CN111160469B - Active learning method of target detection system - Google Patents
Active learning method of target detection system Download PDFInfo
- Publication number
- CN111160469B CN111160469B CN201911396654.XA CN201911396654A CN111160469B CN 111160469 B CN111160469 B CN 111160469B CN 201911396654 A CN201911396654 A CN 201911396654A CN 111160469 B CN111160469 B CN 111160469B
- Authority
- CN
- China
- Prior art keywords
- sample
- target
- information
- sample set
- training
- 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
Classifications
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an active learning method of a target detection system, which comprises the following steps: obtaining a sample training set to carry out model training on a target detection model, and carrying out sample prediction on an unlabeled sample set according to the trained target detection model to obtain a prediction result; respectively calculating the information quantity carried by each unlabeled sample in the unlabeled sample set to obtain a plurality of sample information quantities; sorting unlabeled samples according to the information quantity of the samples to select a target sample with the maximum information quantity; and updating the marked sample set and the unmarked sample set according to the target sample, and sequentially executing model training, sample prediction, information quantity calculation and target sample selection and marking again until the marked sample set meets the stop condition, and stopping training the target detection model. According to the invention, by measuring and evaluating the information quantity of the sample in the training process of the target detection model, the sample which is most effective in model training is selected for manual labeling, and the sample labeling efficiency is improved.
Description
Technical Field
The invention belongs to the technical field of target detection, and particularly relates to an active learning method of a target detection system.
Background
Target detection is a classical task in the field of computer vision, has important application value in the fields of automatic driving, video monitoring, human-computer interaction, face detection and the like, and aims to quickly and accurately locate and identify a specific target in a natural scene, but the task itself is very challenging due to complex and diverse scenes. With the rapid development of deep learning technology in recent years, a convolutional neural network-based target detection algorithm is focused and widely studied, and the target detection result is greatly improved.
In the implementation process of the target detection technology, the labeling operation for the training sample is particularly important, but labeled resources are often limited in the existing target detection technology, and massive unlabeled data are not utilized. Often, a significant amount of time and labor is required to annotate each object in the image, which in turn results in a very time consuming, difficult, and costly overall process.
The method has the advantages that the active learning allows the target detection learning algorithm to select the samples rich in information in the training sample set in a data screening mode, and good training effect can be achieved by only marking part of samples, so that marking efficiency is improved.
Disclosure of Invention
The embodiment of the invention aims to provide an active learning method of a target detection system, which aims to solve the problem of low sample marking efficiency caused by training sample marking manually in the implementation process of the existing target detection technology.
The embodiment of the invention is realized in such a way that an active learning method of a target detection system comprises the following steps:
obtaining a local pre-stored sample training set, wherein the sample training set comprises a marked sample set and an unmarked sample set;
model training is carried out on a target detection model according to the marked sample set, and sample prediction is carried out on the unmarked sample set according to the trained target detection model so as to obtain a prediction result;
respectively calculating the information quantity carried by each unlabeled sample in the unlabeled sample set according to the prediction result and a preset selection strategy to obtain a plurality of sample information quantities;
sorting the unlabeled samples according to the sample information amount, and selecting a target sample with the maximum information amount according to a sorting result;
and respectively updating the marked sample set and the unmarked sample set according to the target sample, and sequentially executing training of the target detection model, sample prediction of the unmarked sample set, calculation of the information amount carried by each unmarked sample and selection and marking of the target sample again according to the updated marked sample set and the unmarked sample set until judging that the updated marked sample set meets a stop condition, and stopping training of the target detection model.
Further, the prediction result includes the number of candidate targets in each unlabeled sample, a probability distribution set of each candidate target corresponding to each classification category in the sample training set, and a corresponding spatial position coordinate set of a boundary frame of each candidate target.
