AI-based intelligent selection system and method for dry products purchased by customers

Abstract

The invention relates to the technical field of dry cargo purchase and selection, and discloses an AI-based intelligent selection system and method for dry cargo products purchased by customers, and the system comprises a dry cargo product selection search module, a dry cargo product selection processing module, and a dry cargo product selection feedback module. According to the single dry goods object frame selection image information of the dry goods product selection site, the intelligent search algorithm and the target dry goods product qualified overall feature image library are combined to comprehensively and intelligently evaluate the single dry goods object selection quality of the dry goods product selection site, and efficient and accurate selection of the dry goods products is achieved; a single dry goods object selection result image of a dry goods product selection site is combined with image feature analysis and a dry goods object frame selection image of the dry goods product selection site for image matching and combination to scientifically construct result image information of the dry goods product selection site, and accurate marking of the selection quality of a single dry goods object of the dry goods product selection site is achieved. And the dry goods purchasing experience and the dry goods purchasing quality of customers are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dry goods purchase and selection, and specifically to an intelligent selection system and method for customers to purchase dry goods products based on AI. Background Art

[0002] Dry goods generally refer to seasonings and foods that have had their moisture removed by methods such as air-drying and sun-drying. Common dry goods include agaric, laver, shiitake mushrooms, red dates, cinnamon, chili peppers, Chinese prickly ash, star anise, fennel, pepper, wolfberries, longans, peanuts, dried tangerine peel, raisins; among them, dry goods are classified into dried fruits, dried vegetables, seasonings, miscellaneous grains, beverages, medicinal diets, and preserved fruits according to their uses; the process of selecting dry goods highly tests the experience of customers, and factors such as the color, shape, smell, origin, shelf life, and brand of dry goods need to be considered when selecting dry goods; how to select qualified and high-quality dry goods products for customers among a wide variety of dry goods products has become an important step in dry goods purchase; existing product recommendation systems cannot intelligently select high-quality dry goods products for customers on-site based on the variety, origin, and brand of dry goods products, nor can they visually distinguish and present the selection results of dry goods products based on image identification, reducing the shopping experience of customers for dry goods products.

[0003] The Chinese patent application with the publication number CN115526674A discloses a commodity recommendation method, storage medium, and product. The commodity demand information of the user is obtained through the user terminal, the recommended commodities are determined based on the commodity demand information, and the recommended commodities are sent to the user terminal for display. The user can quickly select the commodities to be purchased from the recommended commodities according to their own needs; however, the above technical solution cannot select high-quality commodities for the user at the shopping site according to the user's needs, nor can it visually distinguish and feedback high-quality commodities and problematic commodities for the on-site commodities. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] To solve the problem that the existing product recommendation system cannot intelligently select high-quality dry goods products for customers on-site based on the variety, origin, and brand of dry goods products, nor can it visually distinguish and present the selection results of dry goods products based on image identification, which reduces the shopping experience of customers for dry goods products, and to achieve the above purposes of online collecting the feature information of dry goods products and the images of the dry goods product selection site, accurately searching for the calibrated images of single dry goods objects of the target dry goods products, intelligently matching the qualified overall feature image library of the target dry goods products, scientifically generating the framed images of single dry goods objects at the dry goods product selection site, autonomously segmenting the framed images of single dry goods objects at the dry goods product selection site, intelligently analyzing the qualified results of single dry goods objects at the dry goods product selection site, scientifically matching the identification watermark images of the qualified analysis results of single dry goods objects at the dry goods product selection site, accurately generating the selection result images of single dry goods objects at the dry goods product selection site, scientifically constructing the selection result images at the dry goods product selection site, and visually and differentially feedbacking the on-site selection results of dry goods products.

[0006] (II)Technical Solutions

[0007] The present invention is realized through the following technical solutions: An intelligent selection method for customers to purchase dry goods products based on AI, characterized in that the method includes the following steps:

[0008] S1. Collect the feature text data of the target dry goods products and the image data of the dry goods product selection site;

[0009] S2. Perform a search process for the calibrated reference images of single dry goods objects in the on-site images of the target dry goods product selection based on the feature text data of the target dry goods products and the calibrated image data of single dry goods objects of the dry goods products, and generate the calibrated image data of single dry goods objects of the target dry goods products;

[0010] S3. Perform a matching process for the qualified overall feature image library required for the selection analysis of the target dry goods products based on the feature text data of the target dry goods products and the qualified overall feature image library of the dry goods products, and generate the qualified overall feature image library of the target dry goods products;

[0011] S4. Perform a calibration process for single dry goods objects in the on-site images of the target dry goods product selection based on the image data of the dry goods product selection site and the calibrated image data of single dry goods objects of the target dry goods products, generate the framed image data of single dry goods objects at the dry goods product selection site, and perform a single dry goods object image segmentation process on the framed images of single dry goods objects at the dry goods product selection site to generate the framed image data of single dry goods objects at the dry goods product selection site;

[0012] S5. Select the image data of a single dry-goods object at the dry-goods product selection site and analyze and process the quality qualification results of the single dry-goods object selection at the dry-goods product selection site with the qualified overall feature image library of the target dry-goods product to generate the analysis data of the qualified results of the single dry-goods object at the dry-goods product selection site;

[0013] S6. Perform the matching process of the identification watermark image required for different selection quality qualification analysis results of a single dry-goods object at the dry-goods product selection site based on the analysis data of the qualified results of the single dry-goods object at the dry-goods product selection site and the identification watermark image data of the dry-goods object qualified analysis results, generate the identification watermark image data of the qualified analysis results of the single dry-goods object at the dry-goods product selection site, and perform the selection result image identification process of the single dry-goods object at the dry-goods product selection site with the image data of the selected single dry-goods object at the dry-goods product selection site to generate the selection result image data of the single dry-goods object at the dry-goods product selection site;

[0014] S7. Construct the result image data of the dry-goods product selection site and perform the feedback operation of the on-site selection results of the dry-goods product.

[0015] Preferably, the operation steps of collecting the characteristic text data of the target dry-goods product and the image data of the dry-goods product selection site are as follows:

[0016] S11. Input the characteristic text information of the dry-goods product to be purchased by the customer online through the mobile terminal and generate the characteristic text data of the target dry-goods product , the characteristic text data of the target dry-goods product includes the name text data, variety text data, origin text data, and brand text data of the dry-goods product; the mobile terminal includes any one of a smart phone, a smart watch, and a tablet computer;

[0017] Take the on-site selection image information of the dry-goods product to be purchased by the customer in the unpacked state online through the mobile terminal and generate the image data of the dry-goods product selection site .

