CN107704878A - A kind of high-spectral data storehouse semi-automation method for building up based on deep learning - Google Patents
A kind of high-spectral data storehouse semi-automation method for building up based on deep learning Download PDFInfo
- Publication number
- CN107704878A CN107704878A CN201710930972.4A CN201710930972A CN107704878A CN 107704878 A CN107704878 A CN 107704878A CN 201710930972 A CN201710930972 A CN 201710930972A CN 107704878 A CN107704878 A CN 107704878A
- Authority
- CN
- China
- Prior art keywords
- data
- grader
- marked
- mark
- 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.)
- Granted
Links
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
Abstract
The invention discloses a kind of high-spectral data storehouse semi-automation method for building up based on deep learning, comprise the following steps：Using the spectral information of the harvester collection natural scene based on four kinds of different principles, the spectra database not being labeled is established；A part of data are chosen after progress quality examination manually to mark on mass-rent platform；Principle based on deep learning, regard annotation process as a two-value classification problem, using the known partial spectral data collection training for marking true value and choose an optimal classification device, then verified using another part data set, the data not marked can be by grader come automatic marking, it is only necessary to desk checking.The inventive method greatlys save human resources and mark consumes cost, reduces the time established required for the one large-scale intensive spectra database marked, can be easily to the intensive spectra database for calculating markup information known to spectral range offer.
Description
Technical field
The present invention relates to calculate light spectrum image-forming field, more particularly to a kind of large-scale spectra database based on deep learning half
Automatic Building cube method.
Background technology
The feature such as edge, shape, color possessed by object is often used for image segmentation, Bai Ping under natural scene
The scientific researches such as weighing apparatus, wood properly test, target identification, detect and track, but often by background is mixed and disorderly, non-rigid shape deformations, fuzzy, light
Influenceed according to, the factor blocked etc. very big.
Existing supervised learning algorithm shows superior performance, and these learning models are required for greatly quantity of parameters, example
Such as depth convolutional network, with the increase of the number of plies, algorithm model needs a large amount of data with manual annotation to support, excellent number
Missing according to storehouse is to stop the key constraints of current depth convolutional network improving performance.
Spectrum reflects the optical radiation of material, discloses the essential attribute of material, has abundant minutia.Tieed up in spectrum
On degree, existing image lost a large amount of details spy in spectral Dimensions merely with the information of RGB (RGB) three passages
Sign.The fusion of spectral information and spatial information is in fields such as image denoising, image segmentation, target tracking, white balance, scene understandings
Important breakthrough is obtained.But spectra database quantity under natural scene is few, data set is sparse, the age for a long time, can not
Meets the needs of existing research, how to establish a large-scale natural scene spectra database is to calculate light spectrum image-forming field one
Individual urgent need to solve the problem.
The collection and foundation of large scale database with mark need to put into substantial amounts of time and efforts.ImageNet storehouses
Foundation make use of in the world maximum mass-rent platform AMT to still need the human input of more than 1 year, if to establish one more
Large-scale spectra database, then the longer time is needed, this does not obviously catch up with the growth rate of network model depth.Therefore it is anxious
A kind of spectra database method for building up that can save manpower is needed to solve existing problem.
The content of the invention
It is a kind of based on the big of deep learning it is an object of the invention to propose for defect present in above prior art
Type natural scene spectra database semi-automation method for building up.
For the above-mentioned purpose, the present invention adopts the following technical scheme that：
A kind of high-spectral data storehouse semi-automation method for building up based on deep learning, comprises the following steps：
Step 1, by controlling the method for variable while being gathered not using four kinds of hyperspectral imagers based on different principle
Spectroscopic data and aligned RGB color figure with the natural scene under the conditions of illumination, establish small-sized spectra database；
Step 2, put into data pool after being standardized to spectra database, then randomly selected from data pool
A part of data carry out quality inspection, are entered the light source light spectrum collected and standard sources spectrum using spectrum angle matching method
Row compares, if similarity reaches 99%, judges that the spectroscopic data that collects is errorless, then retains, otherwise reject；
Step 3, a part of artificial mark of data progress is randomly selected in the data pool after step 2 screening and obtains true value,
File is marked to correspond with picture name；
Step 4, the data marked are randomly divided into test set and training set, using the method for deep learning to training set
It is trained to obtain two-value grader, then two-value grader is tested using test set, for the confidence of grader output
Degree sets two threshold value Threshold1 and threshold value Threshold2, and threshold value Threshold1>Threshold value Threshold2；If point
Class device output score value is more than threshold value Threshold1, then it is assumed that classification is correct；If grader output score value is less than threshold value
Threshold2, then it is assumed that classification error；If grader output score value is between two threshold values, then it is assumed that classification is fuzzy, will
Corresponding picture puts into next iteration, to be optimized to grader；
Step 5, in order to improve the performance of grader, instructed respectively using VGG, Googlenet and ResNet neutral net
Practice grader, input test collection, statistic discriminance result, then pick out one and differentiate the stable grader of accuracy rate highest；
Step 6, will be marked automatically in the remaining data input step 5 not marked obtains in data pool grader
Note；
Step 7, whether the automatic marking result of checking procedure 6 is qualified, if qualified, data acquisition is entered into known true value
Intensive spectra database, if unqualified, data withdrawal is reused for train grader.
