WO2021115059A1 - 一种信号分类的方法和设备 - Google Patents

一种信号分类的方法和设备 Download PDF

Info

Publication number
WO2021115059A1
WO2021115059A1 PCT/CN2020/129519 CN2020129519W WO2021115059A1 WO 2021115059 A1 WO2021115059 A1 WO 2021115059A1 CN 2020129519 W CN2020129519 W CN 2020129519W WO 2021115059 A1 WO2021115059 A1 WO 2021115059A1
Authority
WO
WIPO (PCT)
Prior art keywords
wavelet
terahertz time
feature vector
entropy
domain
Prior art date
Application number
PCT/CN2020/129519
Other languages
English (en)
French (fr)
Inventor
刘文权
张锐
鲁远甫
李光元
佘荣斌
Original Assignee
深圳先进技术研究院
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 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2021115059A1 publication Critical patent/WO2021115059A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Definitions

  • the present invention relates to the field of signal processing, in particular to a method and equipment for signal classification.
  • Terahertz wave electromagnétique wave with frequency from 0.1THz to 10THz
  • the frequency spectrum covers the vibration and rotation energy levels of a variety of biological macromolecules.
  • Terahertz spectroscopy detection technology has a broad field in the research of biological macromolecule characteristics and medical detection Application prospects. Especially for terahertz spectroscopy detection and imaging applications of biological tissue samples, its equipment is relatively simple and fast compared to magnetic resonance imaging and computed tomography imaging, and can be used for real-time navigation of medical operations, which has attracted more and more researchers' attention.
  • terahertz spectroscopy and imaging have been used in the nature research and imaging recognition of breast cancer tissue, gastric cancer tissue, brain glioma and other biological tissue samples, usually based on the maximum, minimum, or maximum value of the terahertz time-domain signal. Peak-to-peak value, or based on the amplitude value of a certain frequency point in the spectrum after Fourier transform of the signal, or based on the absorption coefficient or refractive index of a certain frequency point calculated by comparing with the reference terahertz signal Size, to distinguish different tissue components and to assist in the differentiation of diseased tissues and normal tissues.
  • the current classification methods for terahertz signals of biological tissues are mostly based on the amplitude of the terahertz time domain signal or the frequency domain energy after Fourier transform, and the frequency domain indicators such as absorption coefficient and refractive index also need to be measured twice and compared with the reference signal. It can only be obtained after the comparison calculation.
  • the key information of entropy which is the complexity feature of biological signals, is not used as an index for classification and recognition.
  • the present invention proposes a signal classification method and device.
  • a feature vector is constructed based on wavelet energy and entropy, and the complexity is reduced while considering energy information. This important information is introduced into the feature vector, which enriches the sample information carried by the feature vector; and by reducing the dimensionality of the feature vector, the classification and recognition speed is improved.
  • the present invention proposes the following specific embodiments:
  • the embodiment of the present invention provides a signal classification method, including:
  • the obtaining the terahertz time-domain signal of the biological tissue includes:
  • the terahertz time-domain signal is obtained by measuring biological tissues.
  • the processing the terahertz time-domain signal to obtain wavelet entropy and wavelet energy includes:
  • the perceptual wavelet packet transform is performed on the terahertz time domain signal to obtain wavelet entropy and wavelet energy.
  • the wavelet entropy is Shannon entropy, or necessary entropy, or logarithmic energy entropy;
  • the wavelet energy is a normalized or unnormalized energy parameter.
  • the constructing a feature vector based on the wavelet energy and the wavelet entropy includes:
  • a feature vector is constructed by combining the features of the wavelet energy and the wavelet entropy.
  • the performing dimensionality reduction processing on the feature vector includes:
  • the dimensionality of the feature vector is reduced through the Laplace feature map.
  • the machine learning classifier includes a support vector machine, or K nearest neighbor, or a decision tree, or an artificial neural network, a deep learning network, or an extreme learning machine, or an integrated learning classifier.
  • it further includes:
  • the processing the terahertz time-domain signal to obtain wavelet entropy and wavelet energy includes:
  • the first part of the terahertz time-domain signal is processed to obtain wavelet entropy and wavelet energy.
  • the embodiment of the present invention also provides a signal classification device, including:
  • the acquisition module is used to acquire the terahertz time-domain signal of the biological tissue
  • the recognition module is used to perform dimensionality reduction processing on the feature vector, and input the dimensionality reduction processed feature vector into a preset machine learning classifier for recognition, so as to realize the recognition of the terahertz time based on the obtained recognition classification result Recognition and classification of domain signals.
  • the acquisition module is used to:
  • the terahertz time domain signal is obtained by measuring biological tissues.
  • the embodiment of the present invention provides a signal classification method and device, the method includes: obtaining a terahertz time domain signal of a biological tissue; processing the terahertz time domain signal to obtain wavelet entropy and wavelet energy; Construct a feature vector based on the wavelet energy and the wavelet entropy; perform dimensionality reduction processing on the feature vector, and input the dimensionality reduction processed feature vector into a preset machine learning classifier for recognition, based on the obtained recognition
  • the classification result realizes the recognition and classification of the terahertz time-domain signal.
  • the feature vector is constructed based on wavelet energy and entropy, and the important information of complexity is introduced into the feature vector while considering the energy information, which enriches the sample information carried by the feature vector; and through The dimensionality reduction of the feature vector improves the classification and recognition speed.
