A method for deblurring based on full-range context-based microscopic defocus imaging

Abstract

The application relates to a microscopic defocus imaging deblurring method based on full-range context, comprising the following steps: constructing a microscopic imaging deblurring neural network model; acquiring a training data set, training the microscopic imaging area deblurring neural network model based on the training data set, and obtaining a trained microscopic imaging area deblurring neural network model; inputting a to-be-processed blurred image into the trained microscopic imaging area deblurring neural network model to obtain a clear image. The application is based on full-range context information, and on the basis of comprehensively considering factors such as microscopic image background stray light, defocus degree and detail texture, the high-quality reconstruction of the defocus degraded image effectively improves the definition and distinguishability of the microscopic image, and provides a powerful guarantee for subsequent accurate observation and quantitative analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of microscopic imaging, in particular to a microscopic defocus imaging deblurring method based on full-range context. BACKGROUND

[0002] With the increasing demand for microscopic observation, microscopic imaging has become an important means to obtain microscopic information in many fields such as biomedical science and material science. However, the microscopic system is often affected by various factors and defocus problems occur, which leads to a significant reduction in the clarity of the captured images, and further limits the accuracy and efficiency of observation and analysis. Once microscopic defocus occurs, it is difficult to identify the structural details in the image, and it may also cause deviation in subsequent quantitative analysis results, which has an adverse effect on high-precision experiments and clinical diagnosis.

[0003] At the same time, with the rapid development of precision manufacturing, nanometer detection and high-resolution microscopic observation, the demand for high-quality and high-definition microscopic images is growing. However, the microscopic imaging environment is usually complex, and factors such as sample thickness, refractive index, and optical system performance can all cause imaging defocus. Especially in multi-level structure or dynamic living body observation, traditional focusing or stack shooting methods often cannot balance efficiency and accuracy, resulting in a large number of images with different degrees of blur. Since the microscopic imaging target is usually small in size and has fine structural details, if the defocus blur cannot be effectively recovered, key information may be missed, and even accurate identification and judgment of the target may be disturbed.

[0004] At present, although the research on microscopic image deblurring in the academic and industrial circles at home and abroad has achieved initial success, most of them are aimed at motion blur or ordinary optical system imaging blur, and there are relatively few methods suitable for microscopic defocus scenes. It should be noted that microscopic defocus and general motion blur or optical scattering phenomenon have significant differences in causes. The former is mainly caused by light path mismatch, sample shift and other factors, and its result is overall or local defocus degradation; while the latter is often caused by object rapid motion, medium scattering and other reasons, and the imaging characteristics are not the same as microscopic defocus. Therefore, directly applying existing deblurring algorithms to the microscopic defocus scene often fails to achieve ideal results. In addition, some deblurring techniques for microscopic images also have certain limitations, such as insufficient consideration of complex interference such as imaging noise and background stray light, or lack of targeted modeling of aberration and off-axis imaging characteristics of the microscopic system, which limits their applicability in high-resolution and high-precision imaging scenarios. SUMMARY

[0005] In order to overcome at least one of the deficiencies in the prior art, the present application provides a microscopic defocus imaging deblurring method based on full-range context.

[0006] In a first aspect, a microscopic defocus imaging deblurring method based on full-range context is provided, comprising:

[0007] A microscopic imaging deblurring neural network model is constructed. The microscopic imaging deblurring neural network model comprises an encoder and a decoder. The encoder is used to gradually extract deep features of an input blurred image, and the decoder is used to gradually restore the deep features back to an image space to obtain a clear image. The encoder comprises a first convolutional layer and a plurality of encoding layers, and the decoder comprises a plurality of decoding layers and a second convolutional layer. The features output by each encoding layer are input into the corresponding decoding layer through a skip connection. The encoding layers and the decoding layers have the same structure, and are all full-range context blocks. The full-range context block is used to capture local features, regional features and global features in the space.

[0008] A training data set is obtained, and the microscopic imaging deblurring neural network model is trained based on the training data set to obtain a trained microscopic imaging deblurring neural network model. The samples in the training data set include blurred images and corresponding real clear images. The blurred images are input into the microscopic imaging deblurring neural network model to obtain predicted clear images. The loss is calculated based on the predicted clear images and the real clear images to train the model.

