Journal of University of Chinese Academy of Sciences

An experimental investigation on the breakup characteristics of liquid metal free jet under a horizontal magnetic field

DONG Quanrun, YANG Juancheng, NI Mingjiu

2022, 39 (5): 577-585. DOI: 10.7523/j.ucas.2022.029

Based on a high-speed photographic system, experiments on three-dimensional free jets of liquid metal in the absence of a magnetic field and in a horizontal magnetic field have been carried out to observe the process of liquid GaInSn jet breakup and droplet formation in an oxygen-free environment with a maximum We of 400 and a maximum Ha of 30. From the results on jet morphology, surface disturbance, and breakup length, we analyzed the characteristics of jet breakup. In the absence of magnetic field, the jet shows nine different morphologies, and the surface disturbance shows two forms:expansion wave and sinusoidal wave; with the increase of We, the disturbance amplitude first decreases and then increases, and the breakup length first increases and then decreases. When a horizontal magnetic field is imposed, the jet shows four typical morphologies, with the leading edge of the jet being flattened in the direction of the vertical magnetic field line and elliptical along the magnetic field line. As the Ha number increases, the jet break length tends to increase overall, but decreases in some operating conditions. The results of this paper have enriched the phenomenon of liquid metal jets under magnetohydrodynamic effects.

Signed-rank-based test for high dimensional mean vector

LIU Yan, LI Shiming, ZHANG Sanguo

2022, 39 (5): 586-592. DOI: 10.7523/j.ucas.2020.0059

Abstract ( 764 ) PDF (0KB) ( 0 )

This work is concerned with tests for one-sample mean vectors under high dimensional cases. Existing high dimensional tests for mean vectors base on the assumption of elliptical distribution have been proposed recently. To extend to more distributions, we propose a signed-rank-based test. The proposed test statistic is robust and scalar-invariant. Asymptotic properties of the test statistic are established. Numerical studies show that the proposed test has a good control of the type-I error and is more efficiency. We also employ the proposed method to analyze an ophthalmic data.

Momentum strategy, momentum crashes, and risk management: an empirical research based on Chinese commodity futures market

WEI Yongfeng, ZHAO Wei

2022, 39 (5): 593-614. DOI: 10.7523/j.ucas.2020.0049

Abstract ( 795 ) PDF (0KB) ( 0 )

This article aims to study the effectiveness of momentum strategies in Chinese commodity futures market, and after judging the existence and causes of momentum crashes, puts forward effective methods to manage the risk of momentum crashes. In this paper, considering the transaction costs, the commodity futures momentum strategy can continuously obtain significant risk-adjusted returns, and further empirical discovery of the momentum crashes phenomenon in Chinese commodity futures market. The reason of momentum crashes is that the loser portfolio has the nature of option-like is more sensitive to market portfolio volatility, which in turn leads to the crashes of momentum portfolios. In order to carry out the risk management of momentum crashes, this paper proposes to construct a dynamic weighted momentum strategy based on target condition stop to manage the risk of momentum crashes. The results show that this method effectively avoids the extreme risk brought about by momentum crashes and obtains higher momentum return and Sharpe ratio.

Predicting sunspot variations through neural network

CHENG Shu, SHI Yaolin, ZHANG Huai

2022, 39 (5): 615-626. DOI: 10.7523/j.ucas.2021.0068

Abstract ( 1556 ) PDF (0KB) ( 0 )

Sunspot variations are the sun's symptoms of strong magnetic perturbations. In this paper, we use long short-term memory neural network and one-dimensional convolution neural network to detect sunspot variations. Here we use three different datasets, including the yearly mean sunspot number (YSSN) from 1700 to 2020, the monthly mean sunspot number (MSSN) from 1749 to 2021 and the monthly mean sunspot areas (MSSA) from 1874 to 2021. First, based on the YSSN dataset, we obtain YSSN for 2021 and the predicted YSSN in the 25th solar cycle appears at 2025 which equals 163.4; Then, based on the MSSN dataset, we obtain MSSN for June 2021 and the predicted YSSN in the 25th solar cycle appears in October 2024 which equals 245.9; Next, based on the MSSA dataset, the predicted MSSA for June 2021 is 73.1; Finally, the latitude is divided into 13 partitions to predict the butterfly diagram, and still, neural network can reconstruct the butterfly diagram. Therefore, neural network can provide a physical perspective for sunspot investigation.

