CN108446330B - Promotion object processing method and device and computer-readable storage medium - Google Patents
Promotion object processing method and device and computer-readable storage medium Download PDFInfo
- Publication number
- CN108446330B CN108446330B CN201810150833.4A CN201810150833A CN108446330B CN 108446330 B CN108446330 B CN 108446330B CN 201810150833 A CN201810150833 A CN 201810150833A CN 108446330 B CN108446330 B CN 108446330B
- Authority
- CN
- China
- Prior art keywords
- promotion
- objects
- object library
- promotion object
- popularization
- 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0242—Determining effectiveness of advertisements
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Databases & Information Systems (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a promotion object processing method and device and a computer readable storage medium, which can identify promotion objects. The promotion object processing method comprises the following steps: crawling promotion objects regularly, generating a first promotion object library, and generating an attribute tag for each promotion object; the crawled popularization objects are subjected to duplicate removal processing regularly to obtain a second popularization object library; and identifying characteristic information in promotion objects in a second promotion object library by adopting a target detection technology based on deep learning, and adding the identified characteristic information into attribute labels of all promotion objects which are the same as the promotion objects in the first promotion object library. Through batch processing, the image processing speed is improved, and researchers can analyze and research massive popularization objects according to the characteristic information of each popularization object to obtain a popularization effect monitoring report.
Description
Technical Field
The present invention relates to image processing technologies, and in particular, to a method and an apparatus for processing a promotion object, and a computer-readable storage medium.
Background
With the continuous development of the Internet + (Internet Plus), especially the wide application of smart phones, more and more Internet advertisements appear in a mode without detecting codes, and for advertisement research, the advertisements are difficult to research and evaluate in a scientific, complete and efficient mode. Meanwhile, with the development of artificial intelligence technology and the maturity of computing vision technology, the content identification of internet advertisement pictures becomes possible, and a solution is urgently needed for the problem that massive internet advertisements without codes cannot be detected.
Disclosure of Invention
In order to solve the technical problem, the invention provides a promotion object processing method, a promotion object processing device and a computer-readable storage medium, which can identify a promotion object.
In order to achieve the purpose of the invention, the invention provides a method for processing a promotion object, which comprises the following steps:
crawling promotion objects regularly to generate a first promotion object library and generating an attribute tag for each promotion object;
the crawled popularization objects are subjected to duplicate removal processing regularly to obtain a second popularization object library;
and identifying characteristic information in promotion objects in a second promotion object library by adopting a target detection technology based on deep learning, and adding the identified characteristic information into attribute labels of all promotion objects which are the same as the promotion objects in the first promotion object library.
Further, the crawling of the promotion objects periodically, generating a first promotion object library, and generating an attribute tag for each promotion object includes:
presetting multiple user attributes, crawling promotion objects at preset positions in the same time period at preset time intervals, downloading the promotion objects, generating a first promotion object library, and generating an attribute tag for each promotion object, wherein the attribute tags at least comprise: and (5) identification of the promotion object.
Further, the periodically removing the duplicate of the crawled popularization object to obtain a second popularization object library, including:
calculating the perception hash value of each promotion object in the first promotion object library, comparing the Hamming distance of the promotion objects pairwise, setting the promotion objects with the Hamming distances smaller than a preset value into a group, and generating a second promotion object library, wherein information in the second promotion object library comprises group identification and identification of each promotion object in the group.
Further, the identifying, by using a deep learning-based target detection technology, feature information in a promotional object in a second promotional object library, and adding the identified feature information to attribute tags of all promotional objects in the first promotional object library that are the same as the promotional object includes:
selecting one promotion object in any group from a second promotion object library, identifying characteristic information in the promotion object by adopting a target detection technology based on deep learning, and adding the identified characteristic information into the attribute tags of the promotion objects in the first promotion object library and the attribute tags of all other promotion objects in the first promotion object library, which belong to the same group with the promotion object; and performing the processing on all the groups in the second promotion object library.
Further, the identifying the feature information in the promotion object by adopting the target detection technology based on deep learning includes:
extracting a feature map in the popularization object by using a set of Convolutional Neural Network (CNN) layers;
the RPN layer of the regional suggestion network judges whether the anchor belongs to the foreground or the background through a classifier, and then corrects the anchor by utilizing a frame regression box regression to obtain accurate suggestion;
pooling ROI Pooling layers in the region of interest to collect feature maps and propulsal, and extracting a propulsal feature map after integrating the information;
and the full-connection full connect layer judges the type of the suggested feature graph, wherein the type is the feature information in the promotion object.
Further, after the fully connected full connect layer determines the category of the suggested feature map, the method further includes: and obtaining the position offset bbox _ pred of each proposal by using the bounding box regression.
Further, the method further comprises: and performing statistical analysis on the attribute labels of all the promotion objects in the first promotion object library to obtain the promotion effect of the promotion objects.
Furthermore, the promotion object is an internet advertisement without codes, and the characteristic information comprises brand information.
