Abstract:The surface reflectance of the Tibetan Plateau is exploited in numerous applications, such as natural resource monitoring, ecological environmental protection, and geoscience research. Typically, the reflectance data of MOD09A1 are affected by detector noise and clouds, producing numerous abnormal pixels and diminishing the integrity and accuracy of remote sensing data. To address these issues, considering the universal geoscience law indicating that neighboring time-series remote sensing images are correlative, and the spectra of adjacent ground objects belonging to the same classification are similar, this paper proposes a deep learning method of surface reflectance reconstruction in the Tibetan Plateau based on incomplete multi-temporal data and land cover classification information. First, based on the multi-temporal reflectance data of MOD09A1 and land cover classification data of MCD12Q1, the basic reflectance image and auxiliary data of the target area are obtained through abnormal pixel removal, effective layer extraction, projection conversion, and mosaic. Subsequently, a deep learning network model is constructed based on the fusion of multi-temporal data and land cover classification information, according to basic principles of the residual network. Third, the deep learning model is trained using cloud mask samples cropped from an area with complete data and augmented training samples generated based on land cover classification and the K-means clustering algorithm. Finally, the trained model is utilized for surface reflectance reconstruction in the area with missing data. Two groups of comparative experiments demonstrate that the proposed method reduces the requirements for the amount and integrity of multi-temporal auxiliary image data and achieves accurate restoration and reconstruction of large-scale surface reflectance in the Tibetan Plateau by combining incomplete multi-temporal data and land cover classification information.
Keywords:surface reflectance;Tibetan plateau;deep learning;MODIS data;reconstruction of missing data
Abstract:High-sensitivity superconducting transition edge sensor (TES) detectors have significant application prospects in the detection of the B-mode polarization of cosmic microwave background (CMB). In this study, we developed a 220 GHz antenna-coupled titanium (Ti) superconducting TES detector array (8×8) using a leg supporting structure. Before the supporting legs were fabricated, we measured the thermal conductance and the noise equivalent power (NEP) of the Ti superconducting TES detector. The measurement results show that the thermal conductance of the Ti superconducting TES detector, primarily determined by the electron-phonon interaction in the superconducting Ti microbridge, is approximately 485.4 pW/K. Moreover, the NEP of the Ti superconducting TES detector is approximately 5×10-17 W/Hz0.5. In principle, the thermal conductance of the Ti superconducting TES detector can be further reduced after its supporting legs are fabricated, which is approximately 38 pW/K. Meanwhile, the thermal fluctuation noise of the detector after its supporting legs are fabricated is approximately 9.2×10-18 W/Hz0.5. Thus, the noise performance can be further improved.
Abstract:In response to the difficulties of processing laser point cloud off-site real-time scanning and on-site absolute coordinate system alignment, a low-cost ground-based LiDAR measurement system is developed by combining the BeiDou/GNSS positioning system and 5G communication technology. The developed measurement system is integrated by LiDAR, a high-precision motor, BeiDou/GNSS receiver module, and a 5G module. LiDAR collects point cloud information, the high-precision motor acquires angle information, BeiDou/GNSS receiver module receives the time, the high-precision motor acquires time stamps using a self-developed time synchronization module, and LiDAR obtains the point cloud and angle file with time tag before transmitting in real-time via 5G communication technology. The terminal performs spatial and temporal alignment of multi-frame point clouds by linear interpolation algorithm through self-developed data pre-processing software, coarse alignment of the multi-station laser point clouds based on BeiDou/GNSS coordinates outdoors, coarse alignment of the multi-station point clouds by single-station feature points in an indoor environment without BeiDou/GNSS, fine alignment using the proximity iteration algorithm to complete overall alignment, and visualization on the self-developed real-time point cloud management and visualization system. The experimental results show that the measurement system achieves real-time off-site scanning and transmission at a transmission rate of 50 Mbit/s by using 5G communication technology. The error of the aligned point cloud is less than 3 mm, which can provide digital infrastructure for real-time applications such as digital twin, monitoring of material and cultural heritage, analysis of construction and operation, and maintenance of large buildings.
