Non-destruction Detection of jelly orange granulation disease using laser Doppler vibrometry
-
摘要:
粒化是柑橘类水果常见的一种内部病害，患有该病害的水果外部特征并不明显，难以从外观上直接识别出来。本文使用微型激光多普勒测振仪（micro-LDV）和共振喇叭搭建了一套声学振动实验装置，将其用于采集“爱媛38号”果冻橙的振动响应信号，然后将一维的振动响应信号转换为振动多域图像，并构建了一个Resnet-Transformer（ResT）网络，用于提取振动多域图像中的深层特征，以识别果冻橙粒化病。本文中，使用振动多域图像分别训练ResT、Resnet50和Vision Transformer（ViT）模型，并将ResT的性能与Resnet50和ViT进行比较；使用振动多域图像纹理特征或振动频谱特征训练偏最小二乘判别分析（PLS-DA）和支持向量机（SVM）模型，并与ResT模型进行性能对比。结果表明，使用振动多域图像训练的ResT模型可以精准识别果冻橙粒化病并且检测准确率为98.61%，模型的F1为0.986、精确率为0.986、召回率为0.986。由上述结果可知，提出的方法可在简单、快速、低成本的前提下准确识别粒化果冻橙。
Abstract:Granulation is a common internal disease of citrus fruits, and it is difficult to identify the fruits with this disease from their external features. In this study, an acoustic vibration experimental setup was constructed using a micro-laser Doppler vibrometer (micro-LDV) and a resonance speaker. This was used to collect vibration response signals of ‘Aiyuan 38’ jelly orange. The one-dimensional vibration response signals were converted into vibration multi-domain images, and a Resnet-Transformer network (ResT) was constructed to extract deeper features from the vibration multi-domain images for identifying granulation disease in jelly oranges. In this paper, the ResT, Resnet50, and Vision Transformer (ViT) models were trained using vibration multi-domain images, and their performances were compared. Then, partial least squares discriminant analysis (PLS-DA) and support vector machine (SVM) models were trained using vibration multi-domain image texture features or vibration spectrum features, and the performance was compared with the ResT model. The results show that the ResT model trained using vibration multi-domain images can achieve accurate identification of jelly orange granulation disease with 98.61% detection accuracy, 0.986 F1, 0.986 precision, and 0.986 recall. The proposed method can accurately identify granulated jelly oranges with simplicity, speed, and low cost.
-
图 1 正常和粒化的果冻橙横切面示例图
Figure 1. Example cross sections of normal and granulated jelly orange
图 3 振动多域图像生成流程图。(a)经过快速傅立叶变换得到的果冻橙振动频谱曲线；(b)果冻橙振动响应信号；(c)振动多域图像生成模块，其包含对时频信号降维的卷积块、生成时频图像的Stockwell transform（ST）块、将时频域信号生成时域或频域图像的格拉姆角场(GAF)块以及归一化块；(d)振动ST时频域图像可视化结果，(e)振动GAF频域图像可视化结果；(f)振动GAF时域图像可视化结果
Figure 3. Flow chart of vibration multi-domain image generation. (a) Vibration spectrum curves of jelly orange obtained by fast Fourier transform, (b) vibration response signal of jelly orange; (c) vibration multi-domain image generation module with contains the convolution block for downscaling the time-frequency signal, the Stockwell transform (ST) block for generating the time-frequency image, the Gramian Angle Field (GAF) block for generating the time-domain or frequency-domain images from the time-frequency signal into, and the normalization block; (d) the result of visualizing the vibration ST time-frequency domain image; (e) the result of visualizing the vibration GAF frequency domain image; (f) the result of visualizing the vibration GAF time-domain image
图 4 ResT模型结构示意图。ResT主要由多个基于CNN的Bottleneck块和基于Transformer的Swin Transformer块组成，分别提取振动多域图像的局部和全局的深层信息
Figure 4. Schematic diagram of ResT model structure. ResT mainly consists of multiple CNN-based Bottleneck blocks and Transformer-based Swin Transformer blocks, respectively, which extract local and global deep information from vibration multi-domain images
图 5 果冻橙不同摆放位置的振动频谱
Figure 5. Vibration spectra of jelly orange at different placement positions
图 7 使用CARS算法从振动频谱中选择有效频率
Figure 7. Selection of effective frequencies from the vibration spectrum using the CARS algorithm
