CN110232445A - A kind of historical relic authenticity identification method of knowledge based distillation - Google Patents
A kind of historical relic authenticity identification method of knowledge based distillation Download PDFInfo
- Publication number
- CN110232445A CN110232445A CN201910526264.3A CN201910526264A CN110232445A CN 110232445 A CN110232445 A CN 110232445A CN 201910526264 A CN201910526264 A CN 201910526264A CN 110232445 A CN110232445 A CN 110232445A
- Authority
- CN
- China
- Prior art keywords
- yolov3
- historical relic
- network
- tiny
- finger
- 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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
Abstract
The present invention relates to historical relic authenticity field, in particular to a kind of historical relic authenticity identification method of knowledge based distillation specifically includes that step 1: before historical relic is put on display, acquiring finger-print region image, make data set;Step 2: configuration YOLOV3 network configures YOLOV3-Tiny network as student network as teacher's network;Step 3: training YOLOV3;Step 4: knowledge based distills training YOLOV3-Tiny;Step 5: after historical relic recycling, resurveying finger-print region image, make test set;Step 6: identifying the historical relic true and false using trained YOLOV3-Tiny.The present invention using accuracy good but slow YOLOV3 as teacher's network, the YOLOV3-Tiny of accuracy difference but fast speed is as student network, carry out knowledge distillation, Internet-supported Study of students is supervised with the target after softening, in the case where holding YOLOV3-Tiny original faster detection speed, its accuracy is greatly promoted, alleviates the hardware resource occupancy during verification retrieval, detection efficiency is improved, appraisal cost has been saved.
Description
Technical field
The present invention relates to historical relic authenticity field, in particular to a kind of historical relic authenticity side of knowledge based distillation
Method.
Background technique
China has a long history, and has the historical relic rarity of a large amount of remnants.In order to carry forward history culture, historical relic in all parts of the country
Collection unit all can periodically carry out historical relic touring exhibition.After activity is put on display, needs to identify historical relic, be broken to prevent historical relic
Bad, replacement.Currently, verification retrieval work is mainly by being accomplished manually, this depends on the personal experience of expert, knowledge, sometimes
It needs to assist detecting by high technology equipment.Such artificial qualification process takes a long time, and needs to put into more manpower and material resources,
And qualification result is subjective.
Summary of the invention
To solve the problems, such as that above-mentioned background technique, the present invention propose a kind of historical relic true and false mirror of knowledge based distillation
Determine method, has the advantages that detection speed is fast, accuracy rate is high.
Technical proposal that the invention solves the above-mentioned problems is: a kind of historical relic authenticity identification method of knowledge based distillation,
It is characterized in that, comprising the following steps:
Step 1: before historical relic is put on display, acquiring finger-print region image, make data set;
Step 2: configuration YOLOV3 network configures YOLOV3-Tiny network as student network as teacher's network;
Step 3: training YOLOV3;
Step 4: knowledge based distills training YOLOV3-Tiny;
Step 5: after historical relic recycling, resurveying finger-print region image, make test set;
Step 6: identifying the historical relic true and false using trained YOLOV3-Tiny.
Further, above-mentioned steps 1 specifically: before historical relic exhibition, select a region as finger-print region on historical relic, not
Under the conditions of illumination, using high-precision camera from the RGB image of multi-angle acquisition finger-print region, figure is marked out using annotation tool
Finger-print region as in makes data set, randomly selects a part of image therein and its mark file as training set, residue
Conduct verify collection.
Further, in above-mentioned steps 2, the yolo layer that the prediction scale of YOLOV3 is 52 × 52 is deleted, only retains 13 × 13
With the prediction of 26 × 26 two scales;For the training set that step 1 obtains, 6 are calculated using K-Means algorithm
Anchorbox replaces the anchorbox of former YOLOV3 and YOLOV3-Tiny.
Further, above-mentioned steps 3 specifically: YOLOV3 network model is obtained into training on data set in step 1, and is saved
Weight file.