[0018] Preferably, the operation steps of performing the search process of the reference image for the frame selection and calibration of a single dry-goods object in the on-site image of the target dry-goods product selection based on the characteristic text data of the target dry-goods product and the frame selection and calibration image data of a single dry-goods object of the dry-goods product to generate the frame selection and calibration image data of a single dry-goods object of the target dry-goods product are as follows:

[0019] S21. Establish a set of frame selection and calibration image data of a single dry-goods object of the dry-goods product , ; where represents the frame selection and calibration image data of a single dry-goods object corresponding to the th type of dry-goods product combination characteristic type, Indicates the maximum number of dry goods product feature types; the dry goods product combination feature type indicates an index identification type mainly formed by a combination of name, variety, origin and brand feature information for searching for a single dry goods object frame selection calibration image of a specific dry goods product; the single dry goods object frame selection calibration image data of the dry goods product indicates a reference image for identifying and framing a single dry goods object image in a selection scene image of different types of dry goods products;

[0020] S22: the target dry goods product feature text data A set of image data for selecting and calibrating a single dry product object of the dry product The image data of a single dry product object selected in the dry product Perform dry goods product feature character matching to search for the target dry goods product feature text data The corresponding single dry goods object of the dry goods product is selected and calibrated by the image data , and generate the target dry product single dry object selection calibration image data through data identification , execute to generate the single dry goods object selection calibration image data of the target dry goods product The specific steps are as follows:

[0021] S221, initialization, updating the maximum number of iterations T and updating the calibration image to search for the pelican population location, the formula for updating the calibration image to search for the pelican population location is as follows: ,in Indicates The calibrated images are searched for the pelican individual in The position of the dimension, i.e. The calibration images search for individual pelicans in dimension The dry goods product single dry goods object box selection calibration image data set The position in the search space, Indicates the position adjustment of random integers, rand indicates a random number in the range [0,1], and Indicated in The upper and lower boundaries of the problem are solved in the dimension The dry goods product single dry goods object box selection calibration image data set Search the search space to find the text data related to the target dry product features The matching dry product single dry product object frame selection calibration image data The number of upper and lower boundaries;

[0022] S222, exploration stage, calibration image search pelican to determine the location of the prey, that is, in the dry goods product single dry goods object frame selection calibration image data set Search the search space to find the text data related to the target dry product features The matching dry product single dry product object frame selection calibration image data Then calibrate the image to search for the individual pelican and select the calibration image data for the single dry goods object where the dry goods product exists The area moves, and the calibration image search pelican adopts the approximate prey strategy to model, so that the algorithm selects the calibration image data set for the single dry object of the dry product The search space is scanned, and the target dry goods product single dry goods object is selected and calibrated to image data The location of the prey in the dry product single dry object box selection calibration image data set The search space is randomly generated, and the formula for the approximate prey strategy is as follows: ,in Indicates the exploration phase after the update The calibrated images are searched for the pelican individual in The position of the dimension, that is, the first The calibration images search for individual pelicans in dimension The dry goods product single dry goods object box selection calibration image data set Search the search space to find the text data related to the target dry product features The matching dry product single dry product object frame selection calibration image data location, For prey in The location of the dimension, that is, the text data of the target dry product characteristics The matching dry product single dry product object frame selection calibration image data The prey is in dimension The dry goods product single dry goods object box selection calibration image data set The position in the search space, is the objective function value of the prey, For the The objective function value of searching for individual pelicans in the calibration images;

[0023] S223, development stage, calibration image search Pelican selects a calibration image data set for a single dry goods object of the dry goods product All the dry goods products in the search space Single dry goods object selection calibration image data Perform a target prey hunting search, that is, select a calibration image data set for a single dry object in the dry product Search the search space to find the text data related to the target dry product features The framed and calibrated image data of a single dry goods object of the dry goods product that matches , model the behavior process of searching for pelicans in the calibrated image. The algorithm checks the positions near the position of the pelican searched in the calibrated image, so that the algorithm converges to a better position. The formula for calculating the pelican searched in the calibrated image during the hunting behavior is as follows: , where represents the position of the th pelican searched in the calibrated image after the update in the development stage in the th dimension, that is, the position of the th pelican searched in the calibrated image after the update in the development stage in the set of framed and calibrated image data of a single dry goods object of the dry goods product with a dimension of ; search for the framed and calibrated image data of a single dry goods object of the dry goods product that matches the target dry goods product feature text data in the search space, the framed and calibrated image data of a single dry goods object of the dry goods product that matches ; the position of, is a random integer of 0 or 2; t is the current iteration number; T is the maximum iteration number;

[0024] S224. When the algorithm meets the maximum iteration number, output the framed and calibrated image data of a single dry goods object of the dry goods product that matches the target dry goods product feature text data ; ;

[0025] S225. Generate the framed and calibrated image data of a single dry goods object of the target dry goods product after data identification from the framed and calibrated image data of a single dry goods object of the dry goods product output in step S224 . .

[0026] Preferably, the operation steps for generating the qualified overall feature image library of the target dry goods product by performing matching processing on the qualified overall feature image library required for selecting the target dry goods product based on the target dry goods product feature text data and the dry goods product qualified overall feature image library are as follows:

[0027] S31. Establish a set of qualified overall feature image libraries of dry goods products , where represents the qualified overall feature image library of dry goods products corresponding to the th type of dry goods product combination feature type; , ; where represents the th qualified overall feature image data of dry goods products in the qualified overall feature image library of dry goods products, represents the qualified overall feature image library of dry goods products​ The maximum value of the number of qualified overall feature images of dry goods products; the qualified overall feature image data of dry goods products represents the overall appearance image information of qualified dry goods products established for different types of dry goods products.

[0028] S32. Use the K-D tree nearest neighbor search algorithm to process the target dry goods product feature text data with the set of qualified overall feature image libraries of dry goods products in the qualified overall feature image library of dry goods products for keyword matching of dry goods product features, search for the qualified overall feature image library of dry goods products corresponding to the target dry goods product feature text data and construct a target qualified overall feature image library of dry goods products , where represents the data of the th target qualified overall feature image of dry goods products.

[0029] Preferably, according to the on-site image data of dry goods product selection and the target dry goods product single dry goods object frame calibration image data, perform single dry goods object frame calibration processing on the on-site image of target dry goods product selection, generate on-site dry goods object frame selection image data of dry goods product selection, and perform single dry goods object image segmentation processing on the on-site dry goods object frame selection image of dry goods product selection. The operation steps for generating on-site single dry goods object frame selection image data of dry goods product selection are as follows:

[0030] S41. Use the FLANN algorithm to perform single dry goods object recognition in the on-site image of dry goods product selection corresponding to the target dry goods product single dry goods object frame calibration image data in the on-site image data of dry goods product selection , and use a closed line to frame and calibrate the recognized single dry goods object in the on-site image of dry goods product selection, and generate on-site dry goods object frame selection image data of dry goods product selection ;

[0031] S42. Perform single dry goods object image feature segmentation extraction processing on the frame selection image of the single dry goods object in the on-site dry goods object frame selection image data of dry goods product selection along the closed line of the frame calibration, and generate a set of on-site single dry goods object frame selection image data of dry goods product selection , ; where represents the on-site dry goods object frame selection image data of dry goods product selection in the Select the image data of a single dry goods object corresponding to the dry goods object at the dry goods product selection site, indicating the image data of the dry goods object selection at the dry goods product selection site and the maximum value of the number of dry goods objects.

