CN114581181A - Method for processing commodities in picture - Google Patents

Abstract

The invention discloses a processing method of commodities in pictures, which can identify, segment and automatically classify commodity objects in the pictures and highlight commodity position information, range information and the like by adopting a machine vision technology based on deep learning. And carrying out commodity SKU object management on commodities contained in any pictures, wherein each identified and labeled commodity SKU object can be further subjected to online ordering processing. The invention firstly separates out commodity objects in mass live pictures, provides a purchasing end user to select commodities in the sea as if the purchasing end user is in a real object spot warehouse, and provides an interactive transaction mode for the purchasing end user to select the commodities one by one to carry out batch purchasing.

Description

Method for processing commodities in picture

Technical Field

The invention relates to the technical field of electronic commerce, in particular to a method for processing commodities in pictures.

Background

In various electronic commerce systems, such as Taobao, Jingdong, Ali baba, and Purchase, whether for enterprise users or consumers, a picture containing a commodity is purchased corresponding to a commodity. And only one picture containing multiple commodities can be sold and processed independently. This is because the purchasing decision for any kind of merchandise generally needs many pictures, accompanied by a lot of written descriptions, and the merchandise is fully displayed from various angles such as vision and connotation knowledge, so that the purchaser can know and recognize the product, thereby obtaining the sales opportunity. Therefore, the industry requirement and practical significance that multiple commodities are displayed in one picture at the same time and are used for independent sales respectively is eliminated.

Further, for the commodities on shelves and sold by any electric commerce system, SKU cataloging and inventory management needs to be performed on the commodities in advance, so that even if a plurality of commodities can be sold on shelves in one picture, each commodity needs to be organized and standardized by the SKU which is pre-cataloged, and then the SKU of the commodity needs to be operated and maintained for the inventory quantity to be actually used for selling by the electric commerce system, so that when the commodities on shelves and sold for a certain picture, each electric commerce system conventionally establishes a SKU for each commodity contained in the system, provides independent display pictures of the commodities, and the operator maintains inventory data of each commodity so as to be used for displaying and selling by the electric commerce system.

The general design concept of the electric commerce systems of generations such as Taobao, Ali, many pieces of information, acquaintance and the like is taken as an industry common sense, and influences practitioners of various industries and corresponding various technical suppliers, but years of industry practice proves that the design of the electric commerce system is difficult to process the actual requirements of the noble metal jewelry batch issuing industry, so that the digitization process of the noble metal batch issuing industry cannot go to end for a long time. The cause of this problem is illustrated below from several perspectives:

1. background of industry status

Precious metal (gold, K gold, platinum and other materials) jewelry wholesale is a B2B transaction scene that thousands of suppliers correspond to dozens of thousands of buyers, the supplier end is a jewelry processing factory with a large number (thousands of Chinese domestic scales or more), the scales of the suppliers are small, the informatization degree is limited by the scale of enterprises, most enterprises do not have complete ERP, PDM and other informatization systems, and most enterprises cannot provide complete, instant and effective data packets (including commodity pictures, specifications and current stock) for the e-commerce system for downstream wholesale retailers.

At present, general suppliers can provide an electronic edition ordering catalog and even an online ordering system, but have no accurate inventory data (reasons are put to the 2 nd expansion), order production cycle commitment is provided, and a channel sales mode of the suppliers produces delivery according to orders or according to market prediction of the suppliers, and funds are raised to produce a batch of free-run spot goods to be sold to various large-batch development halls for downstream purchasers to see edition and select in batches.

Since downstream buyers are distributed all over the country, regular spot purchases to jewelry wholesale markets, visiting and purchasing jewelry exhibition halls, are the most mainstream supply chain channel. Even if a few head suppliers invest in research and development and operate a complete online ordering system, downstream buyers are still dependent on offline exhibition hall purchasing as a main purchasing channel based on various factor constraints. At present, the proportion of purchasing spot goods in batches through real object edition viewing under the line accounts for more than 90% of wholesale trade volume of the noble metal jewelry industry.