Further, the feature information further includes: the display area of the brand information is the percentage of the total area of the picture.
In order to achieve the purpose of the invention, the invention also provides a promotion object processing device, which comprises a crawling module, a duplicate removal module and a processing module, wherein:
the crawling module is used for crawling promotion objects periodically, generating a first promotion object library and generating an attribute tag for each promotion object;
the duplicate removal module is used for periodically carrying out duplicate removal processing on the crawled popularization objects to obtain a second popularization object library;
the processing module is used for identifying the characteristic information in the promotion objects in the second promotion object library by adopting a target detection technology based on deep learning, and adding the identified characteristic information into the attribute labels of all promotion objects in the first promotion object library, which are the same as the promotion objects.
Further, the crawling module crawls popularization objects periodically to generate a first popularization object library and generates an attribute tag for each popularization object, and the method includes the following steps:
the crawling module presets multiple user attributes, crawls popularization objects at preset positions in the same time period at preset time intervals, downloads the popularization objects, generates a first popularization object library, and generates attribute tags for each popularization object, wherein the attribute tags at least comprise: and (5) identification of the promotion object.
Further, the duplication removing module regularly performs duplication removing processing on the crawled popularization object to obtain a second popularization object library, and the duplication removing processing method comprises the following steps:
the duplication eliminating module calculates a perceptual hash value of each promotion object in the first promotion object library, compares the Hamming distance of the promotion objects pairwise, sets the promotion objects with the Hamming distances smaller than a preset value into a group, and generates a second promotion object library, wherein information in the second promotion object library comprises group identification and identification of each promotion object in the group.
Further, the processing module identifies feature information in the promotion objects in the second promotion object library by using a deep learning-based target detection technology, and adds the identified feature information to the attribute tags of all promotion objects in the first promotion object library, which are the same as the promotion objects, and the method includes:
the processing module selects one promotion object in any group from the second promotion object library, adopts a target detection technology based on deep learning to identify the characteristic information in the promotion object, and adds the identified characteristic information to the attribute tags of the promotion objects in the first promotion object library and the attribute tags of all other promotion objects in the first promotion object library, which belong to the same group with the promotion objects; and performing the processing on all the groups in the second promotion object library.
Furthermore, the device further comprises an analysis module, which is used for performing statistical analysis on the attribute tags of all the promotion objects in the first promotion object library to obtain the promotion effect of the promotion objects.
To achieve the object of the present invention, the present invention also provides a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the above-described method.
According to the embodiment of the invention, firstly, popularization objects (picture materials) in the internet are obtained through a network crawler and the like, then duplication removal processing is carried out, finally, a fast R-CNN deep learning system is used for identifying the characteristic information in the picture materials, and the characteristic information is added into an original popularization object library, so that researchers can analyze and research massive popularization objects according to the characteristic information of each popularization object, and a popularization effect monitoring report is obtained. According to the embodiment of the invention, through the duplicate removal processing, each promotion object does not need to be identified when the characteristic information is identified, the same or similar promotion objects can be identified only once, the identified characteristic information of a certain promotion object is added into the attribute tags of all promotion objects same with the promotion objects, and through the batch processing, the image processing speed is greatly improved, especially for scenes of mass promotion objects.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the example serve to explain the principles of the invention and not to limit the invention.
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a second apparatus according to an embodiment of the present invention;
FIG. 3 is a flow chart of fast R-CNN brand detection and brand identification in an exemplary application of the present invention;
FIG. 4 is a diagram of the fast R-CNN neural network in the application example of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that the embodiments and features of the embodiments in the present application may be arbitrarily combined with each other without conflict.
The steps illustrated in the flow charts of the figures may be performed in a computer system such as a set of computer-executable instructions. Also, while a logical order is shown in the flow diagrams, in some cases, the steps shown or described may be performed in an order different than here.
Example one
This embodiment describes a method for processing a promotion object, as shown in fig. 1, including the following steps:
step 11, crawling promotion objects regularly to generate a first promotion object library and generating an attribute tag for each promotion object;
the promotional objects may be, for example, advertising material obtained over the internet, including but not limited to, advertising material obtained from: websites, APP, internet television, etc. Since the internet advertising material is in a large amount, the period of crawling may be in units of minutes, for example, 30 minutes, 60 minutes, 90 minutes, 120 minutes, or the like.
Step 12, regularly carrying out duplicate removal processing on the crawled popularization objects to obtain a second popularization object library;
because the deduplication processing operation amount is large and cannot be realized in real time, the cycle of the deduplication processing is set by day, and for example, the deduplication processing can be set to be performed on a promotion object crawled on the same day at night or in the morning every day.
In this step, the duplication removal may be performed by grouping the pictures determined to be the same, and the second promotion object library stores information of promotion objects in units of groups.
And step 13, identifying the characteristic information in the promotion objects in the second promotion object library by adopting a target detection technology based on deep learning, and adding the identified characteristic information into the attribute labels of all promotion objects in the first promotion object library, which are the same as the promotion objects.
