Deep periscopic waveguide imaging method and imaging system
Through the deep periscope waveguide imaging method, combined with focus judgment and defocus amount judgment model, the problem of poor chip imaging quality is solved, higher imaging accuracy and clarity are achieved, and the optical path design of the packaging machine is simplified.
Patent Information
- Application Number
- CN202510119081.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-24
- Publication Date
- 2025-06-13
AI Technical Summary
The existing chip imaging methods have problems such as poor imaging effects and poor imaging quality during the packaging process, especially in traditional packaging machines, which are complex in imaging optical path design.
The deep periscope waveguide imaging method is used to acquire historical chip images, and focus judgment algorithms and defocus amount judgment models are constructed, the focus position and defocus amount of the target chip image are judged, and the image correction algorithm is used to correct it if necessary.
The complex imaging optical path design in traditional packaging machines is simplified, the recognition accuracy and imaging quality are improved, the obtained images are clearer and more accurate, and the imaging performance of chip packaging machine detection is enhanced.
Smart Images
Figure CN120143396A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of chip imaging technology, and particularly to a deep dive waveguide imaging method and an imaging system. Background Art
[0002] Chip packaging and imaging technology are key areas in semiconductor manufacturing, involving packaging chips into usable components and monitoring and imaging them. Since in this field, an efficient and accurate imaging system is required to ensure the quality and reliability of the packaging process. Therefore, it is necessary to introduce deep dive waveguide imaging technology to improve the quality of images, especially to be able to perform clear imaging even in narrow spaces. Currently, when using deep dive waveguide imaging technology to image chips, due to the complex imaging optical path design in traditional packaging machines, the imaging quality is poor. It can be seen that the existing imaging methods have problems of poor imaging effect and poor imaging quality. Summary of the Invention
[0003] The present invention provides a deep dive waveguide imaging method and an imaging system to solve the problems of poor imaging effect and poor imaging quality existing in the existing imaging methods.
[0004] To achieve the above object, the present invention is realized through the following technical solutions: In a first aspect, the present invention provides a deep dive waveguide imaging method, including: Obtain historical chip images, construct a focus judgment algorithm, and use the focus judgment algorithm to determine the focus position of the historical chip images; Preset a focus position threshold range, compare the focus position with the preset focus position threshold range, and when the focus position is within the preset focus position threshold range, use the historical chip image corresponding to the focus position as a training set image; Construct a defocus amount judgment model, and use the training set images to train the defocus amount judgment model to obtain a trained defocus amount judgment model; Obtain a target chip image, and input the target chip image into the trained defocus amount judgment model to obtain the defocus amount of the target chip image; Compare the defocus amount with a preset defocus amount range. When the defocus amount is within the preset defocus amount range, output the target chip image. When the defocus amount is outside the preset defocus amount range, construct an image correction algorithm, and use the image correction algorithm to correct the target chip image, and output the corrected target chip image.
[0005] Optionally, the step of using the focus judgment algorithm to determine the focus position of the historical chip images includes: Evaluate the edge sharpness of the historical chip images by calculating the gradient information of the historical chip images, and its calculation method satisfies the following relational expression: (1) (2) (3) In the formula, and are the Sobel operators in the horizontal and vertical directions respectively, is the pixel value of the original image, and are the gradients in the horizontal and vertical directions at the pixel point (x, y), are the length and width of the target image respectively, and the size of the target image is ; Introduce the high-frequency information of the picture in formula (3) Improve the capture ability of historical chip images, and its calculation formula satisfies the following relationship: (4) (5) In the formula, F T represents the Fourier transform, H(x, y) represents the high-frequency filtering information of the picture, represents the gradient solution, represents the partial derivative, represents the pixel value of the original image, is an adjustable weight; Calculate the first focal position of the historical chip image through formula (4); Calculate the second focal position of the historical chip image through the discrete cosine transform, and its calculation method satisfies the following relationship: (6) (7) In the formula, B is the DCT coefficient of matrix A, p and q are the index values of the current frequency in the horizontal and vertical directions respectively, represents the adjustment coefficient in the horizontal direction, represents the adjustment coefficient in the vertical direction, represents the representation matrix of the original image in the spatial domain, and each element in it corresponds to the pixel value at a specific position (m rows and n columns) of the image. m represents the number of rows of matrix A, and n represents the number of columns of matrix A, are the length and width of the target image respectively, and the size of the target image is ; Judge the first focal position and the second focal position. When the first focal position is the same as the second focal position, then the first focal position or the second focal position is the focal position of the historical chip image. When the first focal position is different from the second focal position and the difference between them is less than the threshold, then take the average value of the first focal position and the second focal position as the focal position.