2. Checking and settling mode

The design basis of the various E-commerce systems is accurate to one code of a single commodity, but the noble metal jewelry industry has a special phenomenon that a wholesaler is difficult to manage the commodity with one code at low cost, and the daily checking of enterprises in the industry and the trade and settlement units among enterprises are all in a weight recording mode and are not accurate to one code to manage stock and process orders. The reason is mainly based on the following points:

A. one-code management cost is higher, the profit of the precious metal jewelry wholesale industry is low, and after the precious metal jewelry wholesale industry adopts the precious metal jewelry wholesale industry, downstream purchasers do not recognize the cost and cause price overflow.

B. The price of the precious metal material fluctuates frequently and greatly, and the enterprises need to stare at the gram weight clean taken position of the enterprises every day, and each enterprise needs to control the operation risk of the enterprises holding the precious metal material only by keeping the gram weight daily, daily clearance and daily knot of the inventory taken position.

C. The method is characterized in that enterprises with a certain scale enter and warehouse entry of real objects every day, classified display is carried out in exhibition halls, the scale of commodities sold and delivered out of the warehouse by looking at real objects is large, the number of styles is large, the operations at two ends, such as sample-looking purchasing of purchasers and classified settlement of labor cost of suppliers according to purchasing requirements, subpackaging, packaging and delivery and the like, are manually processed by professional specialists, counting of single commodities appears to be unnecessary to increase operation cost on the premise that the purchase, sale and gram weights of all commodities must be accurate, and the method belongs to the process step of optimizing cutting after years of practice, so that fresh settlement and boxing bills in the noble metal jewelry wholesale market have accurate piece number records.

D. The commodities of wholesale enterprises have high mobility, a lot of new commodities are sold in stock and sold at different time periods every day in a mainstream enterprise exhibition hall for offline wholesale operation, individual commodities are purchased empty without being sold at the stock, year, month and month, style iteration is carried out, the commodities are different day and month, even if a considerable number of SKU databases are built in the sold style for which the enterprises do not pay the cost, SKU files are built in a certain period or a certain range of commodity styles, the stocks are refined to be managed to be single pieces, and the situation that the SKU is sold quickly and then continuously produced frequently occurs along with the stop of the sold old commodities and the continuous appearance of new commodities is kept, so that the operation mode of high cost, small profit and cost reduction is realized.

E. The precious metal jewelry old money and new money slightly change in a large number, are continuously and iteratively upgraded, change in rhythm very fast, a commodity seems to be sold every month every year, but put into different batches in different time periods, and have various difference changes all the time, the styles are manually catalogued, under the condition that a PDM database of an upstream supplier is not strictly provided for downstream buyers along with production and supply, long-term, strict classification and coding and non-strict classification of the styles and similar styles are difficult to establish, new and current goods are difficult to identify and enter into correct inventory, and the management complexity of one code is objectively improved.

The current situation of the precious metal jewelry industry is synthesized for the reasons, the constraint conditions of digital operation are summarized, the conventional electronic commerce system which is designed based on one-code management as the basic bottom layer is objectively difficult to popularize and implement in the industry, and enterprises cultivated in the field are restricted by the rules and are difficult to break through the development.

Disclosure of Invention

The invention provides a method for processing commodities in pictures, aiming at the technical problems.

In order to achieve the above purpose, the invention provides the following technical scheme:

a processing method of commodities in pictures comprises the following steps:

s1, deep learning and machine vision recognition are carried out on live pictures by using a pre-trained commodity picture model;

s2, acquiring pre-recorded SKU information of all commodities in the live picture according to the results of deep learning and machine vision recognition;

s3, archiving all live commodities in the graph according to the results of deep learning and machine vision recognition, wherein automatic labeling information, homogeneity information and commodity position information of each live commodity are completely recorded and stored, and the information of each live commodity is combined to generate a complete pre-recorded SKU after being correlated;

s4, when the terminal user is shown in the live picture, dynamic graphic drawing is carried out on the live commodity, the drawing information is derived from the commodity position information collected in the step S3, and the main body position of each live commodity in the picture is highlighted and highlighted, so that the terminal user can further operate all live commodities in the live picture;

s5, establishing a dialogue tracking and real-time draft filing database for all user operations, and using the database for terminal users to edit and operate commodity objects in live commodity pictures at any time;

s6, generating a standby SKU when the live commodity object is collected by the terminal user;

s7, generating a temporary SKU when an end user joins a shopping cart with a live merchandise object

S8, generating formal SKU when the live commodity object has an end user order.