Further, above-mentioned steps 4 are using trained YOLOV3 network as teacher's network, YOLOV3-Tiny network
As student network;Two networks successively execute propagated forward over an input image, and obtaining scale is 13 × 13 × c, 26 × 26
The output of × c, is denoted as out respectivelyt(teacher) and outs(student);YOLOV3-Tiny error is calculated according to formula (1)~(3):
LOSS=α T2·losssoft+(1-α)·losshard (1)
losshard=crossentropy (outs,Target) (3)
In formula, losssoftFor soft object error, losshardFor hard goal error (i.e. the initial error of YOLOV3 network), α
To adjust losssoftWith losshardWeight coefficient, T is vapo(u)rizing temperature;Target represents the original mark of data set, namely hard
Target;Softmax () and crossentropy (), which respectively represent, to be sought softmax functional value and intersects entropy;In order to balance
losssoftWith losshardThe order of magnitude, inlet coefficient β1;In formula 2, by student, the position of teacher's network, confidence level, classification
After predicted value is softened, relative entropy is sought, as soft object error.
Further, above-mentioned steps 5 specifically: after historical relic recycling, resurvey multiple images in the finger-print region, make
For test set.
Further, above-mentioned steps 6 specifically: execute trained YOLOV3-Tiny on the test set that step 5 obtains
Infer, obtains the confidence value of finger-print region, seek the average value of each finger-print region confidence level, if more than given threshold, then
Think not change at this, if the finger-print region does not change, determines historical relic for genuine piece.
Advantages of the present invention:
The present invention is based on the historical relic authenticity identification methods of knowledge distillation, and using accuracy, good but slow YOLOV3 makees
For teacher's network, the YOLOV3-Tiny of accuracy difference but fast speed carries out knowledge distillation, after softening as student network
Target supervise Internet-supported Study of students, in the case where keeping the original faster detection speed of YOLOV3-Tiny, greatly promote it
Accuracy alleviates the hardware resource occupancy during verification retrieval, improves detection efficiency, saved appraisal cost.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow diagrams of the historical relic authenticity identification method of knowledge distillation;
Fig. 2 is modified YOLOV3 network structure;
Fig. 3 is YOLOV3-Tiny network structure;
Fig. 4 is knowledge distillation schematic diagram.
Specific embodiment
To keep the purposes, technical schemes and advantages of embodiment of the present invention clearer, implement below in conjunction with the present invention
The technical solution in embodiment of the present invention is clearly and completely described in attached drawing in mode, it is clear that described reality
The mode of applying is some embodiments of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ability
Domain those of ordinary skill every other embodiment obtained without creative efforts, belongs to the present invention
The range of protection.Therefore, the detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit below and is wanted
The scope of the present invention of protection is sought, but is merely representative of selected embodiment of the invention.Based on the embodiment in the present invention,
Every other embodiment obtained by those of ordinary skill in the art without making creative efforts belongs to this
Invent the range of protection.
Referring to Fig. 1, a kind of historical relic authenticity identification method of knowledge based distillation, comprising the following steps:
Step 1: before historical relic is put on display, acquiring finger-print region image, make data set;
Step 2: configuration YOLOV3 network configures YOLOV3-Tiny network as student network as teacher's network;
Step 3: the training YOLOV3 on training set;
Step 4: knowledge based distills training YOLOV3-Tiny;
Step 5: after historical relic recycling, resurveying finger-print region image, make test set;
Step 6: identifying the historical relic true and false using trained YOLOV3-Tiny;
Step 1: before historical relic is put on display, acquiring finger-print region image and make data set.
Further, above-mentioned steps 1 specifically: before historical relic exhibition, select a region as finger-print region on historical relic, not
Under the conditions of illumination, using high-precision camera from the RGB image of multi-angle acquisition finger-print region, figure is marked out using annotation tool
Finger-print region as in makes data set, randomly selects a part of image therein and its mark file as training set, residue
Conduct verify collection.
Further, in above-mentioned steps 2, the yolo layer that the prediction scale of YOLOV3 is 52 × 52 is deleted, only retains 13 × 13
With the prediction of 26 × 26 two scales, network model is as shown in Figure 2 after deletion.The classification number for modifying YOLOV3 network is m class,
Yolo layers of convolution kernel port number are (m+1+4) × 3, are denoted as c, and modified yolo layer Output Size is 13 × 13 × c, 26 × 26
×c.For the training set that step 1 obtains, calculate 6 anchorbox using K-Means algorithm, replace former YOLOV3 with
The anchorbox of YOLOV3-Tiny.YOLOV3-TINY network model is as shown in Figure 3.