[0006] Optionally, the defocus amount judgment model is a ResNet50 residual network model. The process of training the defocus amount judgment model with the training set images to obtain the trained defocus amount judgment model includes: Divide the training set images at a preset interval to obtain the divided training set images, and classify the divided training set images based on the preset interval to obtain the classified training set images; Input the classified training set images into the defocus amount judgment model to train the model and obtain the trained defocus amount judgment model.
[0007] Optionally, the process of correcting the target chip image using the image correction algorithm includes: Use the selective hourglass mapping diffusion model to generate intermediate degraded images with different degradation degrees, and introduce explicit conditions to strengthen the selectivity of the model. The explicit condition generation function satisfies the following relational formula: (8) In the formula, is the diffusion result at time step t , is the uncorrected degraded image, is the degraded image and the clear image residual, is diffusion coefficient, is the Gaussian noise coefficient, is the Gaussian noise, is the shared distribution coefficient, t is the time step, is the shared distribution term, which can map different degradation types to the same shared distribution, and , , ; As the time step increases, increases from 0 to 0.9, and the weight of the degraded image gradually decreases, making the final state of the diffusion result close to the Gaussian distribution, that is, when , , is the Gaussian distribution, is: (9) The function for gradually correcting the image during the reverse diffusion process is as follows: (10) In the formula, is the residual image predicted by the model; For each time step , …, 1, iterate in sequence. When , output the corrected image When t = T , is: (11) During the model update process, optimize the model parameters at each time step according to the loss function. The loss function is: (12) In the formula, is the loss function of the model, is the mathematical expectation, is the established value during the forward diffusion process, is the prediction result during the reverse diffusion process, is the parameter of the model, and t is the time step; After updating using the gradient descent method, then perform reverse diffusion to obtain a clear restored image.
[0008] In a second aspect, an embodiment of the present application provides a deep diving periscope waveguide imaging system, which is applied to the deep diving periscope waveguide imaging method as described in the first aspect. The system includes: a deep diving periscope component, a honeycomb waveguide image transmission optical fiber, and a focusing judgment device.
[0009] Optionally, one end of the deep diving periscope component is connected to one end of the honeycomb waveguide image transmission optical fiber, and the other end of the honeycomb waveguide image transmission optical fiber is connected to the focusing judgment device.
[0010] Optionally, the system further includes a light condensing device and a diffusion device. The light condensing device is disposed between the deep diving periscope component and the honeycomb waveguide image transmission optical fiber. One end of the light condensing device is connected to the deep diving periscope component, and the other end of the light condensing device is connected to the honeycomb waveguide image transmission optical fiber. The diffusion device is disposed between the honeycomb waveguide image transmission optical fiber and the focusing judgment device. One end of the diffusion device is connected to the honeycomb waveguide image transmission optical fiber, and the other end of the diffusion device is connected to the focusing judgment device; The light condensing device is used to focus the light emitted from the deep diving periscope component; The diffusion device is used to diffuse the light emitted from the honeycomb waveguide image transmission optical fiber.
[0011] Optionally, the system further includes an optical signal amplifier disposed on the cellular waveguide image transmission fiber for amplifying the energy of the light transmitted in the cellular waveguide image transmission fiber.
