CN112964693A - Raman spectrum band region segmentation method - Google Patents
Raman spectrum band region segmentation method Download PDFInfo
- Publication number
- CN112964693A CN112964693A CN202110188585.4A CN202110188585A CN112964693A CN 112964693 A CN112964693 A CN 112964693A CN 202110188585 A CN202110188585 A CN 202110188585A CN 112964693 A CN112964693 A CN 112964693A
- Authority
- CN
- China
- Prior art keywords
- raman spectrum
- network
- important
- spectral band
- raman
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
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
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/65—Raman scattering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Software Systems (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The invention relates to a Raman spectrum band region segmentation method, which comprises the following steps: preprocessing Raman spectrum data, including SG smoothing and baseline removal; according to the Raman characteristic peak, the expert labels important Raman band regions, and a Raman spectrum segmentation model training set and a Raman spectrum segmentation model testing set are constructed; establishing an important spectral band segmentation model of the Raman spectrum by using a deep learning algorithm, extracting Raman spectrum characteristic information in the model by using a 1D convolutional neural network with a residual edge, and predicting an important spectral band region by using an anchor point structure. The method can efficiently and accurately divide the important spectral band region of the Raman spectrum data, has strong practicability, and can accurately position the important characterization region of the Raman spectrum.
Description
Technical Field
The invention relates to the field of spectral informatics, in particular to the technical field of optical detection, and specifically relates to a method for realizing Raman spectrum band region segmentation based on a deep learning model.
Background
Raman spectroscopy has been widely used in various fields such as rock composition detection, biomedical detection, sewage composition detection, and the like. In the raman spectroscopy analysis technology, a raman spectroscopy data analysis method is an important link.
At present, the raman spectral data analysis method mainly focuses on identifying the type of a measured substance, and the identification method mainly includes Principal Component Analysis (PCA) based on machine learning, Linear Discriminant Analysis (LDA), and Support Vector Machine (SVM). However, the current detection method cannot segment important band regions of the raman spectrum, and therefore, the identified raman spectrum result cannot be accurately positioned to the band region corresponding to the characteristic peak, and the detection result of the model cannot be further characterized. In view of the defects in the prior art, the invention provides a deep learning-based Raman spectrum band region segmentation method, which can accurately position an important characterization region of a Raman spectrum.
Disclosure of Invention
The invention aims to design a Raman spectrum band region segmentation method to realize accurate positioning of an important characterization region of a Raman spectrum. Based on the purpose, the invention designs and realizes a Raman spectrum band region segmentation method based on deep learning by utilizing a deep learning method.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a Raman spectrum band region segmentation method comprises the following steps:
step 1: collecting Raman spectrum data by utilizing Raman spectrum equipment, and preprocessing the Raman spectrum data, wherein the preprocessing comprises SG smoothing and baseline removal;
step 2: an expert marks an important spectral band region according to the position of a characteristic peak in a spectrum, and a Raman spectral band segmentation model training data set and a test set are established;
and step 3: establishing a deep learning model, and extracting Raman spectrum characteristic information in the model by using a 1D convolutional neural network with a residual error edge; performing first prediction on an important spectral band region by using a characteristic spectral band region recommendation network of an anchor point structure; intercepting the characteristic diagram by using the first predicted spectral band region, and predicting a final important spectral band region through a classification and regression network;
and 4, step 4: training a deep learning model by using a training set, wherein the deep learning model comprises a backbone feature extraction network, a feature band area recommendation network based on anchor points and a final classification regression network;
and 5: the test set test is used to improve the performance of the deep learning model, then the model is used for segmenting important spectral band regions of the Raman spectrum, and the segmented important spectral band regions are output.
In the scheme, step 1, preprocessing is performed on the raman spectrum data, and the specific steps are as follows:
step 11: the Raman spectrum data is complex and has a large amount of noise, and SG smoothing and baseline removal are required to be carried out on the Raman spectrum data to effectively label important band regions of the Raman spectrum and model training, so that the Raman spectrum noise data is reduced.