Further, in step 3, the initial hyper parameter of YOLOV3 network is set, and sets maximum number of iterations epochmax
With batch processing number batch, the training on training set, calculated on verifying collection after each epoch YOLOV3 precision ratio,
Recall ratio, mAP, and save to training log.(reach maximum number of iterations epochmax) after each training, according to instruction
The coefficient for practicing log adjustment location error function, confidence level error function, error in classification function, finally obtains precision ratio, Cha Quan
The higher training result of rate, mAP saves weight file.
Further, in step 4, using trained YOLOV3 network as teacher's network, YOLOV3-Tiny network is made
For student network.Two networks successively execute propagated forward over an input image, and obtaining scale is 13 × 13 × c, 26 × 26 × c
Output, as shown in figure 4, being denoted as out respectivelyt(teacher) and outs(student).Only YOLOV3-Tiny network is reversely passed
Weight is broadcast and updates, YOLOV3 network does not update weight, to deduction before only executing.Reverse propagated error includes two parts, such as
Shown in formula 1~3.
LOSS=α T2·losssoft+(1-α)·losshard (1)
losshard=crossentropy (outs,Target) (3)
In formula, losssoftFor soft object error, losshardFor hard goal error (i.e. the initial error of YOLOV3 network), α
To adjust losssoftWith losshardWeight coefficient, T is vapo(u)rizing temperature.Target represents the original mark of data set, namely hard
Target.Softmax () and crossentropy (), which respectively represent, to be sought softmax functional value and intersects entropy.In order to balance
losssoftWith losshardThe order of magnitude, inlet coefficient β1.In formula 2, by student, the position of teacher's network, confidence level, classification
After predicted value is softened, relative entropy is sought, as soft object error.
Maximum number of iterations and batch processing number are set, is trained on training set, is being verified after each epoch
Precision ratio, recall ratio, the mAP of YOLOV3-Tiny are calculated on collection, and are saved to training log.After each training, according to
The hyper parameter of training log adjustment YOLOV3-Tiny, finally obtains the higher training result of precision ratio, recall ratio, mAP, saves
Weight file.
Further, step 5 is after historical relic recycling, and finger-print region resurveys multiple images at m, forms test set, uses
In historical relic authenticity.
Further, step 6 executes deduction process using the trained YOLOV3-Tiny network of step 4 on test set, obtains
To the confidence value in every finger image region.The confidence value for counting same place finger-print region, calculates its average value, if m
Average value is all larger than given threshold, then is determined as that historical relic is certified products.
Embodiment:
In the present embodiment, finger-print region at preceding known 5 is put on display in historical relic, size is 5 × 5mm2.In different illumination
Under the conditions of, using EOS 7D Mark II camera carry MP-E 65mm f/2.8 1-5X camera lens, from multiple angles at 5 fingerprint
Region respectively acquires 500 RGB images, and resolution ratio is 5472 × 3648 pixels, and 2500 images are obtained.Use annotation tool
Labelimg marks out the finger-print region in image, obtains the corresponding XML file of every image.Randomly select 2300 therein
Image and its mark file are as training set, and residue is as verifying collection.
Step 2: configuration YOLOV3 network configures YOLOV3-Tiny network as student network as teacher's network;Modification
Network model and parameter;
In the present embodiment, deep learning exploitation environment is needed to configure, CPU i79700K, GPU are
NVIDIAGeForceRTX2080, operating system are Ubuntu 16.04LTS, CUDA10.0, and deep learning frame is
Pytorch。
YOLOV3 network structure is modified, particular content includes: 1) to delete the yolo layer that scale is 52 × 52, is configured in cfg
File deletes corresponding convolutional layer, up-sampling layer, route layers;1) modification anchorbox quantity is 6, in cfg configuration file
Num=6 is modified, 6 anchorbox is recalculated using K-Means algorithm on training set and replaces original value;1) it repairs
Changing classification number is 5, modifies classes=5 in cfg configuration file, and yolo layer of port number is (5+1+4) × 3, i.e., 30, it repairs
Change latter yolo layers having a size of 13 × 13 × 30,26 × 26 × 30.
YOLOV3-Tiny network parameter is modified, particular content includes: 1) that above-mentioned 6 anchorbox replacement is original
Value;2) modification classification number is 5, and classes=5 is modified in cfg configuration file, and yolo layers of port number is (5+1+4) × 3,
I.e. 30, after modification yolo layers having a size of 13 × 13 × 30,26 × 26 × 30.Build YOLOV3 and YOLOV3-Tiny network model.
