Edge wave detection is a key technology in steel plate quality inspection, which is beneficial to improve the production quality of steel products and the efficiency of enterprises. In this letter, a novel algorithm based on visual projection and pixel statistics is proposed to detect the edge wave using 3D camera. The proposed algorithm firstly uses a series of image preprocessing to effectively remove the influence of irrelevant interference, and transforms the colour image into the binary image. Subsequently, three key positions of the edge wave in range image are precisely determined based on visual projection and pixel statistics. Finally, two 3D cameras are used to acquire the range images to evaluate the proposed algorithm's performance. The results show that the proposed algorithm can accurately detect the edge wave instead of human eye detection.
Affective video content analysis is an active topic in the field of affective computing. In general, affective video content can be depicted by feature vectors of multiple modalities, so it is important to effectively fuse information. In this work, a novel framework is designed to fuse information from multiple stages in a unified manner. In particular, a unified fusion layer is devised to combine output tensors from multiple stages of the proposed neural network. With the unified fusion layer, a bidirectional residual recurrent fusion block is devised to model the information of each modality. Moreover, the proposed method achieves state-of-the-art performances on two challenging datasets, i.e., the accuracy value on the VideoEmotion dataset is 55.8%, and the MSE values on the two domains of EIMT16 are 0.464 and 0.176 respectively. The code of UMFN is available at: https://github.com/yunyi9/UMFN.
A novel frequency-to-voltage converter (FVC) based phase-locked loop (PLL) is proposed to overcome the inability of an FVC-based frequency-locked loop (FLL) to lock phase. The proposed dual-loop PLL adds variable phase-locking capability, such that the phase locking angle can vary from 0o – 360o. The additional variable phase-locking can be applied in data communication in the form of phase modulation. The design is targeted for a 0.5-µm CMOS process. The proposed design generates a 480MHz clock from a reference clock of 15MHz. In simulation, the proposed PLL locks within 3.56 µs while consuming 1.61 mW of power.
The motion of targets is well known to result in their defocussing and displacement in SAR imagery, but detection of motion and re-focussing of targets under generic conditions remains of ongoing interest. One class of methods involves forming images of sub-apertures in which motion defocussing will be reduced. In this paper, we use dynamic tomographic image formation methods utilising an optical flow constraint to form a video of SAR sub-aperture images. These retain fine resolution of the full aperture, focussing along-track motion.
This paper presents an energy-efficient digital-to-analog converter (DAC) switching method with low common-mode variations for high resolution successive approximation register (SAR) analog-to-digital converters (ADCs), while enabling to implement resolutions such as 14-bit as compared to the typical 10-bit. The proposed switching method enables high resolution by having a nearly constant common-mode voltage and employing input-swapping to use the reference voltage (Vref) only in the sampling phase. This method eliminates the need for the third reference voltage during the entire DAC switching steps, which reduces the required number of switches even compared to the state-of-the-art methods that use low number of switches. The use of lower number of switches not only lowers the DAC control logic complexity, but also results in a faster operation, lower power, and smaller area. When compared to conventional 10-bit SAR ADCs, the proposed switching method in a 10-bit implementation reduces the average switching energy and area by 93.7 % and 75 %, respectively, while offering high resolution implementation options such as 14 bits.
A common 400V dc bus for industrial motor drives advantageously allows the use of high-performance 600V power semiconductor technology in the inverter drive converter stages and to lower the rated power of the supplying rectifier system. Ideally, this supplying rectifier system features unity power factor operation, bidirectional power flow and nominal power operation in the three-phase and the single-phase grid. This paper introduces a novel bidirectional universal single-/three-phase-input unity power factor differential ac-dc converter suitable for the above mentioned requirements: The basic operating principle and conduction states of the proposed topology are derived and discussed in detail. Then, the main power component voltage and current stresses are determined and simulation results in PLECS are provided. The concept is verified by means of experimental measurements conducted in both three-phase and single-phase operation with a 6kW prototype system employing a switching frequency of 100 kHz and 1200V SiC power semiconductors.