  • FIG. 1 is a schematic flowchart of a method for signal classification according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of terahertz time-domain signal results and corresponding ESR calculation results of fibrous tissue and tumor tissue samples in a signal classification method proposed by an embodiment of the present invention
  • FIG. 3 is a schematic diagram of the ROC curve and the corresponding accuracy, sensitivity and specificity results of the classification and recognition of tumor tissue in a signal classification method proposed by an embodiment of the present invention
  • Fig. 4 is a schematic structural diagram of a signal classification device proposed by an embodiment of the present invention.
  • Embodiment 1 of the present invention discloses a signal classification method, as shown in FIG. 1, including the following steps:
  • Step 101 Obtain terahertz time-domain signals of biological tissues
  • the obtaining the terahertz time-domain signal of the biological tissue in step 101 includes:
  • the terahertz time-domain signal is obtained by measuring biological tissues.
  • Step 102 Process the terahertz time-domain signal to obtain wavelet entropy and wavelet energy
  • the processing of the terahertz time-domain signal in step 102 to obtain wavelet entropy and wavelet energy includes:
  • the perceptual wavelet packet transform is performed on the terahertz time domain signal to obtain wavelet entropy and wavelet energy.
  • the signal processing can be wavelet transform (WT), wavelet packet transform (WPT), or perceptual wavelet packet transform (PWPT), etc.
  • WT wavelet transform
  • WPT wavelet packet transform
  • PWPT perceptual wavelet packet transform
  • the wavelet signal processing method is not limited to the above three methods, as long as the terahertz time-domain signal can be processed to obtain wavelet entropy and wavelet energy.
  • the wavelet entropy is Shannon entropy, sure entropy, or log-energy entropy; specifically, it is not limited to the above. There are several specific embodiments, as long as the wavelet entropy of the complexity information can be represented.
  • Step 103 Construct a feature vector based on the wavelet energy and the wavelet entropy
  • the construction of a feature vector based on the wavelet energy and the wavelet entropy in step 103 includes:
  • a feature vector is constructed by combining the features of the wavelet energy and the wavelet entropy.
  • Step 104 Perform a dimensionality reduction process on the feature vector, and input the feature vector after the dimensionality reduction process into a preset machine learning classifier for recognition, so as to realize the recognition of the terahertz time-domain signal based on the obtained recognition and classification result. Recognition classification.
  • the performing dimensionality reduction processing on the feature vector in step 103 includes:
  • the dimensionality of the feature vector is reduced through the Laplace feature map.
  • the dimensionality reduction process can use principal component analysis (PCA), singular value decomposition (SVD), linear discriminant analysis (LDA), local linear embedding, and Laplace Feature mapping and other methods suitable for dimensionality reduction processing of feature vectors
  • PCA principal component analysis
  • SVD singular value decomposition
  • LDA linear discriminant analysis
  • local linear embedding local linear embedding
  • Laplace Feature mapping other methods suitable for dimensionality reduction processing of feature vectors
  • the machine learning classifier includes a support vector machine, or K nearest neighbor, or decision tree, or artificial neural network, or deep learning network, or extreme learning machine, or ensemble learning classifier.
  • machine learning classifiers including Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree (DT), Artificial Neural Network, Deep Learning Network, Extreme
  • SVM Support Vector Machine
  • KNN K-Nearest Neighbor
  • DT Decision Tree
  • Artificial Neural Network Artificial Neural Network
  • Deep Learning Network Extreme
  • Specific machine learning classifiers can be optimized for their specific parameters through the training set.
  • the method also includes:
  • the second part of the terahertz time domain signal also needs to perform the above steps 101 to 104, specifically based on the second part of the terahertz time domain signal to obtain wavelet entropy and wavelet energy parameters;
  • the feature vector is constructed by wavelet energy and entropy (for example, using the ratio of the two); then the feature vector is reduced in dimensionality, and then the feature vector is input to the machine learning classifier for parameter selection and optimization.
  • the processing the terahertz time-domain signal to obtain wavelet entropy and wavelet energy includes:
  • the first part of the terahertz time-domain signal is processed to obtain wavelet entropy and wavelet energy.
  • Wavelet transform is used to analyze terahertz time-domain signals.
  • Wavelet transform has the advantage of multi-scale resolution and is especially suitable for the analysis of non-stationary signals.
  • the feature vector is constructed based on wavelet energy and wavelet entropy. While considering energy information, the key information of signal complexity is also introduced into the classification of terahertz time-domain signals of biological tissues, and can be further combined with principal component analysis and machine learning algorithms. Achieve effective identification of different biological tissue samples.
  • a transmission terahertz time-domain spectroscopy system is used to collect terahertz time-domain signals from multiple breast fibrous tissue and tumor tissue samples.
  • the sample form is a paraffin-embedded tissue wax block.
  • the thickness of the sample is different. Basically the same ( ⁇ 2mm).
  • a total of 97 terahertz time-domain signals of fibrous tissue samples and 100 terahertz time-domain signals of tumor tissue samples were collected, including 50 terahertz time-domain signals of fibrous tissue and 50 terahertz time-domain signals of tumor tissue.
  • the domain signal is used to train and optimize the machine learning classifier, and the remaining biological tissue terahertz time domain signal is used for classification and recognition evaluation.
  • the wavelet packet transform (WPT) is used to process the terahertz time-domain signal.
  • the wavelet packet transform includes two recursive band-pass filter processes. The calculation method is as follows:
  • T(l) represents the terahertz time-domain signal to be processed
  • J represents the maximum decomposition layer number of wavelet packet transform