Step 3: training YOLOV3 network saves weight file.
The present embodiment is for training YOLOV3 network model.Setting epochmax is 250, batch 8, each epoch
After all calculate precision ratio, recall ratio and mAP on verifying collection, and save as trained log.After primary training, root
The coefficient of position error function, confidence level error function, error in classification function is adjusted according to training log, repeatedly training obtains performance
Preferable network model saves weight file.
Step 4: knowledge based distills training YOLOV3-Tiny
In the present embodiment, YOLOV3 network loads its weight file, and YOLOV3-Tiny network is trained from the beginning.Two
A network successively executes propagated forward over an input image, obtains the output that scale is 13 × 13 × 30,26 × 26 × 30, respectively
It is denoted as outt(teacher) and outs(student).YOLOV3-Tiny network error is calculated then according to formula 1~3, sets relevant parameter
It is as follows: T=4, α=0.6, β1=0.0003.
It sets maximum number of iterations and batch processing number, in training process, only calculates the error and ladder of YOLOV3-Tiny network
Degree, and updates weight, and YOLOV3 network only executes preceding to deduction, does not calculate error and gradient.After each epoch
Precision ratio, recall ratio and the mAP of YOLOV3-Tiny are calculated on verifying collection, and saves as trained log.When primary training terminates
Afterwards, according to the hyper parameter of training log adjustment YOLOV3-Tiny, repeatedly training obtains the network model of better performances, saves power
Weight file.
Step 5: after historical relic recycling, resurveying finger-print region image;
The present embodiment is after historical relic withdrawal, using above-mentioned camera, resurveys finger-print region image at 5 each 100, differentiates
Rate is 5472 × 3648 pixels, obtains totally 500 images and forms test set.
Step 6: identifying the historical relic true and false using trained YOLOV3-Tiny;
Trained YOLOV3-Tiny network is executed into deduction process on test set, every image exports fingerprint region
The confidence value in domain.100 confidence values of every place's finger-print region are averaging, if average value is greater than given threshold 0.95,
Think that the finger-print region does not change, if finger-print region does not change at 5, determines that the historical relic is positive product.
The above description is only an embodiment of the present invention, is not limited the scope of the invention with this, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant systems
Domain is commanded, is included within the scope of the present invention.
Claims (7)
1. a kind of historical relic authenticity identification method of knowledge based distillation, which comprises the following steps:
Step 1: before historical relic is put on display, acquiring finger-print region image, make data set;
Step 2: configuration YOLOV3 network configures YOLOV3-Tiny network as student network as teacher's network;
Step 3: training YOLOV3;
Step 4: knowledge based distills training YOLOV3-Tiny;
Step 5: after historical relic recycling, resurveying finger-print region image, make test set;
Step 6: identifying the historical relic true and false using trained YOLOV3-Tiny.
2. a kind of historical relic authenticity identification method of knowledge based distillation according to claim 1, it is characterised in that:
The step 1 specifically: before historical relic exhibition, select a region as finger-print region on historical relic, in different illumination conditions
Under, using camera from the RGB image of multi-angle acquisition finger-print region, the finger-print region in image is marked out using annotation tool,
Data set is made, randomly selects a part of image therein and its mark file as training set, remaining be used as verifies collection.
3. a kind of historical relic authenticity identification method of knowledge based distillation according to claim 2, it is characterised in that:
In the step 2, the yolo layer that the prediction scale of YOLOV3 is 52 × 52 is deleted, only retains 13 × 13 and 26 × 26 two
The prediction of scale;For the training set that step 1 obtains, 6 anchorbox are calculated using K-Means algorithm, replace original
The anchorbox of YOLOV3 and YOLOV3-Tiny.
4. a kind of historical relic authenticity identification method of knowledge based distillation according to claim 3, it is characterised in that:
The step 3 specifically: YOLOV3 network model is obtained into training on data set in step 1, and saves weight file.
