Photoelectric property prediction method based on perovskite solar cell SEM picture

Abstract

The invention relates to the technical field of cell photoelectric performance prediction, in particular to a photoelectric performance prediction method and system based on a perovskite solar cell SEM picture, and the method comprises the steps: obtaining multi-modal data, obtaining a super-resolution reconstruction picture, obtaining a feature extraction network set, extracting a material microstructure feature vector, and extracting a material composition feature vector. The method comprises the following steps: obtaining preparation process parameters of a thin film preparation method, extracting a processing feature vector, carrying out adaptive feature fusion on a material microstructure feature vector, a material composition feature vector and the processing feature vector to obtain a fused feature vector, obtaining an MLP regression predictor, and carrying out solar cell performance prediction to obtain photoelectric conversion efficiency. And obtaining an optimized solar cell, and completing photoelectric performance prediction based on the SEM picture of the perovskite solar cell based on the optimized solar cell. According to the invention, accurate prediction and microstructure optimization of the photoelectric conversion efficiency of the perovskite solar cell can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the optoelectronic performance of batteries, and particularly to a method and system for predicting the optoelectronic performance based on SEM images of perovskite solar cells. Background Art

[0002] A perovskite solar cell is a new type of solar cell, which is a photovoltaic device with an organic-inorganic hybrid or all-inorganic metal halide material having a perovskite crystal structure as a light absorption layer. An SEM image refers to a picture of the microscopic structure of a sample obtained by a scanning electron microscope.

[0003] Although significant progress has been made in the research of perovskite solar cells, the research methods mainly use single material components, process parameters or limited microscopic structure information as features to predict the cell performance, lacking the effective integration and in-depth mining of multi-modal information such as material components, preparation processes and microscopic structures, and it is difficult to comprehensively reflect the complex characteristics of perovskite solar cells. The resolution of traditional SEM image analysis methods is limited, and it is difficult to clearly observe the key microscopic structure details affecting the performance of perovskite solar cells. Therefore, how to achieve accurate prediction of the optoelectronic conversion efficiency of perovskite solar cells and optimize the microscopic structure. Summary of the Invention

[0004] The present invention provides a method for predicting the optoelectronic performance based on SEM images of perovskite solar cells and a computer-readable storage medium, and its main purpose is to achieve accurate prediction of the optoelectronic conversion efficiency of perovskite solar cells and optimize the microscopic structure.

[0005] To achieve the above object, a method for predicting the optoelectronic performance based on SEM images of perovskite solar cells provided by the present invention includes:

[0006] Obtain multi-modal data, where the multi-modal data includes: the material chemical composition of the perovskite solar cell, the thin film preparation method and the SEM image set, and the SEM image set includes: the cross-section image set of the solar cell and the SEM image set of the perovskite layer surface;

[0007] Successively extract a SEM image of the perovskite layer surface from the SEM image set of the perovskite layer surface, and perform the following operations on each of the extracted SEM images of the perovskite layer surface:

[0008] Perform image preprocessing operations on the SEM image of the perovskite layer surface to obtain an initial SEM image, and obtain a super-resolution reconstruction image based on the initial SEM image and a pre-constructed super-resolution model, where the super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism;

[0009] Obtain a set of feature extraction networks, where the set of feature extraction networks includes: a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network;

[0010] Use the material microstructure feature extraction network in the set of feature extraction networks to extract the material microstructure feature vector from the super-resolution reconstructed image, and use the material composition feature extraction network to extract the material composition feature vector from the material chemical composition;

[0011] Obtain the preparation process parameters of the thin film preparation method, and use the processing technology feature extraction network to extract the processing feature vector from the preparation process parameters;

[0012] Use the pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector, and the processing feature vector to obtain a fused feature vector;

[0013] Obtain an MLP regression predictor, and use the MLP regression predictor to predict the performance of the solar cell for the fused feature vector to obtain the photoelectric conversion efficiency;

[0014] Summarize the photoelectric conversion efficiency to obtain the photoelectric conversion efficiency set corresponding to the set of SEM images of the perovskite layer surface;

[0015] Obtain an optimized solar cell based on the photoelectric conversion efficiency set, and complete the photoelectric performance prediction based on the SEM images of the perovskite solar cell based on the optimized solar cell.

[0016] Optionally, the performing image preprocessing operations on the SEM image of the perovskite layer surface to obtain an initial SEM image includes:

[0017] Perform a picture cropping operation on the SEM image of the perovskite layer surface to obtain a set of cropped image blocks, where the set of cropped image blocks includes six cropped image blocks;

[0018] Perform the following operations on each cropped image block in the set of cropped image blocks:

[0019] Perform normalization processing on the cropped image block to obtain a normalized image block, and perform contrast enhancement operation on the normalized image block using the pre-constructed contrast enhancement technology to obtain a high-contrast image block;

[0020] Stitch the high-contrast image blocks to obtain the reconstructed SEM image corresponding to the set of cropped image blocks, and perform pixel intensity normalization on the reconstructed SEM image to obtain the initial SEM image.

[0021] Optionally, the performing pixel intensity normalization on the reconstructed SEM image to obtain the initial SEM image includes:

[0022] Grayscale the reconstructed SEM image to obtain a grayscale SEM image, where the grayscale SEM image includes an image pixel set, and the image pixel set includes: 256 pixel values;

[0023] Extract an image pixel from the grayscale SEM image in sequence, and perform the following operations on each extracted image pixel:

[0024] Taking the image pixel as the center, construct a local region window, count the total number of image pixels in the local region window, and obtain multiple pixel value counts according to the total number of image pixels;

[0025] Obtain a local region histogram according to multiple pixel value counts, where the horizontal axis of the local region histogram is the pixel value and the vertical axis is the pixel value count;

[0026] Calculate the ratio of each pixel value count in the local region histogram to the total number of image pixels to obtain a probability density function array, where the probability density function array includes multiple ratios;

[0027] Map each ratio in the probability density function array to obtain a mapping function array, where the mapping function array includes multiple updated ratios;

[0028] Use the multiple updated ratios in the mapping function array to update all pixel values in the total number of image pixels to obtain an initial SEM image.

[0029] Optionally, obtaining a super-resolution reconstructed image based on the initial SEM image and a pre-constructed super-resolution model includes:

[0030] Use the pre-constructed two-dimensional discrete cosine transform to extract frequency domain features from the initial SEM image to obtain a frequency domain feature set, where the frequency domain feature set includes multiple frequency domain features;

[0031] Use the frequency domain channel attention mechanism to weight each frequency domain feature in the frequency domain feature set to obtain a feature map;

[0032] Use the self-attention mechanism to perform global dependence modeling on the feature map to obtain a global attention matrix, and obtain a super-resolution reconstructed image according to the global attention matrix.

[0033] Optionally, extracting a material microstructure feature vector from the super-resolution reconstructed image by using the material microstructure feature extraction network in the feature extraction network set includes:

[0034] Obtain the height and width of the super-resolution reconstructed image, and calculate the feature elements in the material microstructure feature vector by using the height and width of the super-resolution reconstructed image. The feature element calculation formula is as follows:

[0035]

[0036] Among them, v 1 k represents the k-th eigen element of the material microstructure feature vector, h represents the height of the super-resolution reconstructed image, w represents the width of the super-resolution reconstructed image, and FeatureMap ij k represents the pixel value of the k-th super-resolution reconstructed image at the i-th row and j-th column;

[0037] Summarize the eigen elements to obtain the material microstructure feature vector.

