CN107704878B - Hyperspectral database semi-automatic establishment method based on deep learning - Google Patents
Hyperspectral database semi-automatic establishment method based on deep learning Download PDFInfo
- Publication number
- CN107704878B CN107704878B CN201710930972.4A CN201710930972A CN107704878B CN 107704878 B CN107704878 B CN 107704878B CN 201710930972 A CN201710930972 A CN 201710930972A CN 107704878 B CN107704878 B CN 107704878B
- Authority
- CN
- China
- Prior art keywords
- data
- classifier
- database
- spectrum
- deep learning
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/58—Extraction of image or video features relating to hyperspectral data
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a hyperspectral database semi-automatic establishment method based on deep learning, which comprises the following steps: collecting spectral information of a natural scene by using a collection device based on four different principles, and establishing a spectral database which is not marked; after quality inspection is carried out, selecting a part of data to be manually marked on a crowdsourcing platform; based on the principle of deep learning, the labeling process is regarded as a binary classification problem, a part of spectral data sets with known labeling truth values are used for training, an optimal classifier is selected, then the other part of data sets are used for verification, unlabeled data can be automatically labeled through the classifier, and only manual inspection is needed. The method greatly saves human resources and the cost consumed by labeling, shortens the time required for establishing a large-scale labeled dense spectrum database, and can conveniently provide the dense spectrum database with known labeling information for the field of calculating spectra.
Description
Technical Field
The invention relates to the field of computational spectrum imaging, in particular to a large-scale spectrum database semi-automatic establishing method based on deep learning.
Background
The characteristics of an object in a natural scene, such as edge, shape, color, etc., are often used for scientific research such as image segmentation, white balance, material detection, target recognition, detection, tracking, etc., but are often greatly affected by factors such as background clutter, non-rigid deformation, blur, illumination, occlusion, etc.
The existing supervised learning algorithms show excellent performance, most of the learning models need a large number of parameters, such as a deep convolutional network, the algorithm models need a large number of data with manual annotations to support along with the increase of the number of layers, and the absence of an excellent database is a main limiting factor for preventing the improvement performance of the existing deep convolutional network.
The spectrum reflects the optical radiation of the substance, reveals the essential properties of the substance, and has abundant detail characteristics. In the spectral dimension, existing images only utilize information of three channels of red, green and blue (RGB), and a large amount of detail features in the spectral dimension are lost. The fusion of the spectral information and the spatial information has made a major breakthrough in the fields of image denoising, image segmentation, target tracking, white balance, scene understanding, and the like. However, the spectral database under the natural scene is small in quantity, sparse in data set and long in the years, and cannot meet the requirements of the existing research, and how to establish a large-scale natural scene spectral database is a difficult problem which needs to be solved urgently in the field of computational spectrum imaging.
The collection and building of large-scale databases with annotations requires a great deal of time and effort. The establishment of the ImageNet library still requires more than one year of human input by utilizing the largest crowdsourcing platform AMT in the world, and if a larger-scale spectrum database is to be established, the time is longer, which obviously cannot keep up with the increase speed of the depth of the network model. Therefore, a method for establishing a spectral database capable of saving manpower is urgently needed to solve the existing problems.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a large natural scene spectral database semi-automatic establishment method based on deep learning.
In order to achieve the purpose, the invention adopts the following technical scheme:
a hyperspectral database semi-automatic building method based on deep learning comprises the following steps:
step 1, collecting spectral data of a natural scene under different illumination conditions and RGB color images aligned with the spectral data by using four hyperspectral imagers based on different principles through a variable control method, and establishing a small spectral database;
step 2, performing standardization processing on the spectrum database, putting the spectrum database into a data pool, then randomly extracting a part of data from the data pool to perform quality inspection, comparing the acquired light source spectrum with a standard light source spectrum by adopting a spectrum angle matching method, if the similarity reaches 99%, judging that the acquired spectrum data is correct, retaining the acquired spectrum data, and otherwise, rejecting the acquired spectrum data;
step 3, randomly extracting a part of data from the data pool screened in the step 2, and carrying out manual annotation to obtain a true value, wherein annotation files correspond to picture names one to one;
step 4, randomly dividing the labeled data into a test set and a training set, training the training set by using a deep learning method to obtain a binary classifier, then testing the binary classifier by using the test set, setting two thresholds Threshold1 and a Threshold Threshold2 for the confidence coefficient output by the classifier, wherein the Threshold Threshold1 is greater than the Threshold Threshold 2; if the output score of the classifier is larger than the Threshold value Threshold1, the classification is considered to be correct; if the output score of the classifier is smaller than the Threshold value Threshold2, the classification is considered to be wrong; if the output value of the classifier is between two thresholds, the classification is considered to be fuzzy, and the corresponding picture is put into the next iteration so as to optimize the classifier;
step 5, in order to improve the performance of the classifier, respectively training the classifier by using neural networks of VGG, Googlenet and ResNet, inputting a test set, counting a judgment result, and then selecting a stable classifier with the highest judgment accuracy;
step 6, inputting the remaining unmarked data in the data pool into the classifier obtained in the step 5 for automatic marking;
and 7, checking whether the automatic labeling result in the step 6 is qualified, if so, recording the data into an intensive spectrum database with a known true value, and if not, recovering the data for training the classifier again.