In the scheme, step 2, important band regions of the raman spectrum are labeled, and the specific steps are as follows:
step 22: adopting target detection marking tool LabelImg to mark important spectral band regions in Raman spectral data, wherein in marking, an expert uses the tool to select the important spectral band regions of the Raman spectral band, records spectral band coordinates, sets a class label of the spectral band, and then stores marked information into an XML format file according to a format protocol of the LabelImg tool, and the method specifically comprises the following steps: target annotation category, target start coordinate X1, and end coordinate X2 (one-dimensional data).
In the scheme, step 3, in the model, a 1D convolutional neural network with a residual edge is used for extracting Raman spectrum characteristic information, and a characteristic band region recommendation network with an anchor point structure is used for carrying out first prediction on an important band region; intercepting the characteristic diagram by using the first predicted spectral band region, and predicting a final important spectral band region through a classification and regression network, wherein the specific steps are as follows:
step 31: a ResNet50 is adopted in a deep learning model backbone network, and a ResNet50 comprises two basic blocks which are named as Conv Block and Identity Block respectively, wherein input and output dimensions of the Conv Block are different, the main function is to change network dimensions, and the Identity Block has the same input and output dimensions and is responsible for deepening the network. Both basic blocks have a residual network structure, which can prevent network degradation.
In the scheme, the main feature extraction network based on ResNet50 compresses Raman spectrum data 4 times to extract features, and the 5 th compression is used for intercepting the feature map.
In the above scheme, step 32: on the basis of step 31, the network is recommended to make a first prediction of the important spectral band using the characteristic spectral band regions of the anchor structures.
In the above scheme, the characteristic band region recommendation network with the anchor point structure is formed by combining one 3 × 1 convolution kernel and two 1 × 1 convolution kernels, and the confidence of the important band and the position of the important band are output after the characteristic diagram obtained in step 31 is input into the network.
Step 33: on the basis of step 32, the feature diagram is intercepted from the output result of the recommended network for the feature band region with the anchor point structure and is input to a classification and regression network, the classification and regression network performs the fifth compression of ResNet50 on the intercepted feature diagram, and then the final prediction result, namely the position of the key band region and the confidence thereof, is output.
The invention intelligently segments the important band region of the Raman spectrum by building a deep learning model, solves the problem that the current method can not accurately position and characterize the important band, and has the following advantages:
1. according to the invention, for the particularity of Raman spectrum data, ResNet50 is used as a characteristic extraction network, and jump connection is introduced at a position different from a common network in a residual error network, so that information of a previous residual error block flows into a next residual error block without being blocked, information flow is improved, and the problems of vanishing gradient and degradation caused by too deep network are avoided.
2. The invention uses an anchor point structure for predicting important band regions, the anchor point structure can cover all Raman spectrum regions, so that the model can be divided into any band regions, and the model adopts a twice prediction mode, wherein the first prediction is used for roughly screening the Raman spectrum band regions, and the second prediction is used for finely modifying the Raman spectrum band regions, thereby improving the accuracy of model division.
3. Aiming at the problems encountered in the division of important band regions of the Raman spectrum, the invention is realized by adopting a TensorFlow deep learning framework programming, is easy to expand and use, and has certain practical application value in Raman spectrum detection.
Drawings
FIG. 1 is a schematic diagram of a deep learning-based Raman spectrum band segmentation model;
FIG. 2Conv Block schematic;
FIG. 3 is a schematic diagram of an Identity Block module;
FIG. 4 is a graph showing the result of region division of important bands of a Raman spectrum.
Detailed Description
The invention is further described with reference to the following figures and examples.
The embodiment takes the Raman spectrum data of the oral tissue as an example, and provides a deep learning-based Raman spectrum band region segmentation method. In the method, a 1D convolutional neural network with a residual edge is used for extracting Raman spectrum characteristic information, and an anchor point structure is used for predicting an important spectral region.