Abstract:To effectively suppress laser speckle and truly achieve a large color gamut, high brightness, and excellent picture quality of laser display and imaging systems, a speckle suppression method based on tracked motion flexible diffractive optical elements (DOEs) is proposed and demonstrated. The design, manufacture, durability, and speckle suppression effect of the flexible DOE loop are studied. First, the optical structure design of the flexible DOE is presented, and the mechanism of dynamic two-dimensional diffraction encoding by the tracked moving flexible DOE loop is described. Next, the preparation of the working stamper, hot pressing, and ring-joining processes of the flexible DOE loop are introduced. A laser projection system was set up to test speckle contrast and durability of flexible DOEs composed of polyvinyl chloride (PVC) and polypropylene (PP). The experimental results show that the proposed laser speckle suppression method based on a flexible DOE loop with tracked motion has the advantages of lower speckle contrast, smaller size, and lower power consumption. The speckle contrast can be reduced to less than 5% for red, green, and blue lasers. During the 30 day experiments, the PVC and PP flexible DOE loops did not have evident deformation and attenuation of the speckle suppression effect. Furthermore, the PP flexible DOE loop had a more apparent and improved durability compared with the PVC flexible DOE loop. The proposed method satisfies the requirements of large color gamut, high brightness, and excellent picture quality and has the advantages of miniaturization, low power consumption, and high reliability.
Keywords:laser speckle suppression;Diffractive Optical Elements (DOE);durability;speckle contrast;Polypropylene (PP)
Abstract:To realize the sensing function of flexible bionic robots, sensors need to be integrated and flexible; otherwise, it is difficult to achieve a good application effect owing to the lack of compliance. To obtain a flexible tactile force/strain sensing fiber based on liquid conductive metal, this study considers as an application carrier a flexible bionic finger that can completely fit the carrier to measure the tactile force of the fingertip and joint angle. Specifically, liquid conductive metal is injected into a prefabricated silicone hose having a certain length instead of conventional microfluidics. This enables the formation of flexible sensing fibers that are tubular and which can be arbitrarily deformed and arranged, thus realizing the detection of different deformation physical quantities (this study utilizes its force and strain sensing properties). This method can considerably reduce the complexity of the microfluidic channel process while ensuring the flexibility and functionality of the device. Experimental results show that the flexible sensing fiber based on liquid conductive metal can be embedded and fully fit the flexible finger structure, and can simultaneously realize the sensing of the tactile force and joint angle changes of the finger as well as accurate tracking at a specific force/angle, demonstrating its role as a flexible sensing unit. The proposed method has excellent potential for application to more types of flexible or software application carriers.
Keywords:tactile force perception;joint angle perception;flexible humanoid fingers;liquid conducting metal
Abstract:This paper proposes a model predictive control (MPC) method for permanent magnet synchronous motors (PMSMs) based on finite control set Gaussian process MPC (FCS-GPMPC) parameter identification to limit the influence of model mismatches on the control system and to improve the current controller performance of control systems in a PMSM. First, the current PMSM prediction model is introduced and the influence of model parameter mismatches on the system performance is analyzed. Secondly, in order to simplify the complex debugging process of hyperparameters in general machine learning parameter identification algorithms, the GPMPC method is proposed. At the same time, the confidence interval of the predicted value is introduced as a real-time evaluation reference for the parameter prediction effect. Finally, the GP parameter identification method is combined with the FCS-MPC to predict the system current after accurately obtaining the identified parameters. The model is updated to improve system robustness and current loop tracking performance. The experimental results show that under the statistical characteristics of the training data, the root mean square error and of the test data are 0.0021 and 0.99, respectively. Under the condition of parameter fluctuation, compared with FCS-MPC, FCS-GPMPC reduces current fluctuation by 30.5% and the average current offset by 19.6%. In addition, for step changes in the reference current, FCS-GPMPC has a better dynamic response. The proposed GP-MPC can effectively suppress the influence of model mismatch on control systems and can improve the performance of the current controller of PMSM control systems.
