Method and device for detecting number of layers of two-dimensional material under microscope based on deep learning

Abstract

The invention discloses a method and device for detecting the number of layers of a two-dimensional material under a microscope based on deep learning, and the method comprises the steps: photographing a high-resolution RGB image of a deposited two-dimensional material through an optical microscope, carrying out the preprocessing, carrying out the transfer learning based on a pre-trained Mask2Former model, and carrying out the training of the two-dimensional material image, thereby achieving the detection of the number of layers of the two-dimensional material. Smoothing the parameters of the model by using an index moving average (EMA) technology; the trained Mask2Former model is applied to a newly-collected two-dimensional material image for segmentation, segmentation masks of all areas are generated, and the number of layers of all the areas is predicted according to the masks; and calculating the area proportion of each layer number region according to the pixel point proportion to obtain the percentage parameter of each layer number. The method is suitable for various 2DP materials, and can be widely applied to research and application of 2DP materials in the fields of flexible electronics, biosensing, drug release, photoelectric devices and the like.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and device for detecting the number of layers of two-dimensional materials under a microscope based on deep learning. Background Art

[0002] Two-dimensional materials have attracted extensive attention in the field of materials science due to their unique properties. Among them, two-dimensional polymers (2DP), as members of the emerging two-dimensional material family, have shown excellent application prospects in the fields of flexible electronics, biosensors and drug release. The number of layers of 2DP has an important influence on its electronic properties, optical properties, etc., especially in the application of semiconductors, optoelectronic devices and other fields. Therefore, accurately and quickly identifying the number of layers of two-dimensional polymers is of great significance in materials science and industrial production, and accurately detecting its number of layers has become a key issue in research. Traditional detection methods, such as transmission electron microscopy (TEM) and atomic force microscopy (AFM), although they can provide accurate layer number information, are complex and costly to operate. In recent years, the application of deep learning technology in image analysis has made significant progress, especially in the field of computer vision, and has gradually become an effective solution. With the development of deep learning, models such as convolutional neural networks (CNN) have made significant progress in the field of image recognition, but they still face three major challenges when processing complex 2DP optical microscopy images: The unique optical interference pattern complexity of 2DP: 2DP materials show unique interference fringes and color changes under an optical microscope. These features are related to the number of layers of the material, but traditional CNN models have difficulty effectively capturing such complex optical properties. Scarcity of 2DP sample annotated data: High-quality 2DP layer annotated data is expensive to obtain and limited in quantity, making it difficult to support the effective training of complex deep learning models; Difficulty integrating multi-scale features of 2DP samples: 2DP materials often show gradient changes in the transition area of ​​different layers, and it is necessary to consider both local microstructure and global context information, which puts higher requirements on model design. For example, the accuracy is insufficient when processing complex and blurred images, and it lacks a certain degree of robustness. Therefore, it is necessary to develop a new method that combines the latest achievements of deep learning to improve detection accuracy and stability.

[0003] At present, there is no deep learning method specifically for optimizing 2DP material properties that can effectively overcome the above three key technical difficulties at the same time, and there is an urgent need to develop a dedicated solution. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a method and device for detecting the number of layers of two-dimensional materials under a microscope based on deep learning. In particular, the method is based on the MaskEMA-2DP model to detect the number of layers of two-dimensional polymers (2DP) under a microscope. This method innovatively designs an enhanced Mask model specifically for 2DP characteristics. On the basis of the Mask2Former of the Transformer architecture, it integrates the innovative application of transfer learning (TransferLearning) and exponential moving average (EMA) algorithm specifically for 2DP material characteristics, overcomes the above three major technical difficulties, can accurately detect the number of 2DP layers through optical microscope images, and has high accuracy, robustness and efficiency.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning, the method comprising the following steps: a) Sample preparation: mixing a two-dimensional material with a solvent, causing the two-dimensional material to swell under the action of the solvent and to be layered under the action of shear force to obtain a plurality of layers of the two-dimensional material, and depositing the two-dimensional material on a substrate; b) Image acquisition: High-resolution RGB images of deposited 2D materials are taken using an optical microscope and pre-processed for deep learning model training; c) Model training: Based on the pre-trained Mask2Former model, after transfer learning, it is used for training two-dimensional material images, and the exponential moving average EMA technology is used to smooth the parameters of the model; d) Layer number detection: Apply the trained Mask2Former model to the newly acquired two-dimensional material image for segmentation, generate segmentation masks for each region, and predict the number of layers in each region based on the masks; e) Quantitative analysis of the number of layers: Calculate the area ratio of each layer area based on the pixel ratio to obtain the percentage parameter of each layer.