[0012] Beneficial effects: The deep diving periscope waveguide imaging method provided by the present invention simplifies the complex imaging optical path design in a traditional packaging machine by combining deep diving periscope and optical waveguide image transmission technologies, improves the recognition accuracy and imaging quality, makes the obtained image clearer and more accurate, thereby enhancing the imaging performance of the chip packaging machine for detection. In this method, an image focusing judgment and an image correction algorithm are innovatively combined, which has the uniqueness of solving the problem of poor imaging quality in the chip packaging machine, greatly simplifies the optical path design of the imaging system in the chip packaging machine, and reduces the complexity of the system. Description of the drawings
[0013] Figure 1 is a flowchart of the deep diving periscope waveguide imaging method according to an embodiment of the present invention; Figure 2 is a schematic diagram of the chip image focusing state according to an embodiment of the present invention; Figure 3 is a schematic flowchart of the preprocessing of the training set image according to an embodiment of the present invention; Figure 4 is a partial structural diagram of the cellular light image transmission according to an embodiment of the present invention; Figure 5 is a schematic structural diagram of the deep diving periscope component according to an embodiment of the present invention; Figure 6 is a flowchart of the construction and training of the RTDN focusing judgment algorithm according to an embodiment of the present invention; Figure 7 is a working flowchart of the imaging system according to an embodiment of the present invention; In the figure, 1, a light condensing device; 2, a diffusion device; 3, a cellular waveguide image transmission fiber; 4, an optical signal amplifier; 5, a focusing judgment device; 6, a deep diving periscope component. Detailed implementation manners
[0014] The technical solutions of the present invention will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0015] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the art to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Similarly, terms such as "a" or "an" do not denote a quantity limitation, but mean that there is at least one. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left" and "right" are only used to represent relative position relationships. When the absolute position of the object being described changes, the relative position relationship also changes accordingly.
[0016] Please refer to Figure 1 , the embodiments of the present application provide a deep-diving periscope waveguide imaging method, including: Obtain historical chip images, construct a focus judgment algorithm, and use the focus judgment algorithm to determine the focus position of the historical chip images; Preset a focus position threshold range, compare the focus position with the preset focus position threshold range, and when the focus position is within the preset focus position threshold range, use the historical chip image corresponding to the focus position as the training set image; Construct a defocus amount judgment model, and use the training set images to train the defocus amount judgment model to obtain a trained defocus amount judgment model; Obtain a target chip image, and input the target chip image into the trained defocus amount judgment model to obtain the defocus amount of the target chip image; Compare the defocus amount with a preset defocus amount range. When the defocus amount is within the preset defocus amount range, output the target chip image. When the defocus amount is outside the preset defocus amount range, construct an image correction algorithm, and use the image correction algorithm to correct the target chip image, and output the corrected target chip image.
[0017] In the above embodiments, by combining deep-diving periscope and optical waveguide image transmission technologies, the complex imaging optical path design in traditional packaging machines is simplified, the recognition accuracy and imaging quality are improved, the obtained images are clearer and more accurate, thereby enhancing the imaging performance of chip packaging machine detection. In this method, the image focus judgment and image correction algorithm are innovatively combined, which has the uniqueness to solve the problem of poor imaging quality in chip packaging machines, greatly simplifies the optical path design of the imaging system in chip packaging machines, and reduces the complexity of the system.
