Image data augmentation method

• Main Author: LUO PEIYUAN
• Format: Patent
• Language: Chinese; English
• Published: 16.11.2018
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention provides an image data augmentation method. The method comprises: S1000, acquiring image data corresponding to category information of a to-be-screened image set; S2000, according to a preset screening strategy and a sample image corresponding to the category information, identifying the image data; S3000, classifying the image data based on the identification result to obtain an image data set; and S4000, carrying out adversarial training on the image data set to obtain a sample data set. The preset screening strategy includes one or more of similarity screening, hash value filtering, and keyword filtering. Therefore, the manual sample data set screening is reduced; the screening efficiency and screening reliability are improved; and the accuracy of the neural network is enhanced. 本发明提供了种图像数据的增广方法，其方法包括：S1000获取待筛选图像集的类别信息对应的图像数据；S2000根据预设筛选策略和所述类别信息对应的样本图像，识别所述图像数据；S3000根据识别结果将所述图像数据进行归类得到图像数据集；所述预设筛选策略包括相似度筛选，哈希值筛选和关键字筛选中的任意种或者多种；S4000对所述图像数据集进行对抗训练得到样本数据集。本发明实现减少人工筛选样本数据集，提升筛选效率和筛选可靠性，提高神经网络的