WO2020062471A1 - 液体识别方法、特征提取方法、液体识别装置及存储装置 - Google Patents

液体识别方法、特征提取方法、液体识别装置及存储装置 Download PDF

Info

Publication number
WO2020062471A1
WO2020062471A1 PCT/CN2018/115080 CN2018115080W WO2020062471A1 WO 2020062471 A1 WO2020062471 A1 WO 2020062471A1 CN 2018115080 W CN2018115080 W CN 2018115080W WO 2020062471 A1 WO2020062471 A1 WO 2020062471A1
Authority
WO
WIPO (PCT)
Prior art keywords
domain signal
signal spectrum
time
spectrum
frequency
Prior art date
Application number
PCT/CN2018/115080
Other languages
English (en)
French (fr)
Inventor
谢芳奇
祁春超
谭信辉
Original Assignee
深圳市华讯方舟太赫兹科技有限公司
华讯方舟科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市华讯方舟太赫兹科技有限公司, 华讯方舟科技有限公司 filed Critical 深圳市华讯方舟太赫兹科技有限公司
Publication of WO2020062471A1 publication Critical patent/WO2020062471A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3581—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation
    • G01N21/3586—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation by Terahertz time domain spectroscopy [THz-TDS]

Definitions

  • the present application relates to the field of liquid identification, and in particular, to a liquid identification method, a feature extraction method, a liquid identification device, and a storage device.
  • the sex detection technology mainly includes Raman spectroscopy, X-ray method and microwave method, but its detection equipment is huge and expensive.
  • THz-TDS Terahertz Time-Domain Spectroscopy
  • THz-TDS Time-Domain Spectroscopy
  • Common THz-TDS systems can be divided into two categories: transmissive and reflective.
  • the transmission THz-TDS technology is used to measure the transmission spectrum of various substances and extract the optical parameters.
  • the work for substance identification has been relatively extensive, and the technology and algorithms are relatively mature.
  • the transmissive configuration needs to be in contact with both ends of the sample, and for liquids, especially polar liquids with strong absorption in the terahertz band, the transmissive THz-TDS technology has encountered a big bottleneck in practicality and versatility.
  • the reflection type configuration exhibits superiority in the problem of strong liquid absorption because the terahertz wave only interacts with the liquid on the inner wall interface of the container.
  • the present application provides a liquid identification field based on the present application, and in particular, relates to a liquid identification method, a feature extraction method, a liquid identification device, and a storage device. It can solve the problem of higher liquid identification cost in the prior art.
  • a technical solution adopted in the present application is to provide a liquid identification method, collecting a first time-domain signal spectrum of a liquid to be measured placed in a container, and performing a preliminary analysis on the first time-domain signal spectrum. Processing to obtain a second time-domain signal spectrum; performing feature processing on the second time-domain signal spectrum to obtain a main feature vector of the second time-domain signal spectrum; inputting the main feature vector into a preset classifier for classification; The classification result of the preset classifier is acquired to obtain the type of the liquid to be detected.
  • another technical solution adopted in the present application is to provide a method for extracting a feature of a liquid, including collecting a first time-domain signal spectrum of the liquid; pre-processing the first time-domain signal spectrum to obtain a first A second time domain signal spectrum; performing feature processing on the second time domain signal spectrum to obtain a feature vector of the second time domain signal spectrum; wherein the feature vector is the above-mentioned main feature vector.
  • the device includes a processor, a memory, and a terahertz transceiver.
  • the processor connects the memory to the terahertz transceiver.
  • the terahertz transceiver is configured to collect a first time-domain signal spectrum of a liquid to be measured placed in a container; and the processor is configured to preprocess the first time-domain signal spectrum to obtain a second time domain Signal atlas; performing feature processing on the second time-domain signal atlas to obtain a main feature vector of the second time-domain signal atlas; inputting the main feature vector into a preset classifier for classification; obtaining the preset classification
  • the classification result is obtained by the device to obtain the type of the liquid to be detected.
  • Another technical solution adopted in the present application is to provide a storage device including a program file capable of implementing the above method.
  • the present application collects the first time-domain signal spectrum of the liquid to be measured by collecting the liquid to be measured, and then pre-processes and features the first time-domain signal spectrum to obtain its The main feature vector is input to the preset classifier for classification, thereby obtaining the type of the liquid to be tested.
  • the first time-domain signal patterns obtained by them are also different, thereby extracting the main feature vector of the first time-domain signal pattern corresponding to the liquid to be measured, and providing a preset classification
  • the device performs recognition calculation on the main feature vector to obtain the type of the liquid to be measured, which can greatly reduce the cost, and has a fast recognition speed and a high recognition accuracy rate.
  • FIG. 1 is a schematic flowchart of a first embodiment of a liquid identification method according to the present application
  • FIG. 2 is a schematic diagram of collecting a time-domain map of a liquid to be measured in the liquid identification method of the present application
  • FIG. 3 is a schematic diagram of a time domain map collected by the liquid identification method of the present application.
  • FIG. 4 is a schematic flowchart of a second embodiment of a liquid identification method according to the present application.
  • FIG. 5 is a schematic flowchart of a third embodiment of a liquid identification method according to the present application.
  • FIG. 6 is a schematic flowchart of a first embodiment of a liquid feature extraction method according to the present application.
  • FIG. 7 is a schematic block diagram of a structure of an embodiment of a liquid identification device of the present application.
  • FIG. 8 is a schematic structural diagram of an embodiment of a storage device of the present application.
  • FIG. 1 is a schematic flowchart of a first embodiment of a liquid identification method according to the present application, which specifically includes the following steps:
  • the liquid containers on the market are mainly metal containers, plastic containers or paper containers.
  • the liquids are mainly safe liquids such as water and beverages and dangerous liquids such as petroleum. Based on the characteristics of reflective terahertz, they are better for plastic containers.