[0006] Furthermore, the image preprocessing includes image denoising, multi-scale contrast enhancement, brightness adaptive correction, data normalization and data labeling.

[0007] Furthermore, the denoising is a denoising method that combines non-local mean filtering and wavelet transform to suppress interference patterns in two-dimensional material images while retaining key information related to the number of layers; the multi-scale contrast enhancement adopts multi-scale adaptive histogram equalization technology to locally optimize the feature differences in different layer areas of two-dimensional materials to make the layer boundaries clearer; the brightness adaptive correction adopts an adaptive brightness correction algorithm based on the optical properties of two-dimensional materials; the data annotation uses SAM to assist in annotating the two-dimensional material image to obtain segmentation results of areas with different number of layers.

[0008] Furthermore, the Mask2Former framework is based on the Transformer architecture and performs image segmentation through a multi-layer decoder to process global and local features in two-dimensional material images.

[0009] Furthermore, the layer number detection method is used to detect the layer number of a two-dimensional polymer 2DP under a microscope.

[0010] Furthermore, the layer prediction results of the Mask2Former model are displayed in the image in a color-coded manner, and different layer areas are marked with different colors.

[0011] Furthermore, the quantitative analysis of the number of layers outputs the percentage of each layer area, and displays the distribution of different layers through statistical charts.

[0012] Furthermore, during the model training process, the following feature weighting mechanism is used for the key features of two-dimensional material layer number detection: 1) Assign a higher EMA weight to the RGB information feature channel of the number of 2D material layers; 2) Dynamically optimize and adjust the weights of the RGB information feature channels based on the training results; 3) Assign higher EMA weights to high-certainty areas identified by the Mask2Former model.

[0013] In a second aspect, the present invention also provides a device for detecting the number of layers of two-dimensional materials under a microscope based on deep learning, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it implements the method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning.

[0014] In a third aspect, the present invention further provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning.

[0015] Compared with the prior art, the present invention has the following beneficial effects: Improve detection accuracy: The MaskEMA-2DP model of the present invention is specially optimized for 2DP optical properties, which greatly improves the accuracy of layer detection. The self-attention mechanism of Mask2Former can deeply capture the global information in the image, so as to accurately segment and identify regions of different levels, overcoming the limitations of traditional methods in complex images.

[0016] Reduce training time: By introducing transfer learning, the present invention effectively reduces the reliance on large-scale annotated data sets and accelerates the convergence of the model. In addition, EMA technology makes the model more stable during the training process, thereby reducing the training time and ensuring that high-precision detection results can be obtained in a shorter time. Through the three-stage transfer learning strategy and 2DP feature weighted EMA technology, the present invention greatly reduces the demand for annotated data. Experiments have shown that only 100-200 annotated 2DP images are needed to achieve good performance, which reduces the data requirement by more than 70% compared to traditional methods.

[0017] Improved robustness and stability: EMA technology can smooth parameter changes during training, so that the model can maintain stable performance when facing different brightness, contrast changes or other experimental conditions. This means that the method has strong adaptability in practical applications and can cope with changing experimental environments.

[0018] The MaskEMA-2DP model of the present invention has a strong adaptability to the changes of 2DP samples under different lighting conditions. Experimental verification shows that under the conditions of ±10% brightness change and ±10% contrast change, the model performance decreases by no more than 3%, which is much better than the existing methods.

[0019] Reduce manual intervention: Traditional layer detection methods usually rely on manual annotation, which is labor-intensive and error-prone. The deep learning model of the present invention can automatically complete image segmentation and layer detection, greatly reducing the time of manual intervention and annotation and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of sample preparation of the present invention.

[0021] Figure 2 Schematic diagram of the model and training framework of the present invention.

[0022] Figure 3 It is a schematic diagram of the image segmentation and layer number detection results of the present invention.

[0023] Figure 4 Schematic diagram of the results of adding EMA and transfer learning to the model of the present invention.

[0024] Figure 5This is a schematic diagram showing the robustness of the model after changing the brightness and contrast of the present invention.