Principle of the invention based on deep learning, regards annotation process as a two-value classification problem, utilizes known mark
Note the partial spectral data collection training of true value and choose an optimal classification device, then carry out carrying out using another part data set
Checking, remaining artificial mark can directly by the spectroscopic data collected by grader come automatic marking, it is only necessary to people
Work is examined.Therefore, compared to traditional spectra database, substantial amounts of human resources can be saved using the method for the present invention
With mark used in cost, the time established required for the one large-scale intensive spectra database marked is greatly reduced
(cycle at least shortening half), can be easily to the intensive spectroscopic data for calculating markup information known to spectral range offer
Storehouse.
Brief description of the drawings
Fig. 1 is the flow chart of the inventive method.
Embodiment
The present invention will be described in detail below in conjunction with the accompanying drawings and the specific embodiments.
As shown in Figure 1, it is a kind of semi-automatic foundation side in high-spectral data storehouse based on deep learning of the present embodiment
Method, including (1) collection spectroscopic data, establish small-sized unlabeled data storehouse；(2) quality examination is carried out, provided spectrum is provided
The accuracy of data；(3) for the ease of using the training of neutral net, it is necessary to be labeled to data, the step for often lead to
Cross the solution of mass-rent platform；(4) the data set training grader for having marked known true value is utilized；(5) an optimal classification device is chosen；
(6) and then the data set to not marking carries out automatic marking；(7) correctness of desk checking mark, and it is incorrect using marking
Data re -training grader；(8) data acquisition of qualified mark is entered into intensive spectra database.This method it is specific
Step is as follows：
1. by controlling the method for variable while utilizing the hyperspectral imager based on different principle of four kinds of technology maturations
The spectroscopic data of the natural scene under different illumination conditions and aligned RGB color figure are gathered, is easy to analyze and compares, build
Found small-sized spectra database.
Wherein, the present embodiment collection device therefor is respectively the PMIS high-resolution light modulated based on prism dispersion and mask
Acquisition Instrument is composed, the CTIS based on tomoscan calculates imaging spectrometer, and spectrometer and traditional pushing away based on coding aperture are swept
The spectrometer of formula.Spectrum coverage is 400-900nm；Spectroscopic data storage format is single band gray-scale map；Spectral resolution
For 1-6nm；Light source includes：Fluorescent lamp, incandescent lamp, iodine-tungsten lamp, LED, at different moments with the sunshine under weather condition.It is natural
Scene refers mainly to scene common in life, specifically includes：People's (the different colours of skin, age, expression), car (bicycle, motorcycle,
Bus, car, lorry, car etc.), furniture (sofa, chair, desk, dining table, wardrobe etc.), it is potted plant (succulent class,
Evergreen broad-leaved class, bamboo, flower etc.), fruits and vegetables (banana, tomato, capsicum, grape, apple, cucumber etc.), animal (ox, sheep, dog, cat,
Bird etc.) and some other living scenes；
2. the data set of pair spectra database is put into data pool after being standardized, then random from data pool
Extract a part of data and carry out quality inspection, using spectrum angle matching method (Spectral Angle Mapping, abbreviation SAM)
Obtained light source light spectrum illumination will be shotacquisition(abbreviation Ia) and standard sources spectrum
illuminationstandard(abbreviation Is) be compared, if similarity reaches 99%, it is believed that the spectroscopic data that it is gathered is errorless,
Then retain, otherwise reject.
Wherein, standardization is that data are normalized：For high spatial resolution hyperspectral remote sensing D
(P, Q, N), wherein P*Q refer to spatial resolution, and N refers to spectrum channel number, because three-dimensional data can not be counted directly using two norms
Calculate, be first transformed into D (P*Q*N) after vector form, calculated further according to formula (1), be then return to the data of three-dimensional：
Wherein, i ∈ (1:P*Q all pixels point in space, j ∈ (1) are referred to:N Spectral dimension) is referred to.