[0009] The blurred image to be processed is input into the trained microscopic imaging deblurring neural network model to obtain a clear image.

[0010] In one embodiment, the encoder comprises five encoding layers, namely a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer and a fifth encoding layer.

[0011] The decoder comprises five decoding layers, namely a first decoding layer, a second decoding layer, a third decoding layer, a fourth decoding layer and a fifth decoding layer.

[0012] The blurred image is input into the first convolutional layer to be converted into spatial features. The spatial features pass through the first encoding layer, the second encoding layer, the third encoding layer, the fourth encoding layer and the fifth encoding layer in sequence to obtain deep features.

[0013] The deep features are input into the first decoding layer. The first features output by the first encoding layer are input into the fourth decoding layer. The second features output by the second encoding layer are input into the third decoding layer. The third features output by the third encoding layer are input into the second decoding layer. The fourth features output by the fourth decoding layer are input into the first decoding layer.

[0014] The features output by the fifth decoding layer are connected with the blurred image residual after passing through the second convolutional layer to obtain a clear image.

[0015] In one embodiment, the full-range context block comprises a first local-range block, a regional-range block, a second local-range block and a global-range block.

[0016] The input of the full-range context block is subjected to local feature enhancement by the first local-range block to obtain first local features; the first local features are normalized and windowed to obtain a first value vector and a first key vector;

[0017] The input of the full-range context block is normalized and windowed to obtain a first query vector;

[0018] The first value vector, the first key vector, and the first query vector are input to the region-range block to obtain region features by using a multi-head self-attention mechanism;

[0019] The region features, the first value vector, the first key vector, and the input of the full-range context block are added to obtain a first addition result; the first addition result is subjected to a normalization layer and an MLP, and is added to the first addition result to obtain a second addition result;

[0020] The second addition result is input to the second local-range block for local feature enhancement to obtain second local features; the second local features are normalized and windowed to obtain a second value vector and a second key vector;

[0021] The second addition result is normalized and windowed to obtain a division result;

[0022] The second addition result is subjected to a max-pooling operation, a traditional attention operation, and a Sigmoid activation function, and is multiplied by the division result to obtain a second query vector;

[0023] The second value vector, the second key vector, and the second query vector are input to the global-range block to obtain global features by using a multi-head self-attention mechanism;

[0024] The global features, the second value vector, the second key vector, and the second addition result are added to obtain a third addition result; the third addition result is subjected to a normalization layer and an MLP, and is added to the third addition result to obtain an output of the full-range context block.

[0025] In one embodiment, the first local-range block and the second local-range block have the same structure, including a deep convolution layer, a point-wise convolution layer, a point-wise convolution layer, and a deep convolution layer connected in sequence.

[0026] In a second aspect, a microscopic defocus imaging deblurring device based on a full-range context is provided, including:

[0027] The model construction module is configured to construct a microscopic imaging deblurring neural network model. The microscopic imaging deblurring neural network model comprises an encoder and a decoder. The encoder is configured to gradually extract deep features of an input blurred image, and the decoder is configured to gradually restore the deep features back to an image space to obtain a clear image. The encoder comprises a first convolutional layer and a plurality of encoding layers, and the decoder comprises a plurality of decoding layers and a second convolutional layer. The features output by each encoding layer are input into a corresponding decoding layer through a skip connection. The encoding layers and the decoding layers have the same structure, and each of the encoding layers and the decoding layers is a full-range context block. The full-range context block is configured to capture local features, regional features and global features in the space.

[0028] The training module is configured to obtain a training data set, train the microscopic imaging deblurring neural network model based on the training data set, and obtain a trained microscopic imaging deblurring neural network model. The samples in the training data set comprise blurred images and corresponding real clear images. The blurred images are input into the microscopic imaging deblurring neural network model to obtain predicted clear images. The loss is calculated based on the predicted clear images and the real clear images to train the model.

[0029] The prediction module is configured to input a to-be-processed blurred image into the trained microscopic imaging deblurring neural network model to obtain a clear image.

[0030] In a third aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the computer program implements the microscopic de-focus imaging deblurring method based on the full-range context.

[0031] In a fourth aspect, a computer program product is provided. The computer program product comprises computer programs / instructions. When the computer programs / instructions are executed by a processor, the computer programs / instructions implement the microscopic de-focus imaging deblurring method based on the full-range context.

