一个集成高斯混合模型-支持向量机-统计模式的适合中低浓度的跃移沙粒识别算法

doi:10.7522/j.issn.1000-694X.2025.00261

摘要/Abstract

摘要：

鉴于目前的风沙颗粒识别算法存在准确率低或时间成本高的问题，提出了一个集成的风沙颗粒识别算法，该算法包括高斯混合模型分割、支持向量机分类与沙粒特征统计模型再确认等环节。风沙图像识别结果表明和已有算法相比较，新算法具有中等召回率（60%~90%）、最高的准确率（80%~95%）和较低的时间成本，为中低浓度的风沙图像识别和追踪提供了新的思路和方法。

关键词: 风沙颗粒, 高斯混合模型, 支持向量机, 统计模式

Abstract:

Current algorithms for saltating particle recognition face challenges in balancing accuracy and computational efficiency. To address this issue， this study proposes an integrated algorithm combining Gaussian Mixture Model （GMM）， Support Vector Machine （SVM） classification， and a statistical feature verification model for sand particle confirmation. Experimental results on sand image recognition demonstrate that the proposed method achieves a moderate recall rate （60%-90%）， the highest accuracy （80%-95%） among existing methods， and significantly reduced computational time. The work provides a new idea and method for the recognition and tracking of aeolian sand images at low to medium concentrations.

Key words: sand particle, Gaussian Mixture Model, Support Vector Machine, probability-distribution

中图分类号:

Q151.94

李浩强, 李惠娟, 杨倩文, 梅凡民. 一个集成高斯混合模型-支持向量机-统计模式的适合中低浓度的跃移沙粒识别算法[J]. 中国沙漠, 2026, 46(3): 255-261.

Haoqiang Li, Huijuan Li, Qianwen Yang, Fanmin Mei. A hybrid algorithm for aeolian saltating particle recognition under low-medium particle concentrations with Gaussian Mixture Model， Support Vector Machine and probability-distribution of saltating particles' geometric and color parameters[J]. Journal of Desert Research, 2026, 46(3): 255-261.

图/表 6

图1 新算法流程

Fig.1 Workflow of the new hybrid scheme for saltating particle recognition

图2 A高斯混合模型检测沙粒数量随着开闭运算核形状及大小变化；B高斯混合模型F1分数在不同开闭运算的核形状及大小情境下随沙粒浓度的变化（括号里斜体数字代表闭运算核大小）

Fig.2 The number of saltating particles identified by Gaussian mixture model varies with the kernel shape and size of the opening and closing for morphological operation； The F1 score of Gaussian mixture model fluctuates with the sand particle concentrations under the various combinations of the kernel shape-size parameters

图3 跃移沙粒宽度、高度、周长及面积的概率分布

Fig.3 Probabilities of widths， heights， perimeters and areas of saltating particles from the L3 video section

表1 确认沙粒的颜色特征依据

Table 1 Color criterions for identifying saltating particle

跃移沙粒颜色特征	颜色空间分量
R	0~48
G	16~96
B	0~48
H	0~89
S	0~160
V	17~96

图4 新算法和发表的风沙颗粒识别算法的比较（GT、DGT、GMM、GMM+SVM分别是指基于Matlab 平台的灰度阈值算法［10］、动态灰度阈值算法［14］、高斯混合模型、高斯混合模型-支持向量机。改进YOLOv5［12］和YOLOv8［11］来自文献）

Fig.4 Comparisons between the new algorithm and other published schemes for saltating particle recognition in recall and accuracy rates （GT， DGT、 GMM、 GMM+SVM correspond to the gray threshold scheme from Matlab ［10］， dynamic threshold scheme ［14］， Gaussian mixture model ［16］ and improved Gaussian mixture model. The data of YOLOv5 and YOLOv8 derive from the Ref.12 and the Ref.11， respectively）

表2 新算法和发表的风沙颗粒识别算法推理时间对比

Table 2 Inference time of the new scheme and the published schemes for saltating particle recognition

算法	运行配置	每帧用时/ms
YOLOv5	RPX 4090	4.2
Sand-YOLOv5	RPX 4090	6.1
YOLOv8	Intel Core i5-8300H	94.5
GMM	Intel Core i5-7300HQ	6.5
GMM+SVM	Intel Core i5-7300HQ	7.9
新算法	Intel Core i5-7300HQ	10.2

参考文献 21

[1]	Bagnold R A.The Physics of Blown Sand and Desert Dunes［M］.London，UK：Methuen，1954：265-270.
[2]	Kok J F， Storelvmo T， Karydis V A，et al.Mineral dust aerosol impacts on global climate and climate change［J］.Nature Reviews Earth & Environment，2023，4（2）：71-86.
[3]	Cheng J， Jiang F， Xue C，et al.Characteristics of the disastrous wind-sand environment along railways in the Gobi area of xinjiang，China［J］.Atmospheric Environment，2015，102：344-354.
[4]	牛清河，屈建军，安志山.甘肃敦煌雅丹地质公园区风蚀气候侵蚀力特征［J］.中国沙漠，2017，37（6）：1066-1070.
[5]	Zhang Y， Wang Y， Jia P.Improving the Delaunay tessellation particle tracking algorithm in the three-dimensional field［J］.Measurement，2014，49：1-14.
[6]	Zhang Y， Wang Y， Jia P.Measuring the kinetic parameters of saltating sand grains using a high-speed digital camera［J］.Science China Physics，Mechanics & Astronomy，2014，57：1137-1143.
[7]	Lin Y， Zhang Y， Jin Y，et al.Three-dimensional particle tracking algorithm based on the special ellipsoids［J］.Measurement，2023，216：112883.
[8]	Zhang Y， Wang Y， Yang B，et al.A particle tracking velocimetry algorithm based on the Voronoi diagram［J］.Measurement Science and Technology，2015，26（7）：075302.
[9]	Zhang Y， Li M， Wang Y，et al.Reinvestigation of the scaling law of the windblown sand launch velocity with a wind tunnel experiment［J］.Journal of Arid Land，2019，11：664-673.
[10]	Mei F M， Zhou H J， Su J，et al.A Kalman filter‐Hungarian algorithm with a postprocessor for tracking aeolian saltating particle in high‐speed video［J］.Earth Surface Processes and Landforms，2024，49（15）：5086-5097.
[11]	Lin C， Mei F M， Zhou H J，et al.A hybrid deep learning scheme with high accuracy for tracking aeolian saltating particles in high-speed pictures［J］.Advanced Powder Technology，2025，36（5）：104862.
[12]	Xi A G， Mei F M， Li H Q，et al.An improved YOLOv5 model for aeolian saltating particle recognition in high-speed videos［J］.Measurement，2025，253：117321.
[13]	Jiang C W， Parteli E J R， Dong Z B，et al.Wind-tunnel experiments of Aeolian sand transport reveal a bimodal probability distribution function for the particle lift-off velocities［J］.Catena，2022，217：106496.
[14]	梅凡民，雒遂，陈金广.一种改进的高浓度风沙图像的动态灰度阈值分割算法［J］.力学学报，2018，50（3）：699-707.
[15]	Zhang W， Wang Y， Lee S J.Simultaneous PIV and PTV measurements of wind and sand particle velocities［J］.Experiments in Fluids，2008，45：241-256.
[16]	Reynolds D A.Gaussian mixture models［J］.Encyclopedia of biometrics，2009，741（659-663）：3.
[17]	Pisner D A， Schnyer D M.Support vector machine［M］//Mechelli A，Vieira S.Machine Learning.London，UK：Academic Press，2020：101-121.
[18]	Cho J， Jung Y， Kim D S，et al.Moving object detection based on optical flow estimation and a Gaussian mixture model for advanced driver assistance systems［J］.Sensors，2019，19（14）：3217.
[19]	Yan A， Li J， Wang Y，et al.Research on moving target detection based on improved Gaussian mixture model［C］//2020 Chinese control and decision conference （CCDC）.IEEE，2020：1168-1173.
[20]	Hou L， Lu K， Yang X，et al.G-rep：Gaussian representation for arbitrary-oriented object detection［J］.Remote Sensing，2023，15（3）：757.
[21]	周虹吉，梅凡民，蒲梦洁，等.基于集成学习的风沙跃移轨迹自动识别模型［J］.中国沙漠，2024，44（6）：70.