Further, in step S1, the live picture is a photograph or streaming live data or 3D real-time modeling data.

Further, in step S1, the automatic labeling information includes a type, a color, a style, a specification, a process, and a manufacturer.

Further, the color forming includes: ten thousand feet, 3D, 5G, 18K, platinum and diamond.

Further, the categories include: pendant, ring, bracelet, earring, hand and foot chain, necklace, bracelet.

Further, in step S3, the article position information includes the relative position of the article in the picture and the boundary range information.

Further, in step S4, the dynamic graphics are highlighted by a rectangular outer frame, a circular outer frame, or an elliptical outer frame.

Further, the highlighting defined in step S4 includes three operation states, namely: the selectable state, the selected state and the current selection focus state, and the three operation states are respectively represented by three drawing and displaying frames with different colors.

Compared with the prior art, the invention has the beneficial effects that:

the invention is different from the design idea that the common E-commerce system must establish commodity styles to compile target commodity SKUs to enable users to place orders, but firstly separates commodity objects in the sea quantity live pictures through machine vision artificial intelligence automatic identification and automatic classification technology to enable purchasing end users to select commodities in the same way as approaching a real object current warehouse, and provides an interactive transaction mode that the purchasing end users select commodities one by one to carry out batch purchasing, so that interactive experience of batch purchasing can be truly reproduced on any place on line and can enter a large centralized warehouse site, the workload of commodity supply and quantity of warehouses is simplified, and the use experience of batch ordering which can indicate what is bought can be realized in a large warehouse exhibition and sale.

The invention has the advantages that:

1. compression space distance: the buyer can check the latest or even instant stock for purchasing without visiting the site of the warehouse exhibition hall of the supplier, thereby saving the travel cost.

2. Time saving: because the commodities are automatically classified by the machine vision of the artificial intelligence, such as the color, the type, the style and the specification of the jewelry industry, and even the existing SKU approximate matching of the commodity, the commodities can be classified and filed in advance, so that a purchaser can simply traverse all stock spot goods of the suppliers like an exhibition hall warehouse, and can filter the commodities according to the classification of the commodities like a Taobao Jingdong type electric merchant, search the stock spot goods in a fuzzy manner, quickly locate a purchasing target, filter the commodities which do not need to be seen, and find specific goods better than picking in person on site.

3. The timeliness value of the live commodity can be extremely good for live broadcasting of the camera, partial timeliness can be sacrificed, the live commodity is updated daily or periodically, and the live commodity is divided into different regions and shot clearly and delicately.

4. The data operation and maintenance cost of the e-commerce platform is greatly reduced, a supplier can maintain an e-commerce back-end system which is very similar to the e-commerce back-end system which is operated in batch under the line on line without spending a large amount of manpower, funds and technical operation and management experience, a client can be supplied for remote and similar to in-person on-site purchasing, and most importantly, the on-line inventory addition and subtraction data of each SKU is not required to be maintained, which is a great burden for each management cost of an operator of a common e-commerce system.

Drawings

In order to more clearly illustrate the embodiments of the present application or technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings can be obtained by those skilled in the art according to the drawings.

Fig. 1 is a flowchart of a processing method for a commodity in a picture according to an embodiment of the present invention.

Fig. 2 is a picture of performing deep learning and machine vision recognition on a live picture according to an embodiment of the present invention.

Fig. 3 and 4 are pre-recorded SKU information acquisition pictures of all the commodities in the live view provided by the embodiment of the invention.

Fig. 5 is an archived picture of all live goods in the figure provided by an embodiment of the present invention.

Fig. 6 is a pendant position information picture provided by the embodiment of the present invention.

Fig. 7 is a picture of a live commodity operation interface provided by the embodiment of the present invention.

Fig. 8 is a diagram illustrating editing and manipulating a commodity object picture in a live commodity picture according to an embodiment of the present invention.

Fig. 9 is a picture of a live product during a collection operation according to an embodiment of the present invention.

FIG. 10 is a photograph of a live item being added to a shopping cart operation according to an embodiment of the present invention.

FIG. 11 is a photograph of a shopping cart removal operation on live merchandise according to an embodiment of the present invention.