The b-value spatiotemporal evolution in southeastern Tibetan Plateau and its implications on regional stress field characteristics

GAO Yajing, LUO Gang, WANG Shaopo, ZHOU Yuanze

2022, 39 (5): 627-638. DOI: 10.7523/j.ucas.2021.0037

Abstract ( 620 ) PDF (0KB) ( 0 )

Stress field is an important factor to assess regional seismic risk. We collected the seismic catalogue data from 1970 to 2019 in southeastern Tibetan Plateau, used the maximum likelihood method to calculate regional b values, and then obtained the spatial and temporal distribution of b values of this region. Based on the negative correlation between seismic b value and stress, we analyzed the stress distribution and variation on the major fault zones in southeastern Tibetan Plateau. The results are as follows. 1) In the center of Xianshuihe-Xiaojiang fault system, the b value of Daliangshan fault zone is lower than that of Anninghe-Zemuhe fault zone. This indicates that the stress on Daliangshan fault zone is greater and its seismic risk is relatively high in the future. 2) The b values in shallow layer (0-20 km) are higher than those of deep layer (20-40 km), which is consistent with the characteristics that confining pressure is low and rock tends to brittle fracture in the shallow, while confining pressure is high and rock tends to ductile deformation in the deep. 3) Before and after the 2008 Wenchuan earthquake, b value of epicenter area had a decrease-rise-decrease process, and this showed the accumulation-release-accumulation process of regional stress. The larger the magnitude of earthquake, the longer the decreasing trend of b value before the earthquake, the greater the impact of earthquake on b value. The closer to the epicenter, the greater the decrease of b value. 4) The b value of Longmenshan fault zone is relatively low at present and this shows that Longmenshan fault zone is accumulating stress.

PolSAR terrain classification based on image segmentation and EM algorithm

CAO Zhe, FENG Shanshan, SUN Xian, HONG Wen

2022, 39 (5): 639-647. DOI: 10.7523/j.ucas.2021.0011

Abstract ( 526 ) PDF (0KB) ( 0 )

In the study of the terrain classification based on the polarimetric synthetic aperture radar (PolSAR), the image segmentation algorithm based on deep neural network has the disadvantages of fuzzy classification boundary, low classification accuracy and complicated calculation caused by the redundancy of high-dimensional feature information. This paper proposes a lightweight segmentation network based on convolutional neural network and EM algorithm called low-rank-reconstruction-net (LRR-Net), which is applied to the terrain classification of fully PolSAR images. Starting from the idea of polarimetric target decomposition, LRR-Net uses the EM algorithm to perform low-rank reconstruction of features, maps the features from high-dimensional space to low-dimensional space, achieving higher classification accuracy while reducing parameters. The model is trained and evaluated in GF-3 fully PolSAR dataset, and the results show that the model complexity is reduced under the guarantee of the classification accuracy.

Fine process method for Gaofen-3 L1A-level image

FANG Hankang, ZHANG Bo, CHEN Weirong, WU Fan, WANG Chao

2022, 39 (5): 648-657. DOI: 10.7523/j.ucas.2021.0005

Abstract ( 1176 ) PDF (0KB) ( 0 )

Level one A (L1A) product of Gaofen-3 SAR satellite is the primary image set delivering for customer. This paper presents a complete workflow to facilitate the post-process of GF-3 L1A images for follow-up scientific research or value-added applications, where robust and precise processing is essential to generate the advanced high-level product concerning radiometric correction and geometric correction. Firstly, to eliminate the statistical bias caused by the null and zero pixel values induced in the quantization of the L1A product, an improved radiometric correction formula is derived based on the equivalent noise coefficient of Gaofen-3 images. Then, to determine the coordinates of image corners, an inverse algorithm supported by RPC parameters is proposed for geometric correction. This algorithm is robust by counting on the orbit direction, look direction, and sampling interval provided in an XML metadata file. Finally, a SAR filter operator is introduced into the resampling step of output results to improve the equivalent look number. Experimental results comparing with the radiometric values of a sentinel-1 image and the geometric accuracy of a sentinel-2 optical image, respectively, validate the accuracy and reliability of this method for L1A product processing.