[0025] Figure 6 A structural diagram of a two-dimensional material layer number detection device under a microscope based on deep learning provided by the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] The present invention discloses a method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning. The specific implementation process of the method is as follows: 1. Sample Preparation like Figure 1 The left figure in the figure shows the experimental process of using high shear mixing equipment for two-dimensional material exfoliation. The high shear agitator causes the layered crystals of the raw material to gradually exfoliate into a single layer or a few layers of two-dimensional material through strong stirring. The solution exfoliation process of 2DP includes mixing the 2DP material with a solvent (such as trifluoroacetic acid TFA), which causes the 2DP material to swell and delaminate under the action of shear force to produce the desired single layer or few layers of material. The equipment in the figure includes a stirrer 1 and a liquid container 2.

[0028] like Figure 1 Middle image: A pipette was used to evenly distribute the stirred solution onto the pre-cleaned and treated SiO2 / Si substrate, which facilitated subsequent image analysis.

[0029] like Figure 1 Right side of the figure: Use a microscope to observe the deposited 2DP sample and take high-resolution images to provide data for model training and layer number analysis. The high-resolution RGB images capture the detailed structural information of the two-dimensional material, providing a basis for subsequent segmentation and layer number detection.

[0030] 2. Image acquisition and preprocessing Microscope image acquisition: High-resolution images of 2D material samples are taken using an optical microscope with a resolution of 1000 dpi and a size of 512x512 pixels. The microscope needs to ensure that the images clearly present the multi-layer structure of the material and have enough details for accurate analysis.

[0031] Preprocessing steps: 2DP-specific denoising: A denoising method that combines non-local mean filtering and wavelet transform, specifically suppressing common interference patterns in 2DP images while retaining key information related to the number of layers; Multi-scale contrast enhancement: Multi-scale adaptive histogram equalization technology is used to perform local optimization on the feature differences of different layers of 2DP to make the layer boundaries clearer. Adaptive brightness correction: adjusts the brightness according to the image quality to adapt to images under different lighting conditions, so that the model can accurately identify the number of layers in various environments. The adaptive brightness correction algorithm designed based on the optical properties of 2DP materials can automatically adjust the optimal observation brightness according to the material characteristics and improve the distinguishability of different layer areas; Data normalization: Normalize the image to unify the pixel value range of the image so that the deep learning model can better learn and adapt to the image features.

[0032] Data annotation: Segment Anything Model (SAM) is used to assist in annotating the two-dimensional polymer image and obtain the segmentation results of regions with different numbers of layers, including single-layer, double-layer, triple-layer and above layers; 3. Model framework and training like Figure 2 As shown in the figure, the MaskEMA-2DP model contains several innovative components designed specifically for 2DP materials. The model includes several main parts: Backbone: First, the image is passed through the backbone network for feature extraction to generate a low-resolution feature map. The local and global information of the image is extracted through the convolutional layer or Transformer network.

[0033] Pixel Decoder: Next, the low-resolution feature map is upsampled through the pixel decoder to generate a high-resolution embedding for each pixel, preserving the image details.

[0034] Transformer Decoder: It uses multiple decoder layers to further optimize image features, helping the model recognize more complex image structures and ultimately achieve accurate image segmentation.

[0035] 2DP-specific transfer learning strategy: To address the scarcity of 2DP labeled data, a three-stage transfer learning path is designed: 1. Basic pre-training: Pre-train the model on large-scale general datasets such as ImageNet to obtain basic visual feature representation capabilities; 2. Real 2DP fine-tuning: Final fine-tuning is performed on real 2DP annotated data to make the model accurately adapt to the characteristics of actual 2DP materials.

[0036] 3.2DP feature weighted EMA technology: Aiming at the key features of 2DP layer detection, an innovative feature weighted EMA mechanism is designed: a. Give higher EMA weights to the key feature channels (RGB information) of the model to identify the number of 2DP layers, and enhance the stability of these features; b. Introduce a dynamic EMA weight adjustment strategy to dynamically adjust the weights of key feature channels based on training results, and adaptively adjust EMA parameters based on training progress and verification performance; c. Combined with the model uncertainty estimation, a higher EMA weight is applied to the high-certainty areas (masked parts) confirmed by the Mask2Former model to further improve the model stability.