  • the penetrability, the container used in this application is a plastic container.
  • the terahertz transceiver 12 in the liquid identification device provided by the present application emits a terahertz wave THZ, and the terahertz wave THZ first contacts the container 110. Due to different media, part of it is reflected by the surface of the container 110 and part of it is transmitted. Then it comes into contact with the liquid to be measured 120 and is reflected by the liquid to be measured 120.
  • the echo carries information about the container and the liquid in the container, including THZ1 reflected by the outer wall of the container and THZ2 reflected by the liquid.
  • the receiver 12 further receives the reflected echo, thereby obtaining a first time-domain signal spectrum as shown in FIG. 3.
  • the first time-domain signal spectrum is further subjected to denoising processing and clipping peak processing to obtain a second time-domain signal spectrum with a higher signal-to-noise ratio.
  • FIG. 4 is a schematic flowchart of a second embodiment of a liquid identification method according to the present application, which is a sub-step of step S12 in FIG. 1, and specifically includes the following steps:
  • Gaussian noise in the first time-domain signal spectrum is removed by using an orthogonal wavelet transform method to obtain a third time-domain signal spectrum.
  • the Gaussian noise in the first time-domain signal spectrum is removed by an orthogonal wavelet transform method, so as to obtain a third time-domain signal spectrum with a higher relative signal-to-noise ratio.
  • the third time domain signal spectrum is intercepted to obtain a second time domain signal spectrum.
  • the maximum reflection peak information with the highest peak value in the third time domain signal spectrum is obtained, specifically, the position information of the peak with the highest peak value in the third time domain signal spectrum, and then based on the maximum reflection peak information in the third time domain.
  • the second largest peak information of the peak is obtained from the signal spectrum; that is, the third largest peak information is obtained from the third time domain signal spectrum after the maximum reflection peak is excluded. After the second peak information is obtained, the second largest peak information is obtained.
  • the interception is performed, that is, the maximum peak is extracted from the third time domain signal spectrum to obtain the second time domain signal spectrum.
  • the first time-domain signal spectrum obtained includes the reflection signal of the container and the reflection signal of the liquid, and what is needed is the reflection signal of the liquid, that is, the second reflection signal, and the first time (the The container reflection) is the peak with the largest peak and the second is the second largest peak. Therefore, a time-domain signal spectrum carrying liquid information is obtained through the above method.
  • the peak with the highest peak value may be the peak reflected by the liquid, so the first and second peaks of the peak value can be obtained, and the time relationship is later.
  • the peaks of T are intercepted to obtain the second time domain signal spectrum.
  • the second time-domain signal spectrum is further characterized to obtain a main feature vector of the second time-domain signal spectrum.
  • FIG. 5 is a schematic flowchart of a third embodiment of the liquid identification method of the present application, which is a sub-step of step S13 of FIG. 1, and specifically includes the following steps:
  • S131 Perform a fast Fourier transform on the second time-domain signal spectrum to obtain a first frequency-domain signal spectrum.
  • the first frequency-domain signal spectrum After performing fast Fourier transform on the second time-domain signal spectrum, the first frequency-domain signal spectrum is obtained. Compared to the second time-domain signal spectrum, the mathematical structure of the first frequency-domain signal spectrum is more obvious and can be passed through the mathematical structure. Intuitively reflect the characteristic information of its second time domain signal spectrum.
  • S132 Perform normalization processing on the frequency spectrum of the first frequency domain signal to obtain the frequency spectrum of the second frequency domain signal.
  • the first frequency domain signal spectrum is normalized to obtain the second frequency domain signal spectrum.
  • the values of the variance and the mean of the first frequency domain signal spectrum are obtained.
  • y i is the frequency spectrum of the first frequency domain signal
  • ⁇ is the average value of the frequency spectrum of the first frequency domain signal
  • ⁇ is the variance of the frequency spectrum of the first frequency domain signal
  • x (n) is the frequency spectrum of the second frequency domain signal
  • a terahertz transceiver before performing acquisition, may also be used to acquire and process the metal reflector to obtain a reference signal. Thereby, it serves as a reference signal for normalization processing.
  • S133 Perform principal component analysis on the frequency spectrum of the second frequency domain signal to extract a main feature vector of the frequency spectrum of the second frequency domain signal.
  • PCA Principal Component Analysis
  • the covariance matrix C and the mean value x of the second frequency domain signal spectrum are obtained; then the eigenvalues D and eigenvectors V of the covariance matrix C are obtained; Specifically, the feature values D are arranged in rows from top to bottom, and then the k-dimensional feature vectors V corresponding to the first k feature values D are formed to form a feature matrix P.
  • D i is the covariance matrix C for the i-th eigenvalue matrix after the descending order, and k ⁇ n; and k is a positive integer.
  • the main feature vector F is obtained by projecting the second frequency domain signal spectrum x (n) onto the feature matrix P.
  • the projection process is shown by the following formula:
  • F is the main eigenvector
  • x (n) is the frequency spectrum of the second frequency domain signal
  • x is the mean value obtained above
  • P is the eigen matrix.
  • the more important feature information of the second frequency domain signal spectrum x (n) can be extracted and projected according to the contribution degree, thereby obtaining the principal feature vector.
  • the main feature vector is input to a preset classifier for classification.
  • the main feature vector is further input into a preset classifier for classification.
  • the preset classifier provided in this application is a feature vector with multiple groups of liquids is performed in a support vector machine classifier (SVM).
  • SVM support vector machine classifier
  • the classifier model obtained through training.
  • the SVM classifier uses a radial kernel function.
  • the SVM (Support Vector Machine) classifier is a machine learning classifier based on the SVM method.