[0018] Optionally, using the focus judgment algorithm to determine the focus position of the historical chip images includes: Evaluate the edge sharpness of the historical chip images by calculating the gradient information of the historical chip images, and its calculation method satisfies the following relational expression: (1) (2) (3) In the formula, and are the Sobel operators in the horizontal and vertical directions respectively, is the pixel value of the original image, and are the gradients in the horizontal and vertical directions at the pixel point (x, y), are the length and width of the target image respectively, and the size of the target image is ; Introduce the high-frequency information of the picture in formula (3) to improve the capture ability of historical chip images, and its calculation formula satisfies the following relational formula: (4) (5) In the formula, F T represents the Fourier transform, H(x, y) represents the high-frequency filtering information of the picture, represents the gradient solution, represents the partial derivative, represents the pixel value of the original image, is the adjustable weight; Calculate the first focal position of the historical chip image through formula (4); Calculate the second focal position of the historical chip image through the discrete cosine transform, and its calculation method satisfies the following relational formula: (6) (7) In the formula, B is the DCT coefficient of matrix A, p and q are the index values of the current frequency in the horizontal and vertical directions respectively, represents the adjustment coefficient in the horizontal direction, represents the adjustment coefficient in the vertical direction, represents the representation matrix of the original image in the spatial domain, and each element in it corresponds to the pixel value at a specific position (m rows and n columns) of the image. m represents the number of rows of matrix A, and n represents the number of columns of matrix A, are the length and width of the target image respectively, and the size of the target image is ; Judge the first focal position and the second focal position. When the first focal position is the same as the second focal position, then the first focal position or the second focal position is the focal position of the historical chip image. When the first focal position is different from the second focal position and the difference between the two is less than the threshold, then the average value of the first focal position and the second focal position is used as the focal position.
[0019] In the above embodiments, during the operation of the chip bonder, mechanical vibrations and other conditions may cause image capture defocus, as Figure 2 shown in Figure 2 Three different focusing states during the sample observation are shown: (a) positive defocus, (b) in focus, and (c) negative defocus. To ensure that images of different focal positions can still be effectively captured under vibration conditions, it is necessary to obtain the historical chip images of these three states, set the defocus distance range to -5 to 5 mm, the collection interval to 0.2 mm, and the picture size to 224 × 224. This setting can ensure that sufficient images of different focusing states are captured within the same field of view.
[0020] Since the chip target is small and not obvious, to effectively judge the focusing state of the image, this application adopts an improved image sharpness evaluation T method and discrete cosine transform (DCT) to provide a more accurate determination of the focusing position.
[0021] The image sharpness evaluation T method is shown in formulas (1)-(3). By introducing the high-frequency information H of the picture into the traditional image sharpness evaluation T method, as shown in formulas (4)-(5), the improved image sharpness evaluation T method is obtained, as shown in formulas (1)-(5).
[0022] The discrete cosine transform (DCT) approximates an image through the sum of a set of cosine functions with different frequencies and amplitudes, as shown in formulas (6)-(7).
[0023] Optionally, the defocus amount judgment model is a ResNet50 residual network model. The trained defocus amount judgment model is obtained by training the defocus amount judgment model with the training set images, including: Dividing the training set images at a preset interval to obtain the divided training set images, and classifying the divided training set images based on the preset interval to obtain the classified training set images; Inputting the classified training set images into the defocus amount judgment model to train the model to obtain the trained defocus amount judgment model.
[0024] In the above embodiments, based on the ResNet50 residual network model, the present application proposes an RTDN focusing judgment method. When constructing the dataset, the defocus distance is re-divided into 11 categories from -5 mm to 5 mm, with an interval of 1 mm for each category. A total of 5 images, including the focal position image and its adjacent images before and after, are grouped into one category, indicating that the defocus distance of this category is 0 mm (i.e., the focal position). In the positive defocus direction, 5 adjacent images at the 1 mm position are grouped into the same category, indicating that their defocus distance is 1 mm, and so on. It can be seen that the category range in the positive defocus direction is from 1 mm to 5 mm, and there are 5 categories in this direction, while the category range in the negative defocus direction is from -1 mm to -5 mm, and there are also 5 categories in this direction.
[0025] The complete process of training this network model is as Figure 3 shown. The collected training set is divided into 11 categories according to the above division criteria, with an interval of 1 mm for each category, and then all are input into the ResNet50 model for training to obtain the best network model, and the result finally output by this model is the defocus amount.
[0026] The ResNet50 model plus the improved image sharpness evaluation T method and discrete cosine transform is the RTDN (Resnet50-T-DCT network) focusing judgment algorithm in the present application. The construction and training process of the RTDN focusing judgment algorithm in the present application is as Figure 6 shown.
[0027] Optionally, use an image correction algorithm to correct the target chip image, including: Use a selective hourglass mapping diffusion model to generate intermediate degraded images with different degradation degrees, and introduce explicit conditions to strengthen the selectivity of the model. The explicit condition generation function satisfies the following relational formula: (8) In the formula, is the diffusion result at time step t , is the uncorrected degraded image, is the degraded image and the residual of the clear image , is 's diffusion coefficient, is the Gaussian noise coefficient, is the Gaussian noise, is the shared distribution coefficient, t is the time step, is the shared distribution term, which can map different degradation types to the same shared distribution, and , , ; As the time step increases, increasing from 0 to 0.9, the weight of the degraded image gradually decreases, making the final state of the diffusion result approach the Gaussian distribution, that is, when when, , is the Gaussian distribution, is: (9) The function for gradually correcting the image during the reverse diffusion process is: (10) In the formula, is the residual image predicted by the model; For each time step , …, 1 iterate in turn, when when, output the corrected image when, t = T when, is: (11) During the model update process, at each time step optimize the model parameters according to the loss function, and the loss function is: (12) In the formula, is the loss function of the model, is the mathematical expectation, is the established value during the forward diffusion process, is the prediction result during the reverse diffusion process, is the parameter of the model, and t is the time step; After updating using the gradient descent method, then perform reverse diffusion to obtain a clear restored image.
[0028] In the above embodiments, considering various problems such as the defocus state of the image and possible distortion, color distortion, and blurring of the image during transmission, it is necessary to repair the image with an algorithm to improve the image quality. Based on the existing diffusion algorithm model, this application can solve image degradation problems such as defocus, distortion, color distortion, and blurring with the same algorithm model to clarify the chip image. The Selective Hourglass Mapping Diffusion (SHM_DiffUIR) model generates intermediate degraded images with different degradation degrees through forward diffusion and gradually restores them through reverse diffusion, training the correction ability of the model. Based on SHM_DiffUIR, a SHM_DiffUIR image correction algorithm is constructed, and the SHM_DiffUIR image correction algorithm is shown in formulas (8)-(12).
[0029] An embodiment of the present application further provides a deep-diving periscope waveguide imaging system, which includes: a deep-diving periscope member 6, a honeycomb waveguide image transmission optical fiber 3, and a focusing judgment device 5.
[0030] Optionally, one end of the deep-diving periscope member 6 is connected to one end of the honeycomb waveguide image transmission optical fiber 3, and the other end of the honeycomb waveguide image transmission optical fiber 3 is connected to the focusing judgment device 5.
[0031] Optionally, the system further includes a light condensing device 1 and a diffusion device 2. The light condensing device 1 is arranged between the deep-diving periscope member 6 and the honeycomb waveguide image transmission optical fiber 3. One end of the light condensing device 1 is connected to the deep-diving periscope member 6, and the other end of the light condensing device 1 is connected to the honeycomb waveguide image transmission optical fiber 3. The diffusion device 2 is arranged between the honeycomb waveguide image transmission optical fiber 3 and the focusing judgment device 5. One end of the diffusion device 2 is connected to the honeycomb waveguide image transmission optical fiber 3, and the other end of the diffusion device 2 is connected to the focusing judgment device 5; The light condensing device 1 is used to focus the light emitted from the deep-diving periscope member 6; The diffusion device 2 is used to diffuse the light emitted from the honeycomb waveguide image transmission optical fiber 3.
[0032] Optionally, the system further includes: an optical signal amplifier 4, which is arranged on the honeycomb waveguide image transmission optical fiber 3 and is used to amplify the energy of the light transmitted in the honeycomb waveguide image transmission optical fiber 3.
[0033] In the above embodiment, a chip image is obtained through the deep-diving periscope member 6. As Figures 4 - 5 shown, the light transmitted from the deep-diving periscope member 6 enters the honeycomb waveguide image transmission optical fiber 3 and propagates in the form of an optical waveguide. There are a light condensing device 1 for gathering the light and transmitting it into the honeycomb waveguide image transmission optical fiber 3 and a diffusion device 2 for diffusing the light in the honeycomb waveguide image transmission optical fiber 3 into an appropriate size at both ends. The structure of the honeycomb waveguide image transmission optical fiber 3 adopts a honeycomb arrangement structure. The honeycomb arrangement can provide a relatively uniform optical fiber distribution, ensure that the optical signal can be transmitted evenly and completely to the entire optical cable, can improve the resistance of the optical cable to external electromagnetic interference, can reduce signal interference and loss, which helps to reduce the attenuation and distortion of the signal during transmission, and the honeycomb arrangement has strong scalability and can relatively easily increase or decrease the number of optical fiber bundles to meet different transmission requirements.
[0034] If the transmission distance is long and the optical signal loss is large, an optical signal amplifier 4 can be connected to the middle section of the waveguide transmission optical fiber to further reduce the energy loss of the light, so as to ensure that the image can be kept as complete as possible after being output from the optical fiber.
[0035] The complete working process of the deep-diving periscope waveguide imaging system is as Figure 7As shown, after the chip enters the encapsulator, the chip image is first obtained through the deep diving periscope structure 6, and then the image is transmitted to the focusing judgment device 5 by the honeycomb waveguide image transmission optical fiber 3. In the focusing judgment device, the defocus amount of the image is first calculated by the RTDN focusing judgment algorithm, and then it is judged whether the image clarity needs to be corrected. When no correction is required, that is, when the defocus amount is within ±0.5 mm, the image can be directly output. When correction is required, that is, when the defocus amount is outside ±0.5 mm, the image is corrected by the SHM_DiffUIR image correction algorithm, and the corrected image is output.
[0036] The above deep diving waveguide imaging system can implement each embodiment of the above deep diving waveguide imaging method and can achieve the same beneficial effects, which will not be elaborated here.
[0037] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations according to the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the protection scope determined by the claims.
Claims
1. A deep periscope waveguide imaging method, characterized in that: include: Acquire historical chip images, build a focus judgment algorithm, and use the focus judgment algorithm to determine the focus position of the historical chip images; Preset a focus position threshold range, compare the focus position with the preset focus position threshold range, and when the focus position is within the preset focus position threshold range, use the historical chip image corresponding to the focus position as a training set image; Constructing a defocus amount judgment model, and using the training set images to train the defocus amount judgment model to obtain a trained defocus amount judgment model; Acquire a target chip image, and input the target chip image into a trained defocus amount judgment model to obtain a defocus amount of the target chip image; The defocus amount is compared with a preset defocus amount range. When the defocus amount is within the preset defocus amount range, a target chip image is output. When the defocus amount is outside the preset defocus amount range, an image correction algorithm is constructed, the target chip image is corrected using the image correction algorithm, and the corrected target chip image is output.
2. The deep periscope waveguide imaging method according to claim 1, characterized in that: The method of determining the focal position of the historical chip image by using a focus judgment algorithm includes: The edge clarity of the historical chip image is evaluated by calculating the gradient information of the historical chip image, and the calculation method satisfies the following relationship: （1） （2） （3） In the formula, , are the Sobel operators in the horizontal and vertical directions respectively, is the original image pixel value, and is the horizontal and vertical gradient at the pixel point (x, y), are the length and width of the target image respectively, and the target image size is ; Introduce high-frequency information of the image into formula (3) Improve the ability to capture historical chip images, and its calculation formula satisfies the following relationship: （4） （5） In the formula, F T represents Fourier transform, H(x,y) represents the high-frequency filtering information of the image, represents the gradient solution, represents partial derivative, Represents the original image pixel value, is an adjustable weight; The first focal position of the historical chip image is calculated by formula (4); The second focal position of the historical chip image is calculated by discrete cosine transform, and the calculation method satisfies the following relationship: （6） （7） In the formula, B is the DCT coefficient of matrix A, p and q are the index values of the current frequency in the horizontal and vertical directions respectively. Represents the adjustment coefficient in the horizontal direction, Represents the adjustment coefficient in the vertical direction, Represents the representation matrix of the original image in the spatial domain, in which each element corresponds to the pixel value of a specific position (m rows and n columns) of the image, m represents the number of rows of matrix A, and n represents the number of columns of matrix A. are the length and width of the target image respectively, and the target image size is ; The first focal position and the second focal position are judged. When the first focal position is the same as the second focal position, the first focal position or the second focal position is the focal position of the historical chip image. When the first focal position is different from the second focal position and the difference between the two is less than a threshold, the average value of the first focal position and the second focal position is used as the focal position.
