A method of retail product recognition based on convolutional neural network

Abstract

The invention discloses a method for recognizing retail goods based on a convolutional neural network. First, a custom-made yolov3 target detector is trained using a general coarse-grained data set, and then an image to be detected is input to obtain a series of primary semantic objects, and then according to A series of rules combine primary semantic objects into high-level semantic objects, and finally obtain the desired target by judging the similarity between the attributes of the target to be detected and the attributes of each high-level semantic object. The invention enables the detectors trained based on general coarse-grained data sets to be used to complete fine-grained classification tasks under certain conditions; compared with traditional methods that directly collect data of target categories for training, the present invention can greatly reduce data collection. Cost and threshold for use in a production environment.

Description

technical field [0001] The invention belongs to target recognition technology, and in particular relates to a retail product recognition method based on a convolutional neural network. Background technique [0002] In the retail industry, the traditional retail mode is to identify retail goods by manually scanning barcodes or QR codes. In recent years, with the deepening of the application of deep learning methods in various fields, artificial intelligence technology has been widely used in people's daily life, and the embodiment in the retail industry is the emergence of unmanned supermarkets or convenience stores. Because there is no cashier, it greatly saves labor costs compared with traditional supermarkets. At the same time, compared with manual cashiers, its unmanned version takes up less space, allowing store owners to arrange more cashiers in the same space. , Improving the cash register efficiency directly. A major component of the cashierless checkout, the item d...

AI Technical Summary

Problems solved by technology

Because DCNN learns hierarchical representation from multiple abstraction levels, it is generally believed that DCNN can implicitly use context information, but the implicit context brings a problem: the detector's ability to recognize specific targets is extremely dependent on the training set

We cannot use existing training results to identify new objects even if they have similar feature attributes to the original object

Images