  • the principle of the SVM method is that it analyzes linearly separable cases. For linearly inseparable cases, by using a non-linear mapping algorithm, Linearly inseparable samples in the low-dimensional input space are transformed into high-dimensional feature spaces to make them linearly separable, which makes it possible to perform linear analysis of the non-linear features of the samples using a linear algorithm in the high-dimensional feature space. And based on the theory of structural risk minimization, an optimal hyperplane is constructed in the feature space, so that the learner is globally optimized, and the expectations in the entire sample space meet a certain upper bound with a certain probability. That is, it has better adaptability and generalization performance, and can better deal with nonlinear problems.
  • the main feature vectors of these combined liquids are extracted as a training group, which can include multi-level labels, such as first-level labels can be safe, including safety and danger, and second-level labels can be types, such as safe beverage-type liquids or edible oils, etc.
  • a training group can include multi-level labels, such as first-level labels can be safe, including safety and danger, and second-level labels can be types, such as safe beverage-type liquids or edible oils, etc.
  • the liquid corresponding to its training group and the labels corresponding to each liquid are input to multiple SVM classifiers for training, and then cross-validation is used to Determine the recognition accuracy of its multiple SVM classifier models, and use the best one as the preset classifier.
  • a deep neural network may also be used.
  • a Deep Neural Network is a neural network with multiple hidden layers. Compared with a traditional classifier, it has a large-scale parallel distributed Structure, so the model has better adaptability and generalization performance, and can better deal with non-linear problems.
  • the main feature vector of the liquid group and its corresponding label are input to a deep neural network for training to obtain a classifier model. , And confirm the recognition accuracy of multiple classifier models through cross-validation, etc., and use the best one as the preset classifier.
  • the recognition result includes safety and type.
  • safety includes safety Or dangerous
  • type includes its specific category, such as beverages, or can be further specific to herbal tea.
  • the type of the liquid to be tested is a safe liquid or a dangerous liquid. If it is in a safe liquid, it can be further obtained that it is a herbal tea in a beverage.
  • the preset classifier provided in this application can also be retrained during work, thereby enhancing its recognition ability and accuracy.
  • FIG. 6 is a schematic flowchart of a first embodiment of a feature extraction method of the present application, which specifically includes the following steps:
  • the first time-domain signal spectrum of the liquid is collected, the first terahertz signal is transmitted to the liquid (stored in the container), and the second terahertz signal reflected by the liquid is received.
  • the first time-domain signal spectrum is further subjected to denoising processing and clipping peak processing to obtain a second time-domain signal spectrum with a higher signal-to-noise ratio.
  • S23 Perform feature processing on the second time-domain signal spectrum to obtain a feature vector of the second time-domain signal spectrum.
  • the second time-domain signal spectrum is further characterized to obtain a feature vector of the second time-domain signal spectrum.
  • FIG. 7 is a schematic block diagram of a liquid identification device according to an embodiment of the present application.
  • the liquid identification device provided in this embodiment specifically includes a processor 10, a memory 11, and a terahertz transceiver 12, where the processor 10 is connected to the memory 11 and the terahertz transceiver 12.
  • the terahertz transceiver 12 is configured to collect a first time-domain signal spectrum of the liquid to be measured.
  • the processor 10 may also be referred to as a CPU (Central Processing Unit).
  • the processor 10 may be an integrated circuit chip and has a signal processing capability.
  • the processor 10 may also be a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, and a discrete hardware component .
  • a general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the processor 10 may be configured to preprocess the first time domain signal spectrum to obtain a second time domain signal spectrum; perform feature processing on the second time domain signal spectrum to obtain a main feature of the second time domain signal spectrum.
  • Vector input the main feature vector into a preset classifier for classification; obtain the classification result of the preset classifier to obtain the type of the liquid to be detected.
  • the module terminals of the foregoing devices may specifically perform the corresponding steps in the foregoing method embodiments, so the details of each module are not described here. For details, refer to the description of the corresponding steps above.
  • FIG. 8 is a schematic structural diagram of an embodiment of a storage device of the present application.
  • the instruction file 21 can be stored in the storage device in the form of a software product, and each record is also recorded.
  • This kind of calculated data includes several instructions for causing a computer device (which may be a personal computer, a server, an intelligent robot, or a network device) or a processor to perform all or part of the steps of the methods in the embodiments of the present application.
  • the instruction file 21 also has certain independence, and can continue to cooperate with the processor 10 to execute related instructions when the operating system or the backup system fails, and will not be replaced, damaged and emptied during the upgrade, boot program upgrade and repair.
  • the aforementioned storage devices include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other media that can store program codes Or terminal devices such as computers, servers, mobile phones, and tablets.
  • this application collects the first time-domain signal spectrum of the liquid to be measured by collecting the liquid to be measured, and then pre-processes and features the first time-domain signal spectrum to obtain its main feature vector, and the main feature is The vector is input to a preset classifier for classification, so as to obtain the type of the liquid to be tested.
  • the first time-domain signal patterns obtained by them are also different, thereby extracting the main feature vector of the first time-domain signal pattern corresponding to the liquid to be measured, and providing a preset classification
  • the device performs recognition calculation on the main feature vector to obtain the type of the liquid to be measured, which can greatly reduce the cost, and has a fast recognition speed and a high recognition accuracy rate.

