Digital image content recognition, device and method for digital image content recognition training

By combining a baseline model and a prototype neural network, and utilizing centroid calculation and image transformation techniques, the problem of image content recognition under extremely imbalanced training data was solved, achieving efficient and accurate image recognition results.

CN112036429BActive Publication Date: 2026-06-30ROBERT BOSCH GMBH
1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ROBERT BOSCH GMBH
Filing Date
2020-06-02
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

In handling extremely imbalanced digital image content recognition tasks, the imbalance of training data in existing technologies leads to poor training performance of artificial neural networks, making it difficult to provide effective pattern recognition.

Method used

A method combining baseline models and prototype neural networks is adopted. The baseline model neural network is used for feature extraction and classification, while the prototype neural network uses the proton set to calculate the centroid and maximize the inter-class distance. The image is processed by transformation techniques such as cropping, mirroring, and rotation to form an end-to-end recognition model.

Benefits of technology

It achieves efficient image content recognition under extremely imbalanced categories, improves the robustness and recognition accuracy of the model, and reduces training time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112036429B_ABST
    Figure CN112036429B_ABST
Patent Text Reader

Abstract

Devices and methods for digital image content recognition and training. A device and computer-implemented method for digital image content recognition, comprising: determining a first candidate class for digital image content by means of a baseline model neural network (110) depending on a digital image, the baseline model neural network (110) including a first feature extractor (114) and a first classifier (116) for classifying the digital image; determining a second candidate class for digital image content by means of a prototype neural network (112), the prototype neural network (112) including a second feature extractor (120) and a second classifier (122) for classifying the digital image; classifying the digital image content into the first candidate class or the second candidate class depending on a comparison of a first confidence score of the first candidate class with a threshold or a comparison of a first confidence score of the first candidate class with a second confidence score of the second candidate class.
Need to check novelty before this filing date? Find Prior Art