This paper proposes an underwater image correction and enhancement algorithm based on generative adversarial networks. In this algorithm, the multi-scale kernel is applied to the improved residual module to construct a generator, which realizes the extraction and fusion of multiple receptive fields feature information. The discriminator design considers the relationship between global information and local details, and establishes a global-region dual discriminator structure, which can ensure the consistency of overall style and edge texture. An unsupervised loss function based on human visual sensory system is proposed. Reference image constraints are not required, and the confrontation loss and the content loss are jointly optimized to obtain better color and structure performance. Experimental evaluations on multiple data sets show that the proposed algorithm can better correct color deviation and contrast, protect details from loss, and is superior to typical algorithms in subjective and objective indexes.
In order to meet the different requirements of circuit targets in various scenarios, an accuracy configurable fast Fourier transform (FFT) processor based on the concept of approximate circuit is proposed. A configurable approximate butterfly unit which can truncate the carry chain is proposed at the butterfly node and a bit-width configurable multiplier is proposed at the rotation factor multiplication node. MATLAB is adopted to develop an error analysis platform. After analyzing the sensitivity of each butterfly node and rotation factor node to approximate calculations, five calculation modes of the accuracy configurable FFT processor are determined, which can achieve dynamic balance among performance, power consumption, and accuracy. Finally, based on the 180 nm complementary metal oxide semiconductor (CMOS) technology of Taiwan Semiconductor Manufacturing Company (TSMC), the proposed processor is implemented with the standard procedure of ultra-large-scale digital integrated circuits. The performance results are obtained by professional electronic design automation (EDA) tools. Compared with the precise mode, the maximum operating frequency of the processor in the approximate mode is increased by 14.33%, and the power consumption is reduced by 15.61% when the operating frequency is 60 MHz.
Aimed at the fact that the chirplet rate parameter of the chirplet transform (CT) cannot match the instantaneous frequency of the signal completely, and that the anti-noise performance of the algorithm is poor, this paper proposes a high-quality local maximum synchrosqueezing chirplet transform (LMSCT) algorithm to improve the deviation of energy diffusion amplitude in CT time-frequency (TF)distribution. The main idea of this algorithm is to reallocate CT frequency points by local maximum synchrosqueezing operation. The experiment results show that the LMSCT algorithm has a higher TF concentration and a strong ability to suppress the interference of noise. The method can maintain a better resolution of TF representation at a low signal-to-noise ratio. In the application analysis of IPIX processing radar signals, the LMSCT algorithm can clearly describe the TF joint distribution characteristic of target signal and determine the distance unit of target, which provides the judgement basis for small target detection of IPIX radar signal in the background of sea clutter.
This paper proposes an end-to-end method that combines deep learning and sensitivity analysis, which can perform gradient back propagation calculation sensitivity on the saved weight information while training the model. The structure and activation function of the depth model are specially designed to adapt to the subsequent sensitivity calculation. The experimental results conducted on a Boston house prices dataset, a track information fusion dataset, and the G function show that the proposed method is more accurate than classical methods such as Sobol’ method when the parameter distribution is uneven, and has a stronger robustness. Compared with the traditional neural network method, the accuracy of the proposed method is higher. The experiment proves that the sample parameter sensitivity obtained by the deep learning model can be used to optimize the model output.
Regarding the disadvantages of existing camera calibration methods, such as external information relative, special camera poses,the need for calibration targets, and complex operations, this paper proposes a real-time self-calibration method based on parallel laser ranging by employing high-precision laser rangefinders to synchronously measure the position of the measured object plane when taking pictures, so that the object plane equation can be solved. The 2D coordinates of at least four sets of corresponding points on object plane and image planes are selected to obtain the homography matrix, which represents the mapping relationship between object and image planes, so as to complete the calibration simply and quickly. A calibration device is developed to validate the accuracy of the proposed self-calibration method in different testing scenarios. The results show that the measurement error of line segments length in the image are between -0.49% and 0.15%, and the average errors are merely -0.14%, which indicates that the parallel laser ranging self-calibration method proposed in this paper is accurate and robust. The causes of measurement error are further investigated by analyzing the influences of laser ranging, laser inclination, and device offset. The error eliminating suggestions are provided to give references for the application of the proposed self-calibration method in the field of image measurement.