This paper proposes a joint optimization method for the imaging algorithm and sampling scheme of sparse spotlight syhthetic aperture radar (SAR) imaging based on deep convolutional neural networks. Traditional compressed sensing (CS) based sparse SAR imaging has been widely studied. Deep learning and sparse unfolding networks have been introduced into sparse SAR imaging, but most current works focus only on the imaging stage and simply adopt the conventional uniform or random down-sampling scheme. Considering that the imaging quality also depends on the sampling pattern besides the imaging algorithm, this paper introduces a learning-based strategy to jointly optimize the sampling scheme and the imaging network parameters of the reconstruction module. In a deep learning-based image reconstruction scheme, joint and continuous optimization of the sampling patterns and convolutional neural network parameters is achieved to improve the image quality. Simulation results based on real SAR image dataset illustrate the effectiveness and superiority of the proposed framework.
The fully passive noise shaping (NS) successive approximation register (SAR) analog-to-digital converters (ADCs) are simple, OTA-free and scaling friendly. Previous passive NS-SAR ADCs rely on the multi-path-input comparator or capacitors stacking to realize the passive gain for compensating the signal attenuation during passive integration. However, the former causes high comparator power consumption, and the latter suffers from additional signal attenuation due to the parasitics and is hard to extend to high-order systems. This work proposes a new fully passive NS-SAR technique, it can realize 2× gain with a simple structure, leading to the reduced comparator power and less parasitics. This technique is also easy to extend to high-order NS-SAR ADCs.
The power consumption of chips has emerged as a major concern with the increased integration of analog circuitry. This work focuses on a two-stage comparator based on a preamplifier with latch for successive approximation analog-to-digital converter. In order to minimize power loss and delay time, the charge steering approach was used in the design of latch as well as preamplifier. The suggested comparator is simulated in SMIC 0.18um process in comparison to the comparator without charge steering mode. The results reveal that the average power consumption is only around 22uW for varied input voltage at a supply voltage of 1.2V, which is relatively lowered by approximately 30%. Meanwhile, delay time is also reduced by about 25%.
In this paper, we investigate the application of Hybrid Representation in Wide-Angle Synthetic Aperture Radar (WASAR) imaging, addressing the challenges of achieving sparse representation in the presence of complex electromagnetic scattering characteristics and highly anisotropic targets. We utilize a Convolutional Neural Network (CNN) to represent two-dimensional data within the same subaperture, while employing dictionary learning for sparse representation across different subapertures. Convolutional Neural Networks (CNNs) excel at learning spatial hierarchies and local dependencies in two-dimensional data, but require a large amount of training data. Isotropic targets within subapertures can be used for training with conventional SAR data, whereas anisotropic targets present challenges in obtaining training samples. To address this, a dictionary for different subapertures is generated from measurements using dictionary learning, eliminating the need for additional training data. By integrating these methods, we propose a novel approach, Hybrid-WASAR, which incorporates two regularization terms into WASAR imaging and employs the Alternating Direction Method of Multipliers (ADMM) to iteratively solve the imaging model. Compared to traditional WASAR imaging techniques, Hybrid-WASAR not only enhances the accuracy of the reconstructed target backscatter coefficients, but also effectively reduces sidelobes and noise, resulting in a significant improvement in overall imaging quality.
Event-based cameras are sensitive to brightness changes and can capture rich temporal information with very high temporal resolution, which has great potential for motion segmentation of moving objects. Under static background, events are only triggered by motion of objects, thereby moving objects can be easily segmented. However, in many real-world applications, events are also be triggered by the motion of camera or background and submerge the ones corresponding to moving objects. In this letter, we propose an event-based motion segmentation method to segment moving small objects in events obtained from the wild. First, motion estimation is performed to align the events triggered by the background. Then, candidate events corresponding to moving objects or moving backgrounds are detected. Finally, motion information is adopted to segment the events of moving small objects from the ones triggered by the background. In addition, we develop the first dataset for event-based motion segmentation of small objects, namely EMSS. Experimental results demonstrate the effectiveness of our method and show that our method can achieve robust motion segmentation of small moving objects in the wild.
We have demonstrated waveguide integration of terahertz quantum cascade lasers (THz QCLs) at frequencies above 4.7 THz. A precision micromachining technique, followed by diamond-turning and electroless-plating has been used to manufacture hollow rectangular waveguides with integrated diagonal feedhorns. We show that surface roughness at the 1μm level is achieved, enabling outcoupling of radiation in the 4.75–5.05 THz band, with a divergence angle of 5° along the plane of the QCL substrate.