And repeating the step 13 until all promotion objects in the second promotion object library are processed.
The target detection technique based on deep learning adopted in step 13 in this embodiment is fast R-CNN. If the promotion object crawled in step 11 is an internet advertisement without codes, the feature information may include brand information, and optionally, may further include: the display area of the brand information is the percentage of the total area of the picture.
By adopting the embodiment of the invention, the identical (or similar) pictures are divided into the same group through repeated processing, so that each popularization object does not need to be identified when identifying the characteristic information, the identical or similar popularization objects can be identified only once, the characteristic information of a certain identified popularization object is added into the attribute labels of all the popularization objects identical to the popularization object, and the image processing speed is greatly improved through batch processing. Through the processing of the steps 11-13, the characteristic information of all the promotion objects can be quickly obtained, and the research personnel can conveniently analyze and research the promotion object reaching effect on the promotion objects.
Optionally, after step 13, further comprising:
and step 14, performing statistical analysis on the attribute labels of all the promotion objects in the first promotion object library to obtain the promotion effect of the promotion objects.
In step 11, the crawling of the promotion objects periodically to generate a first promotion object library and generate an attribute tag for each promotion object includes:
presetting multiple user attributes, crawling promotion objects at preset positions in the same time period at preset time intervals, downloading the promotion objects, generating a first promotion object library, and generating an attribute label for each promotion object, wherein the attribute labels at least comprise: and (5) identification of the promotion object. The crawling operation can be implemented by using a web crawler.
In step 12, the periodically performing deduplication processing on the crawled popularization object to obtain a second popularization object library includes:
calculating a perception hash value of each promotion object in the first promotion object library, comparing the Hamming distances of the promotion objects pairwise, setting the promotion objects with the Hamming distances smaller than a preset value into a group, considering the promotion objects with the Hamming distances smaller than the preset value to be identical or similar, and generating a second promotion object library, wherein information recorded in the second promotion object library comprises a group identification, identifications of the promotion objects in the group, and perception hash values corresponding to the promotion objects.
In step 13, the identifying, by using a deep learning-based target detection technique, feature information in a promotional object in a second promotional object library, and adding the identified feature information to attribute tags of all promotional objects in the first promotional object library that are the same as the promotional object includes:
selecting one promotion object (which can be any one) in any group from a second promotion object library, identifying characteristic information in the promotion object by adopting a target detection technology based on deep learning, and adding the identified characteristic information into the attribute tags of the promotion objects in the first promotion object library and the attribute tags of all other promotion objects in the first promotion object library, which belong to the same group with the promotion object; and performing the above processing on all the groups in the second promotion object library until the characteristic information of any promotion object in each group is identified, and adding the characteristic information into the attribute tags of all the promotion objects in the first promotion object library.
Specifically, the identifying the feature information in the popularization object by using the target detection technology based on deep learning includes:
extracting a feature map in the popularization object by using a set of Convolutional Neural Network (CNN) layers;
the RPN layer of the regional suggestion network judges whether the anchor belongs to the foreground or the background through a classifier, and then corrects the anchor by utilizing a frame regression box regression to obtain accurate suggestion;
pooling ROI Pooling layers in the region of interest to collect feature maps and propulsal, and extracting a propulsal feature map after integrating the information;
and the full-connection full connect layer judges the type of the suggested feature graph, wherein the type is the feature information in the promotion object.
Optionally, after the fully connected full connect layer determines the category of the suggested feature map, the method further includes: and obtaining the position offset bbox _ pred of each progressive by using a frame regression box regression, namely the percentage of the display area of the feature information in the total area of the picture.
Example two
The description in the method embodiment is also applicable to this embodiment, and is not repeated in this embodiment. As shown in fig. 2, the apparatus comprises a crawling module 21, a de-duplication module 22 and a processing module 23, wherein:
the crawling module 21 is configured to crawl the promotion objects periodically, generate a first promotion object library, and generate an attribute tag for each promotion object;
the duplication elimination module 22 is configured to periodically perform duplication elimination processing on the crawled popularization object to obtain a second popularization object library;
the processing module 23 is configured to identify feature information in a promotion object in the second promotion object library by using a deep learning-based target detection technology, and add the identified feature information to attribute tags of all promotion objects in the first promotion object library, which are the same as the promotion object.
In an optional embodiment, the crawling module 21 periodically crawls the promotion objects, generates a first promotion object library, and generates an attribute tag for each promotion object, including:
crawling module 21 and predetermine multiple user attribute, interval preset time interval crawls the popularization object of predetermineeing the position in the same time quantum, downloads the popularization object, generates first popularization object storehouse, for every popularization object generation attribute label, the attribute label includes at least: and (5) identification of the promotion object.
In an optional embodiment, the deduplication module 22 performs deduplication processing on the crawled promotional object periodically to obtain a second promotional object library, including:
the duplication elimination module 22 calculates a perceptual hash value of each promotion object in the first promotion object library, compares hamming distances of the promotion objects two by two, sets the promotion objects with hamming distances smaller than a preset value into a group, and generates a second promotion object library, wherein information in the second promotion object library includes a group identifier and identifiers of the promotion objects in the group.
