Method and device for detecting commodities in advertisement list

Abstract

The embodiment of the invention provides a method and a device for detecting commodities in an advertisement list, and the method comprises the steps: inputting a to-be-detected advertisement list into a trained commodity detection model, and obtaining the commodity information in the to-be-detected advertisement list; inputting the commodity information into a trained commodity comparison model,and extracting feature vectors of commodities; and performing similarity comparison on the feature vector of the commodity and a commodity vector in a preset commodity comparison library, and determining a commodity category of which the similarity is smaller than a preset pre-expenditure so as to complete detection of the commodity in the to-be-detected advertisement list. According to the methodand the device for detecting commodities in an advertisement list, the deep learning neural network is used for processing the advertisement list image, the texture, the edge and other characteristics of the image are automatically extracted and combined, so that the positions of the commodities in the advertisement list are detected, the extracted commodities are vectorized, and the extracted commodities can be quickly compared with the commodities in the commodity library.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and device for detecting commodities in advertisement sheets. Background technique [0002] In recent years, e-commerce has developed rapidly. With the continuous expansion of the market and the continuous increase of various commercial industries, more and more items can be selected, purchased and used by users. In practice, merchants often promote products in the form of billboards. [0003] At present, in the prior art, a detection-based technology is generally used to detect commodities in advertisement sheets. However, this method has the following problems: firstly, the method will be limited by data and fixed categories in the classification stage, and cannot adapt well to the rapidly changing commodity library; secondly, due to the large number of commodities in the advertisement list, it may They will occlude each other. If we simply classify from the...

AI Technical Summary

Problems solved by technology

However, this method has the following problems: firstly, this method will be limited by data and fixed categories in the classification stage, and cannot adapt well to the rapidly changing product library; secondly, due to the large number of products in the advertisement list, it may They will occlude each other. If we simply classify from the minimum bounding box, some images of the products that cause occlusion will be introduced, resulting in too much noise in the input.

Images