  • h( ⁇ ) and g( ⁇ ) represent low-pass and high-pass filters, respectively, Indicates the corresponding p-th subband signal when the number of decomposition layers is j, Yes The low frequency part, Yes The high frequency part.
  • I is the length of the subband signal c.
  • the feature vector is constructed based on the ratio of normalized wavelet energy to Shannon entropy (energy to Shannon entropy ratio, ESR):
  • a Daubechies wavelet (db1) and a maximum decomposition level of 10 are used, and there are a total of 1024 wavelet subband signals.
  • Figure 2 shows the results of the terahertz time-domain signals of fibrous tissue and tumor tissue samples and the corresponding ESR calculation results.
  • amplitudes of the terahertz time-domain signals of the two tissues are relatively small, the ESR of the two tissues is significantly different The difference indicates the feasibility and effectiveness of the feature vector index proposed by the present invention.
  • a principal component analysis (PCA) method is used to perform dimensionality reduction processing on the feature vector, and this embodiment uses the first 10 principal component information for classification and recognition.
  • the classifier is selected and optimized through the training set.
  • KNN K-Nearest Neighbor
  • the K-Nearest Neighbor (KNN) classifier is used in this embodiment, and the terahertz signal classification and recognition results of fibrous tissue and tumor tissue are shown in Figure 3.
  • Receiver operating characteristic curve (receiver operating characteristic, ROC) and area under the curve (AUC) show that this method can effectively classify and identify two tissue samples, and the corresponding accuracy, sensitivity and specificity results also show that this method can Identify and diagnose tumor tissues with high accuracy, low misdiagnosis rate, and low missed diagnosis rate.
  • the invention adopts wavelet transform to analyze the terahertz time-domain signal.
  • the wavelet transform has the advantage of multi-scale resolution and is particularly suitable for the analysis of non-stationary signals. Constructing feature vectors based on wavelet energy and wavelet entropy is different from the previous methods that use terahertz time domain signal amplitude or Fourier transform frequency domain energy and other indicators. While considering energy information, it also considers the complexity of biological samples. The key information is introduced into the terahertz time-domain signal classification of biological tissues, which enriches the sample information carried by the feature vector.
  • this method only needs to perform one terahertz measurement on the sample, which avoids the cumbersome process of two measurements for frequency domain indicators such as absorption coefficient and refractive index and comparison and calculation with reference signals, which improves efficiency and further combines feature vectors.
  • the dimensionality reduction algorithm and machine learning classifier can efficiently identify different biological tissue samples.
  • embodiment 2 of the present invention also discloses a signal classification device, as shown in Figure 4, including:
  • the obtaining module 201 is used to obtain the terahertz time-domain signal of the biological tissue
  • the processing module 202 is configured to process the terahertz time-domain signal to obtain wavelet entropy and wavelet energy;
  • a construction module 203 configured to construct a feature vector based on the wavelet energy and the wavelet entropy
  • the recognition module 204 is configured to perform dimensionality reduction processing on the feature vector, and input the dimensionality reduction processing feature vector into a preset machine learning classifier for recognition, so as to realize the recognition of the terahertz based on the obtained recognition and classification results. Recognition and classification of time domain signals.
  • the obtaining module 201 is configured to:
  • terahertz time domain signal based on the transmission type terahertz time domain spectroscopy system to measure biological tissue
  • the terahertz time-domain signal is obtained by measuring biological tissues.
  • the processing module 202 is configured to:
  • the perceptual wavelet packet transform is performed on the terahertz time domain signal to obtain wavelet entropy and wavelet energy.
  • the wavelet entropy is Shannon entropy, or necessary entropy, or logarithmic energy entropy;
  • the wavelet energy is a normalized or unnormalized energy parameter.
  • the construction module 203 is used for
  • a feature vector is constructed by combining the features of the wavelet energy and the wavelet entropy.
  • the recognition module 204 performs dimensionality reduction processing on the feature vector, including:
  • the dimensionality of the feature vector is reduced through the Laplace feature map.
  • the machine learning classifier includes a support vector machine, or K nearest neighbor, or decision tree, or artificial neural network, or deep learning network, or extreme learning machine, or ensemble learning classifier.
  • it also includes a learning module for:
  • the processing module 202 is configured to:
  • the first part of the terahertz time-domain signal is processed to obtain wavelet entropy and wavelet energy.
  • the embodiment of the present invention provides a signal classification method and device, the method includes: obtaining a terahertz time domain signal of a biological tissue; processing the terahertz time domain signal to obtain wavelet entropy and wavelet energy; Construct a feature vector based on the wavelet energy and the wavelet entropy; perform a dimensionality reduction process on the feature vector, and input the feature vector after the dimensionality reduction process into a preset machine learning classifier for recognition, based on the obtained recognition
  • the classification result realizes the recognition and classification of the terahertz time-domain signal.
  • the feature vector is constructed based on wavelet energy and entropy, and the important information of complexity is introduced into the feature vector while considering the energy information, which enriches the sample information carried by the feature vector; and By reducing the dimensionality of the feature vector, the classification and recognition speed is improved.
  • modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be changed to be located in one or more devices different from the implementation scenario.
  • the modules of the above implementation scenarios can be combined into one module or further divided into multiple sub-modules.