Further, the step of calculating the information amount carried by each unlabeled sample in the unlabeled sample set according to the prediction result and a preset selection policy includes:
respectively calculating the information entropy of each candidate target to obtain an information entropy value;
calculating the intersection ratio of each candidate target boundary frame respectively to obtain an intersection ratio;
and calculating the information quantity carried by each unlabeled sample according to the cross ratio and the information entropy value.
Further, a calculation formula adopted by the calculation of the information amount carried by each unlabeled sample according to the cross ratio and the information entropy is as follows:
or (b)
Or (b)
Or (b)
Wherein x is i For any of the unlabeled samples, the information carried by it has a magnitude of Score (x i ) The boundary box set of the candidate targets of the prediction result isEnt (b) is the information entropy value, ioU (b) is the intersection ratio, alpha is the coefficient of IoU items,x * for the selected target sample.
Further, the calculation formula adopted by the calculation of the information entropy of each candidate target is as follows:
wherein Ent (b) is the entropy of the information of the candidate object b, C is the set of all classification categories in the sample training set,and a probability value of the classification category c corresponding to the candidate target b.
Still further, the step of updating the marked sample set and the unmarked sample set according to the target sample includes:
marking the target sample;
adding the marked target sample into the marked sample set, and deleting the target sample in the unmarked sample set.
Still further, after the step of adding the labeled target sample to the labeled sample set, the method further comprises:
and when judging that the number of the samples in the marked sample set is larger than a number threshold, judging that the marked sample set meets the stop condition.
According to the embodiment of the invention, a better training effect can be achieved by only marking part of samples based on training of the target detection model, the sample marking efficiency is improved, the sample with the highest information is automatically selected by calculating the design of the information carried by each unlabeled sample in the unlabeled sample set according to the prediction result and the preset selection strategy, the accuracy of acquiring the target sample is effectively improved, and the effect of parameter updating of the target detection model is achieved by updating the labeled sample set and the unlabeled sample set.
Drawings
FIG. 1 is a flowchart of an active learning method of an object detection system according to a first embodiment of the present invention;
FIG. 2 is a flowchart of an active learning method of the object detection system according to a second embodiment of the present invention;
FIG. 3 is a block diagram showing the steps of an active learning method of an object detection system according to a second embodiment of the present invention;
FIG. 4 is a schematic diagram of a target detection system according to a third embodiment of the present invention;
the invention will be further described in the following detailed description in conjunction with the above-described figures.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the invention. Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the invention.
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
In order to illustrate the technical scheme of the invention, the following description is made by specific examples.
Example 1
Referring to fig. 1, a flowchart of an active learning method of an object detection system according to a first embodiment of the present invention includes the steps of:
step S10, obtaining a local pre-stored sample training set;
the sample training set comprises a marked sample set and an unmarked sample set, and the proportion of the marked sample set and the unmarked sample set can be set according to requirements;
for example, the obtained sample training set is x= { X 1 ,x 2 ,…,x N Where N is the total number of sample images, there are K classification categories across the entire dataset, c= { C 4 ,c 2 ,…,c K }. Dividing the whole training sample set into a marked sample set L and an unmarked sample set U;
when the training process starts, a small part (10%) of sample data is randomly selected and marked as an initial marked set L, and the rest samples are all used as unlabeled sets U;
step S20, performing model training on a target detection model according to the marked sample set, and performing sample prediction on the unmarked sample set according to the trained target detection model to obtain a prediction result;
training a target detection model by using a data set L, obtaining corresponding parameters after certain training, and applying the learned model to U for prediction to obtain a prediction result, wherein the prediction result comprises probability distribution of each target in each image for K categories and position data of a boundary frame thereof;
step S30, respectively calculating the information quantity carried by each unlabeled sample in the unlabeled sample set according to the prediction result and a preset selection strategy to obtain a plurality of sample information quantities;
based on a prediction result of the unlabeled data set U, calculating a sample information quantity S carried by each unlabeled sample by using a preset selection strategy;
step S40, sorting the unlabeled samples according to the sample information amount, and selecting target samples according to the sorting result;
wherein 5% of the most useful training images are selected by sorting based on the size of the sample information amount S;
step S50, updating the marked sample set and the unmarked sample set according to the target sample;
wherein the training images which are 5% of the most useful training images are marked, and the training images are added into the marked set L and are correspondingly removed from U;
step S60, training of the target detection model, sample prediction for the unlabeled sample set, calculation of information amount carried by each unlabeled sample and selection and marking of the target sample are sequentially executed again according to the updated labeled sample set and the unlabeled sample set until the updated labeled sample set meets a stopping condition, and training of the target detection model is stopped;
the method comprises the steps of training a target detection model, predicting samples of an unlabeled sample set, calculating information quantity, selecting and marking the target samples, so as to achieve the effect of iterative training of the target detection model;
according to the method, the device and the system, the better training effect can be achieved by only marking part of samples based on training of the target detection model, the sample marking efficiency is improved, the information quantity of each unlabeled sample in the unlabeled sample set is calculated according to the prediction result and the preset selection strategy, the sample with the most information quantity is automatically selected, the accuracy of acquiring the target sample is effectively improved, and the parameter updating effect of the target detection model is achieved by updating the labeled sample set and the unlabeled sample set.
