Low Computational Amount Research Articles

High-resolution processing and sigmoid fusion modules for efficient detection of small objects in an embedded system

Recent advances in deep learning realized accurate, robust detection of various types of objects including pedestrians on the road, defect regions in the manufacturing process, human organs in medical images, and dangerous materials passing through the airport checkpoint. Specifically, small object detection implemented as an embedded system is gaining increasing attention for autonomous vehicles, drone reconnaissance, and microscopic imagery. In this paper, we present a light-weight small object detection model using two plug-in modules: (1) high-resolution processing module (HRPM ) and (2) sigmoid fusion module (SFM). The HRPM efficiently learns multi-scale features of small objects using a significantly reduced computational cost, and the SFM alleviates mis-classification errors due to spatial noise by adjusting weights on the lost small object information. Combination of HRPM and SFM significantly improved the detection accuracy with a low amount of computation. Compared with the original YOLOX-s model, the proposed model takes a two-times higher-resolution input image for higher mean average precision (mAP) using 57% model parameters and 71% computation in Gflops. The proposed model was tested using real drone reconnaissance images, and provided significant improvement in detecting small vehicles.

Optimization design method of upper limb exoskeleton cam mechanism’s motion trajectory model

In order to solve the problem caused by the coupling clearance of the upper-limb cam mechanism, based on cam entity structure’s data and motion trajectory characteristics, with the cam axle acceleration αc and the output torque accuracy σT as the optimization targets, a mathematical model of cam mechanism is established in the study. Cam mechanisms was established with the radial basis function (RBF) method to restore the physical model of upper-limb exoskeleton’s motion trajectory model accurately. Then, global optimization was performed with adaptive non-dominated sorting genetic algorithm-II (NSGA-II) to get αc and σT s’ Pareto solution sets and form a data mapping cam mechanism with the dynamic feedback between the motion trajectory model and the entity model. In the verification experiments, the motion trajectory model established with only 37 response points only required a low amount of computation, and the response point verification error was within 3%, indicating the high restoration accuracy. After physical parameters were optimized with NSGA-II, the optimal pareto solution sets of dynamic characteristic parameters of the cam mechanism were obtained. αc and σT were respectively decreased by 52% and 81%. The deviation of αc between the entity model test data and simulation data was 3.84°/s2 and σT was 30.23 N·mm, indicating that cam motion deviation could be adjusted by dynamic data feedback. This study verifies that the dynamic characteristics of the cam mechanism can be optimized by optimizing the material parameters of the rollers, thereby improving the performance of the upper limb exoskeleton.

Integral Sliding Mode-Based Model Predictive Current Control With Low Computational Amount for Three-Level Neutral-Point-Clamped Inverter-Fed PMSM Drives

This article proposes a novel integral sliding mode (ISM)-based model predictive current control (MPCC) strategy with low computational amount for three-level neutral-point-clamped (3L-NPC) inverter-fed PMSM drives. For conventional MPCC for 3L-NPC inverter, 27 voltage vectors are evaluated for prediction and calculation in each control cycle, therefore computational burden is a major hurdle for practical implementation of MPCC. With only calculation of reference voltage and determination of its sector position, a revised MPCC strategy with neutral point potential (NPP) balancing is proposed, which not only dramatically reduces the number of candidate voltage vectors involved in cost function calculation but also effectively suppresses NPP fluctuation of 3L-NPC inverter. To enhance MPCC robustness against motor parameter variations, the integral sliding mode (ISM) is embedded into MPCC for PMSM drives by enforcing sliding mode throughout the entire system response, and thus a novel ISM-based MPCC strategy is put forward. Furthermore, its robust stability is analyzed. The results of simulation experiments confirm that the proposed ISM-based MPCC strategy with NPP balancing for 3L-NPC inverter can enable PMSM drive system to have satisfactory control performance and strong robustness against motor parameter variations.