Detailed Description

In order to make the technical solutions of the present invention better understood by those skilled in the art, the present invention will be further described in detail with reference to the accompanying drawings and examples.

The invention relates to a method for processing commodities in pictures, which comprises the technical methods of positioning, identifying, marking, providing purchasing and the like of a commodity main body in the pictures, wherein the commodities in the pictures are identified by an artificial intelligence technology according to any picture (comprising a frame-by-frame picture in live streaming media, hereinafter referred to as a live picture), and commodity information and the positions of the commodities in the pictures are marked; selecting corresponding commodity classification and online display positions according to the identification result of the picture, wherein each commodity (hereinafter, referred to as a live commodity) in the live picture has a plurality of characteristic information to be used later, so that each commodity is marked independently, an information database is established and all the commodities are stored, and the characteristic information of the commodities in the live picture has homogeneity (the commodities in the same live picture have similar or identical information such as the type, the color, the style, the manufacturer, the storage position and the like); therefore, the identification result (hereinafter referred to as automatic labeling information) of each commodity in the live picture and the homogeneity information (hereinafter referred to as homogeneity information) obtained when the live picture is obtained are combined into a complete information base of each commodity, information such as the relative position and the boundary range of the commodity in the picture is added to be labeled as the position (hereinafter referred to as commodity position information) of the commodity, the information is totally prestored into a special database for storing all the live pictures and the contained commodities (hereinafter referred to as SKU: live picture commodity base), so that each commodity has the live picture, and the commodity has all necessary information for generating the SKU (hereinafter referred to as pre-recorded SKU information), and the pre-recorded SKU information is permanently stored once collected so as to be used for extracting and utilizing subsequent functions; subsequently, in the display area of the live goods, each live goods is obviously displayed, the position range is clearly marked so as to be convenient for selection operation, the user can freely select the live goods to complete further ordering and other operations, the pre-stored pre-recorded SKU information is used as necessary information required by filling orders, and even the live goods SKU for the orders are simultaneously created when the first orders are submitted.

The embodiment of the invention provides a method for processing commodities in pictures, which relates to the identification of commodity pictures, the position positioning of multi-subject commodities, and the determination of when to generate SKUs of the commodities according to operations such as user browsing and the like. As shown in fig. 1, the specific implementation flow is as follows:

preprocessing flow S1:

s101, deep learning and machine vision recognition are carried out on live pictures by using a pre-trained commodity picture model; as shown in fig. 2. The artificial intelligence machine vision automatic labeling square frame accurately selects the position and range information of a plurality of commodities in a wholesale market exhibition hall and inventory, the characters outside the square frame are the information of the color, the type and the style of the commodities identified and judged by the machine vision, and different colors represent a new combination of the color, the type and the style. The basic information of the commodity includes: the product comprises the following components in parts, color, style, specification, process, manufacturer information and the like, wherein the product color comprises the following components: ten thousand feet, 3D, 5G, 18K, platinum and diamond; the product categories include: pendant, ring, bracelet, earring, hand and foot chain, necklace, bracelet.

S102, according to the results of deep learning and machine vision recognition, acquiring pre-recorded SKU information of all commodities in the live picture; the accuracy of pre-recorded SKU information of each live commodity is not screened, and recognition feedback results of a collection person manual filling, deep learning and machine vision recognition training model from live pictures, including the position information and the classification judgment result, are completely collected and recorded, as shown in fig. 3 and 4. The artificial intelligence machine vision automatic labeling square frame accurately selects the position and range information of a plurality of commodities in a wholesale market exhibition hall and inventory, the characters outside the square frame are the information of the color, the type and the style of the commodities identified and judged by the machine vision, and different colors represent a new combination of the color, the type and the style. When the picture is collected, the collector is required to provide homogeneous information as comprehensively and accurately as possible so as to archive the commodity information in the picture.

