A method for amplifying image data

Abstract

A method for amplifying image data is disclosed. Aiming at the problem of imbalance of target detection samples in depth learning, the over-fitting phenomenon caused by the problem is avoided by increasing the amount of data of fewer samples (the category with fewer samples), and the generalization ability of the model after training is improved, at the same time, the amount of image data acquisition and the workload of pre-processing can be reduced. The method comprises the following steps of: determining a category of a desired amplified image; extracting a category mark frame in the image through a labeling software; determining a target area and a noise interference area; changing pixel values of the noise interference area; and generating a new sample. By using this method, new samples can be generated from the existing samples, and the disturbance of noise features to the model can be reduced while the target features are highlighted. Moreover, the generated new samples do not need to be pre-processed again, so the detection model can be trained directly.

Description

technical field [0001] The present invention relates to data augmentation technology, in particular to an image data augmentation method based on deep learning target detection. Background technique [0002] In terms of target detection, deep learning has faster detection speed and higher accuracy than traditional methods. At the same time, deep learning does not need complex feature engineering, just transfer the data set to the network directly. Therefore, deep learning is very dependent on data samples. The larger the amount of data, the better the performance of the model trained by deep learning. How to obtain a large amount of sample data is also one of the focuses of deep learning. [0003] It is inevitable that the collection of sample data is quite cumbersome and requires a lot of time, manpower and material resources. At the same time, the unbalanced number of samples collected in each category will also lead to the problem of over-fitting of the model, which requi...

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to use the existing few-sample images to generate new image data, increase the amount of small-sample image data, solve the over-fitting problem caused by sample imbalance, reduce the influence of noise features on the model, and improve the generality of the model. and reduce the amount of image data collected in the early stage, reduce the workload of image preprocessing such as labeling, and speed up the establishment of training sets

Images