The compact storage system is a new storage technology in the field of intelligent logistics technology, and its most typical feature is that it can realize multi-depth storage of unit loads. In this paper, the path optimization problem of the co-existence of both single command (SC) and dual command (DC) operations of the stacker in the dual-port layout is studied, and the mathematical model of the problem is established with the shortest travel time of the stacker as the objective. A genetic algorithm (GA)-beam search (BS) hybrid optimization algorithm is designed to solve the model, and the optimal individual obtained by the GA is used as the initial path choice of the BS to avoid local optimum. The effectiveness of the model and algorithm is verified by numerical simulation, and the results show that the operation path optimization model of the stacker and the designed solution method can better adapt to the I/O tasks scheduling requirements in the compact three-dimensional warehouse with dual-port layout, get a more reasonable stacker scheduling scheme, and improve the storage efficiency.
Considering the state monitoring problem of flexible joint robots (FJRs) caused by the easy deformation in practice, a distributed receding horizon estimation algorithm based on disturbance blocks is proposed. Based on distributed consistent receding horizon estimation, the proposed algorithm reduces the computational amount and achieves rapidity by designing the disturbance block and applying it to the process disturbance sequence in the estimation window to reduce the variables related to optimization. By analyzing the feasibility and convergence of the proposed algorithm based on the maximum block length, the assumptions are made under which the existence of equivalent solution to the optimization problem of the algorithm is guaranteed, and the results are extended to the case that the process disturbance can be divided into arbitrary blocks. The simulation results show that the proposed algorithm can effectively shorten the computation time without affecting the estimation error compared with the algorithm without disturbance blocks.
In order to solve the problems of difficult feature extraction and low recognition rate of radar signal at low signal-to-noise ratios, a dual channel convolutional neural network model based on Choi-Williams distribution (CWD) and multisynchrosqueezing transform (MSST) is proposed, which obtains two-dimensional time-frequency images by CWD and MSST time-frequency analyses on radar signals. Respectively, the time-frequency images are preprocessed and sequencely fed to a dual channel convolutional neural network for deep feature extraction. Finally, the features acquired by the two channels are fused, and the radar signal is classified and recognized through the convolutional neural network classifier. The simulation results show that when the signal-to-noise ratio is -10 dB, the overall recognition accuracy can reach above 96%, which is excellent at low signal-to-noise ratios.
In order to resolve the conflicts between the communication traffic and the localization accuracy, a self-localization algorithm with adaptive and dynamic observation period for mobile underwater acoustic networks (MUANs) was proposed to improve the localization performance. First, an adaptive and dynamic observation period selection scheme was designed, which could generate a non-uniform observation period vector according to the residual change. Then, based on the non-uniform observation period vector, a self-localization algorithm was proposed, which could precisely predict the trajectory of each mobile node in the network. The simulation results show that the proposed algorithm, which could balance the tradeoff between the localization accuracy and the communication cost, is more suitable for the underwater environment.
Aimed at the problem that the existing K-means clustering (KMC) algorithm is greatly affected by initialization, and the randomly generated clustering center can easily make the clustering result fall into local optimum and stop iterating, resulting in low clustering accuracy and poor robustness, a K-means hybrid iterative clustering algorithm based on memory transfer sailfish optimization (MTSFO-HIKMC) is proposed. First, learning from the existing improvement ideas, the maximum and minimum distance product is introduced to initialize the KMC cluster center, to avoid the uncertainty caused by random initialization. At the same time, in the iterative process, the current optimal solution is made to locally perform adaptive memory transfer correction to solve the problem of poor global optimization ability and insufficient search accuracy caused by the single search path of the sailfish algorithm. Using the Iris, Seeds, CMC and Wine international standard data sets, the MTSFO-HIKMC, the sailfish optimized K-means hybrid iterative clustering (SFO-KMC) algorithm, the introduction of the improved Moth-to-fire K-means cross iterative clustering (IMFO-KMC) algorithm, the KMC algorithm, and the fuzzy C-means (FCM) algorithm are compared and tested. From the obtained convergence curves and performance indicators, it can be seen that the MTSFO-HIKMC algorithm proposed in this paper has a faster convergence speed than IMFO-KMC. Compared with the IMFO-KMC algorithm, the dimensional space has a higher search accuracy. Compared with the KMC algorithm and FCM, it has a higher search accuracy. Compared with the SFO-KMC algorithm, its convergence speed and search accuracy are significantly improved, especially in high-dimensional data sets.