[0032] Compared with the prior art, the microscopic de-focus imaging deblurring method based on the full-range context has the following beneficial effects: the microscopic de-focus imaging deblurring method based on the full-range context starts from the full-range context information, considers the background stray light of the microscopic image, the degree of defocus and the detail texture and other factors, effectively improves the clarity and distinguishability of the microscopic image, and provides a strong guarantee for subsequent accurate observation and quantitative analysis. BRIEF DESCRIPTION OF DRAWINGS

[0033] The present application can be better understood by referring to the following description in conjunction with the accompanying drawings, which form a part of this specification. In the drawings:

[0034] Figure 1 A schematic diagram of a microscopic imaging deblurring neural network model is shown.

[0035] Figure 2 A schematic diagram of a full-range context block is shown;

[0036] Figure 3 A schematic diagram of a local-range block is shown;

[0037] Figure 4 A structural block diagram of a microscopic defocus imaging deblurring device based on a full-range context is shown. DETAILED DESCRIPTION

[0038] In the following, exemplary embodiments of the present application will be described with reference to the accompanying drawings. In the description, not all features of a practical embodiment are described in order to keep the description clear and concise. It should be appreciated that many embodiment-specific decisions can be made in the development of any such practical embodiment in order to achieve the specific goals of the developer, and these decisions can vary from embodiment to embodiment.

[0039] It should also be noted that, in order not to obscure the application with unnecessary details, only the structures of the devices closely related to the solution according to the application are shown in the drawings, and other details not closely related to the application are omitted.

[0040] It should be understood that the present application is not limited to the described embodiments by virtue of the following description with reference to the drawings. In this text, embodiments can be combined with each other, features replaced or borrowed between different embodiments, or one or more features omitted in one embodiment, where possible.

[0041] The embodiment of the present application provides a microscopic defocus imaging deblurring method based on a full-range context, mainly comprising the following steps:

[0042] Step S1, constructing a microscopic imaging deblurring neural network model; the microscopic imaging deblurring neural network model comprises an encoder and a decoder, the encoder is used to extract deep features of an input blurred image step by step, and the decoder is used to restore the deep features back to an image space step by step to obtain a clear image; the encoder comprises a first convolutional layer and a plurality of encoding layers, and the decoder comprises a plurality of decoding layers and a second convolutional layer, features output by each encoding layer are input into a corresponding decoding layer through a skip connection; the encoding layers and the decoding layers have the same structure, and are all full-range context blocks; the full-range context block is used to capture local features, regional features and global features in space.

[0043] Step S2, obtain a training data set, train the microscopic imaging area blur neural network model based on the training data set, and obtain the trained microscopic imaging area blur neural network model; the samples in the training data set include a blur image and a corresponding real clear image; the blur image is input into the microscopic imaging area blur neural network model to obtain a predicted clear image (i.e. a deblurred image); the loss is calculated based on the predicted clear image and the real clear image to train the model.

[0044] Here, the blur image and the corresponding real clear image are used as an image pair for training and testing of the neural network. In view of the requirement of deep learning for a large amount of data and different sizes of existing images, the input image is preprocessed, and multiple training samples are generated from a single image through cropping, which significantly expands the data scale.

[0045] Step S3, input the blur image to be processed into the trained microscopic imaging area blur neural network model to obtain a clear image.

[0046] In this embodiment, by introducing a full-range context modeling technology, local, regional and global features are integrated to provide an efficient and accurate deblurring solution for microscopic imaging, meeting the high-resolution imaging requirements in complex environments.

[0047] In one embodiment, Figure 1 A schematic diagram of the microscopic imaging deblurring neural network model is shown, see Figure 1 The encoder includes five encoding layers, namely a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer and a fifth encoding layer.

[0048] The decoder includes five decoding layers, namely a first decoding layer, a second decoding layer, a third decoding layer, a fourth decoding layer and a fifth decoding layer.

[0049] The blur image is input into the first convolutional layer to convert into spatial features, and the spatial features pass through the first encoding layer, the second encoding layer, the third encoding layer, the fourth encoding layer and the fifth encoding layer in sequence to obtain depth features.