In an optional embodiment, the processing module 23 identifies feature information in a promotional object in a second promotional object library by using a deep learning-based target detection technology, and adds the identified feature information to attribute tags of all promotional objects in the first promotional object library that are the same as the promotional object, including:
the processing module 23 selects one promotional object in any group from the second promotional object library, identifies feature information in the promotional object by using a deep learning-based target detection technology, and adds the identified feature information to the attribute tags of the promotional object in the first promotional object library and the attribute tags of all other promotional objects in the first promotional object library that belong to the same group as the promotional object; and performing the processing on all the groups in the second promotion object library.
The processing module 23 identifies feature information in the promotion object by using a target detection technology based on deep learning, including:
the processing module 23 extracts a feature map in the promotion object by using a set of Convolutional Neural Network (CNN) layers;
the processing module 23 makes the RPN layer of the regional suggestion network judge that the anchor belongs to the foreground or the background through the classifier, and then corrects the anchor by using the frame regression box regression to obtain the accurate suggested deployed;
the processing module 23 pools the ROI pool layer to collect the feature map and the suggested feature map, and extracts the suggested feature map after integrating the information;
the processing module 23 makes the full-connection full connect layer determine the type of the proposed feature map, where the type is the feature information in the promotion object.
Optionally, after the fully connected full connect layer determines the category of the proposed feature map, the processing module 23 further obtains the position offset bbox _ pred of each propofol by using a bounding box regression.
In an optional embodiment, the apparatus further includes an analysis module, configured to perform statistical analysis on the attribute tags of all the promotion objects in the first promotion object library, so as to obtain a promotion effect of the promotion object.
Application example
The example specifically explains the popularization object as an internet code-free advertisement material, and comprises the following steps:
step 1: collecting network advertisement materials without adding codes: in this example, by deploying a distributed web crawler system, various user attributes are simulated, web advertisement material without codes is crawled, and the crawled material is tagged, which specifically includes:
(1) simulating various user attributes to crawl advertisement materials;
the method comprises the steps that a plurality of UA (user agents) are preset in a crawler system, such as age, gender, region, mobile phone model, consumption habit, media habit and the like, and advertisement materials of a target site in the same time period are crawled; downloading advertising material and adding an attribute Label, Spider _ AD _ Label, which includes but is not limited to: SpiderAD _ ID, origin _ URL, AD _ Path, Site _ ID, Media _ Type, Unix _ Time, Area _ ID, AD _ Info, AD _ UA, wherein SpiderAD _ ID represents a unique identifier of an advertisement material, origin _ URL represents an Original URL (Uniform Resource Locator) of the advertisement material, AD _ Path represents a server storage Path of the advertisement material, Site _ ID represents a material source (website or APP), Media _ Tpy represents a Media Type (website, APP, Internet television, etc.) of the advertisement material source, Unix _ Time represents a material crawling Time, Area _ ID represents delivery city information of the advertisement material, and AD _ UA represents UA information used when the advertisement material is crawled. The content of the attribute tags can be increased or decreased according to the requirement of research analysis.
(2) Crawling advertisement materials in different time intervals;
according to the requirement, the crawler can be started to crawl the advertisement materials at intervals of 30 minutes, 60 minutes, 120 minutes and the like.
Step 2: preprocessing material data: using a Perceptual hash algorithm (Perceptual hash algorithm) to perform deduplication on the image material crawled by the crawler at time intervals of days to generate an AD _ Img library, wherein the library includes but is not limited to the following parameters: AD _ Unique _ ID, Spider _ AD _ ID and Img _ Phash, wherein the AD _ Unique _ ID field represents a picture group number with consistent picture content, the Spider _ AD _ ID field is taken from the Spider _ AD _ ID in a Spider _ AD _ Label Label and represents a Unique identifier of an advertisement material, and the Img _ pHash field represents a pHash value of the material; the data preprocessing process specifically comprises the following steps:
(1) calculating the perceptual hash values pHash _ Value of all advertisement materials, and generating an AD _ Img label, wherein the AD _ Img _ ID field is generated in a self-growing mode;
(2) sequentially traversing all the materials, calculating the Hamming distance of the pHash _ Value of the advertisement materials pairwise, if the Hamming distance is smaller than or equal to a preset Value (for example, 0), considering that the two images are the same or similar, grouping the same or similar images into a group, identifying the group number by adopting AD _ Unique _ ID, and enabling the AD _ Unique _ ID to be globally Unique in the system;
(3) and extracting the AD _ Unique _ ID field in the AD _ Img to obtain an AD _ Img _ List, wherein the List only comprises one field of the AD _ Unique _ ID.