Landscapes

  • Physics & Mathematics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Toxicology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

一种液体识别方法、特征提取方法、液体识别装置及存储装置，该识别方法包括采集放置在容器中的待测液体的第一时域信号图谱（S11）；对第一时域信号图谱进行预处理获得第二时域信号图谱（S12）；对第二时域信号图谱进行特征处理以获得第二时域信号图谱的主特征向量（S13）；将主特征向量输入预设分类器进行分类（S14）；获取预设分类器分类结果，以得到待检测液体的类型（S15）。该方法利用不同液体对太赫兹的反射性能不同，其得到的第一时域信号图谱也不同，从而提取出待测液体所对应的第一时域信号图谱的主特征向量，并提供预设分类器来对主特征向量进行识别计算，从而获得待测液体的类型，能够大大缩减成本，且识别速度快，识别准确率高。

Description

液体识别方法、特征提取方法、液体识别装置及存储装置 【技术领域】
本申请涉及液体识别领域，特别是涉及一种液体识别方法、特征提取方法、液体识别装置及存储装置。
【背景技术】
为了消除易燃易爆液体在机场、地铁站、火车站等公共场所引起的安全隐患以及满足快速、准确、无损的检测要求，公安机关迫切需要更加先进的技术来进行安检，已有的非损伤性检测技术主要有拉曼光谱法、X射线法和微波法，但是其检测的设备庞大，且价格昂贵。
太赫兹时域光谱(Terahertz Time-Domain Spectroscopy,THz-TDS)技术是一门相对较新的相干探测技术，已被广泛用于航空航天、生物医学、物质检测、国土安全等领域。常见的THz-TDS***可分为两类：透射式和反射式。利用透射式THz-TDS技术测量各类物质的透射谱，提取光学参数，用于物质识别的工作已相对较多，且技术和算法也较为成熟。然而透射式配置需要和样品两端接触，对液体特别是在太赫兹波段具有很强吸收的极性液体，透射式THz-TDS技术在实用性和通用性上遇到了很大的瓶颈。相对而言，反射式配置由于太赫兹波仅在容器内壁界面上和液体作用，在液体强吸收的问题上展现出了优越性。
【发明内容】
本申请提供一种基于本申请涉及液体识别领域，特别是涉及一种液体识别方法、特征提取方法、液体识别装置及存储装置。能够解决现有技术中液体识别成本较高的问题。
为解决上述技术问题，本申请采用的一个技术方案是：提供一种液体识别方法，采集放置在容器中的待测液体的第一时域信号图谱；对所述第一时域信号图谱进行预处理获得第二时域信号图谱；对所述第二时域信号图谱进行特征 处理以获得所述第二时域信号图谱的主特征向量；将所述主特征向量输入预设分类器进行分类；获取所述预设分类器分类结果，以得到所述待检测液体的类型。
为解决上述技术问题，本申请采用的另一个技术方案是：提供一种液体的特征提取方法，包括采集液体的第一时域信号图谱；对所述第一时域信号图谱进行预处理获得第二时域信号图谱；对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的特征向量；其中，所述特征向量为上述的主特征向量。
为解决上述技术问题，本申请采用的另一个技术方案是：提供一种液体识别装置，该装置包括处理器、存储器以及太赫兹收发器，所述处理器连接所述存储器与所述太赫兹收发器；其中，所述太赫兹收发器用于采集放置在容器中的待测液体的第一时域信号图谱；所述处理器用于对所述第一时域信号图谱进行预处理获得第二时域信号图谱；对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的主特征向量；将所述主特征向量输入预设分类器进行分类；获取所述预设分类器分类结果，以得到所述待检测液体的类型。
为解决上述技术问题，本申请采用的另一个技术方案是：提供一种存储装置，包括能实现上述方法的程序文件。
本申请的有益效果是：区别于现有技术，本申请通过采集待测液体的采集待测液体的第一时域信号图谱，随后对第一时域信号图谱进行预处理与特征处理从而获取其主特征向量，并将主特征向量输入到预设分类器进行分类，从而获取待测液体的类型。通过利用不同液体的对太赫兹的反射性能不同，其得到的第一时域信号图谱也不同，从而提取出待测液体所对应的第一时域信号图谱的主特征向量，并提供预设分类器来对主特征向量进行识别计算，从而获得待测液体的类型，能够大大缩减成本，且识别速度快，识别准确率高。
【附图说明】
图1是本申请液体识别方法的第一实施方式的流程示意图；
图2是本申请液体识别方法中采集待测液体的时域图谱的示意图；
图3是本申请液体识别方法所采集的时域图谱的示意图；
图4是本申请液体识别方法的第二实施方式的流程示意图；
图5是本申请液体识别方法的第三实施方式的流程示意图；
图6是本申请液体特征提取方法的第一实施方式的流程示意图；
图7是本申请液体识别装置一实施方式的结构示意框图；
图8是本申请存储装置一实施方式的结构示意图。
【具体实施方式】
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅是本申请的一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
另外，若本申请实施例中有涉及“第一”、“第二”等的描述，则该“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本实用新型要求的保护范围之内。