The invention is based on the principle of deep learning, the labeling process is regarded as a binary classification problem, a part of spectral data sets with known labeling truth values are used for training, an optimal classifier is selected, then the other part of data sets are used for verification, the rest manual labeling can directly and automatically label the acquired spectral data through the classifier, and only manual inspection is needed. Therefore, compared with the traditional spectrum database, the method of the invention can save a large amount of human resources and cost for labeling, greatly shorten the time (at least half period) required for establishing a large-scale labeled dense spectrum database, and conveniently provide the dense spectrum database with known labeling information for the field of calculating spectra.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, the method for semi-automatically establishing a hyperspectral database based on deep learning of the embodiment includes (1) collecting spectral data and establishing a small unlabeled database; (2) performing quality inspection to ensure the accuracy of the provided spectral data; (3) in order to facilitate training by utilizing a neural network, data needs to be labeled, and the step is usually solved by a crowdsourcing platform; (4) training a classifier by using a data set labeled with a known truth value; (5) selecting an optimal classifier; (6) then, automatically labeling the unlabeled data set; (7) manually checking the correctness of the label, and retraining the classifier by using data with incorrect label; (8) and recording the qualified and labeled data into a dense spectrum database. The method comprises the following specific steps:
1. by means of the variable control method, four technically mature hyperspectral imagers based on different principles are used for collecting spectrum data of natural scenes under different illumination conditions and RGB color images aligned with the spectrum data, analysis and comparison are facilitated, and a small-sized spectrum database is established.
The devices used in the present embodiment are a PMIS high-resolution spectrum acquirer based on prism dispersion and mask modulation, a CTIS computed tomography spectrometer based on tomography, a spectrometer based on coded aperture, and a conventional push-broom spectrometer, respectively. The spectrum shooting range is 400-900 nm; the spectrum data storage format is a single-waveband gray scale map; the spectral resolution is 1-6 nm; the light source includes: fluorescent lamps, incandescent lamps, iodine tungsten lamps, LED lamps, sunlight at different times and weather conditions. The natural scene mainly refers to a common scene in life, and specifically includes: humans (different skin colors, ages, expressions), cars (bicycles, motorcycles, buses, cars, vans, buses, etc.), furniture (sofas, chairs, desks, tables, wardrobes, etc.), potted plants (succulents, evergreen broadleaves, bamboo, flowers, etc.), fruits and vegetables (bananas, tomatoes, peppers, grapes, apples, cucumbers, etc.), animals (cattle, sheep, dogs, cats, birds, etc.), and some other life scenarios;
2. standardizing a data set of a Spectral database, putting the data set into a data pool, randomly extracting a part of data from the data pool for quality inspection, and using a Spectral Angle matching method (SAM for short) to obtain a light source spectrum illminationacquisition(abbreviation I)a) Spectral inversion with standard light sourcestandard(abbreviation I)s) And comparing, if the similarity reaches 99%, determining that the acquired spectral data is correct, retaining, and otherwise, rejecting.
Wherein, the standardization treatment is to normalize the data: for high spatial resolution high spectral resolution data D (P, Q, N), wherein P x Q refers to spatial resolution and N refers to the number of spectral channels, since three-dimensional data cannot be directly calculated by using a two-norm, the three-dimensional data is converted into a vector form and then D (P x Q x N), and then calculated according to the formula (1), and then restored into three-dimensional data:
wherein i belongs to (1: P) Q and j belongs to (1: N) spectral dimensions.