[0032] Preferably, according to the image data of a single dry goods object selection at the dry goods product selection site and the target dry goods product qualified overall feature image library, the operation steps for analyzing and processing the qualified result of a single dry goods object selection at the dry goods product selection site to generate the analysis data of the qualified result of a single dry goods object at the dry goods product selection site are as follows:

[0033] S51. Use the FLANN algorithm to combine the image data set of a single dry goods object selection at the dry goods product selection site wherein the image data of a single dry goods object selection at the dry goods product selection site is sorted in an orderly manner according to the dry goods object number and compared with the target dry goods product qualified overall feature image library wherein the target dry goods product qualified overall feature image data for image feature matching, and generate an analysis data set of the qualified result of a single dry goods object at the dry goods product selection site based on the image feature matching result , where represents the analysis data of the qualified result of a single dry goods object corresponding to the th dry goods object segmented from the image data of the dry goods object selection at the dry goods product selection site;

[0034] When and are successfully matched in image features, it indicates that the quality of the th dry goods product is qualified and meets the selection criteria; then output the analysis data of the qualified result of a single dry goods object at the dry goods product selection site as qualified;

[0035] When and fail to match in image features, it indicates that the quality of the th dry goods product is unqualified and does not meet the selection criteria; then output the analysis data of the qualified result of a single dry goods object at the dry goods product selection site as unqualified.

[0036] Preferably, based on the qualified result analysis data of a single dry-goods object at the dry-goods product selection site and the marked watermark image data of the qualified analysis result of the dry-goods object, perform a matching process on the marked watermark images required for different qualified analysis results of a single dry-goods object at the dry-goods product selection site, generate the marked watermark image data of the qualified analysis result of a single dry-goods object at the dry-goods product selection site, and perform a marked processing of the selection result image for a single dry-goods object at the dry-goods product selection site on the boxed image data of a single dry-goods object at the dry-goods product selection site. The operation steps for generating the selection result image data of a single dry-goods object at the dry-goods product selection site are as follows:

[0037] S61. Establish a set of marked watermark image data for the qualified analysis result of the dry-goods object , where represents the qualified result analysis data of a single dry-goods object at the dry-goods product selection site is the marked watermark image data of the qualified analysis result of the dry-goods object corresponding to when it is qualified. At this time, the marked watermark image of the qualified analysis result of the dry-goods object is a watermark image with the word "qualified"; represents the qualified result analysis data of a single dry-goods object at the dry-goods product selection site is the marked watermark image data of the qualified analysis result of the dry-goods object corresponding to when it is unqualified. At this time, the marked watermark image of the qualified analysis result of the dry-goods object is a watermark image with the word "unqualified";

[0038] S62. Use a two-way search algorithm to match the qualified result analysis data of a single dry-goods object at the dry-goods product selection site in the set of qualified result analysis data of a single dry-goods object at the dry-goods product selection site with the marked watermark image data of the qualified analysis result of the dry-goods object in the set of marked watermark image data of the qualified analysis result of the dry-goods object according to the order of the dry-goods object quantity numbering for keyword matching of the qualified analysis result of a single dry-goods object at the dry-goods product selection site, search for the marked watermark image data of the qualified analysis result of the dry-goods object corresponding to the qualified result analysis data of a single dry-goods object at the dry-goods product selection site , and generate a set of marked watermark image data of the qualified analysis result of a single dry-goods object at the dry-goods product selection site , where represents the marked watermark image data of the qualified analysis result of a single dry-goods object corresponding to the boxed image data of the dry-goods object at the dry-goods product selection site in the th dry-goods object divided; is the marked watermark image data of the qualified analysis result of a single dry-goods object at the dry-goods product selection site;

[0039] S63. The set of marked watermark image data of the qualified analysis result of a single dry-goods object at the dry-goods product selection site ​The watermark image data of the qualified analysis result identification of a single dry goods object at the dry goods product selection site described in According to the numbering of the dry goods objects, the set of cropped image data of a single dry goods object at the dry goods product selection site The cropped image data of a single dry goods object at the dry goods product selection site described in Perform image combination and generate a set of selection result image data of a single dry goods object at the dry goods product selection site , where represents the cropped image data of the dry goods object at the dry goods product selection site The th dry goods object corresponding to the selection result image data of a single dry goods object at the dry goods product selection site; the selection result image data of a single dry goods object at the dry goods product selection site represents the feedback image information for identifying the qualified analysis result of the selection of a single dry goods object at the dry goods product selection site.

[0040] Preferably, the operation steps of constructing the result image data of the dry goods product selection site and performing the feedback operation of the dry goods product on-site selection result are as follows:

[0041] S71. Arrange the selection result image data of a single dry goods object at the dry goods product selection site in the set The selection result image data of a single dry goods object at the dry goods product selection site described in Match and combine the image features with the cropped image of the dry goods object at the dry goods product selection site in the dry goods product selection site according to the numbering order of the dry goods objects to construct the result image data of the dry goods product selection site ; ;

[0042] S72. Perform the feedback operation of the dry goods product on-site selection result through the display screen of the mobile terminal for the generated result image data of the dry goods product selection site .

[0043] The AI-based intelligent selection system for customers to purchase dry goods products is used to implement the AI-based intelligent selection method for customers to purchase dry goods products. The system includes a dry goods product selection search module, a dry goods product selection processing module, and a dry goods product selection feedback module;

[0044] The dry goods product selection search module includes a dry goods product feature information collection unit, a dry goods product selection site image collection unit, a dry goods product single dry goods object cropped calibration image storage unit, a target dry goods product single dry goods object cropped calibration image search unit, a dry goods product qualified overall feature image library storage unit, and a target dry goods product qualified overall feature image library matching unit;

[0045] The dry product feature information acquisition unit collects target dry product feature text data through a mobile terminal; the on-site image acquisition unit for dry product selection collects on-site image data of dry product selection through a mobile terminal; the storage unit for the framed and calibrated image of a single dry product object of the dry product stores the framed and calibrated image data of a single dry product object of the dry product; the search unit for the framed and calibrated image of a single dry product object of the target dry product performs search processing on the framed and calibrated reference image of a single dry product object in the on-site image of the target dry product selection according to the target dry product feature text data and the framed and calibrated image data of a single dry product object of the dry product, and generates the framed and calibrated image data of a single dry product object of the target dry product; the storage unit for the qualified overall feature image library of the dry product stores the qualified overall feature image library of the dry product; the matching unit for the qualified overall feature image library of the target dry product performs matching processing on the qualified appearance feature image library required for the selection analysis of the target dry product according to the target dry product feature text data and the qualified overall feature image library of the dry product, and generates the qualified overall feature image library of the target dry product;

