CN-122023852-A - Card price trend acquisition method and system based on card image recognition

CN122023852ACN 122023852 ACN122023852 ACN 122023852ACN-122023852-A

Abstract

The application discloses a card price trend obtaining method and a card price trend obtaining system based on card image recognition, which belong to the technical field of image data processing, wherein the method comprises the steps of obtaining historical card transaction orders and image data thereof according to a preset period, and carrying out cluster analysis to obtain a plurality of cluster categories; training a deep learning model based on a clustering result, acquiring a card image recognition model, acquiring a card image to be recognized, recognizing and determining the attributive clustering category through the model, generating the historical price trend of the card according to the historical transaction order corresponding to the clustering category, combining unsupervised clustering with supervised deep learning, solving the problem of difficult recognition caused by various types and version differences of the card in the card transaction market, dynamically updating the model to adapt to a new card, realizing the function of accurately acquiring the corresponding market price trend only by shooting the card image, and improving the transaction efficiency and the price transparency.

Inventors

HOU GENG

Assignees

深圳市镖卡科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (10)

1. A card price trend acquisition method based on card image recognition is characterized by comprising the following steps: Acquiring a historical card transaction order every preset time period, and extracting card image data in the historical card transaction order to obtain first target card image data; Clustering the first target card image data to obtain image clustering data of a plurality of clustering categories, and training a deep learning model according to the image clustering data of the plurality of clustering categories to obtain a card image recognition model; acquiring second target card image data to be identified, and scheduling the card image identification model to identify the second target card image data so as to determine target image clustering data; And determining the historical card price trend corresponding to the second target card image data according to the historical card transaction order corresponding to the target image cluster data.
2. The card price trend obtaining method based on card image recognition according to claim 1, wherein clustering the first target card image data to obtain image cluster data of a plurality of cluster categories comprises: Constructing all the first target card image data into a candidate pool, and setting the existing clustering category as null; Randomly taking out first target card image data from the candidate pool, and determining the attribution degree of the first target card image data which is taken out belongs to the existing clustering category; When the attribution degree of the extracted first target card image data to at least one existing clustering category is larger than a preset threshold value, the extracted first target card image data is put into the existing clustering category with the largest attribution degree; when the attribution degree of the extracted first target card image data to all the existing clustering categories is smaller than a preset threshold value, a clustering category is newly established, and the extracted first target card image data is put into the newly established clustering category; and repeating the image data of the first target card in the candidate pool and putting the image data into the clustering categories until the candidate pool is empty, so as to obtain image clustering data of a plurality of clustering categories.
3. The card price trend obtaining method based on card image recognition according to claim 2, further comprising: on the basis of obtaining a plurality of image clustering data, only obtaining a newly-added historical card transaction order after each preset time period, clustering and attributing third target card image data corresponding to the newly-added historical card transaction order, and updating the existing image clustering data.
4. The card price trend obtaining method based on card image recognition according to claim 3, wherein when the number of clustering categories of the existing image clustering data increases, the card image recognition model is updated by adopting transfer learning, an updated card image recognition model is obtained, and in the subsequent card price trend obtaining process, the updated card image recognition model is scheduled to recognize the second target card image data.
5. The card price trend obtaining method based on card image recognition according to claim 1, wherein training of a deep learning model is performed according to image cluster data of the plurality of cluster categories, obtaining a card image recognition model, comprising: assigning a unique label to each clustering category to obtain label data corresponding to image clustering data corresponding to each clustering category; And taking the image clustering data as input, taking tag data corresponding to the image clustering data as expected output, and training a deep learning model by GSGWO to obtain a card image recognition model.
6. The card price trend obtaining method based on card image recognition according to claim 1, wherein obtaining second target card image data to be recognized, and scheduling the card image recognition model to recognize the second target card image data, determining target image cluster data, comprises: Acquiring second target card image data to be identified, and scheduling the card image identification model to identify the second target card image data to obtain probability distribution output by the card image identification model; And determining the maximum probability value in the probability distribution, judging whether the maximum probability value meets a confidence threshold, if so, taking the image clustering data of the clustering class corresponding to the maximum probability value as target image clustering data, otherwise, setting the target image clustering data as null.
7. The card price trend obtaining method based on card image recognition according to claim 6, wherein determining the historical card price trend corresponding to the second target card image data according to the historical card transaction order corresponding to the target image cluster data comprises: setting the historical card price trend corresponding to the second target card image data to be null under the condition that the target image clustering data is null; And under the condition that the target image clustering data is not empty, determining the lowest price trend, the average price trend and the highest price trend of each day or each week according to the historical card trade orders corresponding to the target image clustering data, and obtaining the historical card price trend corresponding to the second target card image data.
8. The card price trend obtaining method based on card image recognition according to claim 7, further comprising, before obtaining the historical card price trend corresponding to the second target card image data: Removing a historical card trade order in which a transaction price abnormal value is located according to the transaction price in the historical card trade order corresponding to the target image cluster data to obtain a target historical card trade order, wherein the transaction price abnormal value is identified by adopting a3 sigma criterion; And determining the lowest price trend, the average price trend and the highest price trend of each day or each week according to the target historical card transaction order to obtain the historical card price trend corresponding to the second target card image data.
9. The card price trend obtaining method based on card image recognition according to claim 1, further comprising, before using the first target card image data and the second target card image data: and carrying out card area extraction, card rotation and/or card size unification on the first target card image data and the second target card image data.
10. Card price trend acquisition system based on card image recognition, characterized by comprising: The historical data acquisition module is used for acquiring a historical card transaction order every preset time period, and extracting card image data in the historical card transaction order to obtain first target card image data; The data deep learning module is used for clustering the first target card image data to obtain image clustering data of a plurality of clustering categories, training a deep learning model according to the image clustering data of the plurality of clustering categories, and obtaining a card image recognition model; the data identification module is used for acquiring second target card image data to be identified, scheduling the card image identification model to identify the second target card image data and determining target image clustering data; and the card price trend acquisition module is used for determining the historical card price trend corresponding to the second target card image data according to the historical card transaction order corresponding to the target image cluster data.