Principle of spectrum angle matching method: the spectral response of each pixel point in the space can be regarded as an N-dimensional vector (N refers to the number of spectral channels), and the confidence value score represents the acquired light source spectrum
And a reference spectrum
The similarity of (c). The calculation formula is as follows:
wherein, the symbol
Representative vector
The two norms of (a).
3. And (3) randomly extracting 60% of data from the data pool screened in the step (2) to perform manual annotation to obtain a true value, wherein the annotated files correspond to the picture names one by one.
In this embodiment, the labeling contents are respectively: name of an object in a scene, date of scene shooting, location of scene shooting position, weather while shooting, light source illmination, spectrometer used, coordinates of a start pixel of the object in space and height and width occupied pixels (x, y, width, height) of the object are stored in a Pascal format.
In order to control the labeling quality, the data sets with known truth values are randomly inserted into the unmarked data sets, when labeling is carried out by using a crowdsourcing platform, the labeling accuracy of the data sets with known truth values is checked, and if the accuracy reaches more than 90%, the labeling task is considered to pass.
In order to further control the accuracy of the labeling, each group of data is divided into three different people for labeling, the labels which conflict with each other are checked, and then the labels which conflict with each other are re-labeled.
4. The process of manual labeling can be similar to the binary classification problem, with the labeled dataset randomly divided into a test set (30% in the labeled dataset) and a training set (70% in the labeled dataset). Training a training set by using a deep learning method to obtain a binary classifier, then testing the classifier by using a test set, and setting two appropriate thresholds Threshold1 and Threshold2(Threshold1> Threshold2, in this embodiment, Threshold1 is selected to be about 80%, and Threshold2 is selected to be about 20%) for confidence level output by the classifier through experiments. If the output score of the classifier is greater than Threshold1, the classification is considered to be correct; if the output score of the classifier is smaller than Threshold2, the classification is considered to be wrong; and if the output value of the classifier is between the two values, the classification is considered to be fuzzy, and the image is put into the next iteration so as to optimize the classifier.
5. In order to improve the performance of the classifier, neural networks such as VGG (19 layers), Googlenet (22 layers) and ResNet (152-1000 layers) are used for training the classifier, a test set is input, the judgment result is counted, and then a stable classifier with the highest judgment accuracy is selected.
6. Inputting the remaining unmarked data sets in the data pool into the classifier obtained in the step 5 for automatic marking;
7. manually checking whether the labeling result of the step 6 is qualified, if so, recording an intensive spectrum database with a known truth value, and if not, recovering the intensive spectrum database for training the classifier again;
8. after the classifier trained by the neural network is provided, automatic labeling can be performed when the natural scene mentioned in the step 1 is collected again, and the manpower required for building a large-scale intensive database is greatly reduced.
Claims (4)
1. A hyperspectral database semi-automatic building method based on deep learning is characterized by comprising the following steps:
step 1, collecting spectral data of a natural scene under different illumination conditions and RGB color images aligned with the spectral data by using four hyperspectral imagers based on different principles through a variable control method, and establishing a small spectral database;
step 2, performing standardization processing on the spectrum database, putting the spectrum database into a data pool, then randomly extracting a part of data from the data pool to perform quality inspection, comparing the acquired light source spectrum with a standard light source spectrum by adopting a spectrum angle matching method, if the similarity reaches 99%, judging that the acquired spectrum data is correct, retaining the acquired spectrum data, and otherwise, rejecting the acquired spectrum data;
step 3, randomly extracting a part of data from the data pool screened in the step 2, and carrying out manual annotation to obtain a true value, wherein annotation files correspond to picture names one to one;
step 4, randomly dividing the labeled data into a test set and a training set, training the training set by using a deep learning method to obtain a binary classifier, then testing the binary classifier by using the test set, setting two thresholds Threshold1 and a Threshold Threshold2 for the confidence coefficient output by the classifier, wherein the Threshold Threshold1 is greater than the Threshold Threshold 2; if the output score of the classifier is larger than the Threshold value Threshold1, the classification is considered to be correct; if the output score of the classifier is smaller than the Threshold value Threshold2, the classification is considered to be wrong; if the output value of the classifier is between two thresholds, the classification is considered to be fuzzy, and the corresponding picture is put into the next iteration so as to optimize the classifier;
step 5, in order to improve the performance of the classifier, respectively training the classifier by using neural networks of VGG, Googlenet and ResNet, inputting a test set, counting a judgment result, and then selecting a stable classifier with the highest judgment accuracy;
step 6, inputting the remaining unmarked data in the data pool into the classifier obtained in the step 5 for automatic marking;
and 7, checking whether the automatic labeling result in the step 6 is qualified, if so, recording the data into an intensive spectrum database with a known true value, and if not, recovering the data for training the classifier again.