Keywords:Permanent Magnet Synchronous Motor (PMSM);Model Predictive Control (MPC);machine learning;Gaussian Process (GP);model mismatch
Abstract:To better understand the method for determining the optimal measurement area of a computerized numerical control (CNC) machine tool in-machine measurement system with different measurement objects and universal application, a ball is selected as the measurement object, and the working principle and error sources of the in-machine measurement system are analyzed. A single error-whitening model is established using a dual-frequency laser interferometer and a beetle antenna search algorithm backpropagation (BAS-BP) neural network measurement. Simultaneously, the comprehensive error model of the measurement system and the spherical measurement error model are established. The differential evolution cuckoo search (DE-CS) algorithm for determining the optimal measurement area search is studied. The search performances of different algorithms are compared, and the optimal performance parameters of the algorithm are determined. An experimental device is set up to determine the optimal measurement area of the sphere, and the corresponding experiments are carried out. Finally, the consistency of the optimal measurement position determined by the algorithm and the actual measurement is compared. The experimental results show that the spherical optimal measurement area obtained by the aforementioned method is consistent with that obtained experimentally, and can be used to determine the optimal measurement area of other measurement objects, such as points, planes, and so on. The proposed system is demonstrated to be universally applicable and capable of determining the optimal measurement area of machine measurement systems.
Abstract:To address the problems of large registration errors and poor adaptability of the traditional iterative closest point (ICP) algorithm when point clouds overlap or partially overlap, an improved registration algorithm based on weighted optimization of matching point pairs is proposed. First, an improved voxel downsampling algorithm is proposed to sample point clouds, which reduces the amount of data and improves the robustness of the algorithm against noise. Then, the improved Sigmoid function is used to assign different weights to the matching point pairs participating in the registration, which overcomes the disadvantage of traditional algorithms that ignore matching point pairs with small distances still have wrong point pairs, while improves the registration accuracy and convergence speed. Finally, a method to solve registration parameters using singular value decomposition (SVD) is proposed to further improve registration accuracy. The registration and noise experiments with different overlapping degrees were performed, and the proposed algorithm was verified by combining the three-dimensional point cloud reconstruction of the crankshaft. The experimental results showed that, compared with the Tr-ICP and AA-ICP algorithms, the error in the proposed algorithm was reduced by approximately 34.1% and 29%, respectively. Further, the registration time was shortened by approximately 16.1% compared with the Tr-ICP algorithm. Hence, compared with traditional algorithms, the proposed algorithm has higher registration accuracy, better applicability, and robustness.
Keywords:point cloud registration;iterative closest point algorithm;Tr-ICP;matching point pairs optimization
Abstract:The compressive sensing (CS)-based space optical remote-sensing (SORS) imaging system can simultaneously perform sampling and compression by using hardware at the sensing stage. The system must reconstruct the original scene during the ship detection task. The scene reconstruction process of CS is computationally expensive, memory intensive, and time-consuming. This paper proposes an algorithm named compressive sensing and improved you only look once (CS-IM-YOLO) for direct ship detection based on measurements obtained by the imaging system. To simulate the block compression sampling process of the imaging system, the convolution measurement layer with the same stride and convolution kernel size is used to perform the convolution operation on the scene, and the high-dimensional image signal is projected into the low-dimensional space to obtain the full-image CS measurements. After obtaining the measurements of the scene, the proposed ship detection network extracts the coordinates of the ship from the measurements. The squeeze-and-excitation Network (SENet) module is imported into the backbone network, and the improved backbone network is used to extract the ship feature information using the measurements. The feature pyramid network is used to enhance feature extraction while fusing the feature information of the shallow, middle, and deep layers, and then to complete predicting the ship's coordinates. CS-IM-YOLO especially connects the convolutional measurement layer and the CS based ship detection network for end-to-end training; this considerably simplifies the preprocessing process. We present an evaluation of the performance of the algorithm by using the HRSC2016 dataset. The experimental results show that the precision of CS-IM-YOLO for detection of ships via CS measurements in SORS scenes is 91.60%, the recall is 87.59%, the F1 value is 0.90, and the AP value is 94.13%. This demonstrates that the algorithm can perform accurate ship detection using the CS measurements of SORS scenes.