Automated Unsupervised 3D Tool-Path Generation Using Stacked 2D Image Processing Technique

Tool-path, feed-rate, and depth-of-cut of a tool determine the machining time, tool wear, power consumption, and realization costs. Before the commissioning and production, a preliminary phase of failure-mode identification and effect analysis allows for selecting the optimal machining parameters for cutting, which, in turn, reduces machinery faults, production errors and, ultimately, decreases costs. For this, scalable high-precision path generation algorithms requiring a low amount of computation might be advisable. The present work provides such a simplified scalable computationally low-intensive technique for tool-path generation. From a three dimensional (3D) digital model, the presented algorithm extracts multiple two dimensional (2D) layers. Depending on the required resolution, each layer is converted to a spatial image, and an algebraic analytic closed-form solution provides a geometrical tool path in Cartesian coordinates. The produced tool paths are stacked after processing all object layers. Finally, the generated tool path is translated into a machine code using a G-code generator algorithm. The introduced technique was implemented and simulated using MATLAB® pseudocode with a G-code interpreter and a simulator. The results showed that the proposed technique produced an automated unsupervised reliable tool-path-generator algorithm and reduced tool wear and costs, by allowing the selection of the tool depth-of-cut as an input.

Soft Set Based Parameter Value Reduction for Decision Making Application

The soft set theory is a completely new mathematical tool for modeling vagueness and uncertainty, which can be applied to decision making. However, in the process of making decision, there are some unnecessary and superfluous information which should be reduced. Normal parameter reduction is a good way to reduce superfluous information, which keeps the entire decision ability. However, the algorithm has a low redundant degree, which involves a great amount of computation. It is not certain that normal parameter reduction has the solution, that is, it has a low success rate of finding reduction. Parameterization value reduction is another reduction method, which improves redundant degree, amount of computation, and success rate of finding reduction. However, this method only considers the best choice, but it does not concern the suboptimal choice, the sequence of choice, and added parameter, that is, it loses some part of decision ability. In order to settle these problems, in this paper, we introduce the parameter value reduction which keeps the entire decision ability while having a very high redundant degree and success rate of finding reduction and low amount of computation. Maximal parameter value reduction is defined as the special cases of parameter value reduction and the related heuristic algorithms are presented, which reaches an extreme degree to reduce the redundant information. The comparison result among maximal parameter value reduction, parameterization value reduction, and normal parameter reduction on 30 datasets shows that the proposed algorithm outperforms parameterization value reduction and normal parameter reduction.

Aluminum agglomeration involving the second mergence of agglomerates on the solid propellants burning surface: Experiments and modeling

The agglomeration of aluminum particles usually occurs on the burning surface of aluminized composite propellants. It leads to low propellant combustion efficiency and high two-phase flow losses. To reach a thorough understanding of aluminum agglomeration behaviors, agglomeration processes, and particles size distribution of Al/AP/RDX/GAP propellants were studied by using a cinephotomicrography experimental technique, under 5MPa. Accumulation, aggregation, and agglomeration phenomena of aluminum particles have been inspected, as well as the flame asymmetry of burning agglomerates. Results reveals that the dependency of the mean and the maximum agglomeration diameter to the burning rate and the virgin aluminum size have the same trend. A second-time mergence of multiple agglomerates on the burning surface is unveiled. Two typical modes of second mergence are concluded, based upon vertical and level movement of agglomerates, respectively. The latter mode is found to be dominant and sometimes a combination of the two modes may occur. A new model of aluminum agglomeration on the burning surface of composite propellants is derived to predict the particulates size distribution with a low computational amount. The basic idea is inspired from the well-known pocket models. The pocket size of the region formed by adjacent AP particles is obtained through scanning electron microscopy of the propellant cross-section coupled to an image processing method. The second mergence mechanism, as well as the effect of the burning rate on the agglomeration processes, are included in the present model. The mergence of two agglomerates is prescribed to occur only if their separation distance is less than a critical value. The agglomerates size distribution resulting from this original model match reasonably with the experimental data. Moreover, the present model gives superior results for mean agglomeration diameter compared to common empirical and pocket models. The average prediction error is lower than 5% for the four propellants tested. Results of this study are expected to provide better insight and enrich in the theoretical frame of aluminum agglomeration.

Real-valued MVDR beamforming using spherical arrays with frequency invariant characteristic

Complex-valued minimum variance distortionless response (MVDR) beamforming for wideband signals has very high computational amount. In this paper, we design a novel real-valued MVDR beamformer for spherical arrays. The dependence of the array steering matrix on source signal directions and frequencies is decoupled using spherical harmonic decomposition. Then a compensation network is designed to solve the frequency dependence of the array response and to get a new array response only determined by the spherical harmonics of the source directions. All frequency bins of wideband signals can be used together instead of being processed independently. By exploiting the property of the conjugate spherical harmonics, a unitary transform can be found to acquire a real-valued frequency invariant steering matrix (FISM). Based on the FISM, real-valued MVDR (RV-MVDR) is developed to obtain good performance with low computational amount. Simulation results demonstrate the performance of our proposed method for beamforming and direction-of-arrival (DOA) estimation by comparing with the complex-valued and real-weighted MVDR methods.