And step 3: the intelligent identification system for the advertisement material identifies brand information in the material: randomly extracting an advertisement material corresponding to the AD _ Unique _ ID in the AD _ Img _ List from the AD _ Img library, inputting the advertisement material into a Faster R-CNN deep learning system, and outputting a group of Brand information AD _ Img _ Brand by the system, wherein the group of Brand information comprises but is not limited to: the AD _ Unique _ ID may identify a group of identical or similar materials, one AD _ Unique _ ID may correspond to multiple groups of AD _ Img _ bands (that is, one material contains multiple pieces of Brand information), the branch _ ID represents the content or category of a Brand in the material, and the probability represents the percentage of the display area of the Brand in the total area of the picture. The system core algorithm is a neural network for detecting the position of the fast R-CNN brand and identifying brand content, a flow chart for detecting the fast R-CNN brand and identifying the brand is shown in figure 3, a composition structure of the fast R-CNN Deep learning neural network is shown in figure 4, the ROI Projection in figure 3 is ROI Projection which represents the Projection of an ROI, the ROI Pooling layer is RoI Pooling layer, Fc represents a Full connection layer for Full connection layer, RoI Feature Vector is RoI Feature Vector, Deep ConvNet, Conv Feature Map, Softmax and Bbox regression has no general Chinese technical terms. The 13 conv layers, 13 relu layers and 4 pooling (Pooling) layers in FIG. 4 are arranged as follows: conv layer, relu layer, pooling layer, conv layer, relu layer, conv layer, relu layer, pooling layer, conv layer, relu layer, poling layer, conv layer, relu layer, conv layer, relu layer. The 2 relu layers and 2 fully connected layers in fig. 4 are arranged in the following way: full connection layer, relu layer, full connection layer, relu layer. Conv, relu, Reshape, Softmax in FIG. 4 have no common Chinese terminology.
The feature identification process can be divided into the following four parts:
(1) the fast RCNN extracts feature maps in the material using a set of CNN (Convolutional Neural Network) layers, which will be shared for subsequent RPN (Region suggested Networks) layers and fully connected layers (FC for short);
(2) an RPN (region Proposal networks) network is used for generating region proposals; the layer judges that anchors belong to forkrounded (foreground) or background through a softmax classifier, and then corrects the anchors by using bounding box regression to obtain accurate spots;
(3) the ROI (region of interest) Pooling layer collects the input feature maps and the prosages, extracts the prosages maps after integrating the information, and sends the prosages to the subsequent full-connection layer to judge the target category;
(4) classification; the Classication part utilizes the obtained generic feature maps to pass through full connect layers, then obtains the category (brand) to which each generic belongs specifically by using a softmax classifier, and outputs a cls _ prob probability vector, namely a material content category, which represents the probability of belonging to a certain brand; meanwhile, a bounding box regression (frame regression) is utilized to obtain the position offset bbox _ pred of each propofol for regression of a more accurate target detection frame, namely, the percentage of the display area of the brand content in the total picture area is obtained.
And 4, step 4: outputting the research result of the effect of the advertisement without adding codes:
(1) the method comprises the steps of reverse network image preprocessing, identification and the like, backtracking step by step, finding an original crawler material corresponding to the AD _ Unique _ ID, and adding a brand information Label to an AD _ Info field of the Spider _ AD _ Label;
(2) inputting the identified Spider _ AD _ Label into an existing sample research system, and outputting an uncoded advertisement effect monitoring report by the sample research system according to requirements, wherein the report can comprise key indexes such as PV (page view ), UV (user view, independent visitor), Reach (coverage), frequency, media attribute, user portrait, ROI (return on investment), advertisement touch effect evaluation and SOV (Share of voice, advertisement occupancy) and the like.
The method for monitoring the release effect of the non-code-added advertisement picture comprises four stages, namely, an initial advertisement material data acquisition stage, wherein a webpage of an electronic commerce, a video website and the like and a non-code-added advertisement picture material of an APP terminal are crawled by simulating various user attributes through a distributed web crawler, and the material is labeled; then entering a data preprocessing stage, and performing operations such as duplicate removal and the like on the crawled materials in the stage; then entering an advertisement material identification stage; inputting the preprocessed advertisement material into a Frast R-CNN-based image detection and identification system, wherein the system identifies and marks brand information and scene information in the advertisement material; and finally, in the stage of calculating the effect of the code-free advertisement delivery, the URL of the advertisement material and the related advertisement attribute are input into a sample data research system, and the detection results of the effect of the code-free advertisement delivery, such as the evaluation of the reach effect of the code-free advertisement, the SOV (Share of voice) and the like, are output. According to the method, the Frast R-CNN deep learning system is used for intelligently identifying the code-free picture materials of the Internet, the effect monitoring requirement of the code-free advertisement of the Internet is effectively met, and the method can be applied to the research on the effect of the code-free advertisement in the Internet.
It will be understood by those of ordinary skill in the art that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical units; for example, one physical component may have multiple functions, or one function or step may be performed by several physical components in cooperation. Some or all of the components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.