Example two
Referring to fig. 2, a flowchart of an active learning method of an object detection system according to a second embodiment of the present invention includes the steps of:
s11, acquiring a local pre-stored sample training set;
wherein the sample training set comprises a marked sample set and an unmarked sample set;
for example, referring to fig. 3, the obtained sample training set is x= { X 1 ,x 2 ,…,x N Where N is the total number of sample images, there are K classification categories across the entire dataset, c= { C 1 ,c 2 ,…,c K }. Dividing the whole training sample set into a marked sample set L and an unmarked sample set U;
when the training process starts, a small part (10%) of sample data is randomly selected and marked as an initial marked set L, and the rest samples are all used as unlabeled sets U;
step S21, performing model training on a target detection model according to the marked sample set, and performing sample prediction on the unmarked sample set according to the trained target detection model to obtain a prediction result;
the prediction result comprises the number of candidate targets in each unlabeled sample, a probability distribution set of each classification category in the sample training set corresponding to each candidate target, and a corresponding space position coordinate set of a boundary frame of each candidate target;
specifically, during the prediction phase, for a given image x i Under the network parameter theta of the target detection algorithm, the output of the model is recorded as the following formula:
wherein N is obj The number of candidate targets obtained by prediction for the image,a probability distribution set corresponding to each classification category for the m candidate prediction target;
the sum of the probabilities for the respective classes is 1, i.e. +.>Box m ＝{(x m ,y m ),(x m_off ,y m_off ) -a corresponding set of spatial position coordinates of a bounding box of the predicted target;
step S31, respectively calculating the information entropy of each candidate target to obtain an information entropy value, and respectively calculating the intersection ratio of each candidate target boundary frame to obtain an intersection ratio;
the calculation formula adopted for calculating the information entropy of each candidate target is as follows:
wherein Ent (b) is the entropy of the information of the candidate object b, C is the set of all classification categories in the sample training set,a probability value of a classification class c corresponding to the candidate target b;
step S41, calculating the information quantity carried by each unlabeled sample according to the cross ratio and the information entropy value to obtain a plurality of sample information quantities;
in this embodiment, in order to define and represent the information of the accuracy, the concept of the intersection ratio IoU, namely Intersection over Union, is introduced first, and is a standard for detecting the positioning accuracy of the corresponding object. IoU is calculated by dividing the overlapping part of two regions by the collective part of two regions, and has the following formula:
wherein, box 4 And Box 2 Two different regions;
in this step, a calculation formula adopted by the calculation of the information amount carried by each unlabeled sample according to the cross ratio and the information entropy value is:
or (b)
Or (b)
Or (b)
Wherein x is i For any of the unlabeled samples, the information carried by it has a magnitude of Score (x i ) The boundary box set of the candidate targets of the prediction result isEnt (b) is the information entropy value, ioU (b) is the intersection ratio, alpha is a coefficient of IoU items, and x * For the selected target sample.