[0038] Optionally, the step of adaptively fusing the material microstructure feature vector, the material composition feature vector, and the processing feature vector by using a pre-constructed adaptive feature vector fusion module to obtain a fusion feature vector includes:

[0039] Calculate the material microstructure weight coefficient by using the material microstructure feature vector, where the calculation formula of the material microstructure weight coefficient is as follows:

[0040] a = v·W + b

[0041] where a represents the material microstructure weight coefficient, v represents the material microstructure feature vector, W represents the learnable parameter of the material microstructure feature vector, and b represents the bias term of the material microstructure feature vector;

[0042] Obtain the material composition weight coefficient based on the material composition feature vector, and obtain the processing weight coefficient based on the processing feature vector;

[0043] Use the adaptive feature vector fusion module to adaptively fuse the material microstructure feature vector, the material composition feature vector, the processing feature vector, the material microstructure weight coefficient, the material composition weight coefficient, and the processing weight coefficient to obtain a fusion feature vector.

[0044] Optionally, the step of predicting the performance of the solar cell by using the MLP regression predictor for the fusion feature vector to obtain the photoelectric conversion efficiency includes:

[0045] Predict the photoelectric conversion efficiency of the solar cell performance by using the MLP regression predictor and the fusion feature vector, where the calculation formula of the photoelectric conversion efficiency is as follows:

[0046] P = f(σ(g 1 (x 1 ), g 2 (x 2 ), g 3 (x 3 )))

[0047] Among them, P represents the photoelectric conversion efficiency, f represents the MLP regression predictor, σ represents the adaptive feature vector fusion module, and g 1 represents the microstructure feature extraction network, and x 1 represents the super-resolution reconstructed image, and g 2 represents the material composition feature extraction network, and x 2 represents the material chemical composition, and g 3 represents the processing technology feature extraction network, and x 3 represents the preparation process parameters.

[0048] Optionally, the obtaining of the optimized solar cell based on the set of photoelectric conversion efficiencies includes:

[0049] Judging whether there is a photoelectric conversion efficiency in the set of photoelectric conversion efficiencies that is not within the preset photoelectric conversion efficiency interval;

[0050] If it is confirmed that there is a photoelectric conversion efficiency in the set of photoelectric conversion efficiencies that is not within the preset photoelectric conversion efficiency interval, then a battery to be optimized is obtained based on the SEM image of the perovskite layer surface, where the battery to be optimized is a perovskite solar cell;

[0051] Optimize the battery to be optimized to obtain a sub-optimal battery, take pictures of the sub-optimal battery to obtain a set of sub-optimal battery pictures, use the set of sub-optimal battery pictures as the set of SEM pictures of the perovskite layer surface, and return to the step of sequentially extracting an SEM picture of the perovskite layer surface from the set of SEM pictures of the perovskite layer surface until the photoelectric conversion efficiency is within the preset photoelectric conversion efficiency interval, and an optimized solar cell is obtained.

[0052] Optionally, the optimizing the battery to be optimized to obtain a sub-optimal battery includes:

[0053] Obtain the equivalent circle diameter of the grains of the battery to be optimized;

[0054] Compare the equivalent circle diameter of the grains with the preset standard grain diameter threshold;

[0055] If it is confirmed that the equivalent circle diameter of the grains is less than the preset standard grain diameter threshold, then obtain BMIMAc ionic liquid, extract an ionic liquid sample from the BMIMAc ionic liquid using a preset sample liquid volume, and use the ionic liquid sample to optimize the grain morphology of the equivalent circle diameter of the grains of the battery to be optimized until the equivalent circle diameter of the grains is greater than or equal to the preset standard grain diameter threshold, and a sub-optimal battery is obtained.

[0056] To achieve the above object, the present invention also provides a photoelectric performance prediction system based on SEM pictures of perovskite solar cells, including:

[0057] A data acquisition module for acquiring multimodal data, where the multimodal data includes: the material chemical composition of a perovskite solar cell, a thin film preparation method, and an SEM image set, and the SEM image set includes: a cross-section image set of the solar cell and an SEM image set of the perovskite layer surface;

[0058] A feature extraction module for sequentially extracting an SEM image of the perovskite layer surface from the SEM image set of the perovskite layer surface and performing the following operations on each of the extracted SEM images of the perovskite layer surface: performing an image preprocessing operation on the SEM image of the perovskite layer surface to obtain an initial SEM image, obtaining a super-resolution reconstructed image based on the initial SEM image and a pre-constructed super-resolution model, where the super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism, obtaining a set of feature extraction networks, where the set of feature extraction networks includes: a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network, using the material microstructure feature extraction network in the set of feature extraction networks to extract a material microstructure feature vector from the super-resolution reconstructed image, and using the material composition feature extraction network to extract a material composition feature vector from the material chemical composition;

[0059] A feature fusion module for obtaining preparation process parameters of the thin film preparation method, using the processing technology feature extraction network to extract a processing feature vector from the preparation process parameters, using a pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector, and the processing feature vector to obtain a fused feature vector, obtaining an MLP regression predictor, using the MLP regression predictor to perform a performance prediction of the solar cell on the fused feature vector to obtain a photoelectric conversion efficiency, summarizing the photoelectric conversion efficiency to obtain a photoelectric conversion efficiency set corresponding to the SEM image set of the perovskite layer surface;

[0060] A photoelectric performance prediction module for obtaining an optimized solar cell based on the photoelectric conversion efficiency set and completing the photoelectric performance prediction based on the SEM image of the perovskite solar cell for the optimized solar cell.

[0061] To solve the above problems, the present invention also provides an electronic device, and the electronic device includes:

[0062] A memory storing at least one instruction;

[0063] A processor for executing the instruction stored in the memory to implement the above-mentioned photoelectric performance prediction method based on the SEM image of the perovskite solar cell.

[0064] To solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned method for predicting the optoelectronic performance based on SEM images of perovskite solar cells.