[0046] The dry product selection processing module includes an on-site dry product object framed image generation unit for dry product selection, an on-site single dry product object framed image segmentation unit for dry product selection, an on-site single dry product object qualified result analysis unit for dry product selection, a storage unit for the marked watermark image of the qualified analysis result of the dry object, a matching unit for the marked watermark image of the qualified analysis result of the on-site single dry product object for dry product selection, and an on-site single dry product object selection result image generation unit for dry product selection;

[0047] The dry product selection site dry object frame selection image generation unit performs single dry object frame selection calibration processing on the target dry product in the target dry product selection site image based on the dry product selection site image data and the single dry object frame selection calibration image data of the target dry product, and generates dry product selection site dry object frame selection image data; the dry product selection site single dry object frame selection image segmentation unit performs single dry object image segmentation processing on the dry product selection site dry object frame selection image based on the dry product selection site dry object frame selection image data, and generates dry product selection site single dry object frame selection image data; the dry product selection site single dry object qualification result analysis unit performs dry product selection site single dry object selection quality qualification result analysis processing based on the dry product selection site single dry object frame selection image data and the target dry product qualification overall feature image library, and generates dry product selection site single dry object qualification result analysis data; the dry object qualification analysis result identification watermark image storage unit is used to store dry object qualification analysis result identification watermark image data; the dry product selection site single dry object qualification analysis result identification watermark image matching unit performs identification watermark image matching processing required for different selection quality qualification analysis results of single dry objects at the dry product selection site based on the dry product selection site single dry object qualification result analysis data and the dry object qualification analysis result identification watermark image data, and generates dry product selection site single dry object qualification analysis result identification watermark image data; the dry product selection site single dry object selection result image generation unit performs selection result image identification processing on the single dry object at the dry product selection site based on the dry product selection site single dry object qualification analysis result identification watermark image data and the dry product selection site single dry object frame selection image data, and generates dry product selection site single dry object selection result image data;

[0048] The dry product selection feedback module includes a dry product selection site selection result image construction unit and a dry product site selection result feedback unit;

[0049] The dry product selection site selection result image construction unit is used to construct dry product selection site result image data; the dry product site selection result feedback unit performs dry product site selection result feedback operations based on the dry product selection site result image data in combination with the mobile terminal display screen.

[0050] (III) Beneficial effects

[0051] The present invention provides an AI-based intelligent selection system and method for customers to purchase dry products. It has the following beneficial effects:

[0052] 1. Obtain the characteristic information of the target dry goods product and the image parameters of the target dry goods product selection site conveniently and accurately through the mobile terminal, providing reliable data support for the intelligent identification of qualified dry goods products; based on the characteristic information of the target dry goods product, combine the artificial intelligence identification algorithm with the scientifically preset image parameters for frame selection and calibration of a single dry goods object of the dry goods product to accurately search for the reference image for frame selection and calibration of the dry goods object in the target dry goods product selection site image, realizing the accurate and efficient identification of the type of dry goods object at the dry goods product selection site, and improving the reliability of dry goods product selection; according to the characteristic information of the target dry goods product, combine the intelligent search algorithm with the standard-set qualified overall characteristic image library of the dry goods product to independently match the qualified dry goods product appearance characteristic image library required for the analysis of the target dry goods product selection, realizing the intelligent matching of the qualified dry goods product image libraries required for different types of dry goods products, and improving the quality of dry goods product selection.

[0053] 2. Through image analysis, accurately identify and frame all the single dry goods object images at the dry goods product selection site, and precisely segment all the frame selection image information of the single dry goods object at the dry goods product selection site, realizing the accurate extraction of the characteristic image information of all the single dry goods objects at the dry goods product selection site; based on the frame selection image information of the single dry goods object at the dry goods product selection site, combine the intelligent search algorithm with the qualified overall characteristic image library of the target dry goods product to comprehensively and intelligently evaluate the selection quality of the single dry goods object at the dry goods product selection site, realizing the efficient and precise selection of dry goods products and improving the accuracy of dry goods product selection; based on the analysis parameters of the qualified results of the single dry goods object at the dry goods product selection site and the watermark image information of the qualified analysis results of the dry goods object, scientifically match the watermark image for marking the selection quality of the single dry goods object at the dry goods product selection site, realizing the intuitive marking of the dry goods product selection quality; based on the image parameters of the watermark image of the qualified analysis results of the single dry goods object at the dry goods product selection site, combine image analysis with the frame selection image parameters of the single dry goods object at the dry goods product selection site to visually mark the selection result image of the single dry goods object at the dry goods product selection site, realizing the intuitive watermark image marking of the selection result of the single dry goods object at the dry goods product selection site and improving the applicability of dry goods selection.

[0054] 3. Combine the selection result image of the single dry goods object at the dry goods product selection site with image feature analysis and the frame selection image of the dry goods object at the dry goods product selection site to scientifically construct the result image information of the dry goods product selection site, realizing the accurate marking of the selection quality type of the single dry goods object at the dry goods product selection site, facilitating the customer to intuitively and accurately identify the quality of the dry goods product; at the same time, combine the mobile terminal display screen to independently and reliably perform the feedback operation of the dry goods product selection result on-site, realizing the convenient and efficient output of the dry goods product selection result, and improving the customer's dry goods purchase experience and dry goods purchase quality. Brief Description of the Drawings

[0055] Figure 1Schematic diagram of the modules of the AI-based intelligent selection system for customers to purchase dry goods products provided by the present invention;

[0056] Figure 2 Flowchart of the AI-based intelligent selection method for customers to purchase dry goods products provided by the present invention. Detailed implementation manners

[0057] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0058] Embodiments of the AI-based intelligent selection system and method for customers to purchase dry goods products are as follows:

[0059] Embodiment 1:

[0060] Please refer to Figure 1 - Figure 2 , an AI-based intelligent selection method for customers to purchase dry goods products, characterized in that the method includes the following steps:

[0061] S1. Collect target dry goods product feature text data and dry goods product selection site image data;

[0062] S2. Perform search processing on the reference image of the single dry goods object in the target dry goods product selection site image according to the target dry goods product feature text data and the single dry goods object bounding box calibration image data of the dry goods product, and generate the single dry goods object bounding box calibration image data of the target dry goods product;

[0063] S3. Perform matching processing on the qualified dry goods product appearance feature image library required for the target dry goods product selection analysis according to the target dry goods product feature text data and the qualified overall feature image library of the dry goods product, and generate the qualified overall feature image library of the target dry goods product;

[0064] S4. Perform bounding box calibration processing on the single dry goods object in the target dry goods product selection site image according to the dry goods product selection site image data and the single dry goods object bounding box calibration image data of the target dry goods product, generate the dry goods object bounding box image data of the dry goods product selection site, and perform single dry goods object image segmentation processing on the dry goods object bounding box image of the dry goods product selection site to generate the single dry goods object bounding box image data of the dry goods product selection site;

[0065] S5. Analyze and process the quality qualified result of the single dry goods object selection on the dry goods product selection site by comparing the selected image data of the single dry goods object on the site with the target dry goods product qualified overall feature image library, and generate the analysis data of the qualified result of the single dry goods object on the dry goods product selection site.

