一种抗噪声干扰的卷积神经网络目标识别训练方法和装置

By randomly flipping and rotating images and adding noise during the training of convolutional neural networks, the dataset is expanded and re-labeled, which solves the problem of decreased recognition accuracy under noise interference, improves recognition accuracy and avoids overfitting.

CN116543251BActive Publication Date: 2026-07-17CENT SOUTH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CENT SOUTH UNIV
Filing Date
2023-04-03
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Convolutional neural networks suffer from decreased accuracy and performance in target recognition when noise interference is present, making it difficult to effectively recognize images containing noise interference.

Method used

The image flipping and rotation angles are determined by generating random numbers, and noise that may exist in real-world application scenarios is added to expand the training dataset. After re-labeling, the convolutional neural network is retrained.

Benefits of technology

This improves the target recognition accuracy of convolutional neural networks under noisy conditions and avoids overfitting, ensuring that the recognition accuracy does not decrease under noisy conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116543251B_ABST
    Figure CN116543251B_ABST
Patent Text Reader

Abstract

本申请涉及一种抗噪声干扰的卷积神经网络目标识别训练方法、装置、计算机设备和存储介质。所述方法包括：通过生成预设范围内的随机数，根据随机数的奇偶性确定是否对原始图像进行翻转，根据随机数的数值大小确定对所述原始图像的进行旋转的角度，再添加预设类别的噪声，得到处理图像，将处理图像重新标注后与原始数据集合并得到新样本集，通过构建好的新样本集对预训练的卷积神经网络进行再训练，得到训练好的抗噪声干扰的卷积神经网络。本发明通过随机旋转、翻转和加入噪声处理相叠加的操作对训练数据集进行扩充，在相同扩充倍数下，本发明方法数据集图像之间的相似度更小，更有利于卷积神经网络的训练，进而提高目标识别准确率。
Need to check novelty before this filing date? Find Prior Art