5. a kind of historical relic authenticity identification method of knowledge based distillation according to claim 4, it is characterised in that:
In the step 4, using trained YOLOV3 network is as teacher's network, YOLOV3-Tiny network is as student
Network;Two networks successively execute propagated forward over an input image, obtain scale be 13 × 13 × c, 26 × 26 × c it is defeated
Out, it is denoted as out respectivelytWith outs;YOLOV3-Tiny error is calculated according to formula (1)~(3):
LOSS=α T2·losssoft+(1-α)·losshard (1)
losshard=crossentropy (outs,Target) (3)
In above formula, losssoftFor soft object error, losshardFor hard goal error, α is to adjust losssoftWith losshardPower
Weight coefficient, T is vapo(u)rizing temperature;Target represents the original mark namely hard goal of data set;Softmax () with
Crossentropy (), which is respectively represented, to be sought softmax functional value and intersects entropy;In order to balance losssoftWith losshardNumber
Magnitude, inlet coefficient β1;In formula 2, after student, the position of teacher's network, confidence level, class prediction value are softened, ask
Relative entropy is taken, as soft object error.
6. a kind of historical relic authenticity identification method of knowledge based distillation according to claim 5, it is characterised in that:
The step 5 specifically: after historical relic recycling, multiple images are resurveyed in the finger-print region, as test set.
7. a kind of historical relic authenticity identification method of knowledge based distillation according to claim 6, it is characterised in that:
The step 6 specifically: trained YOLOV3-Tiny is executed into deduction on the test set that step 5 obtains, is referred to
The confidence value in line region seeks the average value of each finger-print region confidence level, if more than given threshold, then it is assumed that unchanged at this
Change, if the finger-print region does not change, determines historical relic for genuine piece.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910526264.3A CN110232445B (en) | 2019-06-18 | 2019-06-18 | Cultural relic authenticity identification method based on knowledge distillation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910526264.3A CN110232445B (en) | 2019-06-18 | 2019-06-18 | Cultural relic authenticity identification method based on knowledge distillation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110232445A true CN110232445A (en) | 2019-09-13 |
CN110232445B CN110232445B (en) | 2021-03-26 |
Family
ID=67859621
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910526264.3A Active CN110232445B (en) | 2019-06-18 | 2019-06-18 | Cultural relic authenticity identification method based on knowledge distillation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110232445B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112200764A (en) * | 2020-09-02 | 2021-01-08 | 重庆邮电大学 | Photovoltaic power station hot spot detection and positioning method based on thermal infrared image |
CN112308130A (en) * | 2020-10-29 | 2021-02-02 | 成都千嘉科技有限公司 | Deployment method of deep learning network of Internet of things |
CN112348167A (en) * | 2020-10-20 | 2021-02-09 | 华东交通大学 | Knowledge distillation-based ore sorting method and computer-readable storage medium |
CN113158969A (en) * | 2021-05-10 | 2021-07-23 | 上海畅选科技合伙企业(有限合伙) | Apple appearance defect identification system and method |
WO2023015610A1 (en) * | 2021-08-10 | 2023-02-16 | 万维数码智能有限公司 | Artificial intelligence-based method and system for authenticating ancient and modern artwork |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102930251A (en) * | 2012-10-26 | 2013-02-13 | 北京炎黄拍卖有限公司 | Two-dimensional collection data recording and discriminating device and method |
KR20170035362A (en) * | 2015-08-31 | 2017-03-31 | (주)늘푸른광고산업 | Cultural properties guide system using wireless terminal and guide plate for cultural properties guide system using wireless terminal |
CN108287833A (en) * | 2017-01-09 | 2018-07-17 | 北京艺鉴通科技有限公司 | It is a kind of for the art work identification to scheme to search drawing method |
US20180268222A1 (en) * | 2017-03-17 | 2018-09-20 | Nec Laboratories America, Inc. | Action recognition system for action recognition in unlabeled videos with domain adversarial learning and knowledge distillation |
CN109003098A (en) * | 2018-05-24 | 2018-12-14 | 孝昌天空电子商务有限公司 | Agricultural-product supply-chain traceability system based on Internet of Things and block chain |
CN109523282A (en) * | 2018-12-02 | 2019-03-26 | 程昔恩 | A method of constructing believable article Internet of Things |
-
2019
- 2019-06-18 CN CN201910526264.3A patent/CN110232445B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102930251A (en) * | 2012-10-26 | 2013-02-13 | 北京炎黄拍卖有限公司 | Two-dimensional collection data recording and discriminating device and method |