Liquid crystal polymer (LCP) with excellent microwave and millimeter-wave properties is widely applied in high frequency multilayer substrates. The design of an excellent via-hole transition in multilayer circuits board is important to employ as an interconnect to route signal traces on different layers or connect components. Recently, with the operating frequency increasing, the problem of discontinuity in the via-hole transition structure has become increasingly prominent. Therefore, electromagnetic modeling of via-plate-pair structures is essential for the design of microwave and millimeter-wave circuits. In this paper, an efficient and fast via-hole transition modeling method for the ground coplanar waveguide-strip line-ground coplanar waveguide(GCPW-SL-GCPW) structure based on the four-layer LCP circuit board is proposed. By segmented modeling of the multilayer structure and introducing a fast convergence algorithm in the parasitic parameter calculation process, a via-hole lumped parameter equivalent circuit structure is established. Finally, the equivalent circuit model of the GCPW-SL-GCPW structure is quickly constructed based on the microwave network cascade method. Compared with the full-wave simulation results of HFSS high frequency structure simulator, which is a 3D high-frequency electromagnetic software, it is found that this modeling process is simple and fast. The GCPW-SL-GCPW circuit structure has been fabricated using the LCP multilayer process. The test results show that the test results and the equivalent circuit analysis results are highly consistent in the wide frequency range of 10 MHz—40 GHz, which verifies the effectiveness of the via-hole transition modeling method.
The 1S1R storage unit of 3D phase-change memory is composed of ovonic threshold switch selector (OTS) in series with the phase change memory (PCM) device. In order to solve the problems of the current OTS and PCM circuit simulation models, such as not able to accurately simulate the electrical and physical characteristics of devices, and not suitable for confined PCM, a 1S1R spice model based on Verilog-A is proposed. The model simulates the electrical characteristics of OTS and the changes of current, temperature, melting proportion, crystallization proportion and amorphous proportion in the crystallization, melting and quenching of the PCM. The model has a good convergence and fast simulation speed. The simulation results are consistent with the actual test results of the device. Compared with the traditional model, the simulation and integration of confined PCM melting process, crystal nonlinearity, melting resistivity stability and subthreshold nonlinearity, and bidirectional switching characteristics of OTS are realized. The relationship between OTS subthreshold nonlinear parameter and read voltage window is analyzed. It is found that the read window reaches its maximum when OTS threshold current is approximately equal to PCM threshold current. The results of DC simulation of 1S1R cell and transient simulation of array are displayed, providing the basis for circuit design and simulation of 3D phase-change memory.
Supercapacitors have the advantages of fast charging and discharging, high power density, and long life, which are widely used in energy storage systems for new energy vehicles. Reliable operation of the system requires the acquisition of its remaining electricity, which is to estimate its state of charge (SOC). Relying on the equivalent analog circuit model of a single supercapacitor, this paper establishes a state-space model of the capacitor second-order nonlinear system with the multi-capacitor terminal voltage in the model as the state, the capacitor input current as the control input, and the capacitor output voltage as the observation output, and contains the leakage current caused by the self-discharge phenomenon. In order to improve the simulation accuracy, different model parameters were identified to characterize the charging and discharging conditions. In this paper, a nonlinear observer algorithm is used to obtain the internal state of the model to realize the estimation of SOC. The results of the charging and discharging experiment show that considering the leakage factor and establishing the charging and discharging model with different parameters, the dynamic characteristics of the supercapacitor can be better simulated, and the nonlinear observer algorithm has a stable tracking ability.