Spectrum angle matching method principle：The spectral response of each pixel in space can be regarded as a N-dimensional to
Measure (N refers to spectrum channel number), the light source light spectrum of collection is represented with the value of the confidence scoreAnd reference spectraIt is similar
Degree.Calculation formula is as follows：
Wherein, symbolRepresentation vectorTwo norms.
True value, mark text are obtained 3. randomly selecting 60% data in the data pool after step 2 screening and carrying out artificial mark
Part corresponds with picture name.
In the present embodiment, marked content respectively is：Target designation name, scene capture date date, field in scene
Weather whether, light source illumination, spectrometer used, target be in space when scape camera site location, shooting
Starting pixels coordinate and target height and width shared by pixel (x, y, width, height), using Pascal forms store.
In order to control mark quality, by the unlabelled data set of data set radom insertion of known true value, mass-rent is utilized
When platform is labeled, the accuracy of examining the data set of these known true value to mark, if accuracy reaches more than 90%
Think that mark task passes through.
In order to further control the accuracy rate of mark, each group of data are given into three different people and are labeled, checked
Conflicting mark, then mark is re-started to these conflicting marks.
4. the process manually marked can be similar to two-value classification problem, the data set marked is randomly divided into test set
(accounting for 30% in the data set marked) and training set (accounting for 70% in the data set marked).Utilize the side of deep learning
Method is trained to obtain a two-value grader to training set, and then grader is tested using test set, by experiment
Confidence level for grader output sets two suitable threshold value Threshold1 and Threshold2 (Threshold1>
Threshold2, it is about 20%) 80%, Threshold2 is about that Threshold1 is chosen in the present embodiment.If grader exports
Score value is more than Threshold1, then it is assumed that classification is correct；If grader output score value is less than Threshold2, then it is assumed that classification is wrong
By mistake；If grader output score value is positioned between the two, then it is assumed that classification is fuzzy, then this pictures is put into next iteration, with
Just grader is optimized.
5. in order to improve the performance of grader, VGG (19 layers), Googlenet (22 layers) and ResNet (152-1000 are used
Layer) etc. neural metwork training grader, input test collection, statistic discriminance result, then pick out one differentiation accuracy rate highest
Stabilization grader.
6. automatic marking will be carried out in the remaining data set input step 5 not marked obtains in data pool grader；
7. whether the annotation results of desk checking once step 6 are qualified, if qualified, the intensity into known true value is included
Spectra database, if unqualified, withdrawal is reused for training grader；
8. have it is above-mentioned by neural metwork training come out grader and then secondary acquisition step 1 in be previously mentioned from
During right scene, you can automatic marking, greatly reduce and establish the manpower that extensive Method on Dense Type of Data Using place needs.
Claims (4)
1. a kind of high-spectral data storehouse semi-automation method for building up based on deep learning, it is characterised in that comprise the following steps：
Step 1, by controlling the method for variable not shared the same light using four kinds of hyperspectral imager collections based on different principle simultaneously
The spectroscopic data of natural scene according under the conditions of and aligned RGB color figure, establish small-sized spectra database；
Step 2, put into after being standardized to spectra database in data pool, one is then randomly selected from data pool
Divided data carries out quality inspection, is compared the light source light spectrum collected and standard sources spectrum using spectrum angle matching method
It is right, if similarity reaches 99%, judge that the spectroscopic data that collects is errorless, then retains, otherwise reject；
Step 3, a part of artificial mark of data progress is randomly selected in the data pool after step 2 screening and obtains true value, is marked
File corresponds with picture name；
Step 4, the data marked are randomly divided into test set and training set, training set carried out using the method for deep learning
Training obtains two-value grader, and then two-value grader is tested using test set, and the confidence level for grader output is set
Put two threshold value Threshold1 and threshold value Threshold2, and threshold value Threshold1>Threshold value Threshold2；If grader
Output score value is more than threshold value Threshold1, then it is assumed that classification is correct；If grader output score value is less than threshold value Threshold2,
Then think classification error；If grader output score value is between two threshold values, then it is assumed that classification is fuzzy, and corresponding picture is thrown
Enter next iteration, to be optimized to grader；
Step 5, in order to improve the performance of grader, it is respectively trained point using VGG, Googlenet and ResNet neutral net
Class device, input test collection, statistic discriminance result, then pick out one and differentiate the stable grader of accuracy rate highest；
Step 6, automatic marking will be carried out in the remaining data input step 5 not marked obtains in data pool grader；
Step 7, whether the automatic marking result of checking procedure 6 is qualified, if qualified, data acquisition is entered into the intensive of known true value
Type spectra database, if unqualified, data withdrawal is reused for train grader.