[0066] S6. Perform the identification watermark image matching process required for different selection quality qualified analysis results of the single dry goods object on the dry goods product selection site based on the analysis data of the qualified result of the single dry goods object on the dry goods product selection site and the identification watermark image data of the dry goods object qualified analysis result, generate the identification watermark image data of the qualified analysis result of the single dry goods object on the dry goods product selection site, and perform the selection result image identification process on the selected image data of the single dry goods object on the dry goods product selection site, and generate the selection result image data of the single dry goods object on the dry goods product selection site.

[0067] S7. Construct the result image data of the dry goods product selection site and perform the feedback operation of the on-site selection result of the dry goods product.

[0068] Furthermore, please refer to Figure 1 - Figure 2 , the operation steps for collecting the target dry goods product feature text data and the image data of the dry goods product selection site are as follows:

[0069] S11. Input the feature text information of the dry goods product to be purchased by the customer online through the mobile terminal, and generate the target dry goods product feature text data , the target dry goods product feature text data includes the name text data, variety text data, origin text data, and brand text data of the dry goods product; the mobile terminal includes any one of a smart phone, a smart watch, and a tablet computer;

[0070] Take the on-site selection image information of the dry goods product to be purchased by the customer in the unpacked state online through the mobile terminal, and generate the image data of the dry goods product selection site .

[0071] The operation steps for generating the selected calibration image data of a single dry goods object in the on-site image of the target dry goods product by performing a search process on the selected calibration reference image of a single dry goods object in the on-site image of the target dry goods product based on the target dry goods product feature text data and the selected calibration image data of a single dry goods object of the dry goods product are as follows:

[0072] S21. Establish a set of selected calibration image data of a single dry goods object of the dry goods product , ; where represents the selected calibration image data of a single dry goods object corresponding to the th type of dry goods product combination feature type, Represents the maximum value of the number of dry product feature types; the dry product combination feature type represents an index identification type for searching a specific dry product single dry object boxed calibration image mainly formed by combining name, variety, origin, and brand feature information; the dry product single dry object boxed calibration image data represents a reference image for single dry object image recognition and boxing in the selection site images of different types of dry products.

[0073] S22. Compare the target dry product feature text data with the dry product single dry object boxed calibration image data set in the dry product single dry object boxed calibration image data to perform character matching of dry product features, search for the dry product single dry object boxed calibration image data corresponding to the target dry product feature text data , and generate the target dry product single dry object boxed calibration image data through data identification . The specific operation steps for generating the target dry product single dry object boxed calibration image data are as follows:

[0074] S221. Initialization, update the maximum number of iterations T and update the position of the pelican population for calibrating images. The formula for updating the position of the pelican population for calibrating images is as follows: , where represents the position of the th pelican individual for calibrating images in the th dimension, that is, the position of the th pelican individual for calibrating images in the search space of the dry product single dry object boxed calibration image data set with a dimension of . represents a random integer for position adjustment, rand represents a random number in the range of [0, 1], and represent the upper and lower boundaries for solving the problem in the th dimension, that is, the upper and lower boundaries of the number of dry product single dry object boxed calibration image data that match the target dry product feature text data in the search space of the dry product single dry object boxed calibration image data set with a dimension of .

[0075] S222. Exploration stage, the pelican for calibrating images determines the position of the prey, that is, in the dry product single dry object boxed calibration image data set Search for the dry product single dry object bounding box calibration image data that matches the target dry product feature text data in the search space Locate the position of the dry product single dry object bounding box calibration image data, and then calibrate the image to search for the pelican individual to move towards the area where the dry product single dry object bounding box calibration image data exists. Model the calibrated image search pelican using the prey-approaching strategy, so that the algorithm scans the search space of the dry product single dry object bounding box calibration image data set The position of the prey in the search space of the dry product single dry object bounding box calibration image data set is randomly generated. The formula for the prey-approaching strategy is as follows: where represents the position of the th calibrated image search pelican individual in the th dimension after the exploration phase update, that is, the position of the th calibrated image search pelican individual in the dry product single dry object bounding box calibration image data set with a dimension of Search for the dry product single dry object bounding box calibration image data that matches the target dry product feature text data in the search space Locate the position of the dry product single dry object bounding box calibration image data is the position of the prey in the th dimension, that is, the dry product single dry object bounding box calibration image data that matches the target dry product feature text data The position of the prey in the dry product single dry object bounding box calibration image data set with a dimension of in the search space is the objective function value of the prey, is the objective function value of the th calibrated image search pelican individual;

[0076] S223. In the development phase, the calibrated image search pelican conducts target prey hunting search on all the dry product single dry object bounding box calibration image data in the search space of the dry product single dry object bounding box calibration image data set That is, search for the dry product single dry object bounding box calibration image data that matches the target dry product feature text data in the search space of the dry product single dry object bounding box calibration image data set Locate the dry product single dry object bounding box calibration image data ​​​​, model the behavior process of searching for pelicans in the calibrated images. The algorithm checks the positions near the positions of the pelicans searched in the calibrated images, so that the algorithm converges to a better position. The formula for calculating the pelicans searched in the calibrated images during the hunting behavior is as follows: , where represents the position of the -th pelican searched in the calibrated images after the update in the development stage in the -th dimension, that is, the position of the -th pelican searched in the calibrated images after the update in the development stage in the dry product single dry object bounding box calibration image data set with the dimension of Search for the dry product single dry object bounding box calibration image data that matches the target dry product feature text data in the search space . is a random integer of 0 or 2; t is the current iteration number; T is the maximum iteration number;

[0077] S224. When the algorithm meets the maximum iteration number, output the dry product single dry object bounding box calibration image data that matches the target dry product feature text data ;

[0078] S225. Generate the target dry product single dry object bounding box calibration image data by generating data identifiers for the dry product single dry object bounding box calibration image data output in step S224.

[0079] The operation steps for matching the target dry product feature text data with the qualified dry product overall feature image library of the dry product to generate the target dry product qualified overall feature image library are as follows:

[0080] S31. Establish a set of qualified dry product overall feature image libraries , where represents the qualified dry product overall feature image library corresponding to the -th dry product combination feature type; , ; where represents the -th qualified dry product overall feature image data in the qualified dry product overall feature image library , represents the qualified dry product overall feature image library The maximum value of the number of qualified overall feature images of dry goods products; the qualified overall feature image data of dry goods products represents the overall appearance image information of qualified dry goods products established for different types of dry goods products standards.

[0081] S32. Use the K-D tree nearest neighbor search algorithm to process the target dry goods product feature text data with the set of qualified overall feature image libraries of dry goods products in the qualified overall feature image library of dry goods products to perform keyword matching of dry goods product features, and search for the qualified overall feature image library of dry goods products corresponding to the target dry goods product feature text data , and construct the qualified overall feature image library of the target dry goods product , where represents the th qualified overall feature image data of the target dry goods product.