KR20170035362A (en) * | 2015-08-31 | 2017-03-31 | (주)늘푸른광고산업 | Cultural properties guide system using wireless terminal and guide plate for cultural properties guide system using wireless terminal |
CN108287833A (en) * | 2017-01-09 | 2018-07-17 | 北京艺鉴通科技有限公司 | It is a kind of for the art work identification to scheme to search drawing method |
US20180268222A1 (en) * | 2017-03-17 | 2018-09-20 | Nec Laboratories America, Inc. | Action recognition system for action recognition in unlabeled videos with domain adversarial learning and knowledge distillation |
CN109003098A (en) * | 2018-05-24 | 2018-12-14 | 孝昌天空电子商务有限公司 | Agricultural-product supply-chain traceability system based on Internet of Things and block chain |
CN109523282A (en) * | 2018-12-02 | 2019-03-26 | 程昔恩 | A method of constructing believable article Internet of Things |
Non-Patent Citations (2)
Title |
---|
MARIA DE-ARTEAGA ET AL: "Machine Learning for the Developing World", 《ACM TRANSACTIONS ON MANAGEMENT INFORMATION SYSTEMS》 * |
吴旭东等: "智能算法在古陶瓷文物鉴定中的应用", 《技术创新》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112200764A (en) * | 2020-09-02 | 2021-01-08 | 重庆邮电大学 | Photovoltaic power station hot spot detection and positioning method based on thermal infrared image |
CN112200764B (en) * | 2020-09-02 | 2022-05-03 | 重庆邮电大学 | Photovoltaic power station hot spot detection and positioning method based on thermal infrared image |
CN112348167A (en) * | 2020-10-20 | 2021-02-09 | 华东交通大学 | Knowledge distillation-based ore sorting method and computer-readable storage medium |
CN112348167B (en) * | 2020-10-20 | 2022-10-11 | 华东交通大学 | Knowledge distillation-based ore sorting method and computer-readable storage medium |
CN112308130A (en) * | 2020-10-29 | 2021-02-02 | 成都千嘉科技有限公司 | Deployment method of deep learning network of Internet of things |
CN113158969A (en) * | 2021-05-10 | 2021-07-23 | 上海畅选科技合伙企业(有限合伙) | Apple appearance defect identification system and method |
WO2023015610A1 (en) * | 2021-08-10 | 2023-02-16 | 万维数码智能有限公司 | Artificial intelligence-based method and system for authenticating ancient and modern artwork |
Also Published As
Publication number | Publication date |
---|---|
CN110232445B (en) | 2021-03-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110232445A (en) | A kind of historical relic authenticity identification method of knowledge based distillation | |
CN110059694B (en) | Intelligent identification method for character data in complex scene of power industry | |
CN106529565B (en) | Model of Target Recognition training and target identification method and device calculate equipment | |
CN109271374B (en) | Database health degree scoring method and system based on machine learning | |
CN108960135B (en) | Dense ship target accurate detection method based on high-resolution remote sensing image | |
CN109949290A (en) | Pavement crack detection method, device, equipment and storage medium | |
CN108664893A (en) | A kind of method for detecting human face and storage medium | |
CN104951803B (en) | Atmospheric distillation tower jet fuel endpoint flexible measurement method based on dynamic mobile window least square method supporting vector machine | |
CN108416353A (en) | Crop field spike of rice fast partition method based on the full convolutional neural networks of depth | |
CN110910343A (en) | Method and device for detecting pavement cracks and computer equipment | |
CN106371155A (en) | A weather forecast method and system based on big data and analysis fields | |
CN106204597B (en) | A kind of video object dividing method based on from the step Weakly supervised study of formula | |
CN110263835B (en) | Rock category automatic identification method based on deep learning and Bayesian network | |
CN109191386B (en) | BPNN-based rapid Gamma correction method and device | |
CN103745454A (en) | Method for evaluating image processing algorithms or systems based on rank learning | |
CN104252056A (en) | Detection method and device of substrate | |
CN111178438A (en) | ResNet 101-based weather type identification method | |
CN111783516A (en) | Ploughing quality natural grade evaluation method based on deep learning | |
CN107871133A (en) | The recognition methods of the optimization method, pavement disease of rim detection network and system | |
CN107563448A (en) | Sample space clustering method based on near-infrared spectrum analysis | |
CN114140463A (en) | Welding defect identification method, device, equipment and storage medium | |
CN110751170A (en) | Panel quality detection method, system, terminal device and computer readable medium | |
CN110188662A (en) | A kind of AI intelligent identification Method of water meter number | |
CN110070120B (en) | Depth measurement learning method and system based on discrimination sampling strategy | |
Robinson | Automated analysis of simulation output data |
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 |