[0065] To solve the problems described in the background art, the present invention obtains multimodal data, where the multimodal data includes: the material chemical composition of perovskite solar cells, the thin film preparation method, and an SEM image set. The SEM image set includes: a cross-section image set of the solar cell and an SEM image set of the perovskite layer surface. The present invention reflects the characteristics of perovskite solar cells from different perspectives based on the multimodal data. Compared with a single data source, it can describe the state and performance of the battery more comprehensively and accurately, providing a rich and diverse information basis for subsequent analysis and prediction. Sequentially extract an SEM image of the perovskite layer surface from the SEM image set of the perovskite layer surface, and perform the following operations on each of the extracted SEM images of the perovskite layer surface: perform image preprocessing operations on the SEM image of the perovskite layer surface to obtain an initial SEM image, and obtain a super-resolution reconstructed image based on the initial SEM image and a pre-constructed super-resolution model. The super-resolution model includes: a frequency-domain channel attention mechanism and a self-attention mechanism. The present invention can improve the resolution of the initial SEM image through super-resolution reconstruction. The frequency-domain channel attention mechanism can focus on the importance of different frequency channels and highlight the key features in the image. The self-attention mechanism can capture the dependencies between different positions in the initial SEM image, thus presenting more clearly the key microscopic structural details that affect the performance of perovskite solar cells and providing richer and more accurate information for subsequent feature extraction. Obtain a set of feature extraction networks, where the set of feature extraction networks includes: a material microscopic structure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network. The present invention uses different networks to extract features for different types of data, can fully mine the respective key information in the multimodal data, and convert complex data into representative feature vectors, providing effective input for subsequent performance prediction. Use the material microscopic structure feature extraction network in the set of feature extraction networks to extract material microscopic structure feature vectors from the super-resolution reconstructed image, and use the material composition feature extraction network to extract material composition feature vectors from the material chemical composition. The present invention can observe the microscopic structure more clearly through the super-resolution reconstructed image, and the material microscopic structure feature extraction network can convert this microscopic structure information into quantified feature vectors, facilitating subsequent analysis and comparison and helping to reveal the relationship between the microscopic structure and the battery performance. The material chemical composition contains information on various elements and compounds, and the material composition feature extraction network can convert this complex composition information into feature vectors, enabling the influence of the material composition on the battery performance to be expressed and analyzed in a digital form. Obtain the preparation process parameters of the thin film preparation method, and use the processing technology feature extraction network to extract processing feature vectors from the preparation process parameters. The processing technology feature extraction network of the present invention can convert these process parameters into feature vectors, quantify the role of process factors on the battery performance, and provide a basis for optimizing the preparation process.An adaptive feature vector fusion module is used to perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector, and the processing feature vector to obtain a fused feature vector. The present invention can give full play to the advantages of multi-modal data by using the adaptive feature vector fusion module, integrate the information of different types of features, avoid the limitations of single features, and more comprehensively reflect the performance influencing factors of perovskite solar cells. An MLP regression predictor is obtained, and the MLP regression predictor is used to predict the performance of the solar cell for the fused feature vector to obtain the photoelectric conversion efficiency. The present invention can learn the complex relationship between the fused feature vector and the photoelectric conversion efficiency of the solar cell through the MLP regression predictor, so as to achieve accurate prediction of the battery performance. The photoelectric conversion efficiencies are summarized to obtain a set of photoelectric conversion efficiencies corresponding to the set of SEM images of the perovskite layer surface. The present invention uniformly manages and analyzes the battery performance corresponding to all SEM images of the perovskite layer surface, providing comprehensive data support for subsequent optimization decisions. An optimized solar cell is obtained based on the set of photoelectric conversion efficiencies, and the photoelectric performance prediction based on the SEM images of the perovskite solar cell is completed based on the optimized solar cell. Through the analysis of the set of photoelectric conversion efficiencies, the present invention can find out the battery samples with higher photoelectric conversion efficiencies and their corresponding material compositions, preparation processes, and microstructure features. Therefore, the present invention can achieve accurate prediction of the photoelectric conversion efficiency of perovskite solar cells and microstructure optimization., BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 FIG. is a schematic flow chart of a method for predicting the photoelectric performance based on SEM images of perovskite solar cells provided by an embodiment of the present invention;

[0067] Figure 2 FIG. is a functional module diagram of a system for predicting the photoelectric performance based on SEM images of perovskite solar cells provided by an embodiment of the present invention;

[0068] Figure 3 FIG. is a schematic structural diagram of an electronic device for implementing the method for predicting the photoelectric performance based on SEM images of perovskite solar cells provided by an embodiment of the present invention.

[0069] DESCRIPTION OF THE REFERENCE NUMERALS

[0070] 1, electronic device; 10, processor; 11, memory; 12, bus.

[0071] The realization, functional characteristics, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0072] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0073] An embodiment of the present application provides a method for predicting the optoelectronic performance based on SEM images of perovskite solar cells. The execution subject of the method for predicting the optoelectronic performance based on SEM images of perovskite solar cells includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the method for predicting the optoelectronic performance based on SEM images of perovskite solar cells can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0074] Refer to Figure 1 As shown, it is a schematic flowchart of a method for predicting the optoelectronic performance based on SEM images of perovskite solar cells provided by an embodiment of the present invention. In this embodiment, the method for predicting the optoelectronic performance based on SEM images of perovskite solar cells includes:

[0075] S1. Obtain multimodal data, where the multimodal data includes: the material chemical composition of the perovskite solar cell, the thin film preparation method, and the SEM image set.

[0076] Specifically, the SEM image set includes: a solar cell cross-section image set and a perovskite layer surface SEM image set.

[0077] It should be explained that the step of obtaining multimodal data is: obtaining multimodal data from internationally renowned journals, open-source scientific databases, and industry reports. The solar cell cross-section image set refers to a set of images taken along a specific cross-section direction of the solar cell, reflecting the overall structure and the thickness of each layer of the solar cell. The perovskite layer surface SEM image set refers to a set of images showing the surface morphology and grain distribution of the perovskite thin film.

[0078] S2. Sequentially extract a perovskite layer surface SEM image from the perovskite layer surface SEM image set, and perform the following operations on each of the extracted perovskite layer surface SEM images: perform image preprocessing operations on the perovskite layer surface SEM image to obtain an initial SEM image.

[0079] Specifically, the performing image preprocessing operations on the perovskite layer surface SEM image to obtain an initial SEM image includes:

[0080] Perform a picture cropping operation on the perovskite layer surface SEM image to obtain a cropped image block set, where the cropped image block set includes six cropped image blocks;

[0081] Perform the following operations on each cropped image block in the cropped image block set:

[0082] Normalize the cropped image patches to obtain normalized image patches, and perform contrast enhancement operations on the normalized image patches using a pre-built contrast enhancement technique to obtain high-contrast image patches;

[0083] Stitch the high-contrast image patches to obtain the reconstructed SEM image corresponding to the set of cropped image patches, and perform pixel intensity normalization on the reconstructed SEM image to obtain the initial SEM image.

[0084] It should be explained that the set of cropped image patches refers to a set of image patches obtained after performing a cropping operation on the SEM image of the perovskite layer surface. The purpose of performing the image cropping operation on the SEM image of the perovskite layer surface is to remove irrelevant parts and focus on the microscopic structure feature region of the perovskite thin film. The normalization process for the cropped image patches refers to the operation of unifying the sizes of the cropped image patches into a preset standard interval to meet the input requirements of the convolutional neural network.

[0085] Importantly, the normalized image patches refer to the image patches obtained after normalization. The contrast enhancement technique is a type of image processing technique used to improve image quality. The purpose is to make the key features in the normalized image patches more prominent by adjusting the brightness and contrast of the normalized image patches, thereby enhancing the visibility of the microscopic structure. The step of performing contrast enhancement operations on the normalized image patches using a pre-built contrast enhancement technique is a prior art and will not be elaborated here. The high-contrast image patches refer to the image patches obtained after performing contrast enhancement operations on the normalized image patches.

[0086] It should be explained that the stitching of the high-contrast image patches refers to accurately combining multiple cropped and processed high-contrast image patches according to their relative positions in the SEM image of the perovskite layer surface to form a complete and coherent image. The reconstructed SEM image refers to a complete SEM image reconstructed by stitching and combining the high-contrast image patches according to their relative positions in the SEM image of the perovskite layer surface.

[0087] Specifically, the pixel intensity normalization of the reconstructed SEM image to obtain the initial SEM image includes:

[0088] Grayscale the reconstructed SEM image to obtain a grayscale SEM image, where the grayscale SEM image includes an image pixel set, and the image pixel set includes 256 pixel values;

[0089] Successively extract one image pixel from the grayscale SEM image, and perform the following operations on each extracted image pixel:

[0090] Taking the image pixel as the center, construct a local region window, count the total number of image pixels within the local region window, and obtain multiple pixel value counts according to the total number of image pixels;

[0091] Obtain a local area histogram based on multiple pixel value counts, where the horizontal axis of the local area histogram is the pixel value and the vertical axis is the pixel value count;

[0092] Calculate the ratio of each pixel value count in the local area histogram to the total number of image pixels to obtain a probability density function array, where the probability density function array includes multiple ratios;

[0093] Map each ratio in the probability density function array to obtain a mapping function array, where the mapping function array includes multiple updated ratios;

[0094] Use the multiple updated ratios in the mapping function array to update all pixel values in the total number of image pixels to obtain an initial SEM image.