[0082] Through the mutual cooperation of the dry goods product feature information collection unit and the dry goods product selection site image collection unit, the mobile terminal is used to conveniently and accurately obtain the target dry goods product feature information and the target dry goods product selection site image parameters, providing reliable data support for the intelligent identification of qualified dry goods products; the target dry goods product single dry goods object bounding box calibration image search unit accurately searches for the reference image of the dry goods object bounding box in the target dry goods product selection site image according to the target dry goods product feature information combined with the artificial intelligence recognition algorithm and the scientifically preset dry goods product single dry goods object bounding box calibration image parameters, realizing the accurate and efficient identification of the dry goods object type in the dry goods product selection site and improving the reliability of the dry goods product selection; the target dry goods product qualified overall feature image library matching unit independently matches the qualified dry goods product appearance feature image library required for the target dry goods product selection analysis according to the target dry goods product feature information combined with the intelligent search algorithm and the standard-set qualified overall feature image library of dry goods products, realizing the intelligent matching of the qualified dry goods product image libraries required for different types of dry goods products and improving the quality of the dry goods product selection.

[0083] Furthermore, please refer to Figure 1 - Figure 2 , and the operation steps for performing the bounding box calibration processing of a single dry goods object in the target dry goods product selection site image based on the target dry goods product selection site image data and the target dry goods product single dry goods object bounding box calibration image data, generating the dry goods product selection site dry goods object bounding box image data and performing the single dry goods object image segmentation processing of the dry goods product selection site dry goods object bounding box image, and generating the single dry goods object bounding box image data of the dry goods product selection site are as follows:

[0084] S41. Use the FLANN algorithm to select and calibrate the image data based on a single dried product object of the target dried product In the image data of the on-site selection of dried products Perform single dried product object recognition in the corresponding on-site selection image of dried products, and use a closed line to frame and calibrate the recognized single dried product object in the on-site selection image of dried products, and generate the on-site selection image data of dried product object framing ;

[0085] S42. For the on-site selection image data of dried product object framing Perform single dried product object image feature segmentation and extraction processing on the framed image of a single dried product object along the closed line of the frame calibration, and generate a set of on-site selection image data of single dried product object framing , where represents the on-site selection image data of dried product object framing The th on-site selection single dried product object framed image data corresponding to the segmented dried product object, represents the on-site selection image data of dried product object framing The maximum value of the number of dried product objects

[0086] The operation steps for analyzing the qualified result of the single dried product object selection on-site according to the on-site selection single dried product object framed image data and the target dried product qualified overall feature image library are as follows:

[0087] S51. Use the FLANN algorithm to make the on-site selection single dried product object framed image data set The on-site selection single dried product object framed image data in Match the image features with the target dried product qualified overall feature image data in the target dried product qualified overall feature image library in an orderly manner according to the dried product object number, and generate a set of on-site selection single dried product object qualified result analysis data based on the image feature matching results, where represents the on-site selection single dried product object qualified result analysis data corresponding to the th segmented dried product object in the on-site selection image data of dried product object framing ; When

[0088] When and The image feature matching is successful, indicating that the quality of the th dry product meets the quality standard and selection criteria; then output the analysis data of the qualified result of a single dry product object at the dry product selection site as qualified;

[0089] When and the image features do not match successfully, indicating that the quality of the th dry product does not meet the quality standard and selection criteria; then output the analysis data of the qualified result of a single dry product object at the dry product selection site as unqualified.

[0090] According to the analysis data of the qualified result of a single dry product object at the dry product selection site and the marked watermark image data of the qualified analysis result of the dry product object, perform the matching process of the marked watermark image required for the different selection quality qualified analysis results of a single dry product object at the dry product selection site, generate the marked watermark image data of the qualified analysis result of a single dry product object at the dry product selection site, and perform the selection result image marking process of a single dry product object at the dry product selection site with the framed image data of a single dry product object at the dry product selection site. The operation steps for generating the selection result image data of a single dry product object at the dry product selection site are as follows:

[0091] S61. Establish a set of marked watermark image data for the qualified analysis result of the dry product object , where represents the analysis data of the qualified result of a single dry product object at the dry product selection site corresponding marked watermark image data of the qualified analysis result of the dry product object when it is qualified. At this time, the marked watermark image of the qualified analysis result of the dry product object is a watermark image with the word "qualified"; represents the analysis data of the qualified result of a single dry product object at the dry product selection site corresponding marked watermark image data of the qualified analysis result of the dry product object when it is unqualified. At this time, the marked watermark image of the qualified analysis result of the dry product object is a watermark image with the word "unqualified";

[0092] S62. Use the bidirectional search algorithm to match the analysis data of the qualified result of a single dry product object in the set of the analysis data of the qualified result of a single dry product object at the dry product selection site with the marked watermark image data of the qualified analysis result of the dry product object in the set of the marked watermark image data of the qualified analysis result of the dry product object in an orderly manner according to the numbering of the dry product objects, and search for the analysis data of the qualified result of a single dry product object at the dry product selection site Correspondingly, mark the watermark image data of the qualified analysis result of the dry goods object, and generate a set of watermark image data for the qualified analysis result of a single dry goods object at the dry goods product selection site , where represents the image data of the dry goods object frame selection at the dry goods product selection site The th dry goods object corresponding to the watermark image data of the qualified analysis result of a single dry goods object at the dry goods product selection site;

[0093] S63. Combine the watermark image data of the qualified analysis result of a single dry goods object at the dry goods product selection site in the set of watermark image data of the qualified analysis result of a single dry goods object at the dry goods product selection site with the image data of the dry goods object frame selection at the dry goods product selection site according to the numbering of the dry goods object quantity and the set of image data of the dry goods object frame selection of a single dry goods object at the dry goods product selection site in the set of image data of the dry goods object frame selection of a single dry goods object at the dry goods product selection site to perform image combination, and generate a set of image data of the selection result of a single dry goods object at the dry goods product selection site , where represents the image data of the dry goods object frame selection at the dry goods product selection site The th dry goods object corresponding to the image data of the selection result of a single dry goods object at the dry goods product selection site; The image data of the selection result of a single dry goods object at the dry goods product selection site represents the feedback image information for identifying the qualified analysis result of the selection of a single dry goods object at the dry goods product selection site.