The method can efficiently and accurately divide the important band region of the Raman spectrum data, has strong practicability, and can accurately position the important characterization region of the Raman spectrum.
The method comprises the following specific steps:
step 1: collecting Raman spectrum data of tongue tissue and gum tissue in the oral cavity by using Raman spectrum equipment, and carrying out SG smoothing and baseline removal on the Raman spectrum data of the oral tissue to effectively label important band regions of the Raman spectrum and model training, and reducing Raman spectrum noise data;
step 2: preprocessing Raman spectrum data, labeling important band regions of the Raman spectrum data by experts, and establishing a model training data set and a test set, wherein the specific steps are as follows;
step 21: the scale of the Raman spectrum data of the oral tissue related by the invention is 1044 multiplied by 1, and in order to ensure the accuracy of the labeled data, the Raman spectrum data in the range of 350-4000 of Raman frequency shift is intercepted.
Step 22: on the basis of the step 21, a target detection labeling tool LabelImg is adopted to label an important band region in Raman spectrum data, in the labeling, an expert uses the tool to select the important band region of the Raman spectrum, records band coordinates, sets a band type label, and then stores labeled information into an XML format file according to a format protocol of the LabelImg tool, and the method specifically comprises the following steps: target annotation category, target start coordinate X1, and end coordinate X2 (one-dimensional data).
And step 3: the deep learning model provided by the invention is shown in fig. 1 and mainly comprises a main feature extraction network, a regional recommendation network, a classification and regression network. Extracting Raman spectrum characteristic information in a trunk characteristic extraction network by using a 1D convolutional neural network with a residual error edge; performing first prediction on an important spectral band region by using an anchor point structure in a regional recommendation network; and intercepting the characteristic diagram by using the first predicted spectral band region, and predicting a final important spectral band region through a classification and regression network.
Step 31: the deep learning model backbone network adopts ResNet50, and ResNet50 includes two basic blocks named as Conv Block and Identity Block, as shown in FIG. 2 and FIG. 3, wherein Conv Block input and output dimensions are different, and the function is to change network dimensions, and Identity Block input and output dimensions are the same, and the deep learning model backbone network is responsible for deepening the network. Both basic blocks have residual network structures, which can prevent network degradation. The ResNet 50-based backbone feature extraction network performs 4 compressions of Raman spectral data to extract features, and the 5 th compression is used when truncating feature maps.
Step 32: on the basis of the step 31, the original data dimension of the constructed anchor point structure is 1044 × 1, the original data is compressed through a backbone feature extraction network, the dimension output by the compressed feature map is 66 × 1, 3 anchor points are constructed at each feature point of the feature map, the anchor points are respectively {32, 64 and 128} in length, and the constructed anchor point structure can cover all Raman spectrum data and can meet requirements for relatively large and relatively small Raman spectrum band sizes, so that the Raman spectrum key spectrum band area can be conveniently predicted in the follow-up process.
Step 33: on the basis of step 32, the network is recommended to make a first prediction of the important spectral band region using the characteristic spectral band regions of the anchor structures. The characteristic band area recommendation network with the anchor point structure is formed by combining a 3 x1 convolution kernel and two 1 x1 convolution kernels, and the confidence degree and the position of an important spectral band are output after the characteristic diagram obtained in the step 31 is input into the network.
Step 34: on the basis of step 33, the feature map is intercepted from the output result of the recommended network for the feature band region with the anchor point structure and is input to the classification and regression network, the classification and regression network performs the fifth compression of ResNet50 on the intercepted feature map, and then the final prediction result, namely the position of the key band region and the confidence thereof, is output.
And 4, step 4: the deep learning model is trained by using a training set, and the deep learning model comprises a backbone feature extraction network, a feature band area recommendation network based on anchor points and a final classification regression network.
