Data Expansion Method Based on Wavelet Transform

Xiaochun Li,Wen Li,Zhe Liu
DOI: https://doi.org/10.1088/1742-6596/1631/1/012128
2020-09-01
Journal of Physics: Conference Series
Abstract:Abstract For small sample data, data expansion is an effective means to improve the accuracy of target recognition. To image magnification to enhance image resolution as a starting point, the author puts forward two kinds of image magnification method based on wavelet transform. Qualitative analysis on the experimental data and quantitative analysis using average gradient, cross entropy as performance index, show that the proposed image magnification method has obvious advantages in terms of image definition, maintaining image gray scale information. Finally, the author proposes the method of image expansion based on the image magnification data of the four methods, which greatly enriches the diversification of the training set on resolution, clarity, contrast, integrity and so on, and theoretically analyzes the feasibility of improving the recognition accuracy of the data with few samples and small targets.
What problem does this paper attempt to address?