[0050] The depth features are input into the first decoding layer, the first features output by the first encoding layer are input into the fourth decoding layer, the second features output by the second encoding layer are input into the third decoding layer, the third features output by the third encoding layer are input into the second decoding layer, and the fourth features output by the fourth decoding layer are input into the first decoding layer.

[0051] The features output by the fifth decoding layer are connected with the blur image residual after passing through the second convolutional layer to obtain a clear image.

[0052] In this embodiment, the microscopic imaging deblurring neural network model is composed of a multi-layer symmetrical encoder-decoder structure. The encoder is used to gradually extract deep features of the input blurred microscopic image, and the decoder is used to gradually restore the features extracted by the encoder back to the image space, enhance and refine specific feature information, obtain a high-quality clear image, and the two are connected through a jump connection to fuse detailed features and high-level semantic features, avoid information loss, and at the same time enhance the recovery ability of the decoder to local details, thereby improving the performance and deblurring effect of the overall network.

[0053] In one embodiment, Figure 2 A schematic diagram of the full-range context block is shown. The full-range context block includes: a first local-range block, a region-range block, a second local-range block, and a global-range block;

[0054] The input of the full-range context block is subjected to local feature enhancement through the first local-range block to obtain first local features; the first local features are subjected to normalization and window division (using a normalization layer LN and a window division layer WP) to obtain a first value vector V and a first key vector K; here, the input is divided into a plurality of non-overlapping windows, and a shape adjustment operation is used to generate the value vector and the key vector. Here, the input of the full-range context block corresponding to the decoding layer adopts 2 inputs, which are spliced as the input of the full-range context block.

[0055] The input of the full-range context block is subjected to normalization and window division (using a normalization layer LN and a window division layer WP) to obtain a first query vector Q;

[0056] The first value vector, the first key vector, and the first query vector are input to the region-range block to obtain region features by using a multi-head self-attention mechanism (MSA);

[0057] The region features, the first value vector, the first key vector, and the input of the full-range context block are added to obtain a first addition result; the first addition result is subjected to a normalization layer LN and an MLP (Multilayer Perceptron), and is added to the first addition result to obtain a second addition result, which is used to enhance the feature expression after attention, and improve the expression effect and stability of the model.

[0058] The second addition result is input to the second local-range block for local feature enhancement to obtain second local features; the second local features are subjected to normalization and window division (using a normalization layer LN and a window division layer WP) to obtain a second value vector and a second key vector;

[0059] The second addition result is subjected to normalization and window division (using a normalization layer LN and a window division layer WP) to obtain a division result;

[0060] The second addition result is subjected to max pooling Maxpool, a traditional attention operation Attention, and a sigmoid activation function, and is multiplied by the division result to obtain a second query vector;

[0061] The second value vector, the second key vector, and the second query vector are input into a global range block to obtain global features by using a multi-head self-attention mechanism MSA;

[0062] The global features, the second value vector, the second key vector, and the second addition result are added to obtain a third addition result; the third addition result is subjected to a normalization layer LN and an MLP, and is added to the third addition result to obtain an output of the global range context block.

[0063] Figure 3 A schematic diagram of the local range block is shown. The first local range block and the second local range block have the same structure, including sequentially connected deep convolution layers, pointwise convolution layers, pointwise convolution layers, and deep convolution layers.

[0064] Here, first, a 3x3 deep convolution is used to perform spatial context encoding on the input features, then a 1x1 pointwise convolution is used to map the features to a low-dimensional space, then a Relu activation function is used to perform nonlinear transformation on the features, then a 1x1 pointwise convolution is used to adjust the channel dimension to the input channel number, and finally, a 3x3 deep convolution is used to further extract spatial features.

[0065] Specifically, the network model is trained by backpropagating the L1 loss calculated based on the predicted clear image and the real clear image. After the fuzzy microscopic image passes through the neural network, a reconstructed clear image is output. The model gradient is calculated using the L1 loss, so that the network continuously learns the mapping relationship between the fuzzy microscopic image and the clear image, constrains the training process, and updates the model parameters.