Specifically, the information entropy-cross ratio combination type selection strategy has two forms, namely a product form and an addition form:
(1) Product form of information entropy-cross ratio combining type selection strategy:
the product form multiplies the entropy function with the IoU metric, so the selection function takes into account the positioning error of the prediction bounding box. Since a larger value of the information entropy term in the image information metric formula represents a larger amount of information contained, and a smaller value of IoU represents a larger amount of information contained, the IoU term in the following formula takes the form of '1-IoU' such that a larger value represents a larger amount of information contained. The product form of the combined selection strategy includes functions A, B and C;
the information amount of each image is represented in the function (a) by a value having a product of the minimum information entropy and '1-IoU' among all candidate objects of the image. And the information amount of each image in the function (B) is defined by the sum of the product of the information entropy in all candidate objects and '1-IoU'. In comparison with the function (B), the information amount of each image in the function (C) is defined as that the information entropy items and the '1-IoU' items of all candidate targets are first summed up separately, and then the two are multiplied. The 3 product forms finally select the image with the largest information content fraction as the sample to be marked in the next round.
(2) Information entropy-cross-ratio combined addition form of selection strategy:
like most optimized loss functions, the additive form of the combined selection strategy introduces IoU terms as a canonical term into the selection strategy, with larger α meaning that the more important the positioning information occupies in the total information amount. Compared with the product form, the addition form needs to carry out parameter adjustment on the coefficient alpha to balance the proportion between positioning and classifying information, and the optimal value of the coefficient alpha needs to be adjusted according to the condition of an actual data set;
step S51, sorting the unlabeled samples according to the sample information amount, and selecting a target sample with the largest information amount according to the sorting result;
step S61, marking the target sample, adding the marked target sample into the marked sample set, and deleting the target sample in the unmarked sample set;
step S71, training of the target detection model, sample prediction for the unlabeled sample set, calculation of information amount carried by each unlabeled sample and selection and marking of the target sample are sequentially executed again according to the updated labeled sample set and the unlabeled sample set until the updated labeled sample set meets a stopping condition, and training of the target detection model is stopped;
preferably, in the step, when it is determined that the number of samples in the marked sample set is greater than a number threshold, it is determined that the marked sample set satisfies the stop condition;
according to the method, the sample marking efficiency is improved by training the target detection model to achieve the effect of automatically marking the samples, the design of the information amount carried by each unlabeled sample in the unlabeled sample set is calculated according to the prediction result and the preset selection strategy respectively to automatically pick out the sample with the most information amount, the accuracy of acquiring the target sample is effectively improved, the effect of parameter updating of the target detection model is achieved by updating the marked sample set and the unlabeled sample set, and preferably, the combined selection strategy is combined with the original function by introducing IoU items, so that IoU items are weight or regularized items based on an uncertainty sampling formula, the positioning and classifying tasks of a boundary box are regarded as equally important tasks, ioU items play an important role in evaluating the information amount of the sample in the unlabeled data set, and the product form and the addition form of the information entropy-intersection ratio combined selection strategy optimize the selection strategy, so that the selection strategy is effectively avoided, and the data beneficial to the positioning performance of the target classification is improved only is effectively selected. The introduction of IoU in the selection policy function allows for a more comprehensive assessment of the information content of unlabeled samples.
Example III
Referring to fig. 4, a schematic structural diagram of an object detection system 100 according to a third embodiment of the present invention includes: a training set acquisition module 10, a model training module 11, an information amount calculation module 12, a sample ordering module 13, and a sample set updating module 14, wherein:
a training set obtaining module 10, configured to obtain a locally pre-stored sample training set, where the sample training set includes a marked sample set and an unmarked sample set;
the model training module 11 is configured to perform model training on a target detection model according to the labeled sample set, and perform sample prediction on the unlabeled sample set according to the target detection model after training, so as to obtain a prediction result.