Step 41: the loss function required by the training model is a multitask loss function, mainly comprises characteristic band region recommended network loss and classification and regression network loss, wherein the two types of loss comprise classification loss (cls loss) and regression loss (bbox regression loss), and the expression is as follows:
wherein the classification is lostThe method comprises the following steps of including characteristic spectral band region recommended network loss and classification regression network loss, wherein the characteristic spectral band region recommended network loss adopts a two-classification cross entropy loss function, when IoU is larger than 0.7 and marked as a foreground target and IoU is smaller than 0.3 and marked as a background according to an Intersection over Union (IoU), an anchor is selectedThe points participate in the training. And the classification and regression network adopts a multitask cross entropy loss function to participate in training.
Step 42: function of regression lossIn, ti={tx,twAnd represents the predicted offset of an anchor point in the training stage of the recommendation network and the classification regression network of the characteristic spectral band region.And tiThe dimension is the same as the actual offset of the recommended network and the classification regression network relative to the labeled group true in the characteristic spectral band region training stage, and the smooth loss function is used for calculation. And (4) adding the classification loss calculated in the step (41) and the regression loss calculated in the step to obtain a total loss, carrying out network back propagation through a total loss function, setting the learning rate of the model training optimization algorithm to be 0.01 by adopting an Adam method, and finally finishing the whole training of the model.
And 5: and testing the performance of the deep learning model by using the test set, then using the model for segmenting the important band region of the Raman spectrum, and outputting the segmented important band region.
The Raman spectral data training set and test set statistical tables referred to in this example
TABLE 1 statistical table of Raman spectral data set
Data set | Raman spectral data of tongue tissue | Raman spectral data of gingival tissue |
Training set | 719 | 520 |
Test set | 230 | 156 |
Total up to | 949 | 676 |
In order to evaluate the detection result, the invention adopts Precision (Precision) and recall rate (Rcall) indexes, and the calculation formula is as follows:
wherein TP (true Positive) indicates the number of correctly detected Raman bands, FP (false Positive) indicates the number of falsely detected Raman bands, and FN (false negative) indicates the number of undetected Raman bands.
According to the precision and the recall rate, a P-R curve can be obtained, the lower area of the P-R curve is AP (Average precision), each type of sewage outlet correspondingly obtains an AP value, further, the Average precision mAP (mean Average precision) of the two types of Raman spectrum data band segmentation can be calculated, and the precision of the detection algorithm is measured.
Fig. 4 is a graph showing the result of dividing the important bands of the raman spectrum of the gingival tissue, and the result of dividing the important bands of the raman spectrum of the oral tissue is shown in table 2.
TABLE 2 oral tissue Raman spectrum data segmentation result table
Method | AP (tongue tissue) | AP (gingival tissue) | mAP |
The method of the invention | 98.8% | 99.1% | 98.95% |
As can be seen from the table, the AP value of the two data is more than 98 percent, so the invention can accurately segment the important band region of the Raman spectrum of the oral tissue.
The invention has been illustrated by the above examples, but it should be understood that the above examples refer to the division of the important bands of the raman spectrum of oral tissue as an example only for purposes of illustration and description. Therefore, the important band division of Raman spectrum of other substances involved in the technical field and the technical method obtained by logical analysis, reasoning or limited experiments shall fall within the protection scope of the described examples.
Claims (8)
1. A Raman spectrum band region segmentation method is characterized by comprising the following steps:
step 1: collecting Raman spectrum data by utilizing Raman spectrum equipment, and preprocessing the Raman spectrum data, wherein the preprocessing comprises SG smoothing and baseline removal;
step 2: an expert marks an important band region according to the position of a characteristic peak in a spectrum, and a Raman band segmentation model training data set and a test set are established;
and step 3: establishing a deep learning model, and extracting Raman spectrum characteristic information in the model by using a 1D convolutional neural network with a residual error edge; performing first prediction on an important spectral band region by using a characteristic spectral band region recommendation network of an anchor point structure; intercepting the characteristic diagram by using the first predicted spectral band region, and predicting a final important spectral band region through a classification and regression network;
and 4, step 4: training a deep learning model by using a training set, wherein the deep learning model comprises a backbone feature extraction network, a feature band area recommendation network based on anchor points and a final classification regression network;
and 5: the test set test is used to improve the performance of the deep learning model, and then the model is used for segmenting important band regions of the Raman spectrum and outputting the segmented important band regions.
