Table of Content

15 April 2024 Volume 37 Issue 4

  

Multi-Encoder Transformer for End-to-End Speech Recognition

PANG Jiangfei, SUN Zhanquan

Electronic Science and Technology. 2024, 37(4):  1-7.  doi:10.16180/j.cnki.issn1007-7820.2024.04.001

Abstract ( 207 )   HTML ( 14 )   PDF (949KB) ( 100 )  

Figures and Tables | References | Related Articles | Metrics

The current widely used Transformer model has a strong ability to capture global dependencies, but it tends to ignore local feature information at shallow layers. To solve this problem, this study proposes a method using multiple encoders to improve the ability of speech feature extraction. An additional convolutional encoder branch is added to strengthen the capture of local feature information, make up for the neglect of local feature information in shallow Transformer, and effectively realize the integration of global and local dependencies of audio feature sequences. In other words, a multi-encoder model based on Transformer is proposed. Experiments on the open-source Chinese Mandarin data set Aishell-1 show that without an external language model, the proposed Transformer-based multi-encoder model has a relative reduction of 4.00% in character error rate when compared with the Transformer model. On the internal non-public Shanghainese dialect data set, the performance improvement of the proposed model is more obvious, and the character error rate is reduced by 48.24% from 19.92% to 10.31%.

Application of Double-Vector Attitude Determination in Attitude Measurement of Coal Mine Roadheader

XU Yeqian, HUANG Zhe, SHEN Xiaoling, ZHAO Shiyi, LI Jiaxiong, WANG Haosen, XIAO Heng

Electronic Science and Technology. 2024, 37(4):  8-15.  doi:10.16180/j.cnki.issn1007-7820.2024.04.002

Abstract ( 75 )   HTML ( 6 )   PDF (3039KB) ( 45 )  

Figures and Tables | References | Related Articles | Metrics

In view of the problems of high-cost and accumulation of errors in the existing attitude measurement methods of roadheader, an attitude measurement algorithm based on the principle of double-vector attitude determination is proposed. By constructing and sensing the gravity-vector and light-vector respectively in the roadway excavation environment, using inertial inclination measurement and binocular vision measurement techniques, the attitude of the roadheader carrier coordinate frame relative to the roadway navigation coordinate frame can be realized based on the mathematical expression of vector elements in the navigation coordinate frame and the roadheader carrier coordinate frame. A measuring device that consists of an inclinometer and a binocular camera measures the indicated laser and gravity-vectors, and then the attitude of the roadheader can be solved using the double-vector attitude determination algorithm. The static repeatability measurement experiment is designed, and the experimental results show that the repeatability measurement precision of the attitude angles is 0.066 2°. The simulation analysis of the errors that may be introduced using the Monte Carlo method, which shows that the effect of the errors on the yaw, pitch and roll angles are 0.786 4°, 0.454 8° and 0.476 5°, respectively.

Research on Double Layer Ring Equalizer for Lithium Battery Pack of Electric Vehicle

HAN Xinsheng, KAN Jiarong, LING Huiying, WANG Peng, CHENG Qian

Electronic Science and Technology. 2024, 37(4):  16-24.  doi:10.16180/j.cnki.issn1007-7820.2024.04.003

Abstract ( 123 )   HTML ( 4 )   PDF (3025KB) ( 56 )  

Figures and Tables | References | Related Articles | Metrics

The traditional Buck-Boost equalization circuit has long transmission path and low equalization efficiency, and the existing ring equalizer cannot fundamentally solve this problem. In view of this problem, a double-layer ring equalizer for lithium battery based on Buck-Boost and switched capacitor is proposed in this study. The equalization circuit adopts modular balancing method, and adopts Buck-Boost and switching capacitor hybrid circuit to achieve equalization in the bottom module and top module. The circuit structure is analyzed from two aspects of equalization speed and efficiency by graph theory analysis. The results show that the equalization circuit has the advantages of less energy transmission paths and high energy transmission efficiency. The simulation model of the equalization circuit is built on the MATLAB/Simulink power simulation platform. The simulation results show that the equalization speed of the double-layer ring equalizer based on Buck-Boost and switched capacitor is 46% higher than that of the single-layer ring equalizer. Additionally, an experimental prototype of four batteries is established to verify the working principle and system equalization efficiency of the ring equalization circuit. The experimental results show that the equalization efficiency of the double-layer ring equalizer is 26.78% higher than that of single-layer ring equalizer. The simulation and experimental results are consistent with the theoretical analysis, which indicates that the circuit structure is reasonable and effective.

Design of Dual-Frequency Integrated Ultra-Wideband Ground Penetrating Radar Antenna

LIN Xiangyu, ZENG Weihua

Electronic Science and Technology. 2024, 37(4):  25-29.  doi:10.16180/j.cnki.issn1007-7820.2024.04.004

Abstract ( 210 )   HTML ( 9 )   PDF (1869KB) ( 78 )  

Figures and Tables | References | Related Articles | Metrics

In view of the disadvantages of time consuming and low efficiency caused by single frequency band of Ground Penetrating Radar (GPR) products, a dual-frequency integrated composite array antenna with center frequency of 400 MHz and 1 000 MHz and bandwidth of 200~1 500 MHz is designed and implemented, and is applied in Ultra-Wide Band (UWB)GPR system. The array antenna consists of three butterfly antennas with a center frequency of 400 MHz and six butterfly antennas with a center frequency of 1 000 MHz, and has the capability of simultaneous detection of dual frequency bands, which can realize the detection results of two different frequency bands in one survey line, overcome the problem of repeated detection of different frequencies of traditional GPR and enhance the practicability of GPR system. The proposed dual-frequency integrated GPR antenna has the characteristics of ultra-wideband, high gain and narrow beam. Its relative bandwidth is 153%, the maximum peak gain of 17.6 dBi is achieved in the whole bandwidth, and the narrowest half-power beam width is 7.6°. It provides a new antenna scheme for high resolution and high efficiency GPR applications.

Short-Term Load Forecasting Based on CEEMD-ITSA-BiLSTM Combined Model

GAO Dian, ZHANG Jing

Electronic Science and Technology. 2024, 37(4):  30-37.  doi:10.16180/j.cnki.issn1007-7820.2024.04.005

Abstract ( 69 )   HTML ( 3 )   PDF (1013KB) ( 38 )  

Figures and Tables | References | Related Articles | Metrics

Accurate short-term load forecasting of power system is helpful to flexible planning of system resources, reasonable scheduling of units, and improvement of system operation efficiency. In view of the accuracy of load forecasting, this study proposes a short-term load forecasting model based on CEEMD-ITSA-BiLSTM (Complete Ensemble Empirical Mode Decomposition-Improved Tunicate Swarm Algorithm-Bidirectional Long Short-Term Memory). CEEMD decomposition is carried out on the time series load data to obtain several stable IMF (Intrinsic Mode Function), and BiLSTM modeling and prediction are carried out for each IMF. To improve the accuracy of BiLSTM, ITSA algorithm is used to optimize the parameters of the super parameters such as the number of hidden layer nodes, learning rate and training times of BiLSTM, and CEEMD-ITSA-BiLSTM load forecasting model is established. The simulation experiment is conducted with the actual load data, and the single BiLSTM model and the BiLSTM model optimized by different algorithms are compared. The results show that the RMSE (Root Mean Square Error), MAE (Mean Absolute Error) and MAPE (Mean Absolute Percentage Error) error indexes of CEEMD-ITSA-BiLSTM model are increased by 48.54%, 51.32% and 44.78%, respectively when compared with the BiLSTM model, and are significantly lower than other comparison models.

Research on Multiclass Garbage Classification Algorithm Based on Improved MobileNet Network

LIANG Chenye, ZHANG Xuanxiong

Electronic Science and Technology. 2024, 37(4):  38-46.  doi:10.16180/j.cnki.issn1007-7820.2024.04.006

Abstract ( 103 )   HTML ( 8 )   PDF (3591KB) ( 67 )  

Figures and Tables | References | Related Articles | Metrics

view of the large amount of garbage and the fact that a picture contains multiple garbage objects, this study proposes a garbage detection and classification algorithm based on the improved MobileNet network, which integrates the MobileNet network into YOLOv5(You Only Look Oncev5) target detection algorithm. At the same time, the CBAM(Convolutional Block Attention Modul) module is introduced in the backbone to filter meaningful information, and the vision transformer is used to aggregate and form image features. In addition, the weighted bidirectional feature pyramid network is used to distinguish the contribution of different features. At the same time, the ECA(Efficient Channel Attention) module is introduced to combine the image features and transmit them to the prediction layer. Finally, in order to obtain better performance when there is occlusion between garbage targets, soft-NMS(soft-Non Maximum Suppression) method and Alpha-IoU(Alpha-Intersection over Union) loss function is used to predict the extracted features. The experimental results show that the method proposed in this study can realize the location and recognition of multi-target and multi-category garbage., and the mAP(mean Average Percision) value reaches 90.31%, which is 4.95% higher than that of YOLOv5 network, and the processing speed is shortened by about 2.4 seconds. Compared with the Faster R-CNN(Region-based Convolutional Neural Network) algorithm which integrates ResNet(Residual Network) network, the algorithm proposed in this study improves the processing efficiency on the premise of ensuring the accuracy.