Abstract

一种信号分类的方法和设备，该方法包括：获取生物组织的太赫兹时域信号（101）；对所述太赫兹时域信号进行处理，得到小波熵和小波能量特征（102）；基于所述小波能量和所述小波熵构造特征向量（103）；对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类（104）。通过对太赫兹时域信号进行小波变换，基于小波能量与熵构造特征向量，在考虑能量信息的同时将复杂度这一重要信息引入到特征向量中，丰富了特征向量所携带的样品信息；且通过对特征向量进行降维处理，提高了分类识别速度。

Description

一种信号分类的方法和设备 技术领域
本发明涉及信号处理领域，特别涉及一种信号分类的方法和设备。
背景技术
太赫兹波(频率从0.1THz到10THz的电磁波)光子能量低，频谱覆盖多种生物大分子的振动和转动能级，太赫兹波谱检测技术在研究生物大分子特性和医学检测等方面具有广阔的应用前景。尤其是对于生物组织样品的太赫兹波谱检测和成像应用，其设备相对磁共振成像和计算机断层扫描成像等，简便快速，可用于医学手术的实时导航，越来越多地引起研究人员关注。
目前，太赫兹波谱和成像已用于乳腺癌组织、胃癌组织、脑胶质瘤等多种生物组织样品的性质研究和成像识别中，通常是基于太赫兹时域信号的最大值、最小值或峰峰值，或者基于对信号进行傅里叶变换后的频谱中某一个频率点的幅度值，或者基于与参考太赫兹信号进行比对计算后得出的某一个频率点的吸收系数或折射率的大小，来区分不同的组织成分和辅助辨析病变组织和正常组织。进一步，基于上述太赫兹信号特征指标，近期，多种人工智能分析方法，如支持向量机(Support Vector Machine，SVM)，K最近邻(k-Nearest Neighbor，KNN)算法等，也被引入到生物组织样品的太赫兹信号分类识别应用中。
但是与在太赫兹波段具有典型吸收峰的生物大分子不同，生物组织在太赫兹波段没有显著的特征吸收峰，在基于上述时域指标或者傅里叶变换后的频域指标进行分类识别时，不同组织尤其是病变组织样品和正常组织样品之间的对比度有待进一步显著提高。并且已有的方法往往仅利用了某一个时间点或者频率点下的信息，整个太赫兹信号的利用率和相应的区分 度有待进一步提高。
为进一步对不同生物组织样品太赫兹信号进行有效区分，研究人员也提出了一些优化方法。2017年Park等人提出了频域积分的方法对转移性淋巴结进行了有效识别，2018年Cao等人提出了一种太赫兹吸收系数谱分离的方法对肿瘤组织进行识别，2019年Huang等人基于最大信息系数，利用随机森林等算法对不同程度肝损伤进行了自动识别。
但是目前的生物组织太赫兹信号分类方法，大都基于太赫兹时域信号幅度或者傅里叶变换后的频域能量，而吸收系数和折射率等频域指标也需要进行两次测量并与参考信号进行比对计算后才能得到。同时，作为生物信号的复杂度特征-熵值这一关键信息，没有被用来作为分类识别的指标。
发明内容
针对现有技术中的缺陷，本发明提出了一种信号分类的方法和设备，通过对太赫兹时域信号进行小波变换，基于小波能量与熵构造特征向量，在考虑能量信息的同时将复杂度这一重要信息引入到特征向量中，丰富了特征向量所携带的样品信息；且通过对特征向量进行降维处理，提高了分类识别速度。
具体的，本发明提出了以下具体的实施例：
本发明实施例提出了一种信号分类的方法，包括：
获取生物组织的太赫兹时域信号；
对所述太赫兹时域信号进行处理，得到小波熵和小波能量；
基于所述小波能量和所述小波熵构造特征向量；
对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设 的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。
在一个具体的实施例中，所述获取生物组织的太赫兹时域信号，包括：
基于透射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于反射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于衰减全反射太赫兹时域波谱系统测量生物组织得到太赫兹时域信号。
在一个具体的实施例中，所述对所述太赫兹时域信号进行处理，得到小波熵和小波能量，包括：
对所述太赫兹时域信号进行小波变换，得到小波熵和小波能量；或
对所述太赫兹时域信号进行小波包变换，得到小波熵和小波能量；或
对所述太赫兹时域信号进行感知小波包变换，得到小波熵和小波能量。
在一个具体的实施例中，所述小波熵为香农熵、或必然熵、或对数能量熵；
所述小波能量为归一化或未归一化的能量参数。
在一个具体的实施例中，所述基于所述小波能量和所述小波熵构造特征向量，包括：
基于所述小波能量和所述小波熵的比值构造特征向量；或者
通过联合所述小波能量和所述小波熵的特征构造特征向量。
在一个具体的实施例中，所述对所述特征向量进行降维处理，包括：
通过主成分分析法对所述特征向量进行降维；或
通过奇异值分解法对所述特征向量进行降维；或
通过线性判别分析对所述特征向量进行降维；或
通过局部线性嵌入对所述特征向量进行降维；或
通过拉普拉斯特征映射对所述特征向量进行降维。
在一个具体的实施例中，所述机器学习分类器包括支持向量机、或K最近邻、或决策树、或人工神经网络、深度学习网络、或极限学习机、或集成学习分类器。
在一个具体的实施例中，还包括：
对所述太赫兹时域信号进行划分，分为第一部分和第二部分；
通过第二部分的所述太赫兹时域信号训练所述机器学习分类器；
所述对所述太赫兹时域信号进行处理，得到小波熵和小波能量，包括：
对第一部分的所述太赫兹时域信号进行处理，得到小波熵和小波能量。
本发明实施例还提出了一种信号分类的设备，包括：
获取模块，用于获取生物组织的太赫兹时域信号；
处理模块，用于对所述太赫兹时域信号进行处理，得到小波熵和小波能量；
构造模块，用于基于所述小波能量和所述小波熵构造特征向量；
识别模块，用于对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。
在一个具体的实施例中，所述获取模块，用于：
基于透射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于反射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于衰减全反射太赫兹时域波谱系统测量生物组织得到太赫兹时域信号。
