A method and device for obtaining a large-range microscopic imaging image

Abstract

The present invention discloses a method for obtaining a large-range microscopic imaging image, which includes initially collecting a to-be-calibrated image and a base image adjacent to the previous to-be-calibrated image, obtaining the gray-scale distribution function I1(x, y) of the base image and the gray-scale distribution function I2(x, y) of the to-be-calibrated image; substituting the gray-scale distribution function I2(x, y) of the to-be-calibrated image and the gray-scale distribution function I1(x, y) of the base image into a convolutional neural network for training to obtain a high-level feature map F2 representing the gray-scale distribution function I2(x, y) of the to-be-calibrated image and a high-level feature map F1 representing the gray-scale distribution function I1(x, y) of the base image; constructing a gray-scale template matching algorithm to calculate the displacement amounts of the overlapping region in the x and y coordinate directions respectively. By dynamically adjusting the displacement amount of the stage of the imaging control system, the present invention can ensure the real-time and efficient acquisition of the target image while avoiding the inaccuracy of the target image caused by image distortion and the like resulting from complex processing of the acquired images during the image stitching process.

Description

Technical Field

[0001] The present invention relates to the field of microscopic imaging technology, and particularly to a method and device for obtaining large-range microscopic imaging images. Background Art

[0002] In the detection fields such as medicine, biology, and industry, microscopic imaging is an important observation tool.

[0003] The field of view of microscopic imaging shrinks as the magnification increases. The field of view under a low-power microscope is large but the clarity is poor. High-resolution local sample information can be obtained under a high-power microscope, but it is difficult to obtain the overall sample information.

[0004] Therefore, obtaining high-resolution information of the overall sample using a microscopic imaging device is a problem in microscopic imaging.

[0005] In the prior art, one method is:

[0006] Generally, an image stitching algorithm is used to stitch multiple high-definition images to obtain a high-resolution image of the overall sample. The basis of image stitching is that there needs to be a certain overlapping part between the stitched images. Therefore, when taking images, it is necessary to continuously collect images and be able to reasonably overlap a part, so that the images can be stitched better.

[0007] In actual situations: If a person takes images through a microscopic imaging device, not only is the efficiency low, but also due to many uncontrollable factors of the person, the quality of the taken images is unstable and difficult to meet the requirements.

[0008] Another method is to use a traditional method, where a computer control platform is used to move a specified step length and a moving trajectory and then take pictures. When the platform moves, the accuracy of ordinary transmission gears is difficult to meet the precise and effective control requirements, resulting in relatively random errors in the size of the collected image area and the moving range. Directly, it may cause a large displacement of the overlapping area after the platform moves multiple times, and the collected images are difficult to meet the requirements of image stitching, as shown in the accompanying drawings of the specification Figure 1 As shown, theoretically the image moves in the X direction, but due to equipment errors, the moving direction is unstable. Therefore, the collected images result in a reduced and incomplete overlapping area, thus increasing the difficulty of image stitching.

[0009] In view of this, there is an urgent need to design and develop a more accurate, efficient device and method for obtaining high-standard large-range microscopic images. Summary of the Invention

[0010] The object of the present invention is to provide a method and device for obtaining a large-range microscopic imaging image. By adopting the methods of image stitching and control adjustment, on the one hand, stitching is performed using a stitching algorithm; on the other hand, the newly acquired image is evaluated, and the movement parameters of the control system are dynamically adjusted, so that the image acquisition is completed and a complete regional stitching imaging map is presented, solving the problems in the prior art. To achieve the above object, the present invention provides the following technical solutions:

[0011] A method for obtaining a large-range microscopic imaging image, comprising the following steps:

[0012] In the first step, initially collect a to-be-calibrated image and a base image adjacent to the previous to-be-calibrated image, and obtain a base image gray-scale distribution function I1(x, y) and a to-be-calibrated image gray-scale distribution function I2(x, y) respectively, where (x, y) represents the coordinates of any point;

[0013] In the second step, substitute the gray-scale distribution function I2(x, y) of the to-be-calibrated image and the gray-scale distribution function I1(x, y) of the base image into a convolutional neural network for training to obtain a high-level feature map F2 representing the gray-scale distribution function I2(x, y) of the to-be-calibrated image and a high-level feature map F1 representing the gray-scale distribution function I1(x, y) of the base image;

[0014] In the third step, construct a gray-scale template matching algorithm, and after finding the positional relationship of the overlapping region between the high-level feature map F1 and the high-level feature map F2 based on a spatial two-dimensional sliding template, calculate the displacement amounts of the overlapping region in the x and y coordinate directions respectively;

[0015] In the fourth step, calculate the actual displacement amount of the to-be-calibrated image according to the obtained displacement amounts of the overlapping region in the x and y coordinate directions to obtain a precise image;

[0016] In the fifth step, stitch the corrected precise images, form a complete regional image, and then perform classification storage to complete image acquisition and classification storage.