请参阅图1，图1是本申请一种液体识别方法的第一实施例流程示意图，其具体包括如下步骤：
S11，采集放置在容器中的待测液体的第一时域信号图谱。
目前，市面上的液体容器主要有金属容器、塑料容器或者纸质容器，其液体主要有水、饮料等安全液体以及石油等危险液体，基于反射式太赫兹的特性，其对塑料容器有较好的穿透性，本申请采用的容器为塑料容器。
如图2，本申请提供的液体识别装置中的太赫兹收发器12发出太赫兹波THZ，其太赫兹波THZ首先接触到容器110，由于介质不同，其一部分被容器110表面反射，一部分透射，随后接触到待测液体120，从而被待测液体120反射，其中回波携带了关于容器以及容器内液体的信息，包括了被容器外壁反射回来的THZ1与被液体反射回来的THZ2，太赫兹收发器12进一步接收被反射回来的回波，从而获取到如图3所示的第一时域信号图谱。
S12，对第一时域信号图谱进行预处理获得第二时域信号图谱。
在获取到第一时域信号图谱后，进一步对第一时域信号图谱进行去噪处理与截取峰值处理，以获得信噪比更高的第二时域信号图谱。
请参阅图4，图4是本申请液体识别的方法的第二实施例流程示意图，其是图1步骤S12的子步骤，其具体包括如下步骤：
S121，通过正交小波变换法去除第一时域信号图谱中的高斯噪声；以得到第三时域信号图谱。
首先，通过正交小波变换法去除第一时域信号图谱中的高斯噪声，从而得到相对信噪比更高的第三时域信号图谱，
S122，对第三时域信号图谱进行截取化处理，以得到第二时域信号图谱。
进一步的，获取所述第三时域信号图谱中峰值最大的最大反射峰信息，具体是在第三时域信号图谱中峰值最大的峰的位置信息，随后根据最大反射峰信息在第三时域信号图谱中获取峰值第二的次最大峰信息；即在将最大反射峰的进行排除后剩下的第三时域信号图谱中获取峰值最大的反射峰信息，在获取后，将次最大峰信息进行截取，即将该次最大峰从第三时域信号图谱进行提取，从而得到第二时域信号图谱。
由于在采集过程中，所获得的第一时域信号图谱包括容器的反射信号和液体的反射信号，而需要的是液体的反射信号，即其第二次的反射信号，其第一次(被容器反射)为峰值最大的峰，第二次为次最大峰，因此通过上述方法，获取到了携带液体信息的时域信号图谱。
在具体实施例中，对于某些反射性较高的液体而言，其峰值最大的峰可能是液体反射的峰，因此可以获取峰值排名第一和第二的峰，并将时间关系上靠后的峰进行截取从而得到第二时域信号图谱。
S13，对第二时域信号图谱进行特征处理以获得第二时域信号图谱的主特征向量。
在获取到信噪比较高的第二时域信号图谱后，进一步对第二时域信号图谱进行特征化处理，从而获得第二时域信号图谱的主特征向量。
请参阅图5，图5是本申请液体识别方法的第三实施例流程示意图，其是图1步骤S13的子步骤，其具体包括如下步骤：
S131，将第二时域信号图谱进行快速傅里叶变换以获得第一频域信号频谱。
将第二时域信号图谱进行快速傅里叶变换后，获取第一频域信号频谱，相对第二时域信号图谱而言，第一频域信号频谱的数学构造更为明显，能够通过数学结构直观的反应其第二时域信号图谱的特征信息。
S132，对第一频域信号频谱进行归一化处理以获得第二频域信号频谱。
随后对第一频域信号频谱进行归一化处理以获得第二频域信号频谱，首先求出第一频域信号频谱的方差与均值的数值。
随后可以根据如下的公式进行处理：
Figure PCTCN2018115080-appb-000001
其中，y i为第一频域信号频谱，μ为第一频域信号频谱的均值，所述δ为第一频域信号频谱的方差，所述x(n)为第二频域信号频谱，其得到的第二频域信号频谱的数值在[-1,1]之间。
在具体实施例中，在进行采集之前，还可以采用太赫兹收发器对金属反射镜进行采集并处理，获取一个基准化信号。从而作为归一化处理的基准化信号。
S133，对第二频域信号频谱进行主成分分析以提取第二频域信号频谱主特征向量。
主成分分析(Principal Component Analysis，PCA)是一种常用的数据降维、特征选择方法，它可以消除数据之间存在的相关性，进而从多维数据中筛选出对分类贡献率大的特征，减少计算量，提升识别率。
首先获取第二频域信号频谱的协方差矩阵C与均值x；随后获取协方差矩阵C的特征值D与特征向量V；按照特征值D的大小对特征值D对应的特征向量V进行排列，具体是按照特征值D的大小从上到下按行进行排列，随后取前k个特征值D对应的k维特征向量V组成特征矩阵P。
其中，关于k的值的确认公式如下：
Figure PCTCN2018115080-appb-000002
其中，所述D i为协方差矩阵C进行降序排列后的矩阵的第i个特征值，且k≤n；且k为正整数。
进一步的，通过将第二频域信号频谱x(n)向特征矩阵P进行投影得到主特征向量F。投影过程如下公式所示：
Figure PCTCN2018115080-appb-000003
其中，F为主特征向量，x(n)为第二频域信号频谱，x为上述求出的均值，P为特征矩阵。
经过上述主成分分析后，能够根据贡献度的大小将第二频域信号频谱x(n)较为重要的特征信息进行提取，并进行投影后，从而得到主特征向量。
S14，将主特征向量输入预设分类器进行分类。
在获取到主特征向量后，进一步将主特征向量输入到预设分类器中进行分类，本申请提供的预设分类器是具有多组液体的特征向量在支持向量机分类器(SVM)中进行训练而获得的分类器模型，在实施例中，SVM分类器采用径向核函数。
SVM(Support Vector Machine)分类器是一种基于SVM方法的机器学***面，使得学习器得到全局优化，并且在整个样本空间的期望以某个概率满足一定上界。即它具有更好的自适应性和泛化性能，能够更好地处理非线性问题。