And S103, archiving all live commodities in the graph according to the results of deep learning and machine vision identification, wherein automatic labeling information, homogeneity information, commodity position information and the like of each live commodity are completely recorded and stored, and after the information of each live commodity is correlated, the information of each live commodity is combined to generate a complete pre-recorded SKU which is used as a SKU providing information source for subsequently establishing various live commodities. The commodity position information is as follows: left, top, width, height (' 0.3645 ' 0.4885 ' 0.095 ' 0.103 ') or polygon label is the picture path [ space ] x1, y1, x2, y2, x3, y3, x4, y4, label [ space ]

x1, y1 represent the xy coordinates of the first point, an example of which is as follows: 71.7312,201.002,76.7312,240.002,31.7312,295.002,21.7312,345.002,38.7312,391.002,78.7312,414.002,135.731,406.002,169.731,351.002,151.731,288.002,102.731,238.002,108.731,201.002,100.731,186.002,75.7312,191.002,73,195 and a pendant. As shown in fig. 6.

The presentation process S2:

s201, when a live image display terminal user (consumer) displays a live image, performing dynamic graphic drawing (hereinafter, simply referred to as a live product operation interface) on a live product, wherein drawing information is derived from the position information of the product collected in the step S103, and highlighting the main body position of each live product in the image so as to facilitate the terminal user to further operate all live products in the live image; the drawing highlight display mode described herein is a rectangular outer frame, and includes but is not limited to a rectangle in practical application, the pattern is drawn as positions (left, top, width, height) in left, right, up, and down, the four positions are dynamically drawn as a graph, and three operation states are defined herein, that is: the selectable state, the selected state and the state of the current selection focusing on the focus are respectively represented by drawing the rectangular outer frame with three different colors, for example, a red frame is selectable, a green frame is selected and a white frame is the current focus. In addition, the operable object range is determined according to the presented rectangular outer frames, the focus can be selected in the range of each rectangular outer frame by means of manual touch or mouse sliding, and the like, so as to further operate the live commodity object, and the user performs the next operation for the currently selected object, as shown in fig. 7.

S202, based on the live commodity operation interface presented in S201, a conversation tracking and real-time draft archiving database is established for all user operations, and users can edit and operate commodity objects in live commodity pictures at any time. As shown in fig. 8.

Selection determination process S3:

s301, when a terminal user browses a live commodity operation interface of the live image and collects the interested live commodities, a standby SKU is created for the live commodities at the same time, and the standby SKU is only used for long-term attention and tracking use of a terminal user favorite, so that when the pre-recorded SKU information of the live commodities changes due to reasons (such as information deletion, off-shelf and editing change of an operation and maintenance party), the standby SKU in the favorite of a certain terminal user changes along with the change. As shown in fig. 9.

S302, when an end user adds any live commodity in a live commodity operation interface into a shopping cart, the standby SKU is upgraded to be a temporary SKU, and the temporary SKU is a draft state before the end user submits a batch of order demands, so that the difference between the temporary SKU and the standby SKU is that when the pre-recorded SKU information of the live commodity is changed due to reasons (such as information deletion, off-shelf and editing change by an operation and maintenance party), the temporary SKU in the shopping cart of a certain end user cannot be changed along with the change. As shown in fig. 10 and 11.

And S303, when the terminal user selects the live-event commodity and carries out order submitting action, upgrading the main body information of the live-event commodity from the temporary SKU into a formal SKU while generating an order. Formal SKU is generated when an order is initiated after a terminal user submits a first order requirement, and the formal SKU has the same function as a common electronic commerce SKU, so that the formal SKU is generated when a first live commodity order is generated, the quantity of the live commodities is huge, most of the formal SKU cannot generate real order requirements, and in order to save resource consumption and performance pressure caused by the rising of the quantity of the formal SKU, the formal SKU is created for realizing subsequent functions of an electronic commerce system after the first real order is submitted. The coverage area of each live product main body is as shown in fig. 2 and fig. 3, and the live product in each wire frame is a product main body capable of independent action (the labeled information is descriptive characters on the wire frame, but not limited thereto, more information is stored in the live picture product library).

The dynamic graphical rendering used by the live goods manipulation interface to differentiate, mark and select live goods may use the most basic canvas. However, the method is not limited to display and interactive operation technologies logically, and includes all methods that can use other picture graphic frames to draw, mark and select, such as VR virtual reality or XR augmented reality implementations of a 3D engine, and the logic set forth in the method can be implemented.

