CN112328424B - Intelligent anomaly detection method and device for numerical data - Google Patents
Intelligent anomaly detection method and device for numerical data Download PDFInfo
- Publication number
- CN112328424B CN112328424B CN202011396662.7A CN202011396662A CN112328424B CN 112328424 B CN112328424 B CN 112328424B CN 202011396662 A CN202011396662 A CN 202011396662A CN 112328424 B CN112328424 B CN 112328424B
- Authority
- CN
- China
- Prior art keywords
- data
- algorithm
- outliers
- stage
- pool
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
- G06F11/0751—Error or fault detection not based on redundancy
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/20—Drawing from basic elements, e.g. lines or circles
- G06T11/206—Drawing of charts or graphs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses an intelligent anomaly detection method and device for numerical data, wherein the method comprises the following steps: in the data uploading stage, data uploading is realized; a data pool stage, which realizes data storage and data comparison; an algorithm pool stage, which is used for realizing that the system intelligently recommends a plurality of abnormal detection algorithms suitable for current data; an algorithm result integration stage, which is used for summarizing the calculation results of each algorithm and obtaining the final calculation result; an abnormal point judging stage, which realizes the autonomous selection of an abnormal point judging method and makes a judgment; and a detection result visualization stage, which realizes visualization and visual display of data, especially abnormal points. The invention innovatively provides intelligent auxiliary algorithm recommendation, algorithm result integration and abnormal point intelligent judgment and applies the intelligent auxiliary algorithm recommendation, the algorithm result integration and the abnormal point intelligent judgment to the system, thereby greatly simplifying user operation and helping a user to obtain a more accurate and easily observed abnormal detection result in less time.
Description
Technical Field
The invention relates to the field of anomaly detection, in particular to an intelligent anomaly detection method and device for numerical data.
Background
Anomaly detection is the process of detecting objects in a given dataset whose features or behavior differ from those expected, and such objects are referred to as anomalies or outliers. Application scenarios of anomaly detection are many, but at present, no intelligent system or method exists, and an anomaly detection solution can be automatically recommended to a user according to data conditions.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides an intelligent anomaly detection method for numerical data, which comprises the following steps:
the method comprises the following steps: reading a header of data uploaded by a user in a data uploading stage to obtain data content;
step two: comparing the data uploaded in the step one with data in a data pool, and carrying out classification and arrangement according to data characteristic distribution to form a data pool submodule;
step three: under the condition of cold start, the algorithm pool stage recommends an algorithm suitable for current data according to a certain rule, and a user secondarily selects an algorithm needing to be matched and used from the algorithms screened by the system; under the condition of hot start, recommending the optimal algorithm collocation of the rest data sets in the data pool submodule of the data set in the algorithm pool;
step four: the algorithm result integration stage integrates the outliers obtained by the algorithms in the step three;
step five: judging abnormal points of the uploaded data in the first step according to the outliers obtained by calculation in the fourth step;
step six: and visually displaying the detection result judged by the abnormal points in the step five.
Further, the second step is realized by the following sub-steps:
2.1, classifying the uploaded data into a supervised data set and an unsupervised data set according to whether the uploaded data contains a label or not;
2.2 for the supervised data set, classifying the data sets with smaller characteristic dimension span amplitude into a class, and performing dimension reduction processing on the uploaded data if necessary; classifying samples with the edge distribution distance smaller than a threshold value in the supervised data set into one class, and classifying the samples with the condition distribution distance smaller than the threshold value into another class;
2.3 for the unsupervised data set, classifying the data set with smaller characteristic dimension span amplitude into a class, and performing dimension reduction processing on the uploaded data if necessary; and classifying the samples with the edge distribution distance smaller than the threshold value in the unsupervised data set.
Further, the third step is realized by the following sub-steps:
3.1 during cold start, the conditions of the existence of tags and data dimensionality of the data are integrated, and a proper algorithm is screened out; displaying the screened algorithm on an interface for secondary screening by a user;
3.2 when the hot start is carried out, if the same-class data are matched in the step two, skipping the screening step, and directly recommending the same-class data history common algorithm to the user for secondary screening by the user.