Road information extraction and application in the suburban mountainous area based on remote sensing images

CHEN Ruonan, PENG Ling, LIU Yufei, WEI Zhichao, LYU Beiru, CHEN Deyue

2022, 39 (5): 658-667. DOI: 10.7523/j.ucas.2021.0004

Abstract ( 602 ) PDF (0KB) ( 0 )

In recent years, suburban mountain areas have become a good choice for urban residents to go outing. Intensive tourist outings and villagers' production activities bring fire safety hazards to mountains and forests. And road information is vital information for forest fire prevention emergency. However, due to the problems of occlusion, shadow, narrow and multiple branches in suburban mountainous roads, conventional urban road extraction algorithms have poor performance in suburban mountain areas. This paper proposes a road semantic segmentation model and a training method that transforms the binary into a multi-class classification problem, forcing the model to focus on learning spatial distance information to generate road results with better spatial continuity. Then, experiments were carried out on the Yajishan road dataset made by ourselves and the Massachusetts public road dataset respectively to verify the effectiveness of our model and training method. In addition, it is verified that the training method is also applicable to other commonly used semantic segmentation models such as U-Net and DeepLabV3. Finally, this paper also conducts post-processing research based on the above road extraction results to output road surface, road centerline vector data with road width information, and conducts fire truck traffic analysis application in Beijing Yaji Mountain. The research results have alleviated the problem of insufficient road information for commercial electronic maps in the suburban mountain areas with few people, and provided information technology support for forest fire emergency rescue.

3D rock mass point cloud holes detection and filling method based on plane extraction

MA Zhaoyue, XIAO Jun, WANG Ying

2022, 39 (5): 668-676. DOI: 10.7523/j.ucas.2020.0032

Abstract ( 598 ) PDF (0KB) ( 0 )

In rock engineering, the rock point cloud data scanned by a laser scanner always contains holes because of the scanning measurement angle, shadow, occlusion of obstacles, and other factors, which will affect the result of subsequent 3D reconstruction. The existing filling methods mainly focus on the regular point cloud data, and the point cloud hole is filled according to the neighborhood information of the holes, and the experiment result of rock point cloud holes detection and filling are not effective and low efficiency. In this paper, we propose an algorithm for detecting and filling rock point cloud holes based on the plane extraction leverage the characteristics of rock point cloud data. Firstly, an optimized region growing algorithm is applied to extract the plane of the rock point cloud. Then, we traverse all point clouds and retrieve their K-neighborhood point sets. These points are mapped to the corresponding plane, and we calculate the neighborhood angle to detect holes. Finally, we classify the point cloud holes according to the number of corresponding planes of the boundary point set, and the point cloud holes are filled by adding sampling points on the corresponding planes. Our algorithm realizes the process of denoising and plane fitting of point cloud data by plane extraction, simplifys the subsequent hole filling process and reduces the time complexity. Experimental results demonstrate that our algorithm has a higher accuracy of detection and filling, higher operation efficiency, and better filling result for rock-mass point cloud compared to the state-of-the-art approaches.

Accuracy analysis of 3D object detection based on stereo point cloud

LIU Wangchao, LOU Xin

2022, 39 (5): 677-683. DOI: 10.7523/j.ucas.2020.0042

Abstract ( 624 ) PDF (0KB) ( 0 )

3D object detection is a crucial task in autonomous driving. Recently, the accuracy of LiDAR-based 3D object detection algorithms have improved dramatically. However, if we replace the LiDAR data with the depth map that generates from stereo cameras, the detection performance drops a lot. In a recent research, it is found that by transforming the stereo-based depth map to point cloud representation, referred to as pseudo-LiDAR, the performance of 3D object detection can be improved significantly. However, there is still a big gap of accuracy between LiDAR-based and stereo-based algorithms. One of the main reasons is that there are error points in the transformed pseudo-LiDAR data. To study this, we use the LiDAR points cloud to correct the pseudo-LiDAR data, i.e., to detect and exclude the error points. Then we use the optimized pseudo-LiDAR to perform 3D object detection. Experimental results show that the detection accuracy can be improved by as much as 21.02%.