The invention provides a method for processing commodities in pictures, which designs a bottom E-commerce design framework, adopts a machine vision technology based on deep learning to identify, segment and automatically classify commodity objects in the pictures, and highlights position information, range information and the like of the commodities. And carrying out commodity SKU objectification management on commodities contained in any picture, wherein each identified and labeled commodity SKU object can be further subjected to online ordering processing.

The invention is a general method, which establishes an operable commodity object for the marked commodity in any figure, allows the operator to see the figure, can select the following method in the list, and further processes the commodity object, the specific operation includes: selecting, forgoing selection, screenshot, numbering, creating as a minimum purchase unit (SKU), placing a shopping cart, removing a shopping cart, adding favorites, removing favorites, editing names, editing categories, editing specifications, editing notes, and the like. The method can complete the requirement of batch selection and batch purchase of purchasers on the premise of extensive management of commodity SKU (stock keeping unit) one by one fine management of all inventory commodities of a supplier, makes a large amount of fragmentary inventory commodities in a warehouse which are not strictly coded and warehoused in a wholesale market into a visual, operable and orderable link of an e-commerce, puts display and popularizes propaganda commodities firstly, receives the selection of ordering of the purchasers, and generates a strict e-commerce SKU and an e-commerce order after the real requirement of ordering and purchasing so that the supplier can simultaneously carry out on-site real object wholesale activities and receive remote electronic order wholesale of the purchasers, and the process does not need to individually catalog and put each commodity on shelf, does not need to maintain the real-time inventory quantity of the commodities, and can send pictures (or live broadcast live streaming images, actually update real-time pictures) on line for sale at any time in case, the online and offline dynamic inventory does not need to be maintained in the selling process, one set of inventory is in stock, and the remote and local selling requirements of the online and offline wholesale clients and the online wholesale clients are met.

The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: it is to be understood that modifications may be made to the technical solutions described in the foregoing embodiments, or equivalents may be substituted for some of the technical features thereof, but such modifications or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. A processing method of commodities in pictures is characterized by comprising the following steps:

s1, deep learning and machine vision recognition are carried out on live pictures by using a pre-trained commodity picture model;

s2, acquiring pre-recorded SKU information of all commodities in the live picture according to the results of deep learning and machine vision recognition;

s3, archiving all live commodities in the graph according to the results of deep learning and machine vision recognition, wherein automatic labeling information, homogeneity information and commodity position information of each live commodity are completely recorded and stored, and the information of each live commodity is combined to generate a complete pre-recorded SKU after being correlated;

s4, when the terminal user is shown in the live view, dynamic graphic drawing is carried out on the live view commodity, the drawing information is derived from the commodity position information collected in the step S3, and the main body position of each live view commodity in the drawing is highlighted and highlighted, so that the terminal user can further operate all live view commodities in the live view;

s5, establishing a dialogue tracking and real-time draft filing database for all user operations, and using the database for terminal users to edit and operate commodity objects in live commodity pictures at any time;

s6, generating a standby SKU when the live commodity object is collected by the terminal user;

s7, generating a temporary SKU when an end user joins the shopping cart with the live merchandise object

S8, generating formal SKU when the live commodity object has an order of the end user.

2. The method for processing commodities in pictures, as claimed in claim 1, wherein in step S1, the live pictures are photos or live streaming data or 3D real-time modeling data.

3. The method for processing the commodity in the picture according to claim 1, wherein in step S1, the automatic labeling information includes type, color, style, specification, process, manufacturer.

4. The method of claim 2, wherein the coloring comprises: ten thousand feet, 3D, 5G, 18K, platinum and diamond.

5. The method for processing commodities in pictures, as claimed in claim 2, wherein the categories include: pendant, ring, bracelet, earring, hand and foot chain, necklace, bracelet.

6. The method for processing the commodity in the picture according to the claim 1, wherein in the step S3, the commodity position information includes the relative position and the boundary range information of the commodity in the picture.

7. The method as claimed in claim 1, wherein the dynamic graphics in step S4 is highlighted by a rectangular frame, a circular frame, or an elliptical frame.

8. The method for processing the commodity in the picture according to claim 1, wherein the highlighting defined in step S4 includes three operation states: the selectable state, the selected state and the current selection focus state, and the three operation states are respectively represented by three drawing and displaying frames with different colors.

Images