A self-attention-based destruction and construction learning fine-grained image classification method for retail product recognition

摘要

Retail products belonging to the same category usually have extremely similar appearance characteristics such as colors, shapes, and sizes, which cannot be distinguished by conventional classification methods. Currently, the most effective way to solve this problem is fine-grained classification methods, which utilize machine vision + scene to perform fine feature representations on a target local region, thereby achieving fine-grained classification. Fine-grained classification methods have been widely used for recognizing birds, cars, airplanes, and many others. However, the existing fine-grained classification methods still have some drawbacks. In this paper, we propose an improved fine-grained classification method based on self-attention destruction and construction learning (SADCL) for retail product recognition. Specifically, the proposed method utilizes a self-attention mechanism in the destruction and construction of image information in an end-to-end fashion so that to calculate a precise fine-grained classification prediction and large information areas in the reasoning process. We test the proposed method on the Retail Product Checkout (RPC) dataset. Experimental results demonstrate that the proposed method achieved an accuracy above 80% in retail commodity recognition reasoning, which is much higher than the results of other fine-grained classification methods.