| Summary: | 碩士 === 國立中正大學 === 光機電整合工程研究所 === 107 === Two-dimensional (2D) materials with wafer-scale synthesis methods and fascinating properties, have attracted significant interest and triggered revolutions in corresponding device applications. However, the large-area characterization, precision, intelligent automation and high-efficiency detection of nanostructures for two-dimensional materials have not yet reached the industrial level. Therefore, we have successfully developed a set of visible hyperspectral imaging technology to analyze the optical layer characteristics of two-dimensional materials through big data analysis and deep learning.
For the part of the classification algorithm, we try to propose Decision tree(DT), DNN, 1D-CNN and 3D-CNN models to explore the correlation between the accuracy of model recognition and the optical characteristics of 2D materials. The experimental results show that the generalization ability of the 3D-CNN is better than other classification models, and the model is applicable to the feature input of the spectral–spatial information. Therefore, we have a deep understanding of the growth and morphology evolution mechanism of 2D materials, so that we can carry out Data Mining on the data in the cloud database. The research advantage is that the extraction algorithm can be applied to optical images combined with hyperspectral imaging techniques to include optical morphological and spectral characteristics from the edge regions of the film.
The difference between this method and the previous research is that this study does not need to the specific substrate and the image can be given to different DR intervals on the sample via the auto iris shutter, so there is no need to adjust the different color contrast of imaging quality, and does not use the traditional image processing. This experiment uses MoS_2 as the test target. The system has the following advantages:
1.Accurate quantitative analysis of coverage (e.g., identification of thickness, presence of residues/ impurity)
2.Applied to substrates without specific or thickness.
3.Maximum FOV recognition range up to 1.6 mm x 1.2 mm
4.Best recognition resolution up to ~100 nm
5.Best detection time 30 sec/image
6.Non-destructive optical inspection of humanized vision
|
|---|