以此，本发明实施例提出了一种信号分类的方法和设备，该方法包括：获取生物组织的太赫兹时域信号；对所述太赫兹时域信号进行处理，得到小波熵和小波能量；基于所述小波能量和所述小波熵构造特征向量；对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。通过对太赫兹时域信号进行小波变换，基于小波能量与熵构造特征向量，在考虑能量信息同时将复杂度这一重要信息引入到特征向量中，丰富了特征向量所携带的样品信息；且通过对特征向量进行降维处理，提高了分类识别速度。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，应当理解，以下附图仅示出了本发明的某些实施例，因此不应被看作是对范围的限定，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他相关的附图。
图1为本发明实施例提出的一种信号分类的方法的流程示意图；
图2为本发明实施例提出的一种信号分类的方法中纤维组织和肿瘤组织样品的太赫兹时域信号结果和相应的ESR计算结果的示意图；
图3为本发明实施例提出的一种信号分类的方法中对肿瘤组织进行分类识别的ROC曲线和相应的准确度、敏感度和特异度结果的示意图；
图4为本发明实施例提出的一种信号分类的设备的结构示意图。
具体实施方式
在下文中，将更全面地描述本公开的各种实施例。本公开可具有各种实施例，并且可在其中做出调整和改变。然而，应理解：不存在将本公开的各种实施例限于在此公开的特定实施例的意图，而是应将本公开理解为涵盖落入本公开的各种实施例的精神和范围内的所有调整、等同物和/或可选方案。
在本公开的各种实施例中使用的术语仅用于描述特定实施例的目的并且并非意在限制本公开的各种实施例。如在此所使用，单数形式意在也包括复数形式，除非上下文清楚地另有指示。除非另有限定，否则在这里使用的所有术语(包括技术术语和科学术语)具有与本公开的各种实施例所属领域普通技术人员通常理解的含义相同的含义。所述术语(诸如在一般使用的词典中限定的术语)将被解释为具有与在相关技术领域中的语境含义相同的含义并且将不被解释为具有理想化的含义或过于正式的含义，除非在本公开的各种实施例中被清楚地限定。
实施例1
本发明实施例1公开了一种信号分类的方法，如图1所示，包括以下步骤：
步骤101、获取生物组织的太赫兹时域信号；
具体的，步骤101中的所述获取生物组织的太赫兹时域信号，包括：
基于透射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号； 或
基于反射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于衰减全反射太赫兹时域波谱系统测量生物组织得到太赫兹时域信号。
步骤102、对所述太赫兹时域信号进行处理，得到小波熵和小波能量；
具体的，步骤102中的所述对所述太赫兹时域信号进行处理，得到小波熵和小波能量，包括：
对所述太赫兹时域信号进行小波变换，得到小波熵和小波能量；或
对所述太赫兹时域信号进行小波包变换，得到小波熵和小波能量；或
对所述太赫兹时域信号进行感知小波包变换，得到小波熵和小波能量。
具体的，对信号进行的处理，可以是小波变换(wavelet transform，WT)，或是小波包变换(wavelet packet transform，WPT)，或是感知小波包变换(perceptual wavelet packet transform，PWPT)等各种小波信号处理方法，并不限于上述的3种方法，具体的只要能对太赫兹时域信号进行处理得到小波熵和小波能量即可。
在一个具体的实施例中，所述小波熵为香农熵(shannon entropy)、或必然熵(sure entropy)、或对数能量熵(log-energy entropy)；具体的，也并不限于以上的这几种具体的实施例，只要能表征复杂度信息的小波熵就都可以。
步骤103、基于所述小波能量和所述小波熵构造特征向量；
具体的，步骤103中所述基于所述小波能量和所述小波熵构造特征向量，包括：
基于所述小波能量和所述小波熵的比值构造特征向量；或者
通过联合所述小波能量和所述小波熵的特征构造特征向量。
步骤104、对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。
具体的，步骤103中的所述对所述特征向量进行降维处理，包括：
通过主成分分析法对所述特征向量进行降维；或
通过奇异值分解法对所述特征向量进行降维；或
通过线性判别分析对所述特征向量进行降维；或
通过局部线性嵌入对所述特征向量进行降维；或
通过拉普拉斯特征映射对所述特征向量进行降维。
具体的，降维处理，可以采用主成分分析(principal component analysis，PCA)、奇异值分解(Singular Value Decomposition，SVD)、线性判别分析(linear discriminant analysis,LDA)、局部线性嵌入、拉普拉斯特征映射等各种适用于特征向量降维处理的方法
在一个具体的实施例中，所述机器学习分类器包括支持向量机、或K最近邻、或决策树、或人工神经网络、或深度学习网络、或极限学习机、或集成学习分类器。
具体的，机器学习分类器，包括支持向量机(Support Vector Machine，SVM)、K最近邻(k-Nearest Neighbor，KNN)、决策树(Decision tree，DT)、人工神经网络、深度学习网络、极限学习机、集成学习分类器等各种机器学习分类器，具体的机器学习分类器可以通过训练集对其具体参数进行优化选择。
此外，该方法还包括：
对所述太赫兹时域信号进行划分，分为第一部分和第二部分；
通过第二部分的所述太赫兹时域信号训练所述机器学习分类器；
具体的训练，也需要将第二部分的所述太赫兹时域信号执行上述步骤101到步骤104，具体的基于第二部分的所述太赫兹时域信号获取到小波熵和小波能量参数；在通过小波能量与熵(例如利用两者的比值)构造特征向量；继而对特征向量进行降维处理，后面将降维处理的特征向量输入到机器学习分类器中进行参数选择和优化。
所述对所述太赫兹时域信号进行处理，得到小波熵和小波能量，包括：
对第一部分的所述太赫兹时域信号进行处理，得到小波熵和小波能量。
本方案与目前已有方法不同，采用小波变换对太赫兹时域信号进行分析，小波变换具有多尺度分辨率的优点，特别适合于非平稳信号的分析。基于小波能量与小波熵构造特征向量，在考虑能量信息的同时，也把信号复杂度这一关键信息引入到生物组织太赫兹时域信号分类中，且可以进一步结合主成分分析和机器学习算法，实现对不同生物组织样品进行有效识别。
在此，以一个具体的例子来对本申请的方法来进行说明：
具体的，在该例子中，采用透射式太赫兹时域波谱系统对多个乳腺纤维组织和肿瘤组织样品进行太赫兹时域信号采集，样品形式为石蜡包埋的组织蜡块，不同样品的厚度基本一致(～2mm)，总共采集得到97个纤维组织样品太赫兹时域信号和100个肿瘤组织样品太赫兹时域信号，其中50个纤维组织太赫兹时域信号和50个肿瘤组织太赫兹时域信号用来训练和优化机器学习分类器，剩余的生物组织太赫兹时域信号用来进行分类识别评估。