[0066] The same as the full-range context-based microscopic defocus imaging deblurring method, the embodiment also provides a full-range context-based microscopic defocus imaging deblurring device, Figure 4 A structural block diagram of the full-range context-based microscopic defocus imaging deblurring device is shown, which includes:

[0067] Model building module 41 is used to construct a microscopic imaging deblurring neural network model. The microscopic imaging deblurring neural network model includes an encoder and a decoder. The encoder is used to progressively extract deep features from the input blurred image, and the decoder is used to progressively restore the deep features back to the image space to obtain a clear image. The encoder includes a first convolutional layer and multiple encoding layers, and the decoder includes multiple decoding layers and a second convolutional layer. The features output by each encoding layer are input to the corresponding decoding layer through skip connections. The encoding layer and the decoding layer have the same structure, which is a full-range context block. The full-range context block is used to capture local features, regional features, and global features in the space.

[0068] Training module 42 is used to acquire a training dataset and train the fuzzy neural network model of the microscopic imaging area based on the training dataset to obtain the trained fuzzy neural network model of the microscopic imaging area. The samples in the training dataset include fuzzy images and corresponding real sharp images. The fuzzy images are input into the fuzzy neural network model of the microscopic imaging area to obtain the predicted sharp images. The loss is calculated based on the predicted sharp images and the real sharp images to train the model.

[0069] The prediction module 43 is used to input the blurred image to be processed into the trained microscopic imaging area blurred neural network model to obtain a clear image.

[0070] The microscopic defocusing imaging deblurring device based on full-range context in this embodiment has the same inventive concept as the microscopic defocusing imaging deblurring method based on full-range context described above. Therefore, the specific implementation of this device can be found in the embodiment section of the microscopic defocusing imaging deblurring method based on full-range context described above, and its technical effects correspond to the technical effects of the above method, so it will not be repeated here.

[0071] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the above-described method for deblurring microscopic defocusing imaging based on full-range context.

[0072] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the aforementioned method for deblurring microscopic defocusing imaging based on full-range context.

[0073] In summary, this application has the following technical effects: The microscopic defocusing imaging deblurring method based on full-range context of this application starts from full-range context information and comprehensively considers factors such as background stray light, defocus degree and detail texture of microscopic images to reconstruct defocused images with high quality, effectively improving the clarity and recognizability of microscopic images, and providing a strong guarantee for subsequent accurate observation and quantitative analysis.

[0074] The above merely provides the various embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A full-range context-based deblurring method for microscopic defocus imaging, characterized in that, The method comprises the following steps: constructing a microscopic imaging deblurring neural network model; the microscopic imaging deblurring neural network model comprises an encoder and a decoder, the encoder is used for step-by-step extraction of deep features of an input blurred image, and the decoder is used for step-by-step restoration of the deep features back to an image space to obtain a clear image; the encoder comprises a first convolutional layer and a plurality of encoding layers, and the decoder comprises a plurality of decoding layers and a second convolutional layer, the features output by each encoding layer are input into the corresponding decoding layer through a jump connection; the encoding layers and the decoding layers have the same structure, and are all full-range context blocks; the full-range context block is used for capturing local features, regional features and global features in the space; obtaining a training data set, training the microscopic imaging deblurring neural network model based on the training data set to obtain a trained microscopic imaging deblurring neural network model; the samples in the training data set comprise a blurred image and a corresponding real clear image; the blurred image is input into the microscopic imaging deblurring neural network model to obtain a predicted clear image; the loss is calculated based on the predicted clear image and the real clear image to train the model; inputting a to-be-processed blurred image into the trained microscopic imaging deblurring neural network model to obtain a clear image; the full-range context block comprises a first local-range block, a regional-range block, a second local-range block and a global-range block; the input of the full-range context block is subjected to local feature enhancement through the first local-range block to obtain first local features; the first local features are normalized and window-divided to obtain a first value vector and a first key vector; the input of the full-range context block is normalized and window-divided to obtain a first query vector; the first value vector, the first key vector and the first query vector are input into the regional-range block to obtain regional features by using a multi-head self-attention mechanism; the regional features, the first value vector, the first key vector and the input of the full-range context block are added to obtain a first addition result; the first addition result is subjected to a normalization layer and an MLP, and is added to the first addition result to obtain a second addition result; the second addition result is input into the second local-range block for local feature enhancement to obtain second local features; the second local features are normalized and window-divided to obtain a second value vector and a second key vector; the second addition result is normalized and window-divided to obtain a division result; the second addition result is subjected to a max-pooling operation, a traditional attention operation and a Sigmoid activation function, and is multiplied by the division result to obtain a second query vector; the second value vector, the second key vector and the second query vector are input into the global-range block to obtain global features by using a multi-head self-attention mechanism; The global feature, the second value vector, the second key vector and the second addition result are added to obtain a third addition result; the third addition result passes through a normalization layer and an MLP, and is added to the third addition result to obtain an output of the full-range context block.