The prediction result comprises the number of candidate targets in each unlabeled sample, a probability distribution set of each candidate target corresponding to each classification category in the sample training set and a corresponding space position coordinate set of a boundary frame of each candidate target.
And the information amount calculating module 12 is configured to calculate the information amount carried by each unlabeled sample in the unlabeled sample set according to the prediction result and a preset selection policy, so as to obtain a plurality of sample information amounts.
Wherein the information amount calculation module 12 is further configured to: respectively calculating the information entropy of each candidate target to obtain an information entropy value; calculating the intersection ratio of each candidate target boundary frame respectively to obtain an intersection ratio; and calculating the information quantity carried by each unlabeled sample according to the cross ratio and the information entropy value.
Specifically, a calculation formula adopted by the calculation of the information amount carried by each unlabeled sample according to the intersection ratio and the information entropy value is as follows:
or (b)
Or (b)
Or (b)
Wherein x is i For any of the unlabeled samples, the information carried by it has a magnitude of Score (x i ) The boundary box set of the candidate targets of the prediction result isEnt (b) is the information entropy value, ioU (b) is the intersection ratio, alpha is a coefficient of IoU items, and x * For the selected target sample.
Further, a calculation formula adopted by the calculation of the information entropy of each candidate target is as follows:
wherein Ent (b) is the entropy of the information of the candidate object b, C is the set of all classification categories in the sample training set,and a probability value of the classification category c corresponding to the candidate target b.
And the sample sorting module 13 is used for sorting the unlabeled samples according to the sample information quantity, and selecting a target sample with the largest information quantity according to the sorting result.
And a sample set updating module 14, configured to update the marked sample set and the unmarked sample set according to the target sample, and re-execute training of the target detection model, sample prediction for the unmarked sample set, calculation of information carried by each unmarked sample, and selection and marking of the target sample according to the updated marked sample set and the unmarked sample set in sequence, until it is determined that the updated marked sample set meets a stop condition, and stop training of the target detection model.
Wherein the sample set update module 14 is further configured to: marking the target sample; adding the marked target sample into the marked sample set, and deleting the target sample in the unmarked sample set.
Furthermore, the sample set update module 14 is further configured to: and when judging that the number of the samples in the marked sample set is larger than a number threshold, judging that the marked sample set meets the stop condition.
According to the method, the device and the system, the better training effect can be achieved by only marking part of samples based on training of the target detection model, the sample marking efficiency is improved, the information quantity of each unlabeled sample in the unlabeled sample set is calculated according to the prediction result and the preset selection strategy, the sample with the most information quantity is automatically selected, the accuracy of acquiring the target sample is effectively improved, and the parameter updating effect of the target detection model is achieved by updating the labeled sample set and the unlabeled sample set.
It will be appreciated by those skilled in the art that the constituent structures depicted in fig. 4 are not limiting of the object detection system of the present invention and may include more or fewer components than illustrated, or may be combined with certain components, or may be arranged differently, while the active learning method of the object detection system of fig. 1-2 may also be implemented using more or fewer components, or may be combined with certain components, or may be arranged differently, as depicted in fig. 4. The units, modules, etc. referred to in the present invention refer to a series of computer programs capable of being executed by a processor (not shown) in the current object detection system and performing specific functions, which may be stored in a storage device (not shown) of the current object detection system.
The above embodiments describe the technical principle of the present invention, and these descriptions are only for explaining the principle of the present invention and should not be construed in any way as limiting the scope of the present invention. Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein, without departing from the spirit and scope of the invention.