[0095] It should be explained that the technology of grayscale conversion of the reconstructed SEM image in the step of grayscaling the image is an existing technology and will not be elaborated here. The grayscale SEM image refers to the image obtained after grayscaling the reconstructed SEM image. An image pixel refers to the smallest unit that makes up an image. In a grayscale SEM image, each pixel is represented by a specific pixel value to indicate the brightness of the grayscale SEM image. The step of constructing the local area window is as follows: taking the currently extracted image pixel as the center, a rectangular area of a specific size is delimited in the grayscale SEM image as the local area window.

[0096] Exemplarily, the size of the local area window is 3x3. Taking the image pixel as the center, the range is extended by one pixel above, below, to the left, and to the right of the image pixel to form a rectangular area containing 9 pixels.

[0097] It can be understood that obtaining multiple pixel value counts according to the total number of image pixels means traversing all pixels within the local area window, counting the number of times each pixel value appears, and obtaining multiple pixel value counts. The local area histogram refers to a graph used to analyze the brightness distribution of image pixels within the local area window. The probability density function array refers to an array composed of the ratios of each pixel value count in the local area histogram to the total number of image pixels. Mapping each ratio in the probability density function array means using a mapping function to map each ratio in the probability density function array. For example, the mapping function is a linear mapping or a non - linear mapping.

[0098] It should also be explained that the mapping function array refers to an array composed of multiple updated ratios. The initial SEM image refers to the image obtained after updating the pixel set of the grayscale SEM image according to the mapping function array.

[0099] S3. Obtain the super-resolution reconstructed image based on the initial SEM image and the pre-constructed super-resolution model.

[0100] Specifically, the super-resolution model includes: a frequency-domain channel attention mechanism and a self-attention mechanism.

[0101] In detail, the obtaining of the super-resolution reconstructed image based on the initial SEM image and the pre-constructed super-resolution model includes:

[0102] Use the pre-constructed two-dimensional discrete cosine transform to extract the frequency-domain features of the initial SEM image, obtaining a frequency-domain feature set, where the frequency-domain feature set includes multiple frequency-domain features;

[0103] Use the frequency-domain channel attention mechanism to perform feature weighting on each frequency-domain feature in the frequency-domain feature set, obtaining a feature map;

[0104] Use the self-attention mechanism to perform global dependence modeling on the feature map, obtaining a global attention matrix, and obtain the super-resolution reconstructed image according to the global attention matrix.

[0105] It should be explained that the two-dimensional discrete cosine transform refers to a mathematical transformation method widely used in the field of image processing, which is used to transform the image data in the initial SEM image in the two-dimensional spatial domain into the frequency domain. Frequency-domain feature extraction refers to the operation of finding different frequencies of the microscopic structure from the image data after the two-dimensional discrete cosine transform. The frequency-domain feature set refers to a set of features obtained after frequency-domain feature extraction. The frequency-domain feature refers to the amplitude value of a certain frequency in the initial SEM image.

[0106] Importantly, the frequency-domain channel attention mechanism is an attention mechanism. For different frequency-domain features which are different channels, the frequency-domain channel attention mechanism can automatically learn the importance of each channel and assign different weights to each channel. The step of using the frequency-domain channel attention mechanism to perform feature weighting on each frequency-domain feature in the frequency-domain feature set to obtain a feature map is: calculate the weights of different frequency-domain features according to the frequency-domain channel attention mechanism, multiply each frequency-domain feature by the corresponding weight, and obtain a feature map. The step of using the frequency-domain channel attention mechanism to perform feature weighting on each frequency-domain feature in the frequency-domain feature set to obtain a feature map is a prior art and will not be elaborated here.

[0107] It should be noted that the self-attention mechanism refers to a mechanism that can automatically capture the dependencies between different positions in the feature map. By calculating the correlation between each position in the feature map and other positions, a global attention matrix is obtained. The global attention matrix refers to a matrix obtained through the calculation of the self-attention mechanism, which records the correlation weights between each position in the feature map and all other positions. Each element of the matrix represents the weight of one position to another position.

[0108] S4. Obtain a set of feature extraction networks, where the set of feature extraction networks includes: a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network.

[0109] It should be noted that the obtaining of the set of feature extraction networks refers to the collection of feature extraction networks obtained from the historical feature extraction network library. The material microstructure feature extraction network refers to a neural network specifically used to extract material microstructure features from the super-resolution reconstructed images of materials, where the material microstructure features include: grain size and grain boundary morphology.

[0110] The material composition feature extraction network refers to a network used to extract key features from the chemical composition of materials, where the chemical composition of materials includes: element types, element contents, and compound types. The processing technology feature extraction network refers to a network that extracts features from the preparation process parameters of materials, where the preparation process parameters include: processing temperature, processing pressure, processing speed, and processing time.

[0111] S5. Use the material microstructure feature extraction network in the set of feature extraction networks to extract material microstructure feature vectors from the super-resolution reconstructed images, and use the material composition feature extraction network to extract material composition feature vectors from the chemical composition of materials.

[0112] Specifically, the using of the material microstructure feature extraction network in the set of feature extraction networks to extract material microstructure feature vectors from the super-resolution reconstructed images includes:

[0113] Obtain the height and width of the super-resolution reconstructed image, and use the height and width of the super-resolution reconstructed image to calculate the feature elements in the material microstructure feature vector, where the formula for the feature elements is as follows:

[0114]

[0115] where, v 1 k represents the k-th feature element of the material microstructure feature vector, h represents the height of the super-resolution reconstructed image, w represents the width of the super-resolution reconstructed image, FeatureMap ij kDenote the pixel value of the k-th super-resolution reconstructed image at the i-th row and j-th column;

[0116] Summarize the feature elements to obtain the material microstructure feature vector.

[0117] It should be explained that the steps of obtaining the height and width of the super-resolution reconstructed image are as follows: Use an image processing library to obtain the height and width of the super-resolution reconstructed image. Among them, the image processing library technology is an existing technology and will not be elaborated here. The material microstructure feature vector is a vector composed of all feature elements.

[0118] S6. Obtain the preparation process parameters of the film preparation method, and use the processing feature extraction network to extract the processing feature vector from the preparation process parameters.

[0119] It should be explained that the obtaining of the preparation process parameters of the film preparation method refers to the parameters obtained from the historical film preparation methods. The processing feature vector refers to a vector obtained by performing feature extraction on the preparation process parameters of the film preparation method.

[0120] S7. Use the pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, material composition feature vector, and processing feature vector to obtain the fusion feature vector.

[0121] Specifically, the using of the pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, material composition feature vector, and processing feature vector to obtain the fusion feature vector includes:

[0122] Calculate the material microstructure weight coefficient using the material microstructure feature vector. Among them, the calculation formula of the material microstructure weight coefficient is as follows:

[0123] a = v·W + b

[0124] Among them, a represents the material microstructure weight coefficient, v represents the material microstructure feature vector, W represents the learnable parameter of the material microstructure feature vector, and b represents the bias term of the material microstructure feature vector;

[0125] Obtain the material composition weight coefficient based on the material composition feature vector, and obtain the processing weight coefficient based on the processing feature vector;

[0126] Use the adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, material composition feature vector, processing feature vector, material microstructure weight coefficient, material composition weight coefficient, and processing weight coefficient to obtain the fusion feature vector.