Simulation Study on Pulse Width Superposition of Solid-State LTD

YU Tao, RAO Junfeng

Electronic Science and Technology. 2024, 37(4):  47-54.  doi:10.16180/j.cnki.issn1007-7820.2024.04.007

Abstract ( 80 )   HTML ( 2 )   PDF (1236KB) ( 36 )  

Figures and Tables | References | Related Articles | Metrics

A long pulse width solid-state Linear Transformer Driver (LTD) is designed by means of pulse superposition for high voltage, large current and wide range pulses required in biomedical applications of pulsed electromagnetic fields. To break through the pulse width range under the volt-second product limit of the pulse transformer, the LTD increases the output pulse width using parallel superposition. The reset circuit of the pulse transformer is designed to improve the output pulse duty cycle of LTD, so that the multi-pulse superposition of the minimum branch LTD can be achieved. A simulation circuit model is built on Pspice to verify the design principle and the effect. In the case of a single-turn pulse transformer with a volt-second product of 146 V · ms, a pulse output with a single-stage voltage of 300 V and a pulse width of 1.6 ms is realized, which is 3 times of its original volt-second product, and a wider range of pulse extension can be achieved by increasing the number of superimposed pulses.

Workpiece Image Recognition Method Based on Improved ORB-FLANN Algorithm

ZHU Zhihao, LU Zhixu, GUO Yu, GAO Zhi

Electronic Science and Technology. 2024, 37(4):  55-61.  doi:10.16180/j.cnki.issn1007-7820.2024.04.008

Abstract ( 73 )   HTML ( 4 )   PDF (1698KB) ( 50 )  

Figures and Tables | References | Related Articles | Metrics

In view of the problems of low matching rate and long running time of traditional image recognition algorithms, an improved ORB-FLANN(Oriented FAST and Rotated BRIEF-Fast Library for Approximate Nearest Neighbors) based workpiece image recognition method is proposed. The feature description of ORB algorithm and image feature matching algorithm are modified to solve the disadvantages of traditional image recognition algorithm in the case of scale and rotation transformation and reduce the mismatching rate of matching. For the feature points detected by ORB algorithm, SURF (Speeded Up Robust Features) algorithm is used to add orientation information and complete the feature description, so as to obtain the feature points with rotation-scale invariance. FLANN algorithm is combined with bidirectional matching strategy for coarse matching of feature points. Finally, the progressive sampling-congruence algorithm is used to further eliminate the mismatched point pairs and complete the fine matching. The experimental results show that compared with other methods, the improved algorithm can improve the matching accuracy of 2.6%~18.8% and 29.5%~43.9%, respectively, when processing scale and rotation transform images, and the running time is within 4s, improving the efficiency and accuracy of workpiece image recognition.

Instance Segmentation Based on Attention and Image Contour

GU Denghua, GU Chunhua

Electronic Science and Technology. 2024, 37(4):  62-68.  doi:10.16180/j.cnki.issn1007-7820.2024.04.009

Abstract ( 113 )   HTML ( 7 )   PDF (1555KB) ( 63 )  

Figures and Tables | References | Related Articles | Metrics

Based on image contour, the instance segmentation method uses fewer contour nodes to represent an object, which effectively reduces the number of algorithmic parameters and improves its operation efficiency. However, with the segmentation result of poor quality, it is no match for traditional pixel-by-pixel processing segmentation algorithm in terms of accuracy. To improve the accuracy of the algorithm, it is of great necessity to introduce a refined model of the instance segmentation (Attend the Contour snake,AC-snake), which is based on image contour with a combination of attention mechanism. An improved Largekernel+ is added to the backbone network to improve the receptive field of the model and extract richer feature information. The network structure at the contour vertex deformation stage is improved, and the Dual Channel attention (DC-attentio) module is combined to enhance the effective information of contour vertex, reduce the invalid parameters in the training network, and improve the detection accuracy and training speed. The experimental results show that in Cityscapes validation data set, the improved model proposed in this study has improved performance when compared with the original model.

Short-Term Photovoltaic Power Prediction Based on k-sums Quadratic Clustering and Dynamic Combination Learning

WU Jiabao, ZENG Guohui, ZHANG Zhenhua

Electronic Science and Technology. 2024, 37(4):  69-76.  doi:10.16180/j.cnki.issn1007-7820.2024.04.010

Abstract ( 65 )   HTML ( 2 )   PDF (951KB) ( 34 )  

Figures and Tables | References | Related Articles | Metrics

At present, the prediction accuracy of a single model is difficult to remain optimal with power fluctuation. To improve the stability of grid connected system operation and energy saving dispatching of power grid, this study proposes a dynamic learning combination short-term power prediction method based on k-sums hierarchical clustering. The weather types are divided into sunny day A₁, cloudy day A₂, and rainy day B through segmentation clustering using k-sums algorithm. The TCN (Temporal Convolutional Network) is used to extract the time series characteristics of data, and the GRU(Gate Recurrent Unit) structure of the fusion extraction time series characteristics module is established with GRU to achieve the effect of being sensitive to the time series characteristics. After dynamically combining the improved GRU structure with the SVM(Support Vector Machine), the Elastic Net algorithm is adopted to output the best weight value to obtain the final prediction value. The power data of photovoltaic power generation and corresponding meteorological data of a region in Jiangsu are used to verify the proposed method. The results show that the MAE(Mean Absolute Error) of the dynamic combination learning model is 1.888, and the RMSE(Root Mean Squared Error) is 2.403.

