Study On Microstructure Properties Of Energy Materials Based On Micro-hyperspectral Imaging Technology

In recent years,perovskite,as one of the most studied energy materials in the field of materials science,has been widely used in solar cells and other devices due to its good photoelectric performance.However,uneven crystal thickness and inconsistent component ratio are important factors to reduce the performance of the device.Therefore,it is a key problem that how to effectively detect the crystal quality defects of perovskite to improve the performance of perovskite devices.Micro-hyperspectral technology combines advanced microscope technology and spectroscopy technology,which can capture and analyze the spectral information of pixel by pixel in a spatial area.Because it can obtain the unique spectral characteristics of different spatial positions of a single object,it is suitable for material research at micro and nano scales.In this paper,I obtained perovskite crystal（MAPbBr₃）3D data at room temperature by transmission bright field patterns with the use of independent self-built micro-hyperspectral system.I combined with the spectral dimension to study the physical and chemical properties and built the inversion model of perovskite single crystal thickness prediction based on machine learning algorithms to realize nondestructive testing perovskite crystal quality.It shows the application prospect of micro-hyperspectral imaging technology in micro and nano scale materials.The specific works in this paper are as follows:First,in order to study the quality of perovskite crystals,two types of microscopic hyperspectral image acquisition units,reflection and transmission,were constructed using MAPbBr3 prepared by one-step solution self-assembly method.Correlation system detection parameters to obtain high-quality image data.By comparing and analyzing the spatial information and spectral information under different shooting modes,the transmission bright field was finally chosen to image and conduct in-depth research on perovskites.Secondly,after the experimental analysis,it is proposed that the uneven thickness and the inconsistent composition ratio of MAPbBr3 will affect its light absorption properties.The absorption wavelength of MAPbBr3 is proportional to its thickness.When the thickness increases,the absorption of light by the crystal increases,and its absorption peak undergoes a red shift.In addition,in the spectral range before the absorption peak,the composition ratio of MAPbBr3 crystal has a certain dependence on its absorbance.The smaller the percentage of Br atoms in the crystal,the greater the absorption of light.Finally,a variety of preprocessing algorithms are used to transform the original spectrum,and the conventional dimensionality reduction method and the proposed band selection method based on Perovskite Thickness Spectral Characteristics（PTSC）are used to extract characteristic bands.A prediction model for perovskite single crystal thickness was established by means of various regression modeling methods.The results show that the multivariate scattering correction has the best preprocessing effect after the ridge regression model is used to verify the analysis.In the comparison of different machine learning modeling methods,the K-nearest neighbor regression model performs the best.Among them,the proposed PTSC method can extract the optimal modeling variables more effectively than other dimensionality reduction algorithms.It combines the estimation of the K-nearest neighbor regression model.Compared with the K nearest neighbor regression model of the original spectrum,the overall improvement is about 13%.