[0127] It should be understood that the material microscopic weight coefficient refers to the coefficient calculated based on the material microscopic structure feature vector and is used to measure the influence degree of the material microscopic structure feature in the subsequent feature fusion process. The larger the material microscopic weight coefficient, the greater the influence of the material microscopic structure feature on the performance prediction of the solar cell. The learnable parameter is the parameter used to automatically adjust the weight. The bias term is the constant term used to correct the weight coefficient. The method for obtaining the material composition weight coefficient based on the material composition feature vector and the method for obtaining the processing weight coefficient based on the processing feature vector are the same as the method for calculating the material microscopic weight coefficient using the material microscopic structure feature vector, which will not be elaborated here.

[0128] It can be understood that the material composition weight coefficient is used to measure the influence degree of the material composition feature in the feature fusion process. The larger the material composition weight coefficient, the greater the influence of the material composition feature on the performance prediction of the solar cell. The processing weight coefficient is used to measure the influence degree of the processing technology feature in the feature fusion process. The larger the processing weight coefficient, the greater the influence of the processing technology feature on the performance prediction of the solar cell.

[0129] S8. Obtain the MLP regression predictor, use the MLP regression predictor to perform performance prediction of the solar cell on the fusion feature vector, obtain the photoelectric conversion efficiency, summarize the photoelectric conversion efficiency, and obtain the set of photoelectric conversion efficiencies corresponding to the set of SEM pictures of the perovskite layer surface.

[0130] It should be explained that the obtaining of the MLP regression predictor refers to using multiple regression prediction models to perform performance prediction of the solar cell to obtain the optimal regression prediction model, and the optimal regression prediction model is the MLP regression predictor. The set of photoelectric conversion efficiencies refers to the set composed of all photoelectric conversion efficiencies.

[0131] Specifically, the using of the MLP regression predictor to perform performance prediction of the solar cell on the fusion feature vector to obtain the photoelectric conversion efficiency includes:

[0132] Use the MLP regression predictor and the fusion feature vector to predict the photoelectric conversion efficiency of the solar cell performance, where the calculation formula of the photoelectric conversion efficiency is as follows:

[0133] P = f(σ(g 1 (x 1 ), g 2 (x 2 ), g 3 (x 3 )))

[0134] where P represents the photoelectric conversion efficiency, f represents the MLP regression predictor, σ represents the adaptive feature vector fusion module, g 1 represents the microscopic structure feature extraction network, and x1 Denote the super-resolution reconstructed image as g 2 Denote the material composition feature extraction network as x 2 Denote the material chemical composition as g 3 Denote the processing technology feature extraction network as x 3 Denote the preparation process parameters.

[0135] It should be explained that the photoelectric conversion efficiency refers to the ratio of the electric energy output by the perovskite solar cell to the solar energy irradiated on the surface of the perovskite solar cell. The MLP regression predictor is a feedforward artificial neural network model, a predictor composed of an input layer, one or more hidden layers, and an output layer. The role of the MLP regression predictor in the embodiments of the present invention is to predict the photoelectric conversion efficiency of the solar cell according to the fusion feature vector.

[0136] S9. Obtain an optimized solar cell based on the photoelectric conversion efficiency set, and complete the photoelectric performance prediction based on the SEM images of the perovskite solar cell for the optimized solar cell.

[0137] Specifically, the step of obtaining an optimized solar cell based on the photoelectric conversion efficiency set includes:

[0138] Judge whether there is a photoelectric conversion efficiency in the photoelectric conversion efficiency set that is not within the preset photoelectric conversion efficiency interval;

[0139] If it is confirmed that there is a photoelectric conversion efficiency in the photoelectric conversion efficiency set that is not within the preset photoelectric conversion efficiency interval, obtain a battery to be optimized based on the SEM image of the perovskite layer surface, where the battery to be optimized is a perovskite solar cell;

[0140] Optimize the battery to be optimized to obtain a sub-optimal battery, take pictures of the sub-optimal battery to obtain a sub-optimal battery image set, use the sub-optimal battery image set as the SEM image set of the perovskite layer surface, and return to the step of sequentially extracting an SEM image of the perovskite layer surface from the SEM image set of the perovskite layer surface until the photoelectric conversion efficiency is within the preset photoelectric conversion efficiency interval to obtain an optimized solar cell.

[0141] It should be explained that the step of obtaining a battery to be optimized based on the SEM image of the perovskite layer surface is a prior art and will not be elaborated here. The step of optimizing the battery to be optimized to obtain a sub-optimal battery will be given later. The sub-optimal battery refers to the battery obtained after optimizing the battery to be optimized. The step of taking pictures of the sub-optimal battery refers to the operation of taking pictures of the sub-optimal battery using a scanning electron microscope. The sub-optimal battery image set refers to the set of pictures obtained after taking pictures of the sub-optimal battery. The optimized solar cell refers to a perovskite solar cell whose photoelectric conversion efficiency is within the preset photoelectric conversion efficiency interval after optimizing the battery to be optimized.

[0142] Specifically, optimizing the battery to be optimized to obtain a sub-optimal battery includes:

[0143] Obtaining the equivalent circle diameter of the grains of the battery to be optimized;

[0144] Comparing the equivalent circle diameter of the grains with a preset standard grain diameter threshold;

[0145] If it is confirmed that the equivalent circle diameter of the grains is less than the preset standard grain diameter threshold, then obtain BMIMAc ionic liquid, extract an ionic liquid sample from the BMIMAc ionic liquid using a preset sample liquid volume, and use the ionic liquid sample to optimize the grain morphology of the equivalent circle diameter of the battery to be optimized until the equivalent circle diameter of the grains is greater than or equal to the preset standard grain diameter threshold to obtain a sub-optimal battery.

[0146] It should be explained that the step of obtaining the equivalent circle diameter of the grains of the battery to be optimized is as follows: using image processing and analysis software, identify and segment the grains in the SEM image of the perovskite layer surface. The software will calculate the area of each grain, and then according to the circle area formula, inversely deduce the diameter of the circle equal to the area of the grain, that is, the equivalent circle diameter of the grains. The standard grain diameter threshold refers to a preset value, which represents the critical value of the grain size that can make the perovskite solar cell in the photoelectric conversion efficiency range in the perovskite solar cell. Obtaining the BMIMAc ionic liquid means preparing the BMIMAc ionic liquid in the laboratory by chemical synthesis methods.

[0147] Importantly, BMIMAc ionic liquid refers to a specific ionic liquid, whose chemical name is usually 1-butyl-3-methylimidazolium acetate. Ionic liquids have unique physical and chemical properties and are used to improve the grain morphology of the perovskite layer in the optimization of perovskite solar cells. The sample liquid volume refers to a preset liquid volume to ensure that the amount of ionic liquid used in each optimization process is consistent, facilitating the control of experimental conditions and optimization effects. The ionic liquid sample refers to the ionic liquid extracted from the BMIMAc ionic liquid according to the preset sample liquid volume. Grain morphology optimization refers to the operation of increasing the equivalent circle diameter of the grains.