2. The method as claimed in claim 1, wherein in step 3, in order to control the annotation quality, the datasets with known truth values are randomly inserted into the unlabeled datasets, and when manual annotation is performed by using a crowdsourcing platform, the accuracy of annotation of the datasets with known truth values is checked, and if the accuracy reaches more than 90%, the annotation task is considered to pass.
3. The method for semi-automatically building a hyperspectral database based on deep learning according to claim 2, wherein in the step 3, in order to further control the accuracy of labeling, each group of data is labeled multiple times, and labels which conflict with each other are labeled again.
4. The method for semi-automatically building the hyperspectral database based on deep learning of claim 1, wherein in the step 4, the data in the test set accounts for 30% of the labeled data, and the data in the training set accounts for 70% of the labeled data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710930972.4A CN107704878B (en) | 2017-10-09 | 2017-10-09 | Hyperspectral database semi-automatic establishment method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710930972.4A CN107704878B (en) | 2017-10-09 | 2017-10-09 | Hyperspectral database semi-automatic establishment method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107704878A CN107704878A (en) | 2018-02-16 |
| CN107704878B true CN107704878B (en) | 2021-06-22 |
Family
ID=61184836
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710930972.4A Active CN107704878B (en) | 2017-10-09 | 2017-10-09 | Hyperspectral database semi-automatic establishment method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107704878B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647731A (en) * | 2018-05-14 | 2018-10-12 | 宁波江丰生物信息技术有限公司 | Cervical carcinoma identification model training method based on Active Learning |
| CN108764372B (en) * | 2018-06-08 | 2019-07-16 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
| CN109446325B (en) * | 2018-10-22 | 2021-09-14 | 北京云雀智享科技有限公司 | Natural language classifier system with high accuracy |
| CN109979546A (en) * | 2019-04-04 | 2019-07-05 | 成都大学 | Network model analysis platform and construction method based on artificial intelligence number pathology |
| CN110059076A (en) * | 2019-04-19 | 2019-07-26 | 国网山西省电力公司电力科学研究院 | A kind of Mishap Database semi-automation method for building up of power transmission and transformation line equipment |
| CN112613413A (en) * | 2020-12-25 | 2021-04-06 | 平安国际智慧城市科技股份有限公司 | Perishable garbage classification quality determination method and device and computer readable storage medium |
| CN112687369A (en) * | 2020-12-31 | 2021-04-20 | 杭州依图医疗技术有限公司 | Medical data training method and device and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN102541838A (en) * | 2010-12-24 | 2012-07-04 | 日电（中国）有限公司 | Method and equipment for optimizing emotional classifier |
| CN103150578A (en) * | 2013-04-09 | 2013-06-12 | 山东师范大学 | Training method of SVM (Support Vector Machine) classifier based on semi-supervised learning |
| CN103177267A (en) * | 2013-04-22 | 2013-06-26 | 山东师范大学 | Support vector machine semi-supervised learning method in time-frequency joint |
| CN103207913A (en) * | 2013-04-15 | 2013-07-17 | 武汉理工大学 | Method and system for acquiring commodity fine-grained semantic relation |
| CN103258214A (en) * | 2013-04-26 | 2013-08-21 | 南京信息工程大学 | Remote sensing image classification method based on image block active learning |
| CN104166706A (en) * | 2014-08-08 | 2014-11-26 | 苏州大学 | Multi-label classifier constructing method based on cost-sensitive active learning |
| CN105701502A (en) * | 2016-01-06 | 2016-06-22 | 福州大学 | Image automatic marking method based on Monte Carlo data balance |
| CN105760439A (en) * | 2016-02-02 | 2016-07-13 | 西安交通大学 | Figure cooccurrence relation graph establishing method based on specific behavior cooccurrence network |
| CN106779086A (en) * | 2016-11-28 | 2017-05-31 | 北京大学 | A kind of integrated learning approach and device based on Active Learning and model beta pruning |
-
2017