图 9 ResT、Resnet50、ViT、VMIT-SVM、VST-SVM、VMIT-PLS-DA和VST-PLS-DA模型对果冻橙粒化病的识别性能比较
Figure 9. Comparison of the jelly orange granulation disease identification performances of ResT, Resnet50, ViT, VMIT-SVM, VST-SVM, VMIT-PLS-DA, and VST-PLS-DA models
表 1 基于振动多域图像纹理特征的PLS-DA和SVM模型对果冻橙粒化病的分类结果
Table 1. jelly orange granulation disease classification results of PLS-DA and SVM models based on vibrational multi-domain image texture features
模型 实际类别 预测类别 类别
准确率 (%)总体
准确率(%)正常 粒化 SVM 正常 96 5 95.04% 95.13% 粒化 2 41 95.35% PLS-DA 正常 93 8 92.08% 86.81% 粒化 11 32 74.42% 
下载: 导出CSV
表 2 基于振动频谱特征的PLS-DA和SVM模型对果冻橙粒化病的分类结果
Table 2. Jelly orange granulation disease classification results of PLS-DA and SVM models based on vibration spectral features
模型 实际类别 预测类别 类别
准确率总体
准确率正常 粒化 SVM 正常 96 5 95.04% 89.58% 粒化 10 33 76.74% PLS-DA 正常 97 4 92.08% 90.97% 粒化 9 34 74.42%
下载: 导出CSV
表 3 ResT、Resnet50和ViT的训练集和预测集结果
Table 3. Training and prediction set results for ResT, Resnet50 and ViT
(%) 模型 训练集准确率 预测集准确率 正常 粒化 总体 正常 粒化 总体 ResT 100.00 100.00 100.00 99.01 97.67 98.61 Resnet50 100.00 100.00 100.00 98.02 97.67 97.92 ViT 100.00 98.61 99.53 98.02 95.35 97.22
下载: 导出CSV
-
[1] ZHENG Y J, TIAN SH J, XIE L J. Improving the identification accuracy of sugar orange suffering from granulation through diameter correction and stepwise variable selection[J]. Postharvest Biology and Technology, 2023, 200: 112313. doi: 10.1016/j.postharvbio.2023.112313 [2] THEANJUMPOL P, WONGZEEWASAKUN K, MUENMANEE N, et al. Non-destructive identification and estimation of granulation in ‘Sai Num Pung’ tangerine fruit using near infrared spectroscopy and chemometrics[J]. Postharvest Biology and Technology, 2019, 153: 13-20. doi: 10.1016/j.postharvbio.2019.03.009 [3] 陈玥瑶, 夏静静, 韦芸, 等. 近红外光谱法无损检测平谷产大桃品质方法研究[J]. 分析化学,2023,51(3):454-462.CHEN Y Y, XIA J J, WEI Y, et al. Research on nondestructive quality test of Pinggu peach by near-infrared spectroscopy[J]. Chinese Journal of Analytical Chemistry, 2023, 51(3): 454-462. (in Chinese). [4] 于水, 宦克为, 王磊, 等. 基于卷积神经网络的近红外光谱多组分定量分析模型研究[J]. 分析化学,2024,52(5):695-705.YU SH, HUAN K W, WANG L, et al. Multicomponent quantitative analysis model of near infrared spectroscopy based on convolution neural network[J]. Chinese Journal of Analytical Chemistry, 2024, 52(5): 695-705. (in Chinese). [5] JIE D F, WU SH, WANG P, et al. Research on Citrus grandis granulation determination based on hyperspectral imaging through deep learning[J]. Food Analytical Methods, 2021, 14(2): 280-289. doi: 10.1007/s12161-020-01873-6 [6] NAYAK S L, SETHI S, SAHOO R N, et al. Potential of X-ray imaging to detect citrus granulation in different cultivars with progress in harvesting time[J]. Indian Journal of Experimental Biology, 2022, 60(4): 263-268. [7] KADOWAKI M, NAGASHIMA S, AKIMOTO H, et al. Detection of core rot symptom of Japanese pear (Pyrus pyrifolia cv. Kosui) by a nondestructive resonant method[J]. Journal of the Japanese Society for Horticultural Science, 2012, 81(4): 327-331. doi: 10.2503/jjshs1.81.327 [8] ZHANG H, ZHA ZH H, KULASIRI D, et al. Detection of early core browning in pears based on statistical features in vibro-acoustic signals[J]. Food and Bioprocess Technology, 2021, 14(5): 887-897. doi: 10.1007/s11947-021-02613-2 [9] ZHAO K, LI H, ZHA ZH H, et al. Detection of sub-healthy apples with moldy core using deep-shallow learning for vibro-acoustic multi-domain features[J]. Measurement: Food, 2022, 8: 100068. doi: 10.1016/j.meafoo.2022.100068 [10] JIN CH, XIE L J, YING Y B. Eggshell crack detection based on the time-domain acoustic signal of rolling eggs on a step-plate[J]. Journal of Food Engineering, 2015, 153: 53-62. doi: 10.1016/j.jfoodeng.2014.12.011 [11] SUN L, BI X K, LIN H, et al. On-line detection of eggshell crack based on acoustic resonance analysis[J]. Journal of Food Engineering, 2013, 116(1): 240-245. doi: 10.1016/j.jfoodeng.2012.11.001 [12] WANG D CH, FENG ZH, JI SH Y, et al. Simultaneous prediction of peach firmness and weight using vibration spectra combined with one-dimensional convolutional neural network[J]. Computers and Electronics in Agriculture, 2022, 201: 107341. doi: 10.1016/j.compag.2022.107341 [13] 崔笛, 张文, 应义斌. 激光多普勒测振技术在农产品品质检测中的应用[J]. 农业机械学报,2013,44(7):160-164. doi: 10.6041/j.issn.1000-1298.2013.07.027CUI D, ZHANG W, YING Y B. Applications of laser doppler vibrometer technology in nondestructive detection of agro-product quality[J]. Transactions of the Chinese Society for Agricultural Machinery, 2013, 44(7): 160-164. (in Chinese). doi: 10.6041/j.issn.1000-1298.2013.07.027 [14] FATHIZADEH Z, ABOONAJMI M, BEYGI S R H. Nondestructive firmness prediction of apple fruit using acoustic vibration response[J]. Scientia Horticulturae, 2020, 262: 109073. doi: 10.1016/j.scienta.2019.109073 [15] ABBASZADEH R, RAJABIPOUR A, DELSHAD M, et al. Application of vibration response for the nondestructive ripeness evaluation of watermelons[J]. Australian Journal of Crop Science, 2011, 5(7): 920-925. [16] AKAN A, CURA O K. Time-frequency signal processing: Today and future[J]. Digital Signal Processing, 2021, 119: 103216. doi: 10.1016/j.dsp.2021.103216 [17] FU W L, JIANG X H, LI B L, et al. Rolling bearing fault diagnosis based on 2D time-frequency images and data augmentation technique[J]. Measurement Science and Technology, 2022, 34(4): 045005. [18] HUANG X F, LEI Q, XIE T L, et al. Deep transfer convolutional neural network and extreme learning machine for lung nodule diagnosis on CT images[J]. Knowledge-Based Systems, 2020, 204: 106230. doi: 10.1016/j.knosys.2020.106230 [19] TANG X Y, XU Z B, WANG ZH G. A novel fault diagnosis method of rolling bearing based on integrated vision transformer model[J]. Sensors (Basel, Switzerland), 2022, 22(10): 3878. doi: 10.3390/s22103878 [20] LIU H, LIU ZH Y, JIA W Q, et al. Tool wear estimation using a CNN-transformer model with semi-supervised learning[J]. Measurement Science and Technology, 2021, 32(12): 125010. doi: 10.1088/1361-6501/ac22ee [21] ALSHAMMARI H, GASMI K, BEN LTAIFA I, et al. Olive disease classification based on vision transformer and CNN models[J]. Computational Intelligence and Neuroscience, 2022, 2022: 3998193. [22] ZHAO K, ZHA Z H, LI H, et al. Early detection of moldy apple core based on time-frequency images of vibro-acoustic signals[J]. Postharvest Biology and Technology, 2021, 179: 111589. doi: 10.1016/j.postharvbio.2021.111589 [23] ZHANG W, CUI D, YING Y B. The impulse response method for pear quality evaluation using a laser Doppler vibrometer[J]. Journal of Food Engineering, 2015, 159: 9-15. doi: 10.1016/j.jfoodeng.2015.03.013 [24] DING CH Q, WANG D CH, FENG ZH, et al. Integration of vibration and optical techniques for watermelon firmness assessment[J]. Computers and Electronics in Agriculture, 2021, 187: 106307. doi: 10.1016/j.compag.2021.106307 [25] 张弛, 王顺, 关向雨, 等. 激光多普勒测振技术应用的新进展[J]. 激光与光电子学进展,2022,59(19):1900006.ZHANG CH, WANG SH, GUAN X Y, et al. New progress in application of laser doppler vibration measurement technology[J]. Laser & Optoelectronics Progress, 2022, 59(19): 1900006. (in Chinese). [26] WANG D CH, DING CH Q, FENG ZH, et al. Recent advances in portable devices for fruit firmness assessment[J]. Critical Reviews in Food Science and Nutrition, 2023, 63(8): 1143-1154. doi: 10.1080/10408398.2021.1960477 -
下载:
计量
- 文章访问数: 88
- HTML全文浏览量: 41
- PDF下载量: 4
- 被引次数: 0