[0148] The present invention is to solve the problem described in the background technology. The present invention obtains multimodal data, wherein the multimodal data includes: the material chemical composition of the perovskite solar cell, the film preparation method and the SEM picture set, wherein the SEM picture set includes: the solar cell cross-section picture set and the perovskite layer surface SEM picture set. The present invention reflects the characteristics of the perovskite solar cell from different angles based on the multimodal data. Compared with a single data source, it can more comprehensively and accurately describe the state and performance of the battery, and provide a rich and diverse information basis for subsequent analysis and prediction. A perovskite layer surface SEM picture is extracted from the perovskite layer surface SEM picture set in turn, and the following operations are performed on the extracted perovskite layer surface SEM pictures: The SEM image of the surface of the perovskite layer is subjected to image preprocessing operations to obtain an initial SEM image, and a super-resolution reconstructed image is obtained based on the initial SEM image and a pre-built super-resolution model, wherein the super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism. The present invention can improve the resolution of the initial SEM image through super-resolution reconstruction. The frequency domain channel attention mechanism can focus on the importance of different frequency channels and highlight the key features in the image. The self-attention mechanism can capture the dependency between different positions in the initial SEM image, thereby more clearly presenting the key microstructure details that affect the performance of the perovskite solar cell, providing richer and more accurate information for subsequent feature extraction, and obtaining a feature extraction network set, wherein The feature extraction network set includes: a material microstructure feature extraction network, a material composition feature extraction network and a processing technology feature extraction network. The present invention uses different networks to extract features for different types of data, which can fully mine the key information of each multimodal data, convert complex data into representative feature vectors, and provide effective input for subsequent performance prediction. The material microstructure feature extraction network in the feature extraction network set is used to extract material microstructure feature vectors from super-resolution reconstructed images, and the material composition feature extraction network is used to extract material composition feature vectors from material chemical composition. The present invention can observe the microstructure more clearly through super-resolution reconstructed images, and the material microstructure feature can be used to extract material composition feature vectors from material chemical composition. The feature extraction network can convert these microstructure information into quantitative feature vectors, which is convenient for subsequent analysis and comparison, and helps to reveal the relationship between microstructure and battery performance. The chemical composition of materials contains information on a variety of elements and compounds. The material composition feature extraction network can convert these complex component information into feature vectors, so that the influence of material composition on battery performance can be expressed and analyzed in a digital form, and the preparation process parameters of the thin film preparation method are obtained. The processing feature extraction network is used to extract processing feature vectors from the preparation process parameters. The processing feature extraction network of the present invention can convert these process parameters into feature vectors, quantify the effect of process factors on battery performance, and provide a basis for optimizing the preparation process.An adaptive feature vector fusion module is used to perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector, and the processing feature vector to obtain a fused feature vector. The present invention can make full use of the advantages of multi-modal data by using the adaptive feature vector fusion module, integrate the information of different types of features, avoid the limitations of single features, and more comprehensively reflect the performance influencing factors of perovskite solar cells. An MLP regression predictor is obtained, and the MLP regression predictor is used to predict the performance of the solar cell for the fused feature vector to obtain the photoelectric conversion efficiency. The present invention can learn the complex relationship between the fused feature vector and the photoelectric conversion efficiency of the solar cell through the MLP regression predictor, so as to achieve accurate prediction of the battery performance. The photoelectric conversion efficiencies are summarized to obtain a set of photoelectric conversion efficiencies corresponding to the SEM image set of the perovskite layer surface. The present invention uniformly manages and analyzes the battery performance corresponding to all SEM images of the perovskite layer surface, providing comprehensive data support for subsequent optimization decisions. An optimized solar cell is obtained based on the set of photoelectric conversion efficiencies, and the photoelectric performance prediction based on the SEM images of the perovskite solar cell is completed based on the optimized solar cell. Through the analysis of the set of photoelectric conversion efficiencies, the present invention can find the battery samples with higher photoelectric conversion efficiencies and their corresponding material compositions, preparation processes, and microstructure features. Therefore, the present invention can achieve accurate prediction of the photoelectric conversion efficiency of perovskite solar cells and microstructure optimization.,

[0149] As Figure 2 shown, it is a functional module diagram of a photoelectric performance prediction system based on SEM images of perovskite solar cells provided by an embodiment of the present invention.

[0150] The photoelectric performance prediction system 100 based on SEM images of perovskite solar cells according to the present invention can be installed in an electronic device. According to the functions implemented, the photoelectric performance prediction system 100 based on SEM images of perovskite solar cells can include a data acquisition module 101, a feature extraction module 102, a feature fusion module 103, and a photoelectric performance prediction module 104. The modules in the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device;

[0151] The data acquisition module 101 is used to acquire multi-modal data, where the multi-modal data includes: the material chemical composition of the perovskite solar cell, the thin film preparation method, and the SEM image set, where the SEM image set includes: a solar cell cross-section image set and a perovskite layer surface SEM image set;

[0152] The feature extraction module 102 is configured to sequentially extract a SEM image of the perovskite layer surface from the set of SEM images of the perovskite layer surface, and perform the following operations on each of the extracted SEM images of the perovskite layer surface: perform image preprocessing operations on the SEM image of the perovskite layer surface to obtain an initial SEM image, obtain a super-resolution reconstructed image based on the initial SEM image and a pre-constructed super-resolution model, where the super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism, obtain a set of feature extraction networks, where the set of feature extraction networks includes: a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network, use the material microstructure feature extraction network in the set of feature extraction networks to extract a material microstructure feature vector from the super-resolution reconstructed image, and use the material composition feature extraction network to extract a material composition feature vector from the material chemical composition;

[0153] The feature fusion module 103 is configured to obtain the preparation process parameters of the thin film preparation method, use the processing technology feature extraction network to extract a processing feature vector from the preparation process parameters, use a pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector, and the processing feature vector to obtain a fused feature vector, obtain an MLP regression predictor, use the MLP regression predictor to perform solar cell performance prediction on the fused feature vector to obtain a photoelectric conversion efficiency, summarize the photoelectric conversion efficiency, and obtain a set of photoelectric conversion efficiencies corresponding to the set of SEM images of the perovskite layer surface;

[0154] The optoelectronic performance prediction module 104 is configured to obtain an optimized solar cell based on the set of photoelectric conversion efficiencies, and complete the optoelectronic performance prediction based on the SEM images of the perovskite solar cell based on the optimized solar cell.

[0155] Specifically, each module in the optoelectronic performance prediction system 100 based on the SEM images of the perovskite solar cell in the embodiments of the present invention adopts the same technical means as the optoelectronic performance prediction method based on the SEM images of the perovskite solar cell described above Figure 1 and can produce the same technical effects, which will not be elaborated here.

[0156] As Figure 3 shown, it is a schematic structural diagram of an electronic device for implementing the optoelectronic performance prediction method based on the SEM images of the perovskite solar cell provided by an embodiment of the present invention.

[0157] The electronic device 1 may include a processor 10, a memory 11, and a bus 12, and may further include a computer program stored in the memory 11 and executable on the processor 10, such as an optoelectronic performance prediction method program based on the SEM images of the perovskite solar cell.

[0158] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, magnetic disk, optical disc, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as the mobile hard disk of the electronic device 1. In some other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 1. Further, the memory 11 also includes the internal storage unit of the electronic device 1 and also includes an external storage device. The memory 11 can not only be used to store application software installed in the electronic device 1 and various types of data, such as the code of the photoelectric performance prediction method program based on the SEM pictures of perovskite solar cells, etc., but also be used to temporarily store the data that has been output or will be output.

[0159] In some embodiments, the processor 10 may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more Central Processing Units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, connecting all components of the entire electronic device through various interfaces and lines, and by running or executing programs or modules stored in the memory 11 (such as the photoelectric performance prediction method program based on the SEM pictures of perovskite solar cells, etc.), and calling the data stored in the memory 11, to execute various functions of the electronic device 1 and process data.

[0160] The bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is set to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0161] Figure 3 Only the electronic device with components is shown. Those skilled in the art can understand thatFigure 3 The shown structure does not constitute a limitation on the electronic device 1, and it may include fewer or more components than those shown, or combine certain components, or have different component arrangements.

[0162] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for supplying power to each component. Preferably, the power source can be logically connected to the at least one processor 10 through a power management device, so as to implement functions such as charging management, discharging management, and power consumption management through the power management device. The power source may also include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device 1 may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0163] Furthermore, the electronic device 1 may further include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0164] Optionally, the electronic device 1 may further include a user interface. The user interface may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visual user interface.

[0165] The program of the optoelectronic performance prediction method based on the SEM images of perovskite solar cells stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can implement:

[0166] Obtain multimodal data, where the multimodal data includes: the material chemical composition of the perovskite solar cell, the thin film preparation method, and a set of SEM images, where the set of SEM images includes: a set of cross-sectional images of the solar cell and a set of SEM images of the perovskite layer surface;

[0167] Successively extract a SEM image of the perovskite layer surface from the set of SEM images of the perovskite layer surface, and perform the following operations on each of the extracted SEM images of the perovskite layer surface:

[0168] Perform image preprocessing operations on the SEM images of the perovskite layer surface to obtain initial SEM images, and obtain super-resolution reconstructed images based on the initial SEM images and a pre-constructed super-resolution model. Among them, the super-resolution model includes a frequency-domain channel attention mechanism and a self-attention mechanism;

[0169] Obtain a set of feature extraction networks, where the set of feature extraction networks includes a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network;

[0170] Use the material microstructure feature extraction network in the set of feature extraction networks to extract material microstructure feature vectors from the super-resolution reconstructed images, and use the material composition feature extraction network to extract material composition feature vectors from the material chemical composition;

[0171] Obtain the preparation process parameters of the thin film preparation method, and use the processing technology feature extraction network to extract processing feature vectors from the preparation process parameters;

[0172] Use a pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vectors, material composition feature vectors, and processing feature vectors to obtain fused feature vectors;

[0173] Obtain an MLP regression predictor, and use the MLP regression predictor to predict the performance of the solar cell for the fused feature vectors to obtain the photoelectric conversion efficiency;

[0174] Summarize the photoelectric conversion efficiency to obtain the photoelectric conversion efficiency set corresponding to the set of SEM images of the perovskite layer surface;

[0175] Obtain an optimized solar cell based on the photoelectric conversion efficiency set, and complete the photoelectric performance prediction based on the SEM images of the perovskite solar cell for the optimized solar cell.

[0176] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be elaborated here.

[0177] Furthermore, if the modules / units integrated in the electronic device 1 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory).

[0178] The present invention also provides a computer-readable storage medium. The readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the following can be implemented:

[0179] Obtain multimodal data, where the multimodal data includes: the material chemical composition of a perovskite solar cell, a thin film preparation method, and a set of SEM pictures. The set of SEM pictures includes: a set of cross-section pictures of the solar cell and a set of SEM pictures of the perovskite layer surface;

[0180] Successively extract a SEM picture of the perovskite layer surface from the set of SEM pictures of the perovskite layer surface, and perform the following operations on each of the extracted SEM pictures of the perovskite layer surface:

[0181] Perform an image preprocessing operation on the SEM picture of the perovskite layer surface to obtain an initial SEM picture, and obtain a super-resolution reconstruction picture based on the initial SEM picture and a pre-constructed super-resolution model. The super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism;

[0182] Obtain a set of feature extraction networks, where the set of feature extraction networks includes: a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network;

[0183] Use the material microstructure feature extraction network in the set of feature extraction networks to extract a material microstructure feature vector from the super-resolution reconstruction picture, and use the material composition feature extraction network to extract a material composition feature vector from the material chemical composition;

[0184] Obtain the preparation process parameters of the thin film preparation method, and use the processing technology feature extraction network to extract a processing feature vector from the preparation process parameters;

[0185] Use a pre-constructed adaptive feature vector fusion module to perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector, and the processing feature vector to obtain a fusion feature vector;

[0186] Obtain an MLP regression predictor, and use the MLP regression predictor to perform a solar cell performance prediction on the fusion feature vector to obtain a photoelectric conversion efficiency;

[0187] Summarize the photoelectric conversion efficiencies to obtain a set of photoelectric conversion efficiencies corresponding to the set of SEM pictures of the perovskite layer surface;

[0188] Obtain an optimized solar cell based on the set of photoelectric conversion efficiencies, and complete the photoelectric performance prediction based on the SEM pictures of the perovskite solar cell based on the optimized solar cell.

[0189] In several embodiments provided by the present invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and there may be other partitioning methods in actual implementation.

[0190] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0191] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a combination of hardware and software functional modules.

[0192] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting the photoelectric performance of perovskite solar cells based on SEM images, characterized in that: The method comprises: Acquire multimodal data, wherein the multimodal data includes: material chemical composition of perovskite solar cells, thin film preparation methods and SEM picture sets, wherein the SEM picture sets include: solar cell cross-section picture sets and perovskite layer surface SEM picture sets; Extract one perovskite layer surface SEM image from the perovskite layer surface SEM image set in turn, and perform the following operations on the extracted perovskite layer surface SEM images: Perform image preprocessing on the SEM image of the surface of the perovskite layer to obtain an initial SEM image, and obtain a super-resolution reconstructed image based on the initial SEM image and a pre-built super-resolution model, wherein the super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism; Acquire a feature extraction network set, wherein the feature extraction network set includes: a material microstructure feature extraction network, a material composition feature extraction network and a processing technology feature extraction network; The material microstructure feature extraction network in the feature extraction network set is used to extract a material microstructure feature vector from a super-resolution reconstructed image, and the material composition feature extraction network is used to extract a material composition feature vector from a material chemical composition; Acquiring preparation process parameters of the thin film preparation method, and extracting processing feature vectors from the preparation process parameters using the processing feature extraction network; The pre-built adaptive feature vector fusion module is used to adaptively fuse the material microstructure feature vector, material composition feature vector and processing feature vector to obtain a fused feature vector; Obtaining an MLP regression predictor, and using the MLP regression predictor to predict the performance of the solar cell on the fused feature vector to obtain the photoelectric conversion efficiency; Summarize the photoelectric conversion efficiency to obtain the photoelectric conversion efficiency set corresponding to the SEM image set of the perovskite layer surface; Based on the photoelectric conversion efficiency set, the optimized solar cell is obtained, and based on the optimized solar cell, the photoelectric performance prediction based on the SEM image of the perovskite solar cell is completed.

2. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 1, characterized in that: The performing of image preprocessing operation on the SEM image of the surface of the perovskite layer to obtain an initial SEM image includes: Performing an image cropping operation on the SEM image of the surface of the perovskite layer to obtain a cropped image block set, wherein the cropped image block set includes six cropped image blocks; The following operations are performed on each cropped image block in the cropped image block set: The cropped image block is subjected to standardization processing to obtain a standardized image block, and a contrast enhancement operation is performed on the standardized image block using a pre-built contrast enhancement technique to obtain a high-contrast image block; The high-contrast image blocks are spliced ​​to obtain the reconstructed SEM image corresponding to the cropped image block set, and the pixel intensity of the reconstructed SEM image is normalized to obtain the initial SEM image.

3. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 2, characterized in that: The step of normalizing the pixel intensity of the reconstructed SEM image to obtain the initial SEM image comprises: Graying the reconstructed SEM image to obtain a grayscale SEM image, wherein the grayscale SEM image includes an image pixel set, wherein the image pixel set includes: 256 pixel values; Extract one image pixel from the grayscale SEM image one by one, and perform the following operations on each extracted image pixel: Taking the image pixel as the center, construct a local area window, count the total number of image pixels in the local area window, and obtain multiple pixel value numbers according to the total number of image pixels; Obtaining a local area histogram according to the number of pixel values, wherein the horizontal axis of the local area histogram is the pixel value and the vertical axis is the number of pixel values; Calculate the ratio of the number of each pixel value in the local area histogram to the total number of image pixels to obtain a probability density function array, wherein the probability density function array includes multiple ratios; Mapping each ratio in the probability density function array to obtain a mapping function array, wherein the mapping function array includes a plurality of update ratios; All pixel values ​​in the total number of image pixels are updated using multiple update ratios in the mapping function array to obtain an initial SEM image.

4. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 3, characterized in that: The method of obtaining a super-resolution reconstructed image based on the initial SEM image and the pre-built super-resolution model includes: Using a pre-constructed two-dimensional discrete cosine transform to extract frequency domain features from the initial SEM image, a frequency domain feature set is obtained, wherein the frequency domain feature set includes multiple frequency domain features; Using the frequency domain channel attention mechanism, feature weighting is performed on each frequency domain feature in the frequency domain feature set to obtain a feature map; The self-attention mechanism is used to perform global dependency modeling on the feature map to obtain a global attention matrix, and a super-resolution reconstructed image is obtained according to the global attention matrix.

5. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 4, characterized in that: The method of extracting a material microstructure feature vector from a super-resolution reconstructed image using the material microstructure feature extraction network in the feature extraction network set includes: The height and width of the super-resolution reconstructed image are obtained, and the characteristic elements in the material microstructure characteristic vector are calculated using the height and width of the super-resolution reconstructed image. The characteristic element calculation formula is as follows: Among them, v1 k Represents the kth characteristic element of the material microstructure feature vector, h represents the height of the super-resolution reconstructed image, w represents the width of the super-resolution reconstructed image, FeatureMap ij k represents the pixel value of the kth super-resolution reconstructed image in the i-th row and j-th column; The characteristic elements are summarized to obtain the material microstructure characteristic vector.

6. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 5, characterized in that: The method of using the pre-built adaptive feature vector fusion module to adaptively fuse the material microstructure feature vector, the material composition feature vector and the processing feature vector to obtain a fused feature vector includes: The material microstructure characteristic vector is used to calculate the material microscopic weight coefficient, wherein the material microscopic weight coefficient calculation formula is as follows: a=v·W+b Among them, a represents the material microscopic weight coefficient, v represents the material microstructure feature vector, W represents the learnable parameter of the material microstructure feature vector, and b represents the bias term of the material microstructure feature vector; Obtain material composition weight coefficient based on material composition feature vector, and obtain processing weight coefficient based on processing feature vector; The adaptive feature vector fusion module is used to adaptively fuse the material microstructure feature vector, material composition feature vector, processing feature vector, material microscopic weight coefficient, material composition weight coefficient and processing weight coefficient to obtain a fused feature vector.

7. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 6, characterized in that: The method of using the MLP regression predictor to predict the performance of the solar cell on the fused feature vector to obtain the photoelectric conversion efficiency includes: The photoelectric conversion efficiency of solar cell performance is predicted using the MLP regression predictor and fused feature vectors, where the calculation formula for the photoelectric conversion efficiency is as follows: P=f(σ(g1(x1),g2(x2),g3(x3))) Among them, P represents the photoelectric conversion efficiency, f represents the MLP regression predictor, σ represents the adaptive feature vector fusion module, g1 represents the microstructure feature extraction network, x1 represents the super-resolution reconstructed image, g2 represents the material composition feature extraction network, x2 represents the material chemical composition, g3 represents the processing technology feature extraction network, and x3 represents the preparation process parameters.

8. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 7, characterized in that: The method of obtaining an optimized solar cell based on a photoelectric conversion efficiency set comprises: Determining whether there is a photoelectric conversion efficiency that is not within a preset photoelectric conversion efficiency range in the photoelectric conversion efficiency set; If it is confirmed that there is a photoelectric conversion efficiency that is not within the preset photoelectric conversion efficiency range in the photoelectric conversion efficiency concentration, a cell to be optimized is obtained based on the SEM image of the surface of the perovskite layer, wherein the cell to be optimized is a perovskite solar cell; The cell to be optimized is optimized to obtain a suboptimal cell, the suboptimal cell is photographed to obtain a suboptimal cell picture set, the suboptimal cell picture set is used as the perovskite layer surface SEM picture set, and the step of extracting one perovskite layer surface SEM picture from the perovskite layer surface SEM picture set in turn is returned to, until the photoelectric conversion efficiency is within a preset photoelectric conversion efficiency range, to obtain an optimized solar cell.

9. The method for predicting the photoelectric performance of a perovskite solar cell based on a SEM image according to claim 8, characterized in that: The step of optimizing the battery to be optimized to obtain a suboptimal battery includes: Obtain the equivalent circular diameter of the grain of the battery to be optimized; Comparing the size of the grain equivalent circle diameter with a preset standard grain diameter threshold; If it is confirmed that the grain equivalent circle diameter is smaller than the preset standard grain diameter threshold, BMIMAC ionic liquid is obtained, and an ionic liquid sample is extracted from the BMIMAC ionic liquid using a preset sample liquid volume. The ionic liquid sample is used to optimize the grain morphology of the grain equivalent circle diameter of the battery to be optimized until the grain equivalent circle diameter is greater than or equal to the preset standard grain diameter threshold, thereby obtaining a suboptimal battery.

10. A photoelectric performance prediction system based on SEM images of perovskite solar cells, characterized in that: The system comprises: A data acquisition module is used to acquire multimodal data, wherein the multimodal data includes: the material chemical composition of the perovskite solar cell, the film preparation method and the SEM picture set, wherein the SEM picture set includes: the solar cell cross-section picture set and the perovskite layer surface SEM picture set; A feature extraction module is used to extract a perovskite layer surface SEM image from the perovskite layer surface SEM image set in turn, and perform the following operations on the extracted perovskite layer surface SEM images: perform image preprocessing operations on the perovskite layer surface SEM image to obtain an initial SEM image, and obtain a super-resolution reconstructed image based on the initial SEM image and a pre-built super-resolution model, wherein the super-resolution model includes: a frequency domain channel attention mechanism and a self-attention mechanism, and obtain a feature extraction network set, wherein the feature extraction network set includes: a material microstructure feature extraction network, a material composition feature extraction network, and a processing technology feature extraction network, and use the material microstructure feature extraction network in the feature extraction network set to extract a material microstructure feature vector from the super-resolution reconstructed image, and use the material composition feature extraction network to extract a material composition feature vector from the material chemical composition; A feature fusion module is used to obtain the preparation process parameters of the thin film preparation method, extract the processing feature vector from the preparation process parameters using the processing feature extraction network, perform adaptive feature fusion on the material microstructure feature vector, the material composition feature vector and the processing feature vector using a pre-built adaptive feature vector fusion module to obtain a fused feature vector, obtain an MLP regression predictor, use the MLP regression predictor to predict the performance of the solar cell on the fused feature vector, obtain the photoelectric conversion efficiency, summarize the photoelectric conversion efficiency, and obtain a photoelectric conversion efficiency set corresponding to the SEM picture set of the perovskite layer surface; The photoelectric performance prediction module is used to obtain the optimized solar cell based on the photoelectric conversion efficiency set, and complete the photoelectric performance prediction based on the SEM image of the perovskite solar cell based on the optimized solar cell.

Images