2. The method of claim 1, wherein, The encoder comprises five encoding layers, namely a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer and a fifth encoding layer. The decoder comprises five decoding layers, namely a first decoding layer, a second decoding layer, a third decoding layer, a fourth decoding layer and a fifth decoding layer. The blurred image is input into the first convolutional layer to be converted into spatial features, and the spatial features pass through the first encoding layer, the second encoding layer, the third encoding layer, the fourth encoding layer and the fifth encoding layer in sequence to obtain deep features. The deep features are input into the first decoding layer, the first feature output by the first encoding layer is input into the fourth decoding layer, the second feature output by the second encoding layer is input into the third decoding layer, the third feature output by the third encoding layer is input into the second decoding layer, and the fourth feature output by the fourth decoding layer is input into the first decoding layer. The feature output by the fifth decoding layer is connected with the blurred image residual after passing through the second convolutional layer to obtain a clear image.

3. The method of claim 1, wherein, The first local range block and the second local range block have the same structure, comprising a deep convolutional layer, a point-by-point convolutional layer, a point-by-point convolutional layer and a deep convolutional layer connected in sequence.

4. A full-range context-based deblurring apparatus for microscopic defocus imaging, comprising: It comprises: A model construction module is configured to construct a microscopic imaging deblurring neural network model; the microscopic imaging deblurring neural network model comprises an encoder and a decoder, the encoder is configured to gradually extract deep features of an input blurred image, and the decoder is configured to gradually restore the deep features back to an image space to obtain a clear image; the encoder comprises a first convolutional layer and a plurality of encoding layers, and the decoder comprises a plurality of decoding layers and a second convolutional layer; the features output by each encoding layer are input into the corresponding decoding layer through a jump connection; the encoding layers and the decoding layers have the same structure, and are all full-range context blocks; the full-range context block is configured to capture local features, regional features and global features in space; A training module is configured to obtain a training data set, train the microscopic imaging deblurring neural network model based on the training data set, and obtain a trained microscopic imaging deblurring neural network model; the samples in the training data set comprise blurred images and corresponding real clear images; the blurred images are input into the microscopic imaging deblurring neural network model to obtain predicted clear images; the loss is calculated based on the predicted clear images and the real clear images to train the model; A prediction module is configured to input a to-be-processed blurred image into the trained microscopic imaging deblurring neural network model to obtain a clear image. The full-range context block comprises a first local range block, a regional range block, a second local range block and a global range block. The input of the full-range context block is subjected to local feature enhancement by the first local-range block to obtain first local features; the first local features are normalized and windowed to obtain a first value vector and a first key vector; The input of the full-range context block is normalized and windowed to obtain a first query vector; The first value vector, the first key vector, and the first query vector are input into the region-range block to obtain region features by using a multi-head self-attention mechanism; The region features, the first value vector, the first key vector, and the input of the full-range context block are added to obtain a first addition result; the first addition result is subjected to a normalization layer and an MLP, and added to the first addition result to obtain a second addition result; The second addition result is input into the second local-range block to obtain second local features by local feature enhancement; the second local features are normalized and windowed to obtain a second value vector and a second key vector; The second addition result is normalized and windowed to obtain a division result; The second addition result is subjected to a max-pooling operation, a traditional attention operation, and a Sigmoid activation function, and multiplied by the division result to obtain a second query vector; The second value vector, the second key vector, and the second query vector are input into the global-range block to obtain global features by using a multi-head self-attention mechanism; The global features, the second value vector, the second key vector, and the second addition result are added to obtain a third addition result; the third addition result is subjected to a normalization layer and an MLP, and added to the third addition result to obtain an output of the full-range context block.

5. A computer readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the full-range context-based microscopic defocus imaging deblurring method of any one of claims 1-3.

6. A computer program product, characterised in that, The computer program / instructions are executed by a processor to implement the full-range context-based microscopic defocus imaging deblurring method of any one of claims 1-3.

Images