其具体可以通过塑料容器以及不同的液体，如市面上常见的液体，如水、饮料等安全液体，汽油、敌敌畏等危险液体及其混合液体，将其各进行组合形成从而作为初始训练样本，具体的提取这些组合液体的主特征向量作为训练组，，其可以包括多级标签，如一级标签可以安全性，包括安全和危险，二级标签可以为种类，如安全下的饮料型液体或者食用油等，还可以进一步包括三级标签，如饮料型液体中的牛奶等，将其训练组对应的液体及其各液体所对应的标签输入到多个SVM分类器进行训练，随后通过交叉验证的方法来确定其多个SVM分类器模型的识别准确性，并将最佳的那个作为预设分类器。
在其他实施例中，也可以采用深层神经网络，深层神经网络(Deep Neural Network，DNN)是具有多个隐含层的神经网络，相比于传统的分类器，它具有大规模并行的分布式结构，因而模型具有更好的自适应性和泛化性能，能够更好地处理非线性问题，同样将上述液体组的主特征向量及其对应的标签输入到深层神经网络进行训练得到分类器模型，并通过交叉验证来确认多个分类器模型的识别准确性等，并将最佳的那个作为预设分类器。
S15，获取预设分类器分类结果，以得到待检测液体的类型。
因此在获取到主特征向量后，将其输入预设分类器进行分类，则可以得到该主特征向量所对应的待检测液体的识别结果，其识别结果包括安全性和种类，如其安全性包括安全或危险，其种类包括其具体的类别，如饮料类，或者可以进一步具体到凉茶。
进一步的，也就知道待检测液体的类型，是否是安全类液体，还是危险类液体。如果其在安全类液体下，进一步可以得到其是饮料类中的凉茶。
在具体实施例中，本申请提供的预设分类器也可以在工作中进行再次训练， 从而加强其识别能力与准确度。
请参阅图6，图6是本申请特征提取方法的第一实施例流程示意图，其具体包括如下步骤：
S21，采集液体的第一时域信号图谱。
采集液体的第一时域信号图谱，对液体(存储在容器中的)发射第一太赫兹信号，并接收被液体反射回来的第二太赫兹信号。
S22，对第一时域信号图谱进行预处理获得第二时域信号图谱。
在获取到第一时域信号图谱后，进一步对第一时域信号图谱进行去噪处理与截取峰值处理，以获得信噪比更高的第二时域信号图谱。
S23，对第二时域信号图谱进行特征处理以获得第二时域信号图谱的特征向量。
在获取到信噪比较高的第二时域信号图谱后，进一步对第二时域信号图谱进行特征化处理，从而获得第二时域信号图谱的特征向量。
本实施例的具体提取方法在上述实施例中已经有叙述，这里不再赘述，且需要知道的是，本实施例的特征提取方法所提取的特征向量为上述实施例中的主特征向量，其也可以应用于上述任一实施例中的识别方法。
请参阅图7，图7是本申请液体识别装置的一实施方式结构示意框图。
本实施例提供的液体识别装置具体包括处理器10、存储器11、太赫兹收发器12，其中，处理器10连接存储器11与太赫兹收发器12。
在本实施例中，太赫兹收发器12用于采集待测液体的第一时域信号图谱。
其中，处理器10还可以称为CPU(Central Processing Unit，中央处理单元)。处理器10可能是一种集成电路芯片，具有信号的处理能力。处理器10还可以是通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现成可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
在本实施中，处理器10可以用于对第一时域信号图谱进行预处理获得第二时域信号图谱；对第二时域信号图谱进行特征处理以获得第二时域信号图谱的 主特征向量；将主特征向量输入预设分类器进行分类；获取所述预设分类器分类结果，以得到所述待检测液体的类型。
上述设备的模块终端可分别具体执行上述方法实施例中对应的步骤，故在此不对各模块进行赘述，详细请参阅以上对应步骤的说明。
参阅图8，图8为本申请存储装置一实施方式的结构示意图，有能够实现上述所有方法的指令文件21，该指令文件21可以以软件产品的形式存储在上述存储装置中，同时还是记录各种计算的数据，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，智能机器人，或者网络设备等)或处理器(processor)执行本申请各个实施方式方法的全部或部分步骤。
所述指令文件21还具有一定独立性，可以在运行***、备份***发生故障时候继续配合处理器10执行相关指令，在升级、引导程序升级以及修复中不会被替换、损坏以及清空。
而前述的存储装置包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质，或者是计算机、服务器、手机、平板等终端设备。
综上所述，本申请通过采集待测液体的采集待测液体的第一时域信号图谱，随后对第一时域信号图谱进行预处理与特征处理从而获取其主特征向量，并将主特征向量输入到预设分类器进行分类，从而获取待测液体的类型。通过利用不同液体的对太赫兹的反射性能不同，其得到的第一时域信号图谱也不同，从而提取出待测液体所对应的第一时域信号图谱的主特征向量，并提供预设分类器来对主特征向量进行识别计算，从而获得待测液体的类型，能够大大缩减成本，且识别速度快，识别准确率高。
以上仅为本申请的实施方式，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结果或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (18)