Although the embodiments of the present invention have been described above, the above description is only for the convenience of understanding the present invention, and is not intended to limit the present invention. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (15)
1. A method for processing promotion objects, comprising:
crawling promotion objects regularly, generating a first promotion object library, and generating an attribute tag for each promotion object; wherein the promotion object comprises: no code advertisement picture is added to the Internet;
grouping the crawled popularization objects regularly to obtain a second popularization object library;
and identifying characteristic information in promotion objects in a second promotion object library by adopting a target detection technology based on deep learning, and adding the identified characteristic information into attribute labels of all promotion objects which are the same as the promotion objects in the first promotion object library.
2. The method of claim 1,
the periodically crawling promotion objects, generating a first promotion object library, and generating an attribute tag for each promotion object, comprising:
presetting multiple user attributes, crawling promotion objects at preset positions in the same time length at preset time intervals, downloading the promotion objects, generating a first promotion object library, adding preset user attributes for each promotion object, and generating attribute tags, wherein the attribute tags at least comprise: and (5) identification of the promotion object.
3. The method of claim 1,
the regularly grouping the crawled popularization objects to obtain a second popularization object library, and the method comprises the following steps:
calculating the perception hash value of each promotion object in the first promotion object library, comparing the Hamming distance of the promotion objects pairwise, setting the promotion objects with the Hamming distances smaller than a preset value into a group, and generating a second promotion object library, wherein information in the second promotion object library comprises group identification and identification of each promotion object in the group.
4. The method of claim 1,
the identifying the feature information in the promotion objects in the second promotion object library by adopting the target detection technology based on deep learning, and adding the identified feature information into the attribute labels of all promotion objects in the first promotion object library, which are the same as the promotion objects, comprises the following steps:
selecting one promotion object in any group from a second promotion object library, identifying characteristic information in the promotion object by adopting a target detection technology based on deep learning, and adding the identified characteristic information into the attribute tags of the promotion objects in the first promotion object library and the attribute tags of all other promotion objects in the first promotion object library, which belong to the same group with the promotion object;
and performing the processing on all the groups in the second promotion object library.
5. The method of claim 4,
the identifying the characteristic information in the promotion object by adopting the target detection technology based on deep learning comprises the following steps:
extracting a feature map in the popularization object by using a set of Convolutional Neural Network (CNN) layers;
the RPN layer of the regional suggestion network judges that the anchor belongs to the foreground or the background through a classifier, and then corrects the anchor by using a frame regression box regression to obtain accurate suggestion;
pooling ROI Pooling layers in the region of interest to collect feature maps and propulsal, and extracting a propulsal feature map after integrating the information;
and the full-connection full connect layer judges the type of the suggested feature graph, wherein the type is the feature information in the promotion object.
6. The method of claim 5,
after the fully connected connect layer determines the category of the proposed feature map, the method further comprises: and obtaining the position offset bbox _ pred of each proposal by using the bounding box regression.
7. The method of claim 1,
the method further comprises the following steps: and performing statistical analysis on the attribute labels of all the promotion objects in the first promotion object library to obtain the promotion effect of the promotion objects.
8. The method according to any one of claims 1 to 7,
the promotion object is an internet advertisement without a code, and the characteristic information comprises brand information.
9. The method of claim 8,
the feature information further includes: the display area of the brand information is the percentage of the total area of the picture.
10. The utility model provides a popularization object processing apparatus which characterized in that, includes and crawls module, removes heavy module and processing module, wherein:
the crawling module is used for crawling promotion objects periodically, generating a first promotion object library and generating an attribute tag for each promotion object; wherein the promotion object comprises: no code advertisement picture is added to the internet;
the duplication removing module is used for grouping the crawled popularization objects periodically to obtain a second popularization object library;
the processing module is used for identifying the characteristic information in the promotion objects in the second promotion object library by adopting a target detection technology based on deep learning, and adding the identified characteristic information into the attribute labels of all promotion objects in the first promotion object library, which are the same as the promotion objects.
11. The apparatus of claim 10,
the crawling module crawls promotion objects periodically to generate a first promotion object library and generates attribute tags for each promotion object, and the crawling module comprises:
the crawling module presets multiple user attributes, crawls popularization objects at preset positions in the same time length at preset time intervals, downloads the popularization objects, generates a first popularization object library, adds preset user attributes for each popularization object, and generates attribute tags, wherein the attribute tags at least comprise: and (5) identification of the promotion object.
12. The apparatus of claim 10,
the duplication removing module regularly carries out grouping processing on the crawled popularization objects to obtain a second popularization object library, and the method comprises the following steps:
the duplication eliminating module calculates a perceptual hash value of each promotion object in the first promotion object library, compares the Hamming distance of the promotion objects pairwise, sets the promotion objects with the Hamming distances smaller than a preset value into a group, and generates a second promotion object library, wherein information in the second promotion object library comprises group identification and identification of each promotion object in the group.