[0094] Through the cooperation of the dry goods object frame selection image generation unit at the dry goods product selection site and the single dry goods object frame selection image segmentation unit at the dry goods product selection site, accurate identification and frame selection of all single dry goods object images at the dry goods product selection site are achieved based on image analysis, and accurate segmentation of all single dry goods object frame selection image information at the dry goods product selection site is realized, so as to accurately extract all single dry goods object feature image information at the dry goods product selection site; the single dry goods object qualified result analysis unit at the dry goods product selection site comprehensively and intelligently evaluates the selection quality of single dry goods objects at the dry goods product selection site by combining the single dry goods object frame selection image information at the dry goods product selection site with the intelligent search algorithm and the target dry goods product qualified overall feature image library, realizing efficient and accurate selection of dry goods products and improving the accuracy of dry goods product selection; the single dry goods object qualified analysis result identification watermark image matching unit at the dry goods product selection site scientifically matches the single dry goods object selection quality identification watermark image according to the single dry goods object qualified result analysis parameters at the dry goods product selection site and the dry goods object qualified analysis result identification watermark image information, realizing an intuitive identification of the dry goods product selection quality; the single dry goods object selection result image generation unit at the dry goods product selection site visually marks the selection result image of a single dry goods object at the dry goods product selection site according to the single dry goods object qualified analysis result identification watermark image parameters combined with image analysis and the single dry goods object frame selection image parameters at the dry goods product selection site, realizing an intuitive watermark image identification of the single dry goods object selection result at the dry goods product selection site and improving the applicability of dry goods selection.

[0095] Further, please refer to Figure 1 - Figure 2 to construct the result image data of the dry goods product selection site and perform the operation steps of the dry goods product on-site selection result feedback operation as follows:

[0096] S71. Arrange the single dry goods object selection result image data in the single dry goods object selection result image data set at the dry goods product selection site in an orderly manner according to the dry goods object quantity number, and perform image feature matching and combination with the dry goods object frame selection images in the dry goods object frame selection image data at the dry goods product selection site to construct the result image data of the dry goods product selection site ;

[0097] S72. Perform the dry goods product on-site selection result feedback operation on the generated result image data of the dry goods product selection site through the mobile terminal display screen.

[0098] Through the cooperation of the on-site selection result image construction unit and the on-site selection result feedback unit for dry goods products, the selection result images of individual dry goods objects at the dry goods product selection site are combined with image feature analysis and the image of the dry goods object frame selection at the dry goods product selection site for image matching and combination, scientifically constructing the result image information of the dry goods product selection site, realizing the accurate marking of the selection quality type of individual dry goods objects at the dry goods product selection site, facilitating customers to intuitively and accurately identify the quality of dry goods products; at the same time, combining with the mobile terminal display screen, independently and reliably execute the on-site selection result feedback operation of dry goods products, realizing the convenient and efficient output of the dry goods product selection results, and improving the customer's dry goods purchase experience and dry goods purchase quality.

[0099] Embodiment 2:

[0100] Please refer to Figure 1 - Figure 2 , an AI-based intelligent selection system for customers to purchase dry goods products, used to implement an AI-based intelligent selection method for customers to purchase dry goods products. The system includes a dry goods product selection search module, a dry goods product selection processing module, and a dry goods product selection feedback module;

[0101] The dry goods product selection search module includes a dry goods product feature information collection unit, a dry goods product selection site image collection unit, a dry goods product individual dry goods object frame selection and calibration image storage unit, a target dry goods product individual dry goods object frame selection and calibration image search unit, a dry goods product qualified overall feature image library storage unit, and a target dry goods product qualified overall feature image library matching unit;

[0102] The dry goods product feature information collection unit collects the target dry goods product feature text data through the mobile terminal; the dry goods product selection site image collection unit collects the dry goods product selection site image data through the mobile terminal; the dry goods product individual dry goods object frame selection and calibration image storage unit is used to store the dry goods product individual dry goods object frame selection and calibration image data; the target dry goods product individual dry goods object frame selection and calibration image search unit performs search processing on the reference image of the individual dry goods object frame selection and calibration in the target dry goods product selection site image according to the target dry goods product feature text data and the dry goods product individual dry goods object frame selection and calibration image data, generating the target dry goods product individual dry goods object frame selection and calibration image data; the dry goods product qualified overall feature image library storage unit is used to store the dry goods product qualified overall feature image library; the target dry goods product qualified overall feature image library matching unit performs matching processing on the qualified dry goods product appearance feature image library required for the target dry goods product selection analysis according to the target dry goods product feature text data and the dry goods product qualified overall feature image library, generating the target dry goods product qualified overall feature image library;

[0103] The dry product selection and processing module includes a unit for generating the image of the selected dry object at the dry product selection site, a unit for segmenting the image of a single dry object at the dry product selection site, a unit for analyzing the qualified result of a single dry object at the dry product selection site, a unit for storing the identification watermark image of the qualified analysis result of the dry object, a unit for matching the identification watermark image of the qualified analysis result of a single dry object at the dry product selection site, and a unit for generating the selection result image of a single dry object at the dry product selection site;

[0104] The unit for generating the image of the selected dry object at the dry product selection site performs the processing of calibrating the selection of a single dry object in the image of the dry product selection site based on the image data of the dry product selection site and the calibration image data of the selected single dry object of the target dry product, and generates the image data of the selected dry object at the dry product selection site; the unit for segmenting the image of a single dry object at the dry product selection site performs the processing of segmenting the image of a single dry object in the image of the selected dry object at the dry product selection site based on the image data of the selected dry object at the dry product selection site, and generates the image data of the selected single dry object at the dry product selection site; the unit for analyzing the qualified result of a single dry object at the dry product selection site performs the analysis processing of the qualified result of the selection quality of a single dry object at the dry product selection site based on the image data of the selected single dry object at the dry product selection site and the overall feature image library of the qualified target dry product, and generates the analysis data of the qualified result of a single dry object at the dry product selection site; the unit for storing the identification watermark image of the qualified analysis result of the dry object is used to store the image data of the identification watermark image of the qualified analysis result of the dry object; the unit for matching the identification watermark image of the qualified analysis result of a single dry object at the dry product selection site performs the matching processing of the identification watermark image required for different selection quality qualified analysis results of a single dry object at the dry product selection site based on the analysis data of the qualified result of a single dry object at the dry product selection site and the image data of the identification watermark image of the qualified analysis result of the dry object, and generates the image data of the identification watermark image of the qualified analysis result of a single dry object at the dry product selection site; the unit for generating the selection result image of a single dry object at the dry product selection site performs the identification processing of the selection result image of a single dry object at the dry product selection site based on the image data of the identification watermark image of the qualified analysis result of a single dry object at the dry product selection site and the image data of the selected single dry object at the dry product selection site, and generates the image data of the selection result image of a single dry object at the dry product selection site;

[0105] The dry product selection feedback module includes a unit for constructing the selection result image at the dry product selection site and a unit for feeding back the selection result at the dry product site;

[0106] A dry product selection site selection result image construction unit is used to construct dry product selection site result image data; a dry product on-site selection result feedback unit performs a dry product on-site selection result feedback operation based on the dry product selection site result image data in combination with the mobile terminal display screen.