3. The deep periscope waveguide imaging method according to claim 1, characterized in that: The defocus amount judgment model is a ResNet50 residual network model, and the defocus amount judgment model is trained using the training set images to obtain a trained defocus amount judgment model, including: Dividing the training set images according to preset intervals to obtain divided training set images, and classifying the divided training set images based on the preset intervals to obtain classified training set images; The classified training set images are input into the defocus amount judgment model to train the model to obtain a trained defocus amount judgment model.
4. The deep periscope waveguide imaging method according to claim 1, characterized in that: The method of correcting the target chip image by using an image correction algorithm includes: The selective hourglass mapping diffusion model is used to generate intermediate degraded images with different degradation degrees, and explicit conditions are introduced to enhance the selectivity of the model. The explicit condition generation function satisfies the following relationship: （8） In the formula, For the time step t The diffusion result is is the uncorrected degraded image, For degraded images With clear images The residual of for The diffusion coefficient, is the Gaussian noise coefficient, is Gaussian noise, is the shared distribution coefficient, t is the time step, is a shared distribution term, different degradation types can be mapped to the same shared distribution, and , , ; As the time step increases, As the weight of the degraded image increases from 0 to 0.9, it gradually decreases, making the final state of the diffusion result close to the Gaussian distribution. hour, , is a Gaussian distribution, for: （9） The function of gradually correcting the image during the back diffusion process is: （10） In the formula, The residual image predicted by the model; For each time step ,…,1 iterates successively, when When , the corrected image is output ,when t = T hour, for: （11） During the model update process, at each time step Optimize model parameters based on loss function , the loss function is: （12） In the formula, is the loss function of the model, is the mathematical expectation, is the given value in the forward diffusion process, is the prediction result in the reverse diffusion process, is the parameter of the model, t is the time step; After updating using the gradient descent method, a clear restored image is obtained by back diffusion.
5. A deep periscope waveguide imaging system, used to implement the deep periscope waveguide imaging method according to claims 1-4, characterized in that: The system comprises: a deep periscope component (6), a honeycomb waveguide image transmission optical fiber (3), and a focus judgment device (5).
6. The deep periscope waveguide imaging system according to claim 5, characterized in that: The deep periscope component (6) is connected to one end of the honeycomb waveguide image transmission optical fiber (3), and the other end of the honeycomb waveguide image transmission optical fiber (3) is connected to the focus judgment device (5).
7. The deep periscope waveguide imaging system according to claim 5, characterized in that: The system further comprises a light focusing device (1) and a diffusion device (2), wherein the light focusing device (1) is arranged between the deep periscope component (6) and the honeycomb waveguide image transmission optical fiber (3), one end of the light focusing device (1) is connected to the deep periscope component (6), and the other end of the light focusing device (1) is connected to the honeycomb waveguide image transmission optical fiber (3); the diffusion device (2) is arranged between the honeycomb waveguide image transmission optical fiber (3) and the focus judgment device (5), one end of the diffusion device (2) is connected to the honeycomb waveguide image transmission optical fiber (3), and the other end of the diffusion device (2) is connected to the focus judgment device (5); The light focusing device (1) is used to focus the light emitted from the deep periscope component (6); The diffusion device (2) is used to diffuse the light emitted from the honeycomb waveguide imaging optical fiber (3).
8. The deep periscope waveguide imaging system according to claim 5, characterized in that: The system further comprises: an optical signal amplifier (4), wherein the optical signal amplifier (4) is arranged on the honeycomb waveguide imaging optical fiber (3) and is used to amplify the energy of the light transmitted in the honeycomb waveguide imaging optical fiber (3).