Research on Fast 3D Hand Keypoint Detection Algorithm Based on Anchor

QIN Xiaofei, HE Wen, BAN Dongxian, GUO Hongyu, YU Jing

Electronic Science and Technology. 2024, 37(4):  77-86.  doi:10.16180/j.cnki.issn1007-7820.2024.04.011

Abstract ( 76 )   HTML ( 2 )   PDF (4381KB) ( 39 )  

Figures and Tables | References | Related Articles | Metrics

In human-robotcollaboration tasks, hand key point detection provides target point coordinates for the robotic arm.A2J(Anchor-to-Joint) is a representative method of key point detection using anchor points.A2J can achieve better detection effect with depth map input, but it has insufficient ability to acquire global features.In this study, a GLF(Global-Local Feature Fusion) module is designed to fuse the shallow and deep features of the backbone network.In order to improve the detection speed, the backbone network of A2J is replaced with ShuffleNetv2 and reformed, and 3×3 depth separable convolution is replaced with 5×5 depth separable convolution to increase the sensitivity field and effectively improve the backbone network's ability to extract global features.ECA(Efficient Channel Attention) is introduced into the anchor weight estimation branch to improve the network's attention to important anchor points.The results of training and testing on the mainstream data sets ICVL and NYU show that the average error of the proposed method is reduced by 0.09 mm and 0.15 mm, respectively, compared with A2J.The detection rate of 151 frame·s-1 is realized on GTX1080Ti graphics card, which fully meets the real-time requirement of man-machine collaboration task.

Multi-Parameter Prediction of Dissolved Oxygen in Eel Ponds Based on ISSA-LSTM

LIN Binbin, XU Zhen, YUAN Quan, TIAN Zhixin

Electronic Science and Technology. 2024, 37(4):  87-96.  doi:10.16180/j.cnki.issn1007-7820.2024.04.012

Abstract ( 37 )   HTML ( 2 )   PDF (1641KB) ( 29 )  

Figures and Tables | References | Related Articles | Metrics

In order to improve the multi-parameter prediction accuracy of dissolved oxygen, an ISSA-LSTM (Improved Sparrow Search Algorithm-Long and Short-Term Memory Neural Networks) dissolved oxygen prediction model is developed based on the ISSA and LSTM. The model is applied to the prediction of dissolved oxygen in eel breeding ponds at Shanghai academy of agricultural sciences. The sparrow search algorithm is optimized using chaos mapping, lensing imaging backward learning, adaptive adjustment and Cauchy variation. The data are pre-processed by wavelet transform, the input parameters for model training are determined using principal component analysis. The training results show that the correlation coefficient, root mean square error, mean square error and mean absolute error are 0.911, 1.392 mg·L^-1, 1.938 mg·L^-1 and 0.992 mg·L^-1, which are all better than those in the control model. The choice of model input parameters also have an impact on the model prediction results, with the best model predictions using both moderately and strongly correlated parameters with dissolved oxygen as input parameters. The training results provide a new perspective for the development of the dissolved oxygen multi-parameter prediction model.

Research on Network Topology Technology of Close Range Wireless Communication

YANG Jiayi, YAN Jun

Electronic Science and Technology. 2024, 37(4):  97-102.  doi:10.16180/j.cnki.issn1007-7820.2024.04.013

Abstract ( 74 )   HTML ( 1 )   PDF (2080KB) ( 29 )  

Figures and Tables | References | Related Articles | Metrics

In response to the new operational requirements of precision-guided weapons, this study focuses on wireless electrical design as the core technological direction and implements product modular design to address the challenges of flexible networking, resilience against destruction, and dynamic topology. The research adopts an Ad Hoc networking architecture, generalized coding, and passive multiple access techniques to achieve fast, flexible, and stable information exchange. By employing a hybrid routing design, the study divides the metal chamber network into subnets and applies proactive and reactive routing within the subnets, while introducing the concept of landmarks to mitigate the impact of multi-system communication. The specific approach involves predefining the timing and priority of all events, establishing a simplified event type library, and enabling node-to-node backup storage. Additionally, the use of boundary broadcasting protocols facilitates the rapid arrival of destination nodes at gateway nodes. The research findings confirm that the proposed solution meets the requirements of modular design for weapon functionality. Through wireless networking and routing protocol design, the capabilities of flexible networking, resilience against destruction, dynamic topology, and multi-system communication are successfully achieved.