CN-115937603-B - Commodity attribute detection method and device, storage medium and computer equipment

Abstract

The application provides a commodity attribute detection method, a device, a storage medium and computer equipment, wherein the method comprises the steps of obtaining a plurality of commodity pictures of commodities to be detected; the method comprises the steps of inputting each commodity picture into a gesture detection model to obtain gesture key point data corresponding to each commodity picture, screening out first target pictures according to preset picture screening rules and gesture key point data corresponding to each commodity picture, inputting gesture key point data corresponding to each first target picture into a two-class model to obtain classification results of each first target picture, wherein each classification result is used for reflecting whether the corresponding first target picture is suitable for carrying out attribute detection or not, screening out second target pictures suitable for carrying out attribute detection according to the classification results of each first target picture, and detecting predicted commodity attribute values of commodities to be detected based on each second target picture. The application can improve the accuracy of attribute detection.

Inventors

• TAN ZHENGYU
• XU XU

Assignees

• 唯品会(广州)软件有限公司

Dates

Publication Date

20260512

Application Date

20221228

Claims (10)

1. 1. A method for detecting properties of a commodity, the method comprising: acquiring a plurality of commodity pictures of commodities to be detected; acquiring a gesture detection model and a classification model; inputting each commodity picture into the gesture detection model respectively to obtain gesture key point data corresponding to each commodity picture; Screening a first target picture from each commodity picture according to a preset picture screening rule and gesture key point data corresponding to each commodity picture, wherein the specific method comprises the steps of respectively judging whether each commodity picture is a model display picture or not according to the gesture key point data corresponding to each commodity picture, and taking each model display picture as the first target picture; Inputting gesture key point data corresponding to each first target picture into the classification model respectively to obtain classification results of each first target picture, wherein each classification result is used for reflecting whether the corresponding first target picture is suitable for attribute detection; screening second target pictures suitable for attribute detection from the first target pictures according to the classification result of the first target pictures; And detecting the predicted commodity attribute value of the commodity to be detected based on each second target picture.
2. 2. The commodity attribute detection method according to claim 1, wherein the step of detecting the predicted commodity attribute value of the commodity to be detected based on each of the second target pictures includes: determining an attribute to be detected, and acquiring a single-label multi-classification model corresponding to the attribute to be detected; Inputting each second target picture into the single-label multi-classification model to obtain each detection attribute value output by the single-label multi-classification model and the confidence coefficient corresponding to each detection attribute value; And determining the predicted commodity attribute value in each detection attribute value based on each detection attribute value and the confidence corresponding to each detection attribute value.
3. 3. The commodity property detection method according to claim 2, wherein the method further comprises: acquiring an initial commodity attribute value of the commodity to be detected, which is marked manually in advance; and if the initial commodity attribute value is different from the predicted commodity attribute value and the confidence coefficient corresponding to the predicted commodity attribute value is greater than a preset confidence coefficient threshold value, pushing attribute modification information to an information maintainer.
4. 4. The method of claim 2, wherein the single-label multi-classification model is trained using a Focal Loss function as a Loss function.
5. 5. The method for detecting commodity attributes according to claim 1, wherein the step of screening the first target picture from each commodity picture according to a preset picture screening rule and posture key point data corresponding to each commodity picture comprises the steps of: And respectively judging whether each commodity picture is a model display picture or not according to gesture key point data corresponding to each commodity picture, and taking each model display picture as the first target picture.
6. 6. The commodity property detection method according to any one of claims 1 to 5, wherein the step of obtaining a two-classification model comprises: acquiring an initial linear regression model and a training data set, wherein the training data set comprises a plurality of training pictures and a pre-labeled manual classification result of each training picture; inputting each training picture into the gesture detection model to obtain gesture key point data corresponding to each training picture; Inputting gesture key point data corresponding to each training picture into the initial linear regression model to obtain training classification results corresponding to each training picture, and carrying out iterative training on the initial linear regression model according to each training classification result and each manual classification result until a preset training completion condition is met, and obtaining the two classification models.
7. 7. A commodity property detection apparatus, the apparatus comprising: the commodity picture acquisition module is used for acquiring a plurality of commodity pictures of the commodity to be detected; The two-classification model acquisition module is used for acquiring a gesture detection model and a two-classification model; the gesture detection module is used for respectively inputting each commodity picture into the gesture detection model so as to obtain gesture key point data corresponding to each commodity picture; The first target picture screening module is used for screening a first target picture from each commodity picture according to a preset picture screening rule and gesture key point data corresponding to each commodity picture, and specifically comprises the steps of respectively judging whether each commodity picture is a model display picture or not according to the gesture key point data corresponding to each commodity picture, and taking each model display picture as the first target picture; The classification module is used for respectively inputting gesture key point data corresponding to each first target picture into the classification model to obtain a classification result of each first target picture, wherein each classification result is used for reflecting whether the corresponding first target picture is suitable for attribute detection; the second target picture screening module is used for screening second target pictures suitable for attribute detection from the first target pictures according to the classification results of the first target pictures; and the attribute value detection module is used for determining the predicted commodity attribute value of the commodity to be detected according to each second target picture.
8. 8. The article property detection apparatus of claim 7, wherein the property value detection module comprises: A single-label multi-classification model acquisition unit for determining the attribute to be detected, acquiring a single-label multi-classification model corresponding to the attribute to be detected; the detection unit is used for inputting each second target picture into the single-label multi-classification model so as to obtain each detection attribute value output by the single-label multi-classification model and the confidence coefficient corresponding to each detection attribute value; And an attribute value determining unit configured to determine the predicted commodity attribute value among the respective detection attribute values based on the respective detection attribute values and the confidence level corresponding to each detection attribute value.
9. 9. A storage medium having stored therein computer readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the merchandise attribute detection method of any one of claims 1 to 6.
10. 10. A computer device includes one or more processors and a memory; Stored in the memory are computer readable instructions which, when executed by the one or more processors, perform the steps of the merchandise attribute detection method of any one of claims 1 to 6.