Further, the fourth step is realized by the following substeps:
4.1, sequentially processing the original data by using the algorithm screened in the third step and calculating to obtain the corresponding outlier of each data;
4.2 normalizing the outliers of all the data obtained by each algorithm;
4.3, under the cold start environment, averaging the outliers obtained by different algorithms of the same data to be used as the final outlier;
4.4 under the hot start environment, according to the rest of data sets in the same sub-module of the data pool, giving a group of initial outlier labels to the training part of data, adjusting the detection algorithm used by the data, calculating the expected weight of each algorithm by using a neural network, storing the weight result (kernel) to a parameter module in the algorithm pool, and performing weighted integration on all the outliers by using the weight of each algorithm to obtain the final outlier, wherein the weight result (kernel) can be repeatedly used and iterated to achieve a better effect.
Further, the step five is realized by the following sub-steps:
5.1 under the cold starting environment, a user can select one method from a threshold value method and a TopN method to judge the outliers of the four-step integration, and an outlier judging stage simultaneously learns the mode of judging the outliers under the data;
and recommending a threshold value and a TopN value by combining the cantelli inequality, wherein the threshold value and the TopN value are specifically as follows:
wherein Prob represents the probability, yiThe ith algorithm calculation in the fourth expression stepMu is the mean value of the outliers of the step four integration, delta is the variance of the outliers of the step four integration, a is the multiple of delta, a threshold value is calculated with the probability of 0.2 and is taken as a recommended threshold value, and a recommended TopN value is calculated with the threshold value;
5.2 under the hot start environment, if the same-class data is matched in the step two, the abnormal point judgment stage uses the learned abnormal point judgment mode to judge the abnormal points of the outliers of the step four; meanwhile, the user can freely select an abnormal point judgment method to judge the outlier of the step four integration like a cold start mode.
Further, the step six is realized by the following process:
displaying the original data and the corresponding outliers thereof according to the sorting of the outliers from large to small through a table and a histogram; for data in three dimensions and below, additionally drawing a scatter diagram and marking abnormal points by using red; and simultaneously, the chart is subjected to linkage design, and when single data in the table is selected, the selected point is highlighted by the histogram and the scatter diagram.
The invention also provides an intelligent anomaly detection device for numerical data, which comprises the following components from top to bottom:
a task layer which represents a task scene to which the device is applied;
the method comprises a hot start process and a cold start process, wherein the hot start process comprises a data pool and an algorithm pool, the data pool comprises data modules containing the same characteristic data, the algorithm pool comprises algorithm modules and corresponding parameters thereof, and each data module points to a plurality of algorithm modules which are suitable for current data in the algorithm pool; in the cold starting process, a proper algorithm is screened out through a data type, a data dimension and an outlier measurement mode determined by human experience; after the two starting processes are finished, a plurality of outlier values of the corresponding data are obtained;
the anomaly detection layer is used for processing the obtained outliers after the cold and hot starting processes, integrating a plurality of outliers into a final outlier, and judging the anomaly points in the data according to the final outlier;
and in the server layer, the result of the abnormity detection is fed back to the user through the server.
The invention has the beneficial effects that: the invention relates to an intelligent anomaly detection method for numerical data, which is oriented to the anomaly detection field and has the following characteristics:
(1) the invention provides an intelligent anomaly detection method. Through the intelligent flow design, the flow of abnormal detection is greatly optimized, the use threshold of the abnormal detection is reduced, and a better abnormal detection effect can be obtained in a shorter time.
(2) The data pool part related in the invention classifies the data based on a certain strategy so as to carry out mode multiplexing on the similar data in the subsequent process, thereby providing prior experience for users and simultaneously improving the operation efficiency of anomaly detection.
(3) The algorithm result integration part related in the invention can achieve better anomaly detection effect by multiplexing and iterating the algorithm weight for many times, thereby greatly improving the anomaly detection efficiency.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below.
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a logic diagram of the selected portion of the algorithm in the cold start case of the present invention.
FIG. 3 is a logic diagram of the outlier integration portion of the cold start case of the present invention.
FIG. 4 is a logic diagram of the outlier integration portion of the warm boot case of the present invention.
FIG. 5 is a logic diagram illustrating a weight result (kernel) update process in the warm-boot process according to the present invention.
Fig. 6 is a block diagram of the intelligent anomaly detection device for numerical data according to the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, the intelligent anomaly detection method for numerical data according to the present invention includes: in the data uploading stage, data uploading is realized, a header of data uploaded by a user and read data content are read, and the read content is transmitted to a data pool; a data pool stage, which realizes data storage and data comparison; an algorithm pool stage, which is used for realizing that the system intelligently recommends a plurality of abnormal detection algorithms suitable for current data; an algorithm result integration stage, which is used for summarizing the calculation results of each algorithm and obtaining the final calculation result; an abnormal point judging stage, which realizes the autonomous selection of an abnormal point judging method and makes a judgment; and a detection result visualization stage, which realizes visualization and visual display of data, especially abnormal points.
The intelligent anomaly detection method for numerical data specifically comprises the following steps:
the method comprises the following steps: reading a header of data uploaded by a user in a data uploading stage to obtain data content;
step two: comparing the data uploaded in the step one with the data in the data pool, and carrying out classification and arrangement according to the data characteristic distribution to form a data pool submodule;
step three: under the condition of cold start, the algorithm pool stage recommends an algorithm suitable for current data according to a certain rule, and a user can secondarily select an algorithm needing to be matched and used from the algorithms screened by the system; under the condition of hot start, recommending the optimal algorithm collocation of the rest data sets in the data pool submodule of the data set in the algorithm pool;
step four: the algorithm result integration stage integrates the outliers obtained by the algorithms in the step three;
step five: judging abnormal points of the uploaded data in the first step according to the outliers obtained by calculation in the fourth step;
step six: and visually displaying the detection result judged by the abnormal point in the step five.
In the data pool stage, the data pool is composed of different types of sub-modules, and the sub-modules have characteristic distribution differences. Classifying the uploaded data into a supervised data set and an unsupervised data set according to whether the uploaded data contains a label or not; for the supervised data set, classifying the data sets with smaller characteristic dimension span amplitude into one class, and performing dimension reduction processing on the uploaded data if necessary; samples with edge distribution distance smaller than a threshold value in the supervised data set are classified into one class, and samples with conditional distribution distance smaller than the threshold value are classified into another class. For unsupervised data sets, classifying the data sets with smaller characteristic dimension spanning amplitude into a class, and performing dimension reduction processing on uploaded data if necessary; and classifying the samples with the edge distribution distance smaller than the threshold value in the unsupervised data set.
The algorithm pool module is used for automatically recommending an algorithm suitable for the data of the same type for secondary screening of the user when the data are matched with the data of the same type in the data pool; when the data is not matched with the homogeneous data in the data pool, the system autonomous selection algorithm process is as shown in fig. 2: firstly, judging whether the data contains a label, then determining the data dimension (if the dimension reduction operation is carried out, the dimension after dimension reduction is taken as the standard), and screening out a proper algorithm by integrating the conditions of the existence of the label and the data dimension of the data.
In the algorithm result integration stage, the outlier integration under the cold start condition is basically realized as shown in fig. 3: the method comprises the steps of firstly, using m algorithms finally selected from an algorithm pool to calculate m outliers of data uploaded by a user, then carrying out normalization processing on the outliers of all the data obtained by each algorithm, and finally averaging the outliers calculated by different algorithms of the same data, thereby calculating the final outlier of each data. The outlier integration in the hot start case is shown in fig. 4: the method comprises the steps of firstly, using m algorithms finally selected from an algorithm pool to obtain m outliers of data uploaded by a user, then normalizing the outliers of all the data obtained by each algorithm, and finally, using a known weight result (kernel) to perform weighted integration on algorithm results so as to obtain the final outlier of each data. In the warm boot process, the weight result (kernel) update process is shown in fig. 5: after the outlier normalization is completed, if k original data can be matched to a near point in the data pool sub-module, the outliers of the k original data are used as input, the outliers corresponding to the near point are used as training targets, the algorithm corresponding to the kernel enters the neural network to optimize the kernel, and the new kernel is updated to the algorithm pool sub-module.
In the abnormal point judgment stage, under a cold start environment, a user can select one method from a threshold value method and a TopN method to judge the abnormal point of the outlier integrated in the step four, and the abnormal point judgment stage simultaneously learns the abnormal point judgment mode under the data. And recommending a threshold value and a TopN value by combining the cantelli inequality, wherein the threshold value and the TopN value are specifically as follows:
wherein Prob represents the probability, yiAnd expressing the outlier value calculated by the ith algorithm in the fourth step, mu is the mean value of the outliers of the fourth step, delta is the variance of the outliers of the fourth step, a is the multiple of delta, a threshold value is calculated by the probability of 0.2 and is taken as a recommended threshold value, and a recommended TopN value is calculated by the threshold value. Under the hot start environment, if the same-class data are matched in the step two, the abnormal point judgment stage uses the learned abnormal point judgment mode to judge the abnormal points of the outliers integrated in the step four; meanwhile, the user can freely select an abnormal point judgment method to judge the outlier of the step four integration like a cold start mode.
And in the detection result visualization stage, original data and corresponding outliers are displayed in an order from large to small according to the table and the histogram. Meanwhile, for data in three dimensions and below, a scatter diagram is additionally drawn and abnormal points are marked by using red. And simultaneously, the chart is subjected to linkage design, and when single data in the table is selected, the selected point is highlighted by the histogram and the scatter diagram.
As shown in fig. 6, the intelligent anomaly detection device for numerical data according to the present invention adopts the following construction from top to bottom:
the top layer is a task layer and represents a task scene applicable to the device.
The task layer is divided into a hot start process and a cold start process. The hot start process comprises a data pool and an algorithm pool, wherein the data pool comprises data modules containing the same characteristic data, the algorithm pool comprises algorithm modules and corresponding parameters thereof, and each data module points to a plurality of algorithm modules which are suitable for current data in the algorithm pool. The cold start process screens out the appropriate algorithm by data type, data dimension and artificial experience determined outlier measurement. After the two starting processes are finished, a plurality of outlier values of the corresponding data can be obtained.
And the anomaly detection layer processes the data outliers obtained after the cold and hot starting processes. And after integrating the plurality of outliers into a final outlier through a certain method, judging abnormal points in the data according to the final outlier.
And in the server layer, the result of the abnormity detection is fed back to the user through the server.
Claims (5)
1. An intelligent anomaly detection method for numerical data, characterized by comprising the steps of:
the method comprises the following steps: reading a header of data uploaded by a user in a data uploading stage to obtain data content;
step two: comparing the data uploaded in the step one with data in a data pool, and carrying out classification and arrangement according to data characteristic distribution to form a data pool submodule;
step three: under the condition of cold start, the algorithm pool stage recommends an algorithm suitable for current data according to a certain rule, and a user secondarily selects an algorithm needing to be matched and used from the algorithms screened by the system; under the condition of hot start, recommending optimal algorithm collocation of the rest data sets in the algorithm pool in the data pool submodule to which the data set belongs;
step four: the algorithm result integration stage integrates the outliers obtained by the algorithms in the step three;
step five: judging abnormal points of the uploaded data in the first step according to the outliers obtained by calculation in the fourth step;
step six: carrying out visual display on the detection result judged by the abnormal points in the step five;
the second step is realized by the following substeps:
2.1, classifying the uploaded data into a supervised data set and an unsupervised data set according to whether the uploaded data contains a label or not;
2.2 for the supervised data set, classifying the data sets with smaller characteristic dimension span amplitude into a class, and performing dimension reduction processing on the uploaded data if necessary; classifying samples with the edge distribution distance smaller than a threshold value in the supervised data set into one class, and classifying the samples with the condition distribution distance smaller than the threshold value into another class;
2.3 for the unsupervised data set, classifying the data set with smaller characteristic dimension spanning amplitude into a class, and performing dimension reduction processing on the uploaded data when necessary; and classifying the samples with the edge distribution distance smaller than the threshold value in the unsupervised data set.
2. An intelligent anomaly detection method for numerical data according to claim 1, characterized in that said step three is realized by the following sub-steps:
3.1 during cold start, the conditions of the existence of tags and data dimensionality of the data are integrated, and a proper algorithm is screened out; displaying the screened algorithm on an interface for secondary screening by a user;
3.2 when the hot start is carried out, if the same-class data are matched in the step two, skipping the screening step, and directly recommending the same-class data history common algorithm to the user for secondary screening by the user.
3. An intelligent anomaly detection method for numerical data according to claim 1, characterized in that said step four is implemented by the following sub-steps:
4.1, sequentially processing the original data by using the algorithm screened in the third step and calculating to obtain the corresponding outlier of each data;
4.2 normalizing the outliers of all the data obtained by each algorithm;
4.3, under the cold start environment, averaging the outliers obtained by different algorithms of the same data to be used as the final outlier;
4.4 under the hot start environment, according to the rest of data sets in the same sub-module of the data pool, giving a group of initial outlier labels to the training part of data, adjusting the detection algorithm used by the data, calculating the expected weight of each algorithm by using a neural network, storing the weight result (kernel) to a parameter module in the algorithm pool, and performing weighted integration on all the outliers by using the weight of each algorithm to obtain the final outlier, wherein the weight result (kernel) can be repeatedly used and iterated to achieve a better effect.
4. An intelligent anomaly detection method for numerical data according to claim 1, characterized by said step five being implemented by the following sub-steps:
5.1 under the cold starting environment, a user can select one method from a threshold value method and a TopN method to judge the outliers of the four-step integration, and an outlier judging stage simultaneously learns the mode of judging the outliers under the data;
and recommending a threshold value and a TopN value by combining the cantelli inequality, wherein the threshold value and the TopN value are specifically as follows:
wherein Prob represents the probability, yiExpressing the outlier value calculated by the ith algorithm in the fourth step, mu is the mean value of the outliers integrated in the fourth step, delta is the variance of the outliers integrated in the fourth step, a is the multiple of delta, a threshold value is calculated according to the probability of 0.2 and is taken as a recommended threshold value, and meanwhile, a recommended TopN value is calculated according to the threshold value;
5.2 under the hot start environment, if the same-class data are matched in the second step, the abnormal point judgment stage uses the learned abnormal point judgment mode to judge the abnormal points of the outliers integrated in the fourth step; and meanwhile, the user can freely select the outlier judging method to judge the outlier integrated in the step four like a cold starting mode.
5. The intelligent anomaly detection method for numerical data according to claim 1, characterized in that said step six is realized by the following process:
displaying the original data and the corresponding outliers thereof according to the sorting of the outliers from large to small through a table and a histogram; for data in three dimensions and below, additionally drawing a scatter diagram and marking abnormal points by using red; and simultaneously, the chart is subjected to linkage design, and when single data in the table is selected, the selected point is highlighted by the histogram and the scatter diagram.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011396662.7A CN112328424B (en) | 2020-12-03 | 2020-12-03 | Intelligent anomaly detection method and device for numerical data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011396662.7A CN112328424B (en) | 2020-12-03 | 2020-12-03 | Intelligent anomaly detection method and device for numerical data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112328424A CN112328424A (en) | 2021-02-05 |
CN112328424B true CN112328424B (en) | 2022-05-06 |
Family
ID=74301998
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011396662.7A Active CN112328424B (en) | 2020-12-03 | 2020-12-03 | Intelligent anomaly detection method and device for numerical data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112328424B (en) |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3832281B2 (en) * | 2001-06-27 | 2006-10-11 | 日本電気株式会社 | Outlier rule generation device, outlier detection device, outlier rule generation method, outlier detection method, and program thereof |
WO2006002294A2 (en) * | 2004-06-22 | 2006-01-05 | Eric Hirschhorn | Methods and apparatus for online foreign currency exchange |
US10614373B1 (en) * | 2013-12-23 | 2020-04-07 | Groupon, Inc. | Processing dynamic data within an adaptive oracle-trained learning system using curated training data for incremental re-training of a predictive model |
CN107993139A (en) * | 2017-11-15 | 2018-05-04 | 华融融通(北京)科技有限公司 | A kind of anti-fake system of consumer finance based on dynamic regulation database and method |
CN107909472B (en) * | 2017-12-08 | 2020-11-03 | 深圳壹账通智能科技有限公司 | Operation data auditing method, device and equipment and computer readable storage medium |
US20200356544A1 (en) * | 2019-05-07 | 2020-11-12 | Workday, Inc. | False positive detection for anomaly detection |
CN110209560B (en) * | 2019-05-09 | 2023-05-12 | 北京百度网讯科技有限公司 | Data anomaly detection method and detection device |
CN110287238B (en) * | 2019-06-26 | 2022-11-29 | 广东奥博信息产业股份有限公司 | Method and system for detecting abnormal water quality based on priori knowledge |
CN110615001B (en) * | 2019-09-27 | 2021-04-27 | 汉纳森(厦门)数据股份有限公司 | Driving safety reminding method, device and medium based on CAN data |
-
2020
- 2020-12-03 CN CN202011396662.7A patent/CN112328424B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN112328424A (en) | 2021-02-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kukreja et al. | A Deep Neural Network based disease detection scheme for Citrus fruits | |
US10671853B2 (en) | Machine learning for identification of candidate video insertion object types | |
Kao et al. | Visual aesthetic quality assessment with a regression model | |
CN109886335B (en) | Classification model training method and device | |
US20030063779A1 (en) | System for visual preference determination and predictive product selection | |
Kumari et al. | Hybridized approach of image segmentation in classification of fruit mango using BPNN and discriminant analyzer | |
Xia et al. | Dilated multi-scale cascade forest for satellite image classification | |
CN111984824A (en) | Multi-mode-based video recommendation method | |
CN111523421A (en) | Multi-user behavior detection method and system based on deep learning and fusion of various interaction information | |
CN112396428B (en) | User portrait data-based customer group classification management method and device | |
US20230051564A1 (en) | Digital Image Ordering using Object Position and Aesthetics | |
CN112749330B (en) | Information pushing method, device, computer equipment and storage medium | |
CN111222575A (en) | KLXS multi-model fusion method and system based on HRRP target recognition | |
CN112328424B (en) | Intelligent anomaly detection method and device for numerical data | |
GB2604706A (en) | System and method for diagnosing small bowel cleanliness | |
US20220405299A1 (en) | Visualizing feature variation effects on computer model prediction | |
Pereira et al. | Assessing active learning strategies to improve the quality control of the soybean seed vigor | |
CN113254513B (en) | Sequencing model generation method, sequencing device and electronic equipment | |
CN115620083A (en) | Model training method, face image quality evaluation method, device and medium | |
CN114706481A (en) | Live shopping interest degree prediction method based on eye movement characteristics and deep FM | |
CN108287902B (en) | Recommendation system method based on data non-random missing mechanism | |
CN116862626B (en) | Multi-mode commodity alignment method | |
Suhendar et al. | Fruit quality classification using convolutional neural network | |
Wang et al. | Evaluating the Efficiency of the Classifier Method When Analysing the Sales Data of Agricultural Products | |
Wen et al. | When Distortion Meets Perceptual Quality: A Multi-task Learning Pipeline |
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 |