具体的，采用小波包变换(wavelet packet transform，WPT)对太赫兹时域信号进行处理，小波包变换包括两个递归的带通滤波器过程，计算方法如下：
Figure PCTCN2020129519-appb-000001
其中，T(l)表示待处理的太赫兹时域信号，J表示小波包变换的最大分解层数，h(·)和g(·)分别表示低通和高通滤波器，
Figure PCTCN2020129519-appb-000002
表示在分解层数为j时，对应的第p个子带信号，
Figure PCTCN2020129519-appb-000003
是
Figure PCTCN2020129519-appb-000004
的低频部分,
Figure PCTCN2020129519-appb-000005
是
Figure PCTCN2020129519-appb-000006
的高频部分。
对于每个子带信号，通过香农熵(shannon entropy)用来表征复杂度，H(c)的具体计算流程如下：
Figure PCTCN2020129519-appb-000007
Figure PCTCN2020129519-appb-000008
其中，I是子带信号c的长度。
对于每个子带信号，归一化小波能量E(c)计算流程如下:
Figure PCTCN2020129519-appb-000009
进一步，基于归一化小波能量与香农熵的比值(energy to Shannon entropy ratio，ESR)构造特征向量：
Figure PCTCN2020129519-appb-000010
本实施例中，采用多贝西母小波(Daubechies wavelet，db1)和最大分解层数为10，总共有1024的小波子带信号。
附图2是纤维组织和肿瘤组织样品的太赫兹时域信号的结果和相应的ESR计算结果，虽然两种组织的太赫兹时域信号幅度差异较小，但是两种组织的ESR存在着显著的差别，表明本发明所提出特征向量指标的可行性和有效性。进一步，利用主成分分析(principal component analysis，PCA)方法对特征向量进行降维处理，本实施例采用前10个主成分信息进行分类识别。
通过训练集对分类器进行参数选择和优化，本实施例采用K最近邻(k-Nearest Neighbor，KNN)分类器，对纤维组织和肿瘤组织的太赫兹信号分类识别结果如附图3所示，受试者工作特征曲线(receiver operating characteristic，ROC)和曲线下面积(AUC)表明本方法可对两种组织样品进行有效分类识别，相应的准确度、敏感度和特异性结果也表明本方法可以高准确率、低误诊率、低漏诊率的对肿瘤组织进行识别诊断。
本发明采用小波变换对太赫兹时域信号进行分析，小波变换具有多尺度分辨率的优点，特别适合于非平稳信号的分析。基于小波能量与小波熵构造特征向量，与以往方法采用太赫兹时域信号幅度或者傅里叶变换后的频域能量等指标不同，在考虑能量信息的同时，也把复杂度这一生物样品的关键信息引入到生物组织太赫兹时域信号分类中，丰富了特征向量所携带的样品信息。并且本方法只需对样品进行一次太赫兹测量，避免了吸收系数和折射率等频域指标需要进行两次测量并与参考信号进行比对计算的繁琐过程，提高了效率，且进一步结合特征向量的降维算法和机器学习分类器，可对不同生物组织样品进行高效识别。
实施例3
为了对本申请方案进行进一步的说明，本发明实施例2还公开了一种 信号分类的设备，如图4所示，包括：
获取模块201，用于获取生物组织的太赫兹时域信号；
处理模块202，用于对所述太赫兹时域信号进行处理，得到小波熵和小波能量；
构造模块203，用于基于所述小波能量和所述小波熵构造特征向量；
识别模块204，用于对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。
在一个具体的实施例中，所述获取模块201，用于：
基于透射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于反射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
基于衰减全反射太赫兹时域波谱系统测量生物组织得到太赫兹时域信号。
在一个具体的实施例中，所述处理模块202，用于：
对所述太赫兹时域信号进行小波变换，得到小波熵和小波能量；或
对所述太赫兹时域信号进行小波包变换，得到小波熵和小波能量；或
对所述太赫兹时域信号进行感知小波包变换，得到小波熵和小波能量。
在一个具体的实施例中，所述小波熵为香农熵、或必然熵、或对数能量熵；
所述小波能量为归一化或未归一化的能量参数。
在一个具体的实施例中，所述构造模块203，用于
基于所述小波能量和所述小波熵的比值构造特征向量；或者
通过联合所述小波能量和所述小波熵的特征构造特征向量。
在一个具体的实施例中，所述识别模块204对所述特征向量进行降维处理，包括：
通过主成分分析法对所述特征向量进行降维；或
通过奇异值分解法对所述特征向量进行降维；或
通过线性判别分析对所述特征向量进行降维；或
通过局部线性嵌入对所述特征向量进行降维；或
通过拉普拉斯特征映射对所述特征向量进行降维。
在一个具体的实施例中，所述机器学习分类器包括支持向量机、或K最近邻、或决策树、或人工神经网络、或深度学习网络、或极限学习机、或集成学习分类器。
在一个具体的实施例中，还包括学习模块，用于：
对所述太赫兹时域信号进行划分，分为第一部分和第二部分；
通过第二部分的所述太赫兹时域信号训练所述机器学习分类器；
在此情况下，所述处理模块202，用于：
对第一部分的所述太赫兹时域信号进行处理，得到小波熵和小波能量。
以此，本发明实施例提出了一种信号分类的方法和设备，该方法包括：获取生物组织的太赫兹时域信号；对所述太赫兹时域信号进行处理，得到小波熵和小波能量；基于所述小波能量和所述小波熵构造特征向量；对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域 信号的识别分类。通过对太赫兹时域信号进行小波变换，基于小波能量与熵构造特征向量，在考虑能量信息的同时将复杂度这一重要信息引入到特征向量中，丰富了特征向量所携带的样品信息；且通过对特征向量进行降维处理，提高了分类识别速度。
本领域技术人员可以理解附图只是一个优选实施场景的示意图，附图中的模块或流程并不一定是实施本发明所必须的。
本领域技术人员可以理解实施场景中的装置中的模块可以按照实施场景描述进行分布于实施场景的装置中，也可以进行相应变化位于不同于本实施场景的一个或多个装置中。上述实施场景的模块可以合并为一个模块，也可以进一步拆分成多个子模块。
上述本发明序号仅仅为了描述，不代表实施场景的优劣。
以上公开的仅为本发明的几个具体实施场景，但是，本发明并非局限于此，任何本领域的技术人员能思之的变化都应落入本发明的保护范围。