- 2017-10-09 CN CN201710930972.4A patent/CN107704878B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN102541838A (en) * | 2010-12-24 | 2012-07-04 | 日电（中国）有限公司 | Method and equipment for optimizing emotional classifier |
| CN103150578A (en) * | 2013-04-09 | 2013-06-12 | 山东师范大学 | Training method of SVM (Support Vector Machine) classifier based on semi-supervised learning |
| CN103207913A (en) * | 2013-04-15 | 2013-07-17 | 武汉理工大学 | Method and system for acquiring commodity fine-grained semantic relation |
| CN103177267A (en) * | 2013-04-22 | 2013-06-26 | 山东师范大学 | Support vector machine semi-supervised learning method in time-frequency joint |
| CN103258214A (en) * | 2013-04-26 | 2013-08-21 | 南京信息工程大学 | Remote sensing image classification method based on image block active learning |
| CN104166706A (en) * | 2014-08-08 | 2014-11-26 | 苏州大学 | Multi-label classifier constructing method based on cost-sensitive active learning |
| CN105701502A (en) * | 2016-01-06 | 2016-06-22 | 福州大学 | Image automatic marking method based on Monte Carlo data balance |
| CN105760439A (en) * | 2016-02-02 | 2016-07-13 | 西安交通大学 | Figure cooccurrence relation graph establishing method based on specific behavior cooccurrence network |
| CN106779086A (en) * | 2016-11-28 | 2017-05-31 | 北京大学 | A kind of integrated learning approach and device based on Active Learning and model beta pruning |
Non-Patent Citations (2)
| Title |
|---|
| 《一种主动学习和协同训练相结合的半监督微博情感分类方法》;毕秋敏等;《现代图书情报技术》;20150131(第1期);第38-44页 * |
| 《基于多分类器决策的VN组合自动标注》;陈丽江;《计算机工程》;20080331;第34卷(第5期);第79-81页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107704878A (en) | 2018-02-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107704878B (en) | Hyperspectral database semi-automatic establishment method based on deep learning | |
| CN110059694B (en) | Intelligent identification method for character data in complex scene of power industry | |
| Aquino et al. | vitisBerry: An Android-smartphone application to early evaluate the number of grapevine berries by means of image analysis | |
| CN108764372B (en) | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set | |
| CN106446933B (en) | Multi-target detection method based on contextual information | |
| CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
| CN110309746A (en) | High-grade information security area list data information extracting method without communication interconnection | |
| CN110569747A (en) | method for rapidly counting rice ears of paddy field rice by using image pyramid and fast-RCNN | |
| CN109886153B (en) | Real-time face detection method based on deep convolutional neural network | |
| Lv et al. | A visual identification method for the apple growth forms in the orchard | |
| CN111178177A (en) | Cucumber disease identification method based on convolutional neural network | |
| Keller et al. | Soybean leaf coverage estimation with machine learning and thresholding algorithms for field phenotyping | |
| CN114758132B (en) | Fruit tree disease and pest identification method and system based on convolutional neural network | |
| CN114140665A (en) | Dense small target detection method based on improved YOLOv5 | |
| CN108615401B (en) | Deep learning-based indoor non-uniform light parking space condition identification method | |
| CN113409355A (en) | Moving target identification system and method based on FPGA | |
| CN114332942A (en) | Night infrared pedestrian detection method and system based on improved YOLOv3 | |
| CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
| CN113111716A (en) | Remote sensing image semi-automatic labeling method and device based on deep learning | |
| CN105404682B (en) | A kind of book retrieval method based on digital image content | |
| Yuan et al. | Sensitivity examination of YOLOv4 regarding test image distortion and training dataset attribute for apple flower bud classification | |
| TWI696958B (en) | Image adaptive feature extraction method and its application | |
| CN114445665A (en) | Hyperspectral image classification method based on Transformer enhanced non-local U-shaped network | |
| CN112699901A (en) | Plant species identification system based on Internet of things | |
| CN109299295B (en) | Blue printing layout database searching method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |