Data Augmentation Technology for Improving the Recognition Accuracy of Target Image

Abstract

<p align="justify">Relevant studies have pointed out that public has paid highly attention on the accuracy of neural network algorithm as it is widely applied in recent years. According to the present practice, it is quite difficult to collect related data when applying neural network algorithm. Besides, problems of trifles and complication exists in data image labeling process, which leads to a bad impact on the recognition accuracy of targets. In this article, analyzes are conducted on the relevant data from the perspective of data image processing with neural network algorithm as the core of this work. Besides, corresponding data augmentation technology is also put forward. Generally speaking, this technology has effectively realized the simulation under different shooting and lighting conditions by flipping, transforming and changing the pixel positions of the related original images, which contributes to the expansion of database types and promotes the robustness of detection work.</p>