2. a kind of high-spectral data storehouse semi-automation method for building up based on deep learning according to claim 1, it is special
Sign is, in the step 3, in order to control mark quality, and data set that the data set radom insertion of known true value is not marked
In, when manually being marked using mass-rent platform, the accuracy of examining the data set of these known true value to mark, if accuracy
Reach more than 90% and then think that mark task passes through.
3. a kind of high-spectral data storehouse semi-automation method for building up based on deep learning according to claim 2, it is special
Sign is, in the step 3, in order to further control the accuracy rate of mark, each group of data are repeatedly marked, and to phase
The mark mutually to conflict re-starts mark again.
4. a kind of high-spectral data storehouse semi-automation method for building up based on deep learning according to claim 1, it is special
Sign is, in the step 4, the data of test set account for the 30% of the data marked, and the data of training set account for the number marked
According to 70%.
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 true CN107704878A (en) | 2018-02-16 |
| CN107704878B 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) |
Cited By (6)
| 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 |
| CN109446325A (en) * | 2018-10-22 | 2019-03-08 | 北京云雀智享科技有限公司 | A kind of natural language classifier system of 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 |
| WO2019233297A1 (en) * | 2018-06-08 | 2019-12-12 | Oppo广东移动通信有限公司 | Data set construction method, mobile terminal and 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 |
|---|
| 毕秋敏等: "《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》", 《现代图书情报技术》 * |
| 陈丽江: "《基于多分类器决策的VN组合自动标注》", 《计算机工程》 * |
Cited By (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 |
| WO2019233297A1 (en) * | 2018-06-08 | 2019-12-12 | Oppo广东移动通信有限公司 | Data set construction method, mobile terminal and readable storage medium |
| CN109446325A (en) * | 2018-10-22 | 2019-03-08 | 北京云雀智享科技有限公司 | A kind of natural language classifier system of high accuracy |
| 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 |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107704878B (en) | 2021-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107704878A (en) | A kind of high-spectral data storehouse semi-automation method for building up based on deep learning | |
| CN110472575B (en) | Method for detecting ripeness of tomatoes stringing together based on deep learning and computer vision | |
| CN106951836B (en) | crop coverage extraction method based on prior threshold optimization convolutional neural network | |
| CN110276386A (en) | A kind of apple grading method and system based on machine vision | |
| CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
| CN108195767B (en) | Estuary wetland foreign species monitoring method | |
| CN107392130A (en) | Classification of Multispectral Images method based on threshold adaptive and convolutional neural networks | |
| CN107808375B (en) | Merge the rice disease image detecting method of a variety of context deep learning models | |
| CN108648169A (en) | The method and device of high voltage power transmission tower defects of insulator automatic identification | |
| CN109684906A (en) | The method of detection red turpentine beetle based on deep learning | |
| CN103852034B (en) | A kind of method for measuring perendicular | |
| Rong et al. | Pest identification and counting of yellow plate in field based on improved mask r-cnn | |
| CN111199192A (en) | Method for detecting integral maturity of field red globe grapes by adopting parallel line sampling | |
| CN109919007A (en) | A method of generating infrared image markup information | |
| CN112147078B (en) | Multi-source remote sensing monitoring method for crop phenotype information | |
| CN111462058A (en) | Method for quickly detecting effective ears of rice | |
| CN114140665A (en) | Dense small target detection method based on improved YOLOv5 | |
| CN106228136A (en) | Panorama streetscape method for secret protection based on converging channels feature | |
| CN111724354A (en) | Image processing-based method for measuring spike length and small spike number of multiple wheat | |
| Yuan et al. | Research on vegetation information extraction from visible UAV remote sensing images | |
| CN114937266A (en) | Hard shell clam biological sign identification method based on YOLOX-S | |
| Tompalski et al. | A comparison of LiDAR and image-derived canopy height models for individual tree crown segmentation with object based image analysis | |
| Deng et al. | Tree crown recognition algorithm on high spatial resolution remote sensing imagery | |
| Kuikel et al. | Individual banana tree crown delineation using unmanned aerial vehicle (UAV) images | |
| CN103871065B (en) | Vegetation canopy layer aggregation effect quantitative evaluation method based on hemispherical videos |
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 |