2. A method as claimed in claim 1, wherein the step 1 of preprocessing the raman spectrum data comprises the following steps:
step 11: the Raman spectrum data is complex and has a large amount of noise, and SG smoothing and baseline removal are required to be carried out on the Raman spectrum data to effectively label important band regions of the Raman spectrum and model training, and the Raman spectrum noise data are subtracted.
3. The method for splitting the band regions of the raman spectrum according to claim 1, wherein in the step 2, important band regions of the raman spectrum are labeled, and the specific steps are as follows:
step 22: adopting target detection marking tool LabelImg to mark important spectral band regions in Raman spectral data, using the tool to frame and select the important spectral band regions of Raman spectral band, recording spectral band coordinates, setting category labels of spectral band, then storing marked information into XML format file according to format protocol of LabelImg tool, specifically comprising: target annotation category, target start coordinate X1, and end coordinate X2 (one-dimensional data).
4. The method of claim 1, wherein in step 3, the model extracts the raman spectrum feature information by using a 1D convolutional neural network with residual edges, and uses a feature band region recommendation network of an anchor structure to perform a first prediction on an important band region; intercepting the characteristic diagram by using the first predicted spectral band region, and predicting a final important spectral band region through a classification and regression network, wherein the specific steps are as follows:
step 31: a ResNet50 is adopted in a deep learning model backbone network, and a ResNet50 comprises two basic blocks which are named as Conv Block and Identity Block respectively, wherein the input and output dimensions of the Conv Block are different, the network dimension is changed, and the Identity Block is the same in input and output dimension and is responsible for deepening the network.
5. A Raman spectrum band region segmentation method as claimed in claim 4, wherein the main feature extraction network based on ResNet50 performs 4 compression on Raman spectrum data to extract features, and the 5 th compression is used in feature map truncation.
6. A Raman spectral band region segmentation method according to claim 4, wherein step 32: on the basis of step 31, the important spectral band regions are predicted for the first time by using the characteristic spectral band region recommendation network of the anchor structure.
7. A Raman spectral band region segmentation method according to claim 6, wherein the characteristic band region recommendation network with the anchor point structure is formed by combining a 3 x1 convolution kernel and two 1 x1 convolution kernels, and the confidence of the important spectral band and the position of the important spectral band are output after the characteristic diagram obtained in step 31 is input into the network.
8. A method of raman spectral band region segmentation according to claim 6, characterized in that step 33: on the basis of step 32, the feature diagram is intercepted from the output result of the feature band area recommendation network with the anchor point structure and is input to the classification and regression network, the classification and regression network performs the fifth compression of ResNet50 on the intercepted feature diagram, and then the final prediction result, namely the position of the key band area and the confidence thereof, is output.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110188585.4A CN112964693A (en) | 2021-02-19 | 2021-02-19 | Raman spectrum band region segmentation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110188585.4A CN112964693A (en) | 2021-02-19 | 2021-02-19 | Raman spectrum band region segmentation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112964693A true CN112964693A (en) | 2021-06-15 |
Family
ID=76285034
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110188585.4A Pending CN112964693A (en) | 2021-02-19 | 2021-02-19 | Raman spectrum band region segmentation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112964693A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114417937A (en) * | 2022-01-26 | 2022-04-29 | 山东捷讯通信技术有限公司 | Deep learning-based Raman spectrum denoising method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615024A (en) * | 2018-12-28 | 2019-04-12 | 东北大学 | A kind of Rock Species intelligence Division identification and localization method |