  1. 一种基于反射式太赫兹的液体识别方法，其特征在于，所述方法包括：
    采集放置在容器中的待测液体的第一时域信号图谱；
    对所述第一时域信号图谱进行预处理获得第二时域信号图谱；
    对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的主特征向量；
    将所述主特征向量输入预设分类器进行分类；
    获取所述预设分类器分类结果，以得到所述待检测液体的类型；
    其中，所述预设分类器是通过将获取到的预设数量的多组液体的主特征向量在支持向量机分类器或者深层神经网络进行训练而获得的分类器模型。
  2. 根据权利要求1所述的识别方法，其特征在于，所述对所述第一时域信号图谱进行预处理获得第二时域信号图谱包括：
    通过正交小波变换法去除所述第一时域信号图谱中的高斯噪声；
    以得到第三时域信号图谱；
    对所述第三时域信号图谱进行截取化处理，以得到第二时域信号图谱。
  3. 根据权利要求2所述的识别方法，其特征在于，所述对所述第三时域信号图谱进行截取化处理，以得到第二时域信号图谱包括：
    获取所述第三时域信号图谱中峰值最大的最大反射峰信息；
    根据所述最大反射峰信息在所述第三时域信号图谱中获取峰值第二的次最大峰信息；
    将所述次最大峰信息进行提取，以得到所述第二时域信号图谱。
  4. 根据权利要求1所述的识别方法，其特征在于，所述对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的主特征向量包括：
    将所述第二时域信号图谱进行快速傅里叶变换以获得第一频域信号频谱；
    对所述第一频域信号频谱进行归一化处理以获得第二频域信号频谱；
    对所述第二频域信号频谱进行主成分分析以提取所述第二频域信号频谱主 特征向量。
  5. 根据权利要求4所述的识别方法，其特征在于，所述对所述第一频域信号频谱进行归一化处理以获得第二频域信号频谱包括:
    根据如下公式对对所述第一频域信号频谱进行归一化处理以获得第二频域信号频谱:
    Figure PCTCN2018115080-appb-100001
    其中，所述y i为第一频域信号频谱，所述μ为第一频域信号频谱的均值，所述δ为第一频域信号频谱的方差，所述x(n)为第二频域信号频谱。
  6. 根据权利要求4所述的识别方法，其特征在于，所述对所述第二频域信号频谱进行主成分分析以提取所述第二频域信号频谱主特征向量包括：
    获取所述第二频域信号频谱的协方差矩阵与均值；
    获取所述协方差矩阵的特征值与特征向量；
    按照所述特征值的大小对所述特征值对应的所述特征向量进行排列，并取其前k个最大的特征值所对应的k维特征向量组成以获得特征矩阵；
    根据如下公式求出所述k的值；
    Figure PCTCN2018115080-appb-100002
    其中，所述D i为降序排列后矩阵的第i个特征值，且k≤n；且k为正整数；
    根据如下公式将第二频域信号频谱向特征矩阵进行投影得到所述主特征向量：
    Figure PCTCN2018115080-appb-100003
    其中，所述F为所述主特征向量，所述x(n)为所述第二频域信号频谱，所述
    Figure PCTCN2018115080-appb-100004
    为所述均值，所述P为所述特征矩阵。
  7. 根据权利要求1所述的识别方法，其特征在于，所述采集放置在容器中的待测液体的第一时域信号图谱包括：
    对所述放置在容器中的待测液体发射太赫兹波；
    获取经由所述容器与所述待测液体进行反射的回波；
    将所述回波的图谱作为所述待测液体的第一时域信号图谱。
  8. 一种基于反射式太赫兹的液体识别方法，其特征在于，所述方法包括：
    采集放置在容器中的待测液体的第一时域信号图谱；
    对所述第一时域信号图谱进行预处理获得第二时域信号图谱；
    对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的主特征向量；
    将所述主特征向量输入预设分类器进行分类；
    获取所述预设分类器分类结果，以得到所述待检测液体的类型。
  9. 根据权利要求8所述的识别方法，其特征在于，所述对所述第一时域信号图谱进行预处理获得第二时域信号图谱包括：
    通过正交小波变换法去除所述第一时域信号图谱中的高斯噪声；
    以得到第三时域信号图谱；
    对所述第三时域信号图谱进行截取化处理，以得到第二时域信号图谱。
  10. 根据权利要求9所述的识别方法，其特征在于，所述对所述第三时域信号图谱进行截取化处理，以得到第二时域信号图谱包括：
    获取所述第三时域信号图谱中峰值最大的最大反射峰信息；
    根据所述最大反射峰信息在所述第三时域信号图谱中获取峰值第二的次最大峰信息；
    将所述次最大峰信息进行提取，以得到所述第二时域信号图谱。
  11. 根据权利要求8所述的识别方法，其特征在于，所述对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的主特征向量包括：
    将所述第二时域信号图谱进行快速傅里叶变换以获得第一频域信号频谱；
    对所述第一频域信号频谱进行归一化处理以获得第二频域信号频谱；
    对所述第二频域信号频谱进行主成分分析以提取所述第二频域信号频谱主特征向量。
  12. 根据权利要求11所述的识别方法，其特征在于，所述对所述第一频域 信号频谱进行归一化处理以获得第二频域信号频谱包括:
    根据如下公式对对所述第一频域信号频谱进行归一化处理以获得第二频域信号频谱:
    Figure PCTCN2018115080-appb-100005
    其中，所述y i为第一频域信号频谱，所述μ为第一频域信号频谱的均值，所述δ为第一频域信号频谱的方差，所述x(n)为第二频域信号频谱。
  13. 根据权利要求11所述的识别方法，其特征在于，所述对所述第二频域信号频谱进行主成分分析以提取所述第二频域信号频谱主特征向量包括：
    获取所述第二频域信号频谱的协方差矩阵与均值；
    获取所述协方差矩阵的特征值与特征向量；
    按照所述特征值的大小对所述特征值对应的所述特征向量进行排列，并取其前k个最大的特征值所对应的k维特征向量组成以获得特征矩阵；
    根据如下公式求出所述k的值；
    Figure PCTCN2018115080-appb-100006
    其中，所述D i为降序排列后矩阵的第i个特征值，且k≤n；且k为正整数；
    根据如下公式将第二频域信号频谱向特征矩阵进行投影得到所述主特征向量：
    Figure PCTCN2018115080-appb-100007
    其中，所述F为所述主特征向量，所述x(n)为所述第二频域信号频谱，所述
    Figure PCTCN2018115080-appb-100008
    为所述均值，所述P为所述特征矩阵。
  14. 根据权利要求8所述的识别方法，其特征在于，所述采集放置在容器中的待测液体的第一时域信号图谱包括：
    对所述放置在容器中的待测液体发射太赫兹波；
    获取经由所述容器与所述待测液体进行反射的回波；
    将所述回波的图谱作为所述待测液体的第一时域信号图谱。
  15. 根据权利要求8所述的液体识别方法，其特征在于，所述预设分类器是通过将获取到的预设数量的多组液体的主特征向量在支持向量机分类器或者深层神经网络进行训练而获得的分类器模型。
  16. 一种基于反射式太赫兹的液体特征提取的方法，其特征在于，所述方法包括：
    采集液体的第一时域信号图谱；
    对所述第一时域信号图谱进行预处理获得第二时域信号图谱；
    对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的特征向量；
    其中，所述特征向量为权利要求1-8任一项所述的主特征向量。
  17. 一种液体识别装置，其特征在于，所述液体识别装置包括处理器、存储器以及太赫兹收发器，所述处理器连接所述存储器与所述太赫兹收发器；
    其中，所述太赫兹收发器用于采集放置在容器中的待测液体的第一时域信号图谱；
    所述处理器用于对所述第一时域信号图谱进行预处理获得第二时域信号图谱；对所述第二时域信号图谱进行特征处理以获得所述第二时域信号图谱的主特征向量；将所述主特征向量输入预设分类器进行分类；获取所述预设分类器分类结果，以得到所述待检测液体的类型。
  18. 一种存储装置，其特征在于，所述存储装置包括能实现权利要求1-17中任一项权利要求的程序文件。
PCT/CN2018/115080 2018-09-29 2018-11-12 液体识别方法、特征提取方法、液体识别装置及存储装置 WO2020062471A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811150314.4A CN109406445A (zh) 2018-09-29 2018-09-29 液体识别方法、特征提取方法、液体识别装置及存储装置
CN201811150314.4 2018-09-29

