanwhile, the conversion efficiency of the pump power to the signal power is increased. The influence of filter attenuation on the gain is also discussed. It is shown that, by properly selecting the parameters of the fiber, signal wave, pump wave, and filter, the optimized gain can be obtained, which is extremely useful for the optical communication systems.
All-optical wavelength conversion is one of the key technologies in all optical signal processing. A wavelength
conversion technology based on semiconductor optical amplifier nonlinear polarization rotation effect is proposed. By
theoretical analysis and simulation, we achieve a 320Gbit/s signal wavelength conversion. We show that inverted and
non-inverted wavelength conversion can be realized. However, the ER and Q-factor are improved.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.