CN109977812A (en) * | 2019-03-12 | 2019-07-05 | 南京邮电大学 | A kind of Vehicular video object detection method based on deep learning |
CN111553337A (en) * | 2020-04-27 | 2020-08-18 | 南通智能感知研究院 | Hyperspectral multi-target detection method based on improved anchor frame |
CN111563408A (en) * | 2020-03-09 | 2020-08-21 | 浙江中海达空间信息技术有限公司 | High-resolution image landslide automatic detection method with multi-level perception characteristics and progressive self-learning |
WO2020221990A1 (en) * | 2019-04-30 | 2020-11-05 | Huawei Technologies Co., Ltd. | Facial localisation in images |
-
2021
- 2021-02-19 CN CN202110188585.4A patent/CN112964693A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615024A (en) * | 2018-12-28 | 2019-04-12 | 东北大学 | A kind of Rock Species intelligence Division identification and localization method |
CN109977812A (en) * | 2019-03-12 | 2019-07-05 | 南京邮电大学 | A kind of Vehicular video object detection method based on deep learning |
WO2020221990A1 (en) * | 2019-04-30 | 2020-11-05 | Huawei Technologies Co., Ltd. | Facial localisation in images |
CN111563408A (en) * | 2020-03-09 | 2020-08-21 | 浙江中海达空间信息技术有限公司 | High-resolution image landslide automatic detection method with multi-level perception characteristics and progressive self-learning |
CN111553337A (en) * | 2020-04-27 | 2020-08-18 | 南通智能感知研究院 | Hyperspectral multi-target detection method based on improved anchor frame |
Non-Patent Citations (2)
Title |
---|
M. ERZINA等: "Precise cancer detection via the combination of functionalized SERS surfaces and convolutional neural network with independent inputs", 《SENSORS AND ACTUATORS B: CHEMICAL》 * |
SHAOQING REN等: "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114417937A (en) * | 2022-01-26 | 2022-04-29 | 山东捷讯通信技术有限公司 | Deep learning-based Raman spectrum denoising method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105447206B (en) | New comment object identifying method and system based on word2vec algorithms | |
CN101567043B (en) | Face tracking method based on classification and identification | |
Aversano et al. | A new text-independent method for phoneme segmentation | |
CN107123088B (en) | A kind of method of automatic replacement photo background color | |
CN111340130A (en) | Urinary calculus detection and classification method based on deep learning and imaging omics | |
CN105184292A (en) | Method for analyzing and recognizing structure of handwritten mathematical formula in natural scene image | |
CN114820625B (en) | Automobile top block defect detection method | |
CN103824090A (en) | Adaptive face low-level feature selection method and face attribute recognition method | |
CN107818321A (en) | A kind of watermark date recognition method for vehicle annual test | |
CN111210402A (en) | Face image quality scoring method and device, computer equipment and storage medium | |
CN111260645A (en) | Method and system for detecting tampered image based on block classification deep learning | |
CN112964693A (en) | Raman spectrum band region segmentation method | |
CN105760865A (en) | Facial image recognizing method capable of increasing comparison correct rate | |
CN104978569A (en) | Sparse representation based incremental face recognition method | |
CN110825896A (en) | Trademark retrieval system and method | |
CN102722732A (en) | Image set matching method based on data second order static modeling | |
CN111860500B (en) | Shoe stamp wearing area detection and edge tracing method | |
CN101344928B (en) | Method and apparatus for confirming image area and classifying image | |
CN105740399A (en) | Time series data processing method and apparatus | |
CN111626313B (en) | Feature extraction model training method, image processing method and device | |
CN116363422A (en) | Fruit classification method, system, computer and storage medium based on support vector machine | |
CN115984639A (en) | Intelligent detection method for fatigue state of part | |
CN111259914B (en) | Hyperspectral extraction method for characteristic information of tea leaves | |
CN114067208A (en) | Shallow hyperspectral remote sensing image classification method based on over-parameterization depth convolution kernel | |
CN110472583B (en) | Human face micro-expression recognition system based on deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210615 |
|
RJ01 | Rejection of invention patent application after publication |