[0037] Based on the model design of this step, the present invention proposes for the first time a MaskEMA-2DP model specifically for 2DP layer number detection, which has made breakthroughs in solving the following technical difficulties: 1.2DP optical property perception mechanism: Aiming at the unique optical interference pattern of 2DP materials, a special optical feature extraction process is designed (2DP-specific denoising and multi-scale contrast enhancement process in pre-processing), which can effectively identify and enhance the interference fringes and color change characteristics of 2DP with different layers; 2. Multi-scale feature fusion architecture: Innovatively designed hierarchical feature fusion mechanism, combined with EMA and transfer learning, to achieve effective integration of multi-scale features of 2DP materials; 3.2DP sample efficient learning strategy: To address the problem of scarce 2DP labeled data, transfer learning and EMA technology are integrated to greatly improve the learning efficiency and stability of the model under small sample conditions.

[0038] The MaskEMA-2DP model of the present invention is based on the Mask2Former architecture, but is specially optimized for 2DP materials. It can effectively capture the layer-related features in the image, thereby achieving accurate segmentation of regions with different numbers of layers in the 2DP material.

[0039] 2DP-specific optimization of transfer learning and EMA technology: In order to address the problems of insufficient 2DP labeled data and training stability, this paper designs a transfer learning and EMA collaborative optimization mechanism for 2DP characteristics: 1.2DP-specific pre-training strategy: First, pre-train the model on a general dataset to obtain basic feature representation capabilities, and finally make fine adjustments on real 2DP data to form an efficient three-stage transfer learning path; 2.2DP feature weighted EMA: Innovatively proposed weighted EMA technology for 2DP key features, giving higher smoothing weights to feature channels that are critical to the number of recognition layers, improving the stability of the model for key features; 3. Dynamic batch normalization calibration: In view of the variability between 2DP samples, a dynamic batch normalization parameter adjustment mechanism is designed to enable the model to better adapt to the characteristic changes of different batches of 2DP samples.

[0040] These technologies optimized for 2DP significantly reduce the need for large-scale labeled data, enabling the model to quickly learn the layer features of 2DP under small sample conditions and maintain the stability of the training process.

[0041] 4. Image segmentation and layer detection like Figure 3 As shown in the figure, the trained MaskEMA-2DP model is used to input optical microscope images for prediction, demonstrating the performance of the Mask2Former model in different layer detection tasks. After the image is segmented, the number of layers is classified according to the pixels in different segmented areas. Each area is classified into different numbers of layers (such as single layer, double layer, etc.) according to the color label, and the percentage of each type of area is quantitatively analyzed. Through this process, quantitative support can be provided for the performance optimization of 2DP materials.

[0042] like Figure 3 As shown in a)-d) in the figure, the segmentation results of images with different number of layers (1L, 2L, 3L, etc.) are shown. The segmentation results of each image clearly mark the different layer areas, and the colors correspond to the layer classification, which is convenient for intuitive comparison.

[0043] like Figure 3 As shown in Figures e)-f), the statistical data of the segmentation results are given, including the area ratio of each layer region (such as single layer accounts for 15.32%, double layer accounts for 52.88%, etc.). This quantitative analysis helps to understand the layer distribution of two-dimensional materials under different experimental conditions and supports experimental optimization.

[0044] Model performance evaluation: like Figure 4 As shown in a)-c), a) is the original image, b) is the prediction result using mask2former, and c) is the prediction result of the model after using EMA. The results in the figure show that EMA enhances the model's ability to predict impurities.

[0045] like Figure 4 As shown in d), the EMA technique improves the model performance. Without EMA, the mIoU (Mean Intersection over Union) value during model training is low. After using EMA, the mIoU value of the model is significantly improved. The results in the figure show that the EMA technique enhances the model's ability to capture details and improves the accuracy of layer detection.

[0046] like Figure 4 As shown in e), the training efficiency of different models is demonstrated. After adding transfer learning, the Mask2Former model requires fewer training iterations than other models (such as U-Net, PSPNet, etc.), showing its efficient training performance.

[0047] 5. Experimental verification and robustness analysis like Figure 5 As shown in Figure 2, the robustness of the Mask2Former model under different brightness and contrast conditions is demonstrated. Experiments show that despite changes in image brightness and contrast, the model can still maintain high detection accuracy.

[0048] like Figure 5 As shown in a)-e) in Figure 1, by changing the brightness and contrast of the image, the performance of the model shows good stability. Under different processing conditions, the model can still accurately segment the regions of each layer.

[0049] like Figure 5 As shown in f), the prediction accuracy of the model under different conditions is demonstrated, indicating that the model can still maintain high accuracy when dealing with changes in brightness and contrast, verifying its wide applicability in practical applications.