[0107] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based intelligent selection method for customers to purchase dry goods products, characterized in that: The method comprises the following steps: S1. Collecting target dry goods product feature text data and dry goods product selection site image data; S2, performing a search process for a single dry goods object frame selection calibration reference image in a target dry goods product selection scene image based on the target dry goods product feature text data and the single dry goods object frame selection calibration image data of the dry goods product, and generating the single dry goods object frame selection calibration image data of the target dry goods product; S3, performing matching processing on the appearance feature image library of qualified dry goods products required for target dry goods product selection analysis based on the target dry goods product feature text data and the dry goods product qualified overall feature image library, to generate a target dry goods product qualified overall feature image library; S4, performing a single dry goods object frame selection calibration process in the target dry goods product selection scene image according to the dry goods product selection scene image data and the single dry goods object frame selection calibration image data of the target dry goods product, generating dry goods object frame selection image data at the dry goods product selection scene, and performing a single dry goods object image segmentation process on the dry goods object frame selection image at the dry goods product selection scene, generating single dry goods object frame selection image data at the dry goods product selection scene; S5, performing analysis and processing of qualified results of selection quality of individual dry goods objects at the dry goods product selection site based on the frame selection image data of individual dry goods objects at the dry goods product selection site and the qualified overall feature image library of the target dry goods products, and generating qualified result analysis data of individual dry goods objects at the dry goods product selection site; S6, performing identification watermark image matching processing required for the qualified analysis results of different selection qualities of individual dry goods objects at the dry goods product selection site based on the qualified result analysis data of the individual dry goods objects at the dry goods product selection site and the identification watermark image data of the qualified analysis results of the dry goods objects, generating the qualified analysis result identification watermark image data of the individual dry goods objects at the dry goods product selection site, and performing selection result image identification processing of the individual dry goods objects at the dry goods product selection site with the frame selection image data of the individual dry goods objects at the dry goods product selection site, generating the selection result image data of the individual dry goods objects at the dry goods product selection site; S7. Construct dry goods product selection on-site result image data and perform dry goods product selection on-site result feedback operation.

2. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 1 is characterized by: The S1 comprises the following steps: S11. Inputting the characteristic text information of the dry goods product to be purchased by the customer online through the mobile terminal, and generating the characteristic text data of the target dry goods product ; Use mobile terminals to capture the on-site image information of the unpackaged dry goods products that customers want to purchase, and generate on-site image data of dry goods product selection .

3. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 2 is characterized by: The S2 comprises the following steps: S21. Establish a single dry product object selection and calibration image data set ; S22, the With the Middle The image data of the single dry goods object selected by the dry goods product corresponding to the dry goods product combination feature type Perform dry goods product feature character matching and search for the The corresponding , and generate the target dry product single dry product object selection calibration image data through data identification , execute to generate the The specific steps are as follows: S221, initialization, updating the maximum number of iterations T and updating the calibration image to search for the location of the pelican population; S222, exploration stage, calibration image search pelican to determine the location of the prey, that is, in the Search the search space for Matching the The image is then calibrated to search for individual pelicans located in the The algorithm moves in the area and models the approaching prey strategy of the pelican in the calibration image search, so that the algorithm The search space is scanned, and the target is described The location of the prey is described in The search space is randomly generated; S223, development stage, calibration image search pelican Search space for all the Conducting a target hunting search, i.e. Search the search space for Matching the ,This behavior process of searching the pelican in the calibration image is modeled, and the algorithm checks the position near the pelican position in the calibration image, so that the algorithm converges to a better position; S224, when the algorithm meets the maximum number of iterations, the output is the same as described Matching the ; S225: the output in step S224 After data identification, the target dry goods product single dry goods object selection calibration image data is generated .

4. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 3 is characterized by: The S3 comprises the following steps: S31. Establish a collection of qualified overall feature image libraries for dry goods products ; S32, using the KD tree nearest neighbor search algorithm to With the Middle A qualified overall feature image library of dry goods products corresponding to the combination feature types of dry goods products Match the dry goods product feature keywords and search for the The corresponding , and build a qualified overall feature image library of target dry goods products .

5. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 4 is characterized by: The S4 comprises the following steps: S41, using the FLANN algorithm according to the In the A single dry goods object is identified in the corresponding dry goods product selection scene image, and the identified single dry goods object is calibrated using a closed line frame selection in the dry goods product selection scene image, and dry goods object frame selection image data of the dry goods product selection scene is generated ; S42, the The framed image of a single dry product object in the image is segmented and extracted along the closed line marked by the frame selection, and a framed image data set of a single dry product object at the dry product product selection site is generated. .

6. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 5 is characterized by: The S5 comprises the following steps: S51, using FLANN algorithm to middle The middle segment Dry goods product selection scene corresponding to the dry goods object single dry goods object frame selection image data According to the number of dry goods objects, they are numbered in order and described Middle Target dry goods product qualified overall feature image data Perform image feature matching and generate a data set of qualified results analysis of individual dry goods objects at the dry goods product selection site based on the image feature matching results ; when and If the image feature matching is successful, the qualified result analysis data of a single dry goods object at the dry goods product selection site will be output. To be qualified; when and If the image features are not matched successfully, the qualified result analysis data of a single dry goods object at the dry goods product selection site will be output. Is unqualified.

7. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 6 is characterized by: The S6 comprises the following steps: S61. Establishing a watermark image data set for qualified analysis results of dry goods objects ,in Indicates the The corresponding dry goods object qualified analysis result identification watermark image data when it is qualified, at this time, the dry goods object qualified analysis result identification watermark image is a qualified word watermark image; Indicates the The watermark image data of the qualified analysis result of the dry goods object corresponding to the unqualified state, at this time, the watermark image of the qualified analysis result of the dry goods object is the watermark image of the word "unqualified"; S62, using a bidirectional search algorithm to As stated in According to the number of dry goods objects, they are numbered in order and described The qualified analysis result identification watermark image data of the dry goods object is used to match the keyword of the qualified analysis result of a single dry goods object on the dry goods product selection site, and the qualified analysis result of the dry goods object is searched out. The corresponding dry goods object qualified analysis result identification watermark image data, and generate a single dry goods object qualified analysis result identification watermark image data set at the dry goods product selection site ; S63, the middle The middle segment Dry goods product selection site corresponding to each dry goods object, qualified analysis result identification watermark image data of a single dry goods object According to the number of dry goods objects and the As stated in Combine images and generate a dataset of image data of a single dry goods object selection result at the dry goods product selection site .

8. The AI-based intelligent selection method for customers to purchase dry goods products according to claim 7 is characterized by: The S7 comprises the following steps: S71, the middle The middle segment Image data of the selection result of a single dry goods object at the scene of the dry goods product selection corresponding to the dry goods object According to the number of dry goods objects, they are numbered in order and described In the on-site dry goods product selection, the dry goods object frame selection image is combined with image feature matching to construct the dry goods product selection on-site result image data ; S72, generating the Carry out the feedback of on-site selection results of dry goods products.

9. An AI-based intelligent selection system for customers to purchase dry goods products, used to implement any one of claims 1-8, wherein the AI-based intelligent selection method for customers to purchase dry goods products is characterized in that: The system comprises a dry goods product selection search module, a dry goods product selection processing module, and a dry goods product selection feedback module.

Images