Claims (5)
1. An active learning method of an object detection system, the method comprising:
obtaining a local pre-stored sample training set, wherein the sample training set comprises a marked sample set and an unmarked sample set;
model training is carried out on a target detection model according to the marked sample set, and sample prediction is carried out on the unmarked sample set according to the trained target detection model so as to obtain a prediction result;
respectively calculating the information quantity carried by each unlabeled sample in the unlabeled sample set according to the prediction result and a preset selection strategy to obtain a plurality of sample information quantities;
sorting the unlabeled samples according to the sample information amount, and selecting a target sample with the maximum information amount according to a sorting result;
updating the marked sample set and the unmarked sample set according to the target samples, and re-executing training of the target detection model, sample prediction for the unmarked sample set, calculation of information amount carried by each unmarked sample and selection and labeling of the target samples according to the updated marked sample set and the unmarked sample set in sequence, until judging that the updated marked sample set meets a stop condition, stopping training of the target detection model, wherein a prediction result comprises the number of candidate targets in each unmarked sample, a probability distribution set of each candidate target corresponding to each classification category in the sample training set, and a corresponding spatial position coordinate set of a boundary frame of each candidate target, and the step of calculating the information amount carried by each unmarked sample in the unmarked sample set according to the prediction result and a preset selection strategy comprises the following steps of:
respectively calculating the information entropy of each candidate target to obtain an information entropy value;
calculating the intersection ratio of each candidate target boundary frame respectively to obtain an intersection ratio;
and calculating the information quantity carried by each unlabeled sample according to the cross ratio and the information entropy value.
2. The method of active learning of an object detection system according to claim 1, wherein a calculation formula used for calculating an amount of information carried by each of the unlabeled exemplars according to the intersection ratio and the information entropy is:
or (b)
Or (b)
Or (b)
Wherein x is i For any of the unlabeled samples, the information carried by it has a magnitude of Score (x i ) The boundary box set of the candidate targets of the prediction result isEnt (b) is the information entropy value, ioU (b) is the intersection ratio, alpha is a coefficient of IoU items, and x * For the selected target sample.
3. The method for active learning of an object detection system according to claim 1, wherein a calculation formula adopted for calculating the information entropy of each candidate object is:
wherein Ent (b) is the entropy of the information of the candidate object b, C is the set of all classification categories in the sample training set,and a probability value of the classification category c corresponding to the candidate target b.
4. The method of active learning of a target detection system of claim 1, wherein the step of updating the marked sample set and the unmarked sample set, respectively, from the target sample comprises: marking the target sample; adding the marked target sample into the marked sample set, and deleting the target sample in the unmarked sample set.
5. The method of active learning of a target detection system of claim 4, wherein after the step of adding the labeled target sample to the labeled sample set, the method further comprises: and when judging that the number of the samples in the marked sample set is larger than a number threshold, judging that the marked sample set meets the stop condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911396654.XA CN111160469B (en) | 2019-12-30 | 2019-12-30 | Active learning method of target detection system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911396654.XA CN111160469B (en) | 2019-12-30 | 2019-12-30 | Active learning method of target detection system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111160469A CN111160469A (en) | 2020-05-15 |
| CN111160469B true CN111160469B (en) | 2023-08-08 |
Family
ID=70559099
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911396654.XA Active CN111160469B (en) | 2019-12-30 | 2019-12-30 | Active learning method of target detection system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111160469B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111666993A (en) * | 2020-05-28 | 2020-09-15 | 平安科技（深圳）有限公司 | Medical image sample screening method and device, computer equipment and storage medium |
| CN111783844B (en) * | 2020-06-10 | 2024-05-28 | 广东正扬传感科技股份有限公司 | Deep learning-based target detection model training method, device and storage medium |
| CN111898739B (en) * | 2020-07-30 | 2024-02-20 | 平安科技（深圳）有限公司 | Data screening model construction method, data screening method, device, computer equipment and storage medium based on meta learning |
| CN112149721B (en) * | 2020-09-10 | 2023-11-17 | 南京大学 | Target detection method for reducing labeling requirements based on active learning |
| CN112434809B (en) * | 2021-01-26 | 2021-04-06 | 成都点泽智能科技有限公司 | Active learning-based model training method and device and server |
| CN113221875B (en) * | 2021-07-08 | 2021-09-21 | 北京文安智能技术股份有限公司 | Target detection model training method based on active learning |
| CN113642610B (en) * | 2021-07-15 | 2024-04-02 | 南京航空航天大学 | Distributed asynchronous active labeling method |
| CN113723467A (en) * | 2021-08-05 | 2021-11-30 | 武汉精创电子技术有限公司 | Sample collection method, device and equipment for defect detection |
| CN114898171B (en) * | 2022-04-07 | 2023-09-22 | 中国科学院光电技术研究所 | Real-time target detection method suitable for embedded platform |
| CN116385818B (en) * | 2023-02-09 | 2023-11-28 | 中国科学院空天信息创新研究院 | Training method, device and equipment of cloud detection model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657744A (en) * | 2015-01-29 | 2015-05-27 | 中国科学院信息工程研究所 | Multi-classifier training method and classifying method based on non-deterministic active learning |
| CN109492026A (en) * | 2018-11-02 | 2019-03-19 | 国家计算机网络与信息安全管理中心 | A kind of Telecoms Fraud classification and Detection method based on improved active learning techniques |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650780B (en) * | 2016-10-18 | 2021-02-12 | 腾讯科技（深圳）有限公司 | Data processing method and device, classifier training method and system |
-
2019
- 2019-12-30 CN CN201911396654.XA patent/CN111160469B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657744A (en) * | 2015-01-29 | 2015-05-27 | 中国科学院信息工程研究所 | Multi-classifier training method and classifying method based on non-deterministic active learning |
| CN109492026A (en) * | 2018-11-02 | 2019-03-19 | 国家计算机网络与信息安全管理中心 | A kind of Telecoms Fraud classification and Detection method based on improved active learning techniques |
Non-Patent Citations (1)
| Title |
|---|
| Lili Yin等.Incorporate active learning to semi-supervised industrial fault classification.Journal of Process Control.2019,第88-97页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111160469A (en) | 2020-05-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111160469B (en) | Active learning method of target detection system | |
| CN109741332B (en) | Man-machine cooperative image segmentation and annotation method | |
| US10417524B2 (en) | Deep active learning method for civil infrastructure defect detection | |
| CN111784685B (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN110598620B (en) | Deep neural network model-based recommendation method and device | |
| CN107330027B (en) | Weak supervision depth station caption detection method | |
| CN104537359B (en) | A kind of vehicle target detection method and device | |
| RU2014127535A (en) | AUTOMATION OF ANALYSIS OF THE IMAGE | |
| US20200125898A1 (en) | Methods and systems of segmentation of a document | |
| CN110599453A (en) | Panel defect detection method and device based on image fusion and equipment terminal | |
| CN114998595B (en) | Weak supervision semantic segmentation method, semantic segmentation method and readable storage medium | |
| CN110992307A (en) | Insulator positioning and identifying method and device based on YOLO | |
| CN114781514A (en) | Floater target detection method and system integrating attention mechanism | |
| CN115147745A (en) | Small target detection method based on urban unmanned aerial vehicle image | |
| CN116823793A (en) | Device defect detection method, device, electronic device and readable storage medium | |
| CN114299036B (en) | Electronic element detection method and device, storage medium and electronic equipment | |
| CN111768247A (en) | Order-placing rate prediction method, device and readable storage medium | |
| CN114429577A (en) | Flag detection method, system and equipment based on high beacon strategy | |
| CN113887655A (en) | Model chain regression prediction method, device, equipment and computer storage medium | |
| CN111539390A (en) | Small target image identification method, equipment and system based on Yolov3 | |
| CN115439700B (en) | Image processing method and device and machine-readable storage medium | |
| CN115661542A (en) | Small sample target detection method based on feature relation migration | |
| CN116091784A (en) | Target tracking method, device and storage medium | |
| CN110909797B (en) | Image detection method and device, equipment and storage medium | |
| CN112860736A (en) | Big data query optimization method and device and readable storage medium |
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 |