Hyperspectral target detection method based on filter tensor analysis

YANG Shuai, JI Luyan, GENG Xiurui

2022, 39 (5): 684-694. DOI: 10.7523/j.ucas.2020.0060

Abstract ( 505 ) PDF (0KB) ( 0 )

Most of the existing target detection algorithms for hyperspectral images treat each band indiscriminately, so the physical information of the image band cannot be fully utilized. In this paper, the hyperspectral images are firstly divided into several different waveband ranges (such as visible light, near infrared, shortwave infrared, etc.) according to the different imaging mechanisms. A recently developed multi-temporal target detection algorithm:FTA (filter tensor analysis) is introduced into the hyperspectral target detection by combining the different waveband ranges of the hyperspectral images with the time-phase dimension of multi-temporal remote sensing data correspondingly. Based on the new approach, a band-divided FTA algorithm for single-temporal hyperspectral images is proposed. Experiments on hyperspectral images prove that the band-divided FTA algorithm can achieve better results in target detection than the traditional single-phase target detection algorithm.

A saliency-based ship target detection method in high sea state SAR images

ZHANG Ziqi, WANG Xiaolong

2022, 39 (5): 695-703. DOI: 10.7523/j.ucas.2020.0050

Abstract ( 660 ) PDF (0KB) ( 0 )

In this study, a ship target detection method in high sea state SAR image, named Itti-SAR, based on saliency is proposed, which consists of two stages:saliency image extraction and connectivity judgment. In the stage of saliency map extraction, taking the characteristics of SAR images into consideration, the improved direction feature and consistency feature are introduced into the traditional visual attention model to construct a saliency model suitable for SAR images, which realizes the extraction of ship target saliency maps for high sea state SAR images. In the stage of connectivity judgment, the density constraint is used to judge the connectivity of salient areas to prevent the detection of a single target into multiple, thereby further reducing false alarms. The experimental results on several SAR images verify the effectiveness of the method. The experimental results show that compared with the classical CFAR algorithm, the proposed method has the advantages of high precision, high recall rate and independent of prior knowledge.

Throughput analysis of UAV-assisted cellular networks by Matern hardcore point processes

LIU Mengbing, QIU Ling, LIANG Xiaowen

2022, 39 (5): 704-711. DOI: 10.7523/j.ucas.2020.0053

Abstract ( 450 ) PDF (0KB) ( 0 )

Unmanned aerial vehicles (UAVs) have been expected to coexist with conventional terrestrial cellular networks and become an important component to support high rate transmissions. This paper presents an analytical framework for evaluating the throughput performance of a downlink two-tier heterogeneous network. The locations of terrestrial base stations (BSs) are modeled by Poisson point processes. Considering the minimum distance constraint among UAVs, Matern hardcore point processes is utilized to model the locations of UAVs. Tools of stochastic geometry are invoked to derive more tractable expressions for average data rates of users. With the analytical results, we discuss the optimal combinations of UAVs' height and power control factor. The result shows that an appropriate power control factor can effectively maximize UAV users' average data rate as well as guaranteeing the BS users' performance under our proposed model.

Spoon network: a new network structure for Landsat imagery cloud detection

WANG Shuli, TANG Hairong, JI Luyan

2022, 39 (5): 712-720. DOI: 10.7523/j.ucas.2020.0036

Abstract ( 563 ) PDF (0KB) ( 0 )

In view of the shortcomings of the neural network model for remote sensing image cloud detection, such as the loss of detail information, the high cloud miss detection rate and the complexity of calculation caused by the insufficient utilization of spectral information, this paper proposes a new and lightweight network called spoon net (S-Net), which is applied to the cloud detection of Landsat remote sensing image. S-Net is divided into two stages. In the first stage, the convolution kernel of 1×1 is used to extract image spectral features to avoid image details being blurred; in the second stage, the encoder decoder framework is used to extract image spatial features, and group convolution is introduced to convolute each layer of spectral channels extracted in the first stage separately to maintain spectral features and reduce model parameters. The model is trained and evaluated in Landsat8 biome dataset, and the results show that the model has a great advantage in memory and time, and achieves an accuracy of 95%.

Current Issue