Publications (1)

Publication Number Publication Date
WO2020062471A1 true WO2020062471A1 (zh) 2020-04-02

Family

ID=65465645

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/115080 WO2020062471A1 (zh) 2018-09-29 2018-11-12 液体识别方法、特征提取方法、液体识别装置及存储装置

Country Status (2)

Country Link
CN (1) CN109406445A (zh)
WO (1) WO2020062471A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114088656A (zh) * 2020-07-31 2022-02-25 中国科学院上海高等研究院 太赫兹光谱物质识别方法及***、存储介质及终端

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110044839A (zh) * 2019-04-25 2019-07-23 华太极光光电技术有限公司 瓶装液体的检测方法与检测***
CN112485218A (zh) * 2020-11-05 2021-03-12 电子科技大学中山学院 一种基于人工神经网络的太赫兹危险液体识别方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120191371A1 (en) * 2011-01-25 2012-07-26 University Of Washington Through Its Center For Commercialization Terahertz spectroscopy of rough surface targets
CN103941254A (zh) * 2014-03-03 2014-07-23 中国神华能源股份有限公司 一种基于地质雷达的土壤物性类别识别方法和装置
CN104897605A (zh) * 2015-06-16 2015-09-09 中国人民解放军国防科学技术大学 一种基于改进支持向量机的太赫兹谱分类识别方法
CN106645014A (zh) * 2016-09-23 2017-05-10 上海理工大学 基于太赫兹光谱的物质识别方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6646519B2 (ja) * 2016-05-20 2020-02-14 浜松ホトニクス株式会社 全反射分光計測装置及び全反射分光計測方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120191371A1 (en) * 2011-01-25 2012-07-26 University Of Washington Through Its Center For Commercialization Terahertz spectroscopy of rough surface targets
CN103941254A (zh) * 2014-03-03 2014-07-23 中国神华能源股份有限公司 一种基于地质雷达的土壤物性类别识别方法和装置
CN104897605A (zh) * 2015-06-16 2015-09-09 中国人民解放军国防科学技术大学 一种基于改进支持向量机的太赫兹谱分类识别方法
CN106645014A (zh) * 2016-09-23 2017-05-10 上海理工大学 基于太赫兹光谱的物质识别方法

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114088656A (zh) * 2020-07-31 2022-02-25 中国科学院上海高等研究院 太赫兹光谱物质识别方法及***、存储介质及终端

Also Published As

Publication number Publication date
CN109406445A (zh) 2019-03-01

Similar Documents

Publication Publication Date Title
WO2020062471A1 (zh) 液体识别方法、特征提取方法、液体识别装置及存储装置
Liu et al. Dimensionality reduction for identification of hepatic tumor samples based on terahertz time-domain spectroscopy
Zand et al. Texture classification and discrimination for region-based image retrieval
Yadav et al. Multiresolution local binary pattern variants based texture feature extraction techniques for efficient classification of microscopic images of hardwood species
Li et al. Rotating machine fault diagnosis using dimension reduction with linear local tangent space alignment
Yu et al. Raisin quality classification using least squares support vector machine (LSSVM) based on combined color and texture features
Singh et al. Wavelet analysis of signals in agriculture and food quality inspection
Zhu et al. Gabor feature-based apple quality inspection using kernel principal component analysis
CN109470720B (zh) 液体识别方法、向量提取方法、液体识别装置及存储介质
Luo et al. Visualization of vibrational spectroscopy for agro-food samples using t-Distributed Stochastic Neighbor Embedding
Li et al. Classification for glucose and lactose terahertz spectrums based on SVM and DNN methods
Yuan et al. Fingerprint liveness detection using multiscale difference co-occurrence matrix
Mentouri et al. Steel strip surface defect identification using multiresolution binarized image features
Chelali et al. Face recognition using MLP and RBF neural network with Gabor and discrete wavelet transform characterization: a comparative study
Yin et al. Non-destructive detection of foreign contaminants in toast bread with near infrared spectroscopy and computer vision techniques
Bionda et al. Deep autoencoders for anomaly detection in textured images using CW-SSIM
Liu et al. Classification of early mechanical damage over time in pears based on hyperspectral imaging and transfer learning
Nemade et al. Co-occurrence patterns based fruit quality detection for hierarchical fruit image annotation
Jiang et al. Detection and recognition of veterinary drug residues in beef using hyperspectral discrete wavelet transform and deep learning
Al-Awadhi et al. Honey adulteration detection using hyperspectral imaging and machine learning
Kagawade et al. A new scheme of polar Fast Fourier Transform Code for iris recognition through symbolic modelling approach
Buisman et al. The log-Gabor method: Speech classification using spectrogram image analysis
Aryanmehr et al. Efficient deep CNN-based gender classification using Iris wavelet scattering
Drzewiecki et al. Applicability of multifractal features as global characteristics of WorldView-2 panchromatic satellite images
Nguyen et al. LBP-and-ScatNet-based combined features for efficient texture classification

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18934536

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18934536

Country of ref document: EP

Kind code of ref document: A1