Keywords:ship detection oriented to compressive sensing measurements;compressive sensing;deep learning;joint training optimization
Abstract:Point cloud registration technology is the core technology of point cloud data processing. The quality of the point cloud will influence the registration effect of point cloud registration. An excellent quality point cloud can improve registration accuracy, spatial integrity, and slam performance. Therefore, assessing the quality of point cloud data has significant objective value. The point cloud data obtained by the sensor contains noise such as systematic and nonsystematic errors. In this case, point cloud data processing becomes crucial. However, there is no more objective method to evaluate the treatment effect. A quantitative evaluation method of multidimensional point cloud structure similarity is proposed. This method compares the point cloud data before and after filtering with the standard data. The mean, standard deviation, and covariance of all point coordinates on the three-dimensional coordinate axis are compared. Subsequently, the structural similarity values on the three coordinate axes are weighted. Finally, the similarity and correlation degree of the three-dimensional structure is obtained. Then, it realizes the evaluation of point cloud filtering, point cloud sparse, and point cloud data quality. The method is also verified by experiments to improve registration accuracy. Experiments demonstrate its capacity to evaluate the quality of the 3D point cloud. It can evaluate the quality of the point cloud obtained under different noise types and processing methods. It provides a reference for point cloud registration. This method improves both the accuracy and efficiency of cloud point registration, as well as its quality.
Abstract:The target detection algorithm based on the convolutional neural network is developing rapidly, and with the increase in computational complexity, requirements for device performance and power consumption are increasing. To enable the target detection algorithm to be deployed on embedded devices, this study proposes a Yolo v3-SPP target detection system based on the ZYNQ platform by using a hardware and software co-design approach and hardware acceleration of the algorithm through FPGA. The system is deployed on the XCZU15EG chip, and the required power consumption, hardware resources, and performance of the system are analyzed. The network model to be deployed is first optimized and trained on the Pascal VOC 2007 dataset, and finally, the trained model is quantified and compiled using the Vitis AI tool to make it suitable for deployment on the ZYNQ platform. To select the best configuration scheme, the impact of each configuration on hardware resources and system performance is explored. The system power consumption (W), detection speed (FPS), mean value of average precision (mAP) for each category, output error, etc. are also analyzed. The experimental results show that the detection speed is 38.44 FPS and 177 FPS for Yolo V3-SPP and Yolo V3-Tiny network structures, respectively, with mAPs of 80.35% and 68.55%, on-chip power consumption of 21.583 W, and board power consumption of 23.02 W at 300 M clock frequency and input image size of (416,416). This shows that the proposed target detection system meets the requirements of embedded devices for deploying neural network models with low power consumption, real-time, and high detection accuracy.
Abstract:To solve the problems of high redundancy of behavior feature extraction and inaccurate localization of behavior boundary of R-C3D, an improved behavior detection network (RS-STCBD) based on residual shrinkage and spatio-temporal context is proposed. First, the residual shrinkage structure and soft threshold operation are integrated into the residual module of 3D-ResNet, and a unit of 3D residual shrinkage with channel-adaptive soft thresholds (3D-RSST) is designed. Moreover, multiple 3D-RSSTs are cascaded to construct a feature extraction network to adaptively eliminate redundant information such as noise and background in behavioral features. Second, instead of single convolution, multi-layer convolutions are embedded into the proposed subnet to increase the temporal dimension receptive field of the temporal proposal fragments. Finally, a non-local attention mechanism is introduced into the behavior classification subnet to obtain the spatio-temporal context information of behavior by capturing remote dependencies among high-quality behavior proposals. Experimental results on THUMOS14 and ActivityNet1.2 datasets show that the mAP@0.5 values of the improved network reach 36.9% and 41.6%, which are 8.0% and 14.8% higher than those of R-C3D, respectively. The behavior detection method based on the improved network, which increases the accuracy of behavior boundary localization and behavior classification, is beneficial and enhances the quality of human-robot interaction in natural scenes.