13. The apparatus of claim 10,
the processing module adopts a target detection technology based on deep learning to identify the characteristic information in the promotion objects in the second promotion object library, and adds the identified characteristic information into the attribute labels of all promotion objects in the first promotion object library, which are the same as the promotion objects, and the method comprises the following steps:
the processing module selects one promotion object in any group from the second promotion object library, adopts a target detection technology based on deep learning to identify the characteristic information in the promotion object, and adds the identified characteristic information to the attribute tags of the promotion objects in the first promotion object library and the attribute tags of all other promotion objects in the first promotion object library, which belong to the same group with the promotion objects;
and performing the processing on all the groups in the second promotion object library.
14. The apparatus of claim 10,
the device also comprises an analysis module which is used for carrying out statistical analysis on the attribute labels of all the promotion objects in the first promotion object library to obtain the promotion effect of the promotion objects.
15. A computer-readable storage medium having stored thereon computer instructions, which when executed by a processor, perform the steps of the method of any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810150833.4A CN108446330B (en) | 2018-02-13 | 2018-02-13 | Promotion object processing method and device and computer-readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810150833.4A CN108446330B (en) | 2018-02-13 | 2018-02-13 | Promotion object processing method and device and computer-readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108446330A CN108446330A (en) | 2018-08-24 |
| CN108446330B true CN108446330B (en) | 2022-05-13 |
Family
ID=63192496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810150833.4A Active CN108446330B (en) | 2018-02-13 | 2018-02-13 | Promotion object processing method and device and computer-readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446330B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109740729B (en) * | 2018-12-14 | 2020-12-22 | 安徽寒武纪信息科技有限公司 | Operation method, device and related product |
| CN109447719B (en) * | 2018-12-17 | 2020-10-16 | 杭州柚子街信息科技有限公司 | Target promoted commodity automatic determination method, device, medium and electronic equipment |
| CN113971592B (en) * | 2021-12-23 | 2022-04-19 | 成都易播科技有限公司 | Supervision evaluation method, system and device for promotion information release main body |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2013254921A1 (en) * | 2013-11-07 | 2015-05-21 | Canon Kabushiki Kaisha | Method, apparatus and system for determining a label for a group of individuals represented in images |
| KR20160059403A (en) * | 2014-11-18 | 2016-05-26 | 주식회사 솔루엠 | Apparatus for generating advertisement image using display of a plurality of electronic information labels and advertisement method by thereof |
| CN105677844A (en) * | 2016-01-06 | 2016-06-15 | 北京摩比万思科技有限公司 | Mobile advertisement big data directional pushing and user cross-screen recognition method |
| CN105825396A (en) * | 2016-03-11 | 2016-08-03 | 合网络技术（北京）有限公司 | Co-occurrence-based advertisement label clustering method and system |
| CN105913275A (en) * | 2016-03-25 | 2016-08-31 | 哈尔滨工业大学深圳研究生院 | Clothes advertisement putting method and system based on video leading role identification |
| WO2017019646A1 (en) * | 2015-07-24 | 2017-02-02 | Videoamp, Inc. | Sequential delivery of advertising content across media devices |
| WO2017019643A1 (en) * | 2015-07-24 | 2017-02-02 | Videoamp, Inc. | Targeting tv advertising slots based on consumer online behavior |
| CN106383887A (en) * | 2016-09-22 | 2017-02-08 | 深圳市博安达信息技术股份有限公司 | Environment-friendly news data acquisition and recommendation display method and system |
| CN107203598A (en) * | 2017-05-08 | 2017-09-26 | 广州智慧城市发展研究院 | A kind of method and system for realizing image switch labels |
| CN107358264A (en) * | 2017-07-14 | 2017-11-17 | 深圳市唯特视科技有限公司 | A kind of method that graphical analysis is carried out based on machine learning algorithm |
| CN107545271A (en) * | 2016-06-29 | 2018-01-05 | 阿里巴巴集团控股有限公司 | Image-recognizing method, device and system |
| CN107562742A (en) * | 2016-06-30 | 2018-01-09 | 苏宁云商集团股份有限公司 | A kind of image processing method and device |
| CN107636646A (en) * | 2015-08-03 | 2018-01-26 | 谷歌有限责任公司 | Facility grappling is carried out using the imaging of geo-location |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100082427A1 (en) * | 2008-09-30 | 2010-04-01 | Yahoo! Inc. | System and Method for Context Enhanced Ad Creation |
| US9400925B2 (en) * | 2013-11-15 | 2016-07-26 | Facebook, Inc. | Pose-aligned networks for deep attribute modeling |
-
2018
- 2018-02-13 CN CN201810150833.4A patent/CN108446330B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2013254921A1 (en) * | 2013-11-07 | 2015-05-21 | Canon Kabushiki Kaisha | Method, apparatus and system for determining a label for a group of individuals represented in images |
