Target detection, identification and training method and device, and storage medium

Abstract

The invention provides a target detection, identification and training method and device, and a storage medium, and the method comprises the steps: marking a first training data set based on a preset data marking rule, and training a data marking model according to the marked first training data set; labeling the second training data set according to the trained data labeling model, and constructing a target detection training data set; and training a target detection model according to the target detection training data set, obtaining a difference value between a training output result of the target detection model and a pre-obtained real label, and updating a network weight of the target detection model according to the difference value so as to enable the difference value to meet a preset threshold condition. According to the method disclosed by the invention, the labeling efficiency can be effectively improved, and the small target object detection accuracy is high.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular, to a target detection and identification method, apparatus, device and storage medium. Background technique [0002] In recent years, with the rapid development of the national economy, the types of commodities on the market are also growing rapidly. Facing the complex and massive types of trademarks, the ability to control the infringement of commodities is particularly important. As my country attaches more and more importance to independent intellectual property rights, the compliance and legality testing of various commodity logos has been gradually developed and perfected, while some do not have obvious logos, and only rely on the image of the product and the pattern on the surface to represent the brand's identity. Commodities have slipped through the net, and infringements of commodity patterns are not uncommon. [0003] At present, the realization sche...

AI Technical Summary

Problems solved by technology

[0005] However, using the method of deep learning classification, although the characteristics of commodity trademarks can be automatically extracted and classified, the recognition error rate will rise linearly in the small target scene, and there are many small target scenes in the real environment. At the same time, the classification model training process A large number of data sets are required, and the labeling of the data sets is time-consuming and labor-intensive, which seriously restricts the scalability of deep learning classification methods in the identification of commodity trademark categories

[0006] To sum up, the existing methods cannot meet the business needs of low labeling, high time consumption, and accurate identification of commodity patterns

Images