Claims (10)

  1. 一种信号分类的方法，其特征在于，包括：
    获取生物组织的太赫兹时域信号；
    对所述太赫兹时域信号进行处理，得到小波熵和小波能量；
    基于所述小波能量和所述小波熵构造特征向量；
    对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。
  2. 如权利要求1所述的一种信号分类的方法，其特征在于，所述获取生物组织的太赫兹时域信号，包括：
    基于透射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
    基于反射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
    基于衰减全反射太赫兹时域波谱系统测量生物组织得到太赫兹时域信号。
  3. 如权利要求1所述的一种信号分类的方法，其特征在于，所述对所述太赫兹时域信号进行处理，得到小波熵和小波能量，包括：
    对所述太赫兹时域信号进行小波变换，得到小波熵和小波能量；或
    对所述太赫兹时域信号进行小波包变换，得到小波熵和小波能量；或
    对所述太赫兹时域信号进行感知小波包变换，得到小波熵和小波能量。
  4. 如权利要求1或3所述的一种信号分类的方法，其特征在于，所述小波熵为香农熵、或必然熵、或对数能量熵；
    所述小波能量为归一化或未归一化的能量参数。
  5. 如权利要求1所述的一种信号分类的方法，其特征在于，所述基于所述小波能量和所述小波熵构造特征向量，包括：
    基于所述小波能量和所述小波熵的比值构造特征向量；或者
    通过联合所述小波能量和所述小波熵的特征构造特征向量。
  6. 如权利要求1所述的一种信号分类的方法，其特征在于，所述对所述特征向量进行降维处理，包括：
    通过主成分分析法对所述特征向量进行降维；或
    通过奇异值分解法对所述特征向量进行降维；或
    通过线性判别分析对所述特征向量进行降维；或
    通过局部线性嵌入对所述特征向量进行降维；或
    通过拉普拉斯特征映射对所述特征向量进行降维。
  7. 如权利要求1所述的一种信号分类的方法，其特征在于，所述机器学习分类器包括支持向量机、或K最近邻、或决策树、或人工神经网络、或深度学习网络、或极限学习机、或集成学习分类器。
  8. 如权利要求1所述的一种信号分类的方法，其特征在于，还包括：
    对所述太赫兹时域信号进行划分，分为第一部分和第二部分；
    通过第二部分的所述太赫兹时域信号训练所述机器学习分类器；
    所述对所述太赫兹时域信号进行处理，得到小波熵和小波能量，包括：
    对第一部分的所述太赫兹时域信号进行处理，得到小波熵和小波能量。
  9. 一种信号分类的设备，其特征在于，包括：
    获取模块，用于获取生物组织的太赫兹时域信号；
    处理模块，用于对所述太赫兹时域信号进行处理，得到小波熵和小波能量；
    构造模块，用于基于所述小波能量和所述小波熵构造特征向量；
    识别模块，用于对所述特征向量进行降维处理，并将降维处理后的特征向量输入预设的机器学习分类器中进行识别，以基于得到的识别分类结果实现对所述太赫兹时域信号的识别分类。
  10. 如权利要求9所述的一种信号分类的设备，其特征在于，所述获取模块，用于：
    基于透射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
    基于反射式太赫兹时域波谱系统测量生物组织得到太赫兹时域信号；或
    基于衰减全反射太赫兹时域波谱系统测量生物组织得到太赫兹时域信号。
PCT/CN2020/129519 2019-12-11 2020-11-17 一种信号分类的方法和设备 WO2021115059A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911269591.1A CN111027488A (zh) 2019-12-11 2019-12-11 一种信号分类的方法和设备
CN201911269591.1 2019-12-11

Publications (1)

Publication Number Publication Date
WO2021115059A1 true WO2021115059A1 (zh) 2021-06-17

Family

ID=70206129

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/129519 WO2021115059A1 (zh) 2019-12-11 2020-11-17 一种信号分类的方法和设备

Country Status (2)