| KR20160059403A (en) * | 2014-11-18 | 2016-05-26 | 주식회사 솔루엠 | Apparatus for generating advertisement image using display of a plurality of electronic information labels and advertisement method by thereof |
| WO2017019643A1 (en) * | 2015-07-24 | 2017-02-02 | Videoamp, Inc. | Targeting tv advertising slots based on consumer online behavior |
| WO2017019646A1 (en) * | 2015-07-24 | 2017-02-02 | Videoamp, Inc. | Sequential delivery of advertising content across media devices |
| CN107636646A (en) * | 2015-08-03 | 2018-01-26 | 谷歌有限责任公司 | Facility grappling is carried out using the imaging of geo-location |
| CN105677844A (en) * | 2016-01-06 | 2016-06-15 | 北京摩比万思科技有限公司 | Mobile advertisement big data directional pushing and user cross-screen recognition method |
| CN105825396A (en) * | 2016-03-11 | 2016-08-03 | 合网络技术（北京）有限公司 | Co-occurrence-based advertisement label clustering method and system |
| CN105913275A (en) * | 2016-03-25 | 2016-08-31 | 哈尔滨工业大学深圳研究生院 | Clothes advertisement putting method and system based on video leading role identification |
| CN107545271A (en) * | 2016-06-29 | 2018-01-05 | 阿里巴巴集团控股有限公司 | Image-recognizing method, device and system |
| CN107562742A (en) * | 2016-06-30 | 2018-01-09 | 苏宁云商集团股份有限公司 | A kind of image processing method and device |
| CN106383887A (en) * | 2016-09-22 | 2017-02-08 | 深圳市博安达信息技术股份有限公司 | Environment-friendly news data acquisition and recommendation display method and system |
| CN107203598A (en) * | 2017-05-08 | 2017-09-26 | 广州智慧城市发展研究院 | A kind of method and system for realizing image switch labels |
| CN107358264A (en) * | 2017-07-14 | 2017-11-17 | 深圳市唯特视科技有限公司 | A kind of method that graphical analysis is carried out based on machine learning algorithm |
Non-Patent Citations (4)
| Title |
|---|
| TRANSFERRING CNNS TO MULTI-INSTANCE MULTI-LABEL CLASSIFICATION ON SMALL DATASETS;Dong Mingzhi等;《2017 24TH IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)》;20171231;第1332-1336页 * |
| 基于机器学习的监控视频行人检测与追踪系统的设计与实现;吕云翔等;《工业和信息化教育》;20161130;第66-72页 * |
| 基于标签的个性化广告精准营销系统设计与实现;张国燕;《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》;20131215;I138-362 * |
| 基于贝叶斯方法的图像标注研究与系统实现;张广;《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》;20160615;I138-1331 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108446330A (en) | 2018-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220172476A1 (en) | Video similarity detection method, apparatus, and device | |
| CN107944450B (en) | License plate recognition method and device | |
| CN101587622B (en) | Forest rocket detecting and identifying method and apparatus based on video image intelligent analysis | |
| CN112153483B (en) | Information implantation area detection method and device and electronic equipment | |
| Chen et al. | Learning-based spatio-temporal vehicle tracking and indexing for transportation multimedia database systems | |
| CN108446330B (en) | Promotion object processing method and device and computer-readable storage medium | |
| CN112258254B (en) | Internet advertisement risk monitoring method and system based on big data architecture | |
| US20210304796A1 (en) | Data processing method and system, storage medium, and computing device | |
| CN113382279A (en) | Live broadcast recommendation method, device, equipment, storage medium and computer program product | |
| CN108932509A (en) | A kind of across scene objects search methods and device based on video tracking | |
| CN113627402B (en) | Image identification method and related device | |
| CN111127507A (en) | Method and system for determining throwing object | |
| CN112733666A (en) | Method, equipment and storage medium for collecting difficult images and training models | |
| CN114708426A (en) | Target detection method, model training method, device, equipment and storage medium | |
| CN115393751A (en) | Data processing method, storage medium and electronic device | |
| Sumari et al. | Towards practical implementations of person re-identification from full video frames | |
| Ding et al. | Mit-avt clustered driving scene dataset: Evaluating perception systems in real-world naturalistic driving scenarios | |
| CN115705685A (en) | Image data set labeling method and device and electronic equipment | |
| CN113239931A (en) | Logistics station license plate recognition method | |
| CN114639076A (en) | Target object detection method, target object detection device, storage medium, and electronic device | |
| CN112418026A (en) | Vehicle violation detection method, system and device based on video detection | |
| CN111539390A (en) | Small target image identification method, equipment and system based on Yolov3 | |
| CN116310985A (en) | Video stream data-based abnormal data intelligent identification method, device and equipment | |
| CN115115979A (en) | Identification and replacement method of component elements in video and video recommendation method | |
| CN113536928A (en) | High-efficiency unsupervised pedestrian re-identification method and device |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210122 Address after: Room 2020, 2 / F, building 27, No. 25, North Third Ring Road West, Haidian District, Beijing 100089 Applicant after: Beijing minglue Zhaohui Technology Co.,Ltd. Address before: 100012 Room 302, block B, Ruipu building, No.15, Hongjunying South Road, Beiyuan, Chaoyang District, Beijing Applicant before: BEIJING XINSIGHT TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |