Image data processing method and device

Abstract

The embodiment of the present invention discloses an image data processing method and device, which relate to the technical field of big data and can reduce the labor cost consumed in the labeling process of massive image data. The present invention includes: obtaining image data according to the acquired keywords; performing clustering processing on the image data, and obtaining the cluster centers corresponding to the keywords; screening out the image data matching the cluster centers, and passing The keywords mark the image data matching the cluster centers, and import the marked image data into the sample library. The invention is suitable for automatic identification of training data for image recognition engines or image processing applications.

Description

technical field [0001] The invention relates to the field of big data technology, in particular to an image data processing method and device. Background technique [0002] At present, big data technology is widely used in various national defense and civil projects, especially in the operation and maintenance optimization of major e-commerce platforms. In recent years, as the data scale and complexity of major e-commerce platforms have exploded geometrically, how to effectively use massive user data, transaction data, merchant data, etc. has become a research topic for major e-commerce platforms. focus. [0003] Among them, especially in image recognition using deep learning models, a large number of successful solutions have been designed and developed. The general ideas of these solutions include: using a large amount of accurately labeled image data as basic samples, and combining massive sample data to train and calculate the model, so as to optimize the model in a ta...

AI Technical Summary

Problems solved by technology

[0004] It can be seen that in the current scheme, although the processing of massive sample data has been realized, some sample data still need to be labeled manually, and because the amount of data is soaring every year, the workload of manual labeling is increasing. Large, labor costs are rising rapidly

In addition, the artificially labeled sample data will still be subject to human subjective influence, making some of the pre-labeled data basic quality not high, or the generalization ability is not strong, making the basic sample made by looking for similar data in the vicinity based on the manually labeled data Poor quality, which ultimately leads to poor performance of the optimized model in real business environments

Images