Country Link
CN (1) CN111027488A (zh)
WO (1) WO2021115059A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116541686A (zh) * 2022-11-01 2023-08-04 河海大学 基于多域特征融合极限学习机的电能质量扰动分类方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111027488A (zh) * 2019-12-11 2020-04-17 深圳先进技术研究院 一种信号分类的方法和设备
CN112910813B (zh) * 2021-04-10 2022-09-06 青岛科技大学 基于lda-knn的水声信号自动调制识别方法
CN113569966B (zh) * 2021-07-30 2023-10-10 太原理工大学 基于gaf-深度学习的煤岩界面比例识别分类方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103345641A (zh) * 2013-07-16 2013-10-09 杭州电子科技大学 基于小波包熵和支持向量机的手部肌电信号动作识别方法
CN103411912A (zh) * 2013-05-22 2013-11-27 首都师范大学 一种利用THz-TDS结合模糊规则专家系统鉴定中草药的方法
CN109101890A (zh) * 2018-07-16 2018-12-28 中国科学院自动化研究所 基于小波变换的电能质量扰动识别方法及装置
CN109785272A (zh) * 2018-12-19 2019-05-21 核工业北京地质研究院 一种矿物光谱特征提取中最优小波基选取的方法
CN111027488A (zh) * 2019-12-11 2020-04-17 深圳先进技术研究院 一种信号分类的方法和设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103411912A (zh) * 2013-05-22 2013-11-27 首都师范大学 一种利用THz-TDS结合模糊规则专家系统鉴定中草药的方法
CN103345641A (zh) * 2013-07-16 2013-10-09 杭州电子科技大学 基于小波包熵和支持向量机的手部肌电信号动作识别方法
CN109101890A (zh) * 2018-07-16 2018-12-28 中国科学院自动化研究所 基于小波变换的电能质量扰动识别方法及装置
CN109785272A (zh) * 2018-12-19 2019-05-21 核工业北京地质研究院 一种矿物光谱特征提取中最优小波基选取的方法
CN111027488A (zh) * 2019-12-11 2020-04-17 深圳先进技术研究院 一种信号分类的方法和设备

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HENIGULI WUMAIER ET AL.: "A Gearbox Fault Detecting Method Based on Morlet Wavelet Transform and MLP Neural Network", CHINESE JOURNAL OF ELECTRON DEVICES, vol. 39, no. 4, 31 August 2016 (2016-08-31), pages 834 - 840, XP055819987 *
LIU WENQUAN, ZHANG RUI, LING YU, TANG HONGPING, SHE RONGBIN, WEI GUANGLU, GONG XIAOJING, LU YUANFU: "Automatic recognition of breast invasive ductal carcinoma based on terahertz spectroscopy with wavelet packet transform and machine learning", BIOMEDICAL OPTICS EXPRESS, vol. 11, no. 2, 1 February 2020 (2020-02-01), United States, pages 971, XP055819981, ISSN: 2156-7085, DOI: 10.1364/BOE.381623 *
LIU WENQUAN; ZHANG RUI; LU YUANFU; SHE RONGBIN; ZHOU KAI; FANG BEIHUA; WEI GUANGLU; LI GUANGYUAN: "Classification of terahertz pulsed signals from breast tissues using wavelet packet energy feature exaction and machine learning classifiers", SPIE PROCEEDINGS, vol. 11196, 18 November 2019 (2019-11-18), US, pages 1119606 - 1119606-8, XP060127480, ISBN: 978-1-5106-3673-6, DOI: 10.1117/12.2537277 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116541686A (zh) * 2022-11-01 2023-08-04 河海大学 基于多域特征融合极限学习机的电能质量扰动分类方法
CN116541686B (zh) * 2022-11-01 2024-03-15 河海大学 基于多域特征融合极限学习机的电能质量扰动分类方法

Also Published As

Publication number Publication date
CN111027488A (zh) 2020-04-17

Similar Documents

Publication Publication Date Title
WO2021115059A1 (zh) 一种信号分类的方法和设备
Liu et al. Automatic recognition of breast invasive ductal carcinoma based on terahertz spectroscopy with wavelet packet transform and machine learning
CN108088834B (zh) 基于优化反向传播神经网络的包虫病血清拉曼光谱诊断仪
Avci et al. A new method based on adaptive discrete wavelet entropy energy and neural network classifier (ADWEENN) for recognition of urine cells from microscopic images independent of rotation and scaling
CN106096571B (zh) 一种基于emd特征提取和稀疏表示的细胞分类方法
WO2019181845A1 (ja) 生体組織解析装置、生体組織解析プログラムおよび生体組織解析方法
Lederman et al. Alternating diffusion for common manifold learning with application to sleep stage assessment
Fang et al. Fast discrimination of tumor and blood cells by label-free surface-enhanced Raman scattering spectra and deep learning
Vidyarthi et al. Texture based feature extraction method for classification of brain tumor MRI
Almuntashri et al. Gleason grade-based automatic classification of prostate cancer pathological images
Alheejawi et al. Automated melanoma staging in lymph node biopsy image using deep learning
Mabwa et al. Mid-infrared spectral classification of endometrial cancer compared to benign controls in serum or plasma samples
Purandare et al. Infrared spectroscopy with multivariate analysis segregates low-grade cervical cytology based on likelihood to regress, remain static or progress
Aytekin et al. Covid-19 detection from respiratory sounds with hierarchical spectrogram transformers
CN112580486B (zh) 一种基于雷达微多普勒信号分离的人体行为分类方法
Vimalajeewa et al. Early detection of ovarian cancer by wavelet analysis of protein mass spectra
Tawhid et al. Brain data mining framework involving entropy topography and deep learning
Mamun et al. Recognition of human skin diseases using inception-v3 with transfer learning
Gour et al. Application of deep learning techniques for prostate cancer grading using histopathological images
Jusman et al. Automated cervical precancerous cells screening system based on Fourier transform infrared spectroscopy features
Jothi et al. Abnormality classification of brain tumor in MRI images using multiclass SVM
Bouguerra et al. Brain Tumor Classification Based Deep Transfer Learning Approaches
Singh et al. An analysis of detection and diagnosis of different classes of skin diseases using artificial intelligence-based learning approaches with hyper parameters
Umadevi et al. Diagnosis of Diabetes from Tongue Image Using Versatile Tooth-Marked Region Classification
Liu et al. Classification of terahertz pulsed signals from breast tissues using wavelet packet energy feature exaction and machine learning classifiers

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: 20897977

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: 20897977

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205N DATED 19/01/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20897977

Country of ref document: EP

Kind code of ref document: A1