Description

Card price trend acquisition method and system based on card image recognition Technical Field The application belongs to the technical field of image data processing, and particularly relates to a card price trend acquisition method and system based on card image recognition. Background With the popularity of trading card games, various ball star cards and collection cards, the market for second hand card transactions is increasingly large. The price of the card is influenced by various factors such as quality, version, printing batch, rarity and the like, and the fluctuation is large. At present, users are mainly faced with the problems that (1) the types of cards are huge as the sea, and cards with the same name can have a plurality of different versions (such as different pictures, heavy printing plates, sales promotion plates and the like), the users are difficult to accurately match specific versions of the cards in hands by text searching, and the queried price trend is inaccurate. (2) The existing inquiry tools depend on the fact that users manually input complex card numbers or names, are complex in operation, and are easy to input errors for non-professional users. (3) New cards in the market are continuously released, and if the existing identification system adopts a fixed classification label, the new cards cannot be identified in time, so that the system applicability is poor. Disclosure of Invention The application provides a card price trend acquisition method and system based on card image recognition, which are used for solving the problem of inaccurate card price trend query in the prior art. In one aspect, the application provides a card price trend acquisition method based on card image recognition, which comprises the following steps: Acquiring a historical card transaction order every preset time period, and extracting card image data in the historical card transaction order to obtain first target card image data; Clustering the first target card image data to obtain image clustering data of a plurality of clustering categories, and training a deep learning model according to the image clustering data of the plurality of clustering categories to obtain a card image recognition model; acquiring second target card image data to be identified, and scheduling the card image identification model to identify the second target card image data so as to determine target image clustering data; And determining the historical card price trend corresponding to the second target card image data according to the historical card transaction order corresponding to the target image cluster data. In one possible implementation manner, clustering the first target card image data to obtain image cluster data of a plurality of cluster categories includes: Constructing all the first target card image data into a candidate pool, and setting the existing clustering category as null; Randomly taking out first target card image data from the candidate pool, and determining the attribution degree of the first target card image data which is taken out belongs to the existing clustering category; When the attribution degree of the extracted first target card image data to at least one existing clustering category is larger than a preset threshold value, the extracted first target card image data is put into the existing clustering category with the largest attribution degree; when the attribution degree of the extracted first target card image data to all the existing clustering categories is smaller than a preset threshold value, a clustering category is newly established, and the extracted first target card image data is put into the newly established clustering category; and repeating the image data of the first target card in the candidate pool and putting the image data into the clustering categories until the candidate pool is empty, so as to obtain image clustering data of a plurality of clustering categories. In one possible embodiment, the method further comprises: on the basis of obtaining a plurality of image clustering data, only obtaining a newly-added historical card transaction order after each preset time period, clustering and attributing third target card image data corresponding to the newly-added historical card transaction order, and updating the existing image clustering data. In a possible implementation manner, when the number of clusters of the existing image cluster data increases, the card image recognition model is updated by adopting transfer learning, so as to obtain an updated card image recognition model, and in the subsequent card price trend obtaining process, the updated card image recognition model is scheduled to recognize the second target card image data. In a possible implementation manner, training a deep learning model according to the image clustering data of the plurality of clustering categories to obtain a card image recognition model includes: assigning a unique label