[0050] like Figure 5 As shown in g), the model evaluation under different image processing conditions (such as a 10% increase in brightness, a 10% decrease in contrast, etc.) is shown. The results show that even under different image changes, the Mask2Former model can still accurately identify the number of layers, and the performance under all conditions maintains a high mIoU and prediction accuracy.

[0051] 6. Final results and optimization After training, the model can automatically identify the number of layers in two-dimensional material images and perform quantitative analysis, greatly improving the efficiency of data analysis. Ultimately, the segmentation results and layer ratios output by the model provide data support for material optimization.

[0052] Corresponding to the aforementioned embodiment of a method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning, the present invention also provides an embodiment of a device for detecting the number of layers of two-dimensional materials under a microscope based on deep learning.

[0053] See also Figure 6 An embodiment of the present invention provides a device for detecting the number of two-dimensional material layers under a microscope based on deep learning, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement a method for detecting the number of two-dimensional material layers under a microscope based on deep learning in the above embodiment.

[0054] An embodiment of a two-dimensional material layer detection device under a microscope based on deep learning provided by the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located reading the corresponding computer program instructions in the non-volatile memory into the internal memory for execution. From a hardware perspective, if Figure 6 As shown, it is a hardware structure diagram of any device with data processing capability where a two-dimensional material layer number detection device under a microscope based on deep learning provided by the present invention is located, except Figure 6 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0055] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0056] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0057] An embodiment of the present invention also provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning in the above embodiment is implemented.

[0058] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0059] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning.

[0060] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0061] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0062] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning, characterized in that: The method comprises the following steps: a) Sample preparation: mixing a two-dimensional material with a solvent, causing the two-dimensional material to swell under the action of the solvent and to be layered under the action of shear force to obtain a plurality of layers of the two-dimensional material, and depositing the two-dimensional material on a substrate; b) Image acquisition: High-resolution RGB images of deposited 2D materials are taken using an optical microscope and pre-processed for deep learning model training; c) Model training: Based on the pre-trained Mask2Former model, after transfer learning, it is used for training two-dimensional material images, and the exponential moving average EMA technology is used to smooth the parameters of the model; d) Layer number detection: Apply the trained Mask2Former model to the newly acquired two-dimensional material image for segmentation, generate segmentation masks for each region, and predict the number of layers in each region based on the masks; e) Quantitative analysis of the number of layers: Calculate the area ratio of each layer area based on the pixel ratio to obtain the percentage parameter of each layer.

2. The method according to claim 1, characterized in that The image preprocessing includes image denoising, multi-scale contrast enhancement, brightness adaptive correction, data normalization and data labeling.

3. The method according to claim 2, characterized in that The denoising is a denoising method that combines non-local mean filtering and wavelet transform to suppress interference patterns in two-dimensional material images while retaining key information related to the number of layers; the multi-scale contrast enhancement adopts multi-scale adaptive histogram equalization technology to locally optimize the feature differences in different layer areas of two-dimensional materials to make the layer boundaries clearer; the brightness adaptive correction adopts an adaptive brightness correction algorithm based on the optical properties of two-dimensional materials; the data annotation uses SAM to assist in annotating the two-dimensional material image to obtain segmentation results of areas with different number of layers.

4. The method according to claim 1, characterized in that The Mask2Former framework is based on the Transformer architecture and performs image segmentation through a multi-layer decoder to process global and local features in two-dimensional material images.

5. The method according to claim 1, characterized in that The layer number detection method is used for detecting the layer number of a two-dimensional polymer 2DP under a microscope.

6. The method according to claim 1, characterized in that The layer prediction results of the Mask2Former model are displayed in the image in a color-coded manner, and different layer areas are marked with different colors.

7. The method according to claim 1, characterized in that The quantitative analysis of the number of layers outputs the percentage of each layer area and displays the distribution of different layers through statistical charts.

8. The method according to claim 1, characterized in that During the model training process, the following feature weighting mechanism is used for the key features of two-dimensional material layer detection: 1) Assign a higher EMA weight to the RGB information feature channel of the number of 2D material layers; 2) Dynamically optimize and adjust the weights of the RGB information feature channels based on the training results; 3) Assign higher EMA weights to high-certainty areas identified by the Mask2Former model.

9. A device for detecting the number of layers of two-dimensional materials under a microscope based on deep learning, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, a method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, a method for detecting the number of layers of two-dimensional materials under a microscope based on deep learning as described in any one of claims 1 to 8 is implemented.

Images