A dual-polarized symmetrically cross-slotted square patch (SCSSP) antenna with multimode resonance is proposed for 5G millimeter-wave broadband applications. A symmetrical cross-slot is etched on the square patch surface to change the original E-field distribution where a whole square patch is cut into four disconnected equal parts. For both polarizations, this etching also tunes the resonance frequencies of the two modes to be sufficiently close to each other. Therefore, the antenna can realize good impedance performance within the wide designated bandwidth. Dual-polarized radiation of the SCSSP is vertically excited by three bow-tie-shaped slots and fed by two orthogonal substrate integrated coaxial lines. Simulated and measured results show that the fabricated prototype achieves a broad overlapped impedance bandwidth of 50.0% (24-40 GHz), isolation higher than 30 dB between the two input ports, stable radiation pattern, and low cross-polarization over the operating band. Moreover, the proposed SCSSP antenna with compact size, planar shape, and simple vertical feeding is very well-suited for two-dimensional array design.
Robots with telepresence capabilities are typically employed for tasks where human presence is not feasible due to geography, safety risks like fire or radiation exposure, or other factors like any epidemic disease. Time delay is a significant consideration in controlling a telepresence robot. This study proposes a deep learning-based approach to compensate for the delay by predicting the behaviour of the teleoperator. We integrate a recurrent neural network (RNN) based on the Long Short-Term Memory (LSTM) architecture with the reinforcement learning-based Deep Deterministic Policy Gradient (DDPG) algorithm. The proposed method predicts the teleoperator’s angular and linear controlling commands by using data gathered by embedded sensors on the specially designed and built telepresence robot. Simulations and experiments assess the operation of the proposed technique in Gazebo simulation and MATLAB with ROS integration, which shows 2.3% better response in the presence of static and dynamic obstacles.
Pavement distress classification is a vital step for automatic pavement inspection and maintenance. Recently, patch-based approaches have achieved promising performances and thus extensive attention in this field. However, these methods simply assume that all patches contribute equally to the distress classification, leading to weakly discriminating abilities of models. Moreover, their tedious processes also leads to a low efficiency in inference. In this letter, we present a novel patch-based pavement distress classification approach named Deep Patch Soft Selective Learning (DPS$^2$L), which addresses these issues. Similar to other patch-based approaches, DPS$^2$L partitions the pavement images into patches and aggregates the patch features to accomplish the task. To address the first issue, we introduce a succinct Soft Patch Feature Selection Network (SPFSN) to assess the importance of each patch to the distress classification with a score based on its feature. These scores will be considered as patch-wise weights for feature aggregation. In such a manner, the most discriminative patches are selected in a soft way, and thereby benefit the final classification. To address the inference efficiency issue, knowledge distillation is leveraged to transfer the classification knowledge from DPS$^2$L to the image-based approaches, such as EfficientNet-B3. This distilled model enables incorporating both the advantages of patch-based approaches in classification performance and the advantages of image-based approaches in inference efficiency. Extensive experiments on a large-scale pavement image dataset named CQU-BPDD demonstrates the superiority of our methods over baselines regardless of performance or efficiency.
Low-cost air pollutant sensors suffer several interferences due to the variation of climatic elements. Recent studies look for calibration solutions based on different regression and classification machine learning algorithms. The present work brings together the implementation and extraction of performance metrics from these algorithms in a single open-source tool. Both the input data and parameters for each algorithm are automatically configured. This feature makes the tool compatible with any input dataset and removes the need to interact with complex codes.
We present a low-power area-efficient subarray beamforming receiver (RX) structure for a miniaturized 3-D ultrasound imaging system. Given that the delay-and-sum (DAS) and digitization functions consume most of the area and power in the receiver, the beamforming successive approximation register (SAR) analog-to-digital converter (ADC) shares its capacitive digital-to-analog converter (CDAC) with the delay cells. As a result, the delay cells implemented with capacitors are embedded in the CDAC with significant area reduction, further eliminating the need for power-hungry ADC buffers. Furthermore, the dual reference 10-bit SAR ADC reduces the area of CDAC by 32 times, achieving a switching energy reduction of 98.3%, compared to the conventional SAR ADC. As a result, the proposed beamforming SAR ADC, simulated using a 0.18 μm CMOS process, consumes 230 μW per channel, significantly reducing the per channel capacitance.