Local Smoothness Enforced Cost Volume Regularization for Fast Stereo Correspondence

A novel cost volume regularization algorithm is proposed. It successfully combines weighted cost aggregation in local stereo methods and smoothness constraints in global stereo methods. By enforcing local smoothness during the weighted cost aggregation on the horizontal tree structure, cost distribution can be well regularized. Accurate disparity results with sharp boundary can be generated while keeping extremely low computational amount. A new post-processing method is also proposed. The whole stereo correspondence algorithm achieves competitive performance on the Middlebury stereo testbed in terms of both accuracy and speed.

Sports Video Segmentation using Spectral Clustering

With the rapid development of the computer and multimedia technology, the video processing technique is applied to the field of sports in order to analyze the sport video. For sports video analysis, how to segment the sports video image has become an important research topic. Nowadays, the algorithms for video image segmentation mainly include neural network, K-means and so on. However, the accuracy and speed of these algorithms for moving objects segmentation are not satisfied, and easily influenced by the irregular movement of the object and illumination, etc. In view of this, this paper proposes an algorithm for object segmentation in sports video image sequence, based on the spectral clustering. This algorithm simultaneously considers the pixel level visual feature and the edge information of the neighboring pixels to make the calculation of similarity is more intuitive and not affected by factors such as image texture. When clustering the image feature, the proposed method: (1) preprocesses video image sequence and extracts the image feature. (2)Using weight function to build and calculate the similar matrix between pixels. (2) Extract feature vector. (3) Perform clustering using spectral clustering algorithm to segment the sports video image. The experimental results indicate that the method proposed in this paper has the advantages, such as lower complexity, high computational effectiveness, low computational amount, and so on. It can get better extraction effects on video image.

The use of geometric meshes in product integration Simpson's rules

This paper concerns with the use of geometric meshes in product integration Simpson's rules for the numerical evaluation of weakly singular weighted integrals. We provide a criterion for selecting the mesh in order to achieve a desired accuracy with a low amount of computation. Numerical experiments indicate a reduction in computational costs with respect to the same rules with uniform meshes.

An English auction scheme in the online transaction environment

Internet technology in the recent years has progressed with great strides, and has transcended physical boundaries to achieve a global community. It has been an efficient tool in the development of modern communication, electronic commerce, and various living applications. Under an Internet environment, a reliable and high-performing English auction scheme is presented in the research involving three parties, namely the Registration Manager, Auction Manager, and Bidder. The Registration Manager identifies and authenticates the bidder. The Auction Manager issues the bidding rights and maintains order during the auction. The proposed scheme has the following features: anonymity, traceability, no framing, unforgeability, non-repudiation, fairness, public verifiability, unlinkability among different auction rounds, linkability in a round of auction, efficiency of bidding, one-time registration, and easy revocation. Given the Internet environment, significant importance is attached to time costs in transmitting bidding data. Hence, the bulletin board method is used to enable both the registration and auction managers to declare the necessary parameters. Furthermore, the elliptic curve cryptosystem, owing to its low computational amount and small key size, is applied to the scheme. Consequently, the auction-manager server load can be effectively reduced, while simultaneously significantly increasing bidding efficiency.

The Weber problem with regional demand

This paper is devoted to the study of the Regional Weber Problem, an extension of the Weber problem which allows the demand not be concentrated onto a finite set of points. The most serious drawback of this formulation, from a resolution viewpoint, is the high computational cost involved in the evaluation of the objective function. A new approach is proposed, which requires a low amount of computation and where it is possible to control the error on the approximation. This approximation suggests a new methodology to solve the problem. This methodology is compared with the existing ones, showing its relevance from a practical point of view.

Fuzzy logic control scheme for on-line stabilization of multi-machine power system

The paper presents a fuzzy logic control scheme to enhance the overall stability of a multi-machine power system. Several simple control rules are prepared to control generating units. Desired stabilizing signals are determined according to the rules together with the speed/acceleration states of generating units at every sampling time. The proposed fuzzy logic control scheme is easy to implement, and requires a low amount of computation because of its simple control rules, and required data. The proposed fuzzy logic stabilizer can be easily set up by using a micro-computer with the functions of A/D and D/A conversion. The efficiency of the proposed fuzzy logic stabilizers is demonstrated through simulations using a 10-machine and 39-bus power system.