Description

Commodity attribute detection method and device, storage medium and computer equipment Technical Field The present application relates to the field of artificial intelligence, and in particular, to a method and apparatus for detecting a commodity attribute, a storage medium, and a computer device. Background In the e-commerce platform, in order to facilitate users to quickly understand commodities and to perform personalized pushing on the users, attribute values of various attributes of each commodity need to be determined. Taking clothing commodities as an example, the clothing commodities can comprise commodity attributes such as layout attributes, clothing length attributes, style attributes, sleeve length attributes and the like, and a user can quickly know the commodities through the commodity attributes. However, the conventional commodity attribute detection method has a problem of low accuracy. Disclosure of Invention The present application aims to solve at least one of the above technical drawbacks, especially the technical drawbacks of the prior art with low detection accuracy. In a first aspect, an embodiment of the present application provides a method for detecting a commodity attribute, where the method includes: acquiring a plurality of commodity pictures of commodities to be detected; acquiring a gesture detection model and a classification model; inputting each commodity picture into the gesture detection model respectively to obtain gesture key point data corresponding to each commodity picture; screening a first target picture from each commodity picture according to a preset picture screening rule and gesture key point data corresponding to each commodity picture; Inputting gesture key point data corresponding to each first target picture into the classification model respectively to obtain classification results of each first target picture, wherein each classification result is used for reflecting whether the corresponding first target picture is suitable for attribute detection; screening second target pictures suitable for attribute detection from the first target pictures according to the classification result of the first target pictures; And detecting the predicted commodity attribute value of the commodity to be detected based on each second target picture. In one embodiment, the step of detecting the predicted commodity attribute value of the commodity to be detected based on each of the second target pictures includes: determining an attribute to be detected, and acquiring a single-label multi-classification model corresponding to the attribute to be detected; Inputting each second target picture into the single-label multi-classification model to obtain each detection attribute value output by the single-label multi-classification model and the confidence coefficient corresponding to each detection attribute value; And determining the predicted commodity attribute value in each detection attribute value based on each detection attribute value and the confidence corresponding to each detection attribute value. In one embodiment, the commodity attribute detection method further includes: acquiring an initial commodity attribute value of the commodity to be detected, which is marked manually in advance; and if the initial commodity attribute value is different from the predicted commodity attribute value and the confidence coefficient corresponding to the predicted commodity attribute value is greater than a preset confidence coefficient threshold value, pushing attribute modification information to an information maintainer. In one embodiment, the single-label multi-classification model is trained using a Focal Loss function as a Loss function. In one embodiment, the step of screening the first target picture from each commodity picture according to a preset picture screening rule and gesture key point data corresponding to each commodity picture includes: And respectively judging whether each commodity picture is a model display picture or not according to gesture key point data corresponding to each commodity picture, and taking each model display picture as the first target picture. In one embodiment, the step of obtaining the classification model includes: acquiring an initial linear regression model and a training data set, wherein the training data set comprises a plurality of training pictures and a pre-labeled manual classification result of each training picture; inputting each training picture into the gesture detection model to obtain gesture key point data corresponding to each training picture; Inputting gesture key point data corresponding to each training picture into the initial linear regression model to obtain training classification results corresponding to each training picture, and carrying out iterative training on the initial linear regression model according to each training classification result and each manual classification result until a preset trai