The sources reconstruction method (SRM) is widely applied to antenna measurement, antenna diagnostics and imaging, etc. It is a near field to far field transform, which utilizes near-field data to determine an equivalent magnetic current source over a fictitious flat surface enclosing the antenna under test. But the existing SRM is low in accuracy, since it does not have a proper probe correction algorithm. First, this paper introduces an integral equation connecting the equivalent magnetic current and the signal received by the probe according to the reciprocity theorem. Secondly, the integral equation is converted to a matrix equation by the method of moments and the magnetic current is determined by solving the matrix equation. Finally, the magnetic current leads to the pattern of the antenna under test readily. As an example, a horn is measured with a wave-guide probe and the receiving signals are processed by the proposed method to obtain the pattern, whose root mean square errors (RMSE) are only 1/3 of those of the pattern obtained by the SRM without probe correction. Therefore, the proposed probe correction algorithm is accurate and effective for the SRM.
Under the few-shots condition caused by non-cooperative scenes, robust extraction of communication emitter features and accurate identification of targets are the difficulties and hotspots of current research. Aimed at the problem of emitter identification under the few-shots condition of orthogonal frequency division multiplexing (OFDM) signals, this paper proposes a non-cooperative target identification method based on phase/time domain flipping data augmentation and source domain instance-based transfer learning. The data set is expanded by different domain flipping data augmentation methods, and the improved residual network is applied to achieve the purpose of promoting the identification rate of the OFDM emitter. Then, transfer learning is introduced to strengthen the generalization ability of the identification model. The experimental results show that the data augmentation method can significantly improve the OFDM emitter identification rate under the few-shots condition. Furthermore, the transfer learning method accelerates the convergence speed, slightly increases the recognition rate, and improves robustness of the model.
Grammatical error correction (GEC) is a low-resource task, which requires annotations with high costs and is time consuming in training. In this paper, the MASS-GEC is proposed to solve this problem by transferring learning from a pre-trained language generation model, and masked sequence is proposed to sequence pre-training for language generation (MASS). In addition, specific preprocessing and postprocessing strategies are applied to improve the performance of the GEC system. Finally, this system is evaluated on two public datasets and a competitive performance is achieved compared with the state-of-the-art work with limited resources. Specifically, this system achieves 57.9 in terms of F0.5 score which emphasizes more on precision on the CoNLL2014 task. On the JFLEG task, the MASS-GEC achieves 59.1 in terms of GLEU score which measures the n-gram coincidence between the output of the model and the correct answer manually annotated. This paper provides a new perspective that the low-resource problem in GEC can be solved well by transferring the general language knowledge from the self-supervised pre-trained language model.
45 steel has the problems of low wear resistance and poor corrosion resistance. CoCrFeMnNiMox (x=0.00, 0.25, 0.50, 0.75, 1.00) high-entropy alloy coating was prepared on 45 steel by laser cladding. The influence of Mo on the microstructure and properties of coating were explored in detail. The results show that the CoCrFeMnNiMox high-entropy alloy coating is composed of a single face-centered cubic (FCC)solid-solution phase. The microstructure of the Mo-containing coating is a typical dendritic and interdendritic structure, which is caused by the heterogeneous nucleation of the molten pool during the solidification process. The microhardness of the coating increases with the increase of x, and the maximum microhardness of the Mo1.00 coating is 2.391 GPa. Quantitative calculations show that solution strengthening is the main reason for the increase of microhardness. With the increase of Mo mass fraction, the wear mechanism evolves from adhesive wear to abrasive wear and oxidative wear. The Mo1.00 coating has the lowest volume wear rate (0.68×10-4 mm3/(N·m)). The influence of the passivation process on the corrosion resistance of coating was analyzed based on the point defect model theory. The addition of the Mo element increases the dehydration rate of the passivation behavior of coating, which makes the oxide layer thicker, and thereby improving the corrosion resistance of coating. The corrosion mechanism of coatings is intergranular corrosion. Mo0.75 coating has the smallest self-corrosion current density and the most positive self-corrosion potential, which are 3.69×10-6 A/cm2 and -0.432 V, respectively.