[0017] As an improvement to the method for obtaining a large-range microscopic imaging image of the present invention, in the first step, after respectively obtaining the base image gray-scale distribution function I1(x, y) and the to-be-calibrated image gray-scale distribution function I2(x, y), and before substituting them into the convolutional neural network for training,

[0018] it is also necessary to perform filtering processing on them based on Gaussian filtering to reduce the noise of the acquired images.

[0019] As an improvement to the method for obtaining a large-range microscopic imaging image of the present invention, in the second step, the convolutional neural network adopts an Alextnet neural network or a VGG16 neural network.

[0020] As an improvement to a large-scale microscopic imaging image acquisition method of the present invention, in the third step, after obtaining the high-level feature map F2 and the high-level feature map F1 based on the convolutional neural network, when calculating the displacement of the overlapping area in the coordinate x and y directions:

[0021] S3-1, first, the absolute square difference MAD algorithm is used to construct the grayscale template matching algorithm:

[0022] S3-11, respectively obtaining the image module size T of the image to be calibrated in the displacement area and the image module size s of the reference image in the displacement area;

[0023] S3-12, calculating the average difference D of the image to be calibrated relative to the reference image in the displacement area:

[0024] in,

[0025] 1≤i≤m-M+1, 1≤j≤n-N+1

[0026] Where M and N are the length and width of the module image of the area to be registered, respectively, and their values ​​are related to the displacement distance plus the displacement step in each direction; when D(i, j) is the smallest, the current (i, j) is the actual displacement value;

[0027] S3-2, secondly, calculate the position of the maximum overlapping area of ​​the high-level feature map F1 and the high-level feature map F2 in the grayscale distribution function I2(x, y) of the calibration image, and obtain the positional relationship of the overlapping area of ​​the high-level feature map F1 and the high-level feature map F2:

[0028] F1(x,y)=F2(x+Δ' x ,y+Δ' y )

[0029] S3-3, finally, obtaining the displacement Δ' of the overlapping area in the x-direction x And the displacement Δ' of the overlapping area in the y direction y .

[0030] As an improvement to the large-scale microscopic imaging image acquisition method of the present invention, in the fourth step, before calculating the actual displacement of the image to be calibrated, it is also necessary to evaluate the acquisition effect of the displacement of the overlapped area in the coordinate x and y directions to dynamically correct the displacement, wherein:

[0031] The steps to establish a collection effect evaluation include:

[0032] S4-1. The minimum step size of the custom overlapping region in the x and y directions is Z, and the impact on the next image acquisition during step size adjustment is determined. If adjustment is required, the adjustment amplitude is obtained, and step S4-2 is executed; otherwise, according to step S4-3, the actual displacement of the image to be calibrated is directly calculated.

[0033] S4-2. Calculate the overlapping region f(I1(x,y), I2(x,y)) of the base image and the image to be calibrated:

[0034] S4-21. First, set the threshold value δ as the threshold of the minimum spliceable overlapping region between the base image and the image to be calibrated.

[0035] S4-22. Determine the size relationship between the overlapping region S of the base image and the image to be calibrated and the threshold value δ. If it is greater than, return to the fifth step to splice the accurate image; otherwise, calculate the adjustment parameters:

[0036] In the X direction: n1 = Δ x / z, n2 = Δ x / z + 1;

[0037] In the Y direction: m1 = Δ x / z, m2 = Δ x / z + 1;

[0038] Calculate the overlapping areas of the reference image and the corrected image to be calibrated respectively:

[0039] S1 = f(I1(x,y), I2(x + n1*z, y + m1*z))

[0040] S2 = f(I1(x,y), I2(x + n2*z, y + m1*z))

[0041] S3 = f(I1(x,y), I2(x + n1*z, y + m2*z))

[0042] S4 = f(I1(x,y), I2(x + n2*z, y + m2*z))

[0043] Among them, compare the sizes of these four values, and select the largest one as the displacement correction direction parameter for the next step;

[0044] S4-3. Calculate the actual displacement of the image to be calibrated:

[0045] Δx = ε·Δ’x and Δy = ε·Δ’y

[0046] In the formula, ε represents the proportional relationship between the sizes of the high-level feature map F1, high-level feature map F2 and the size of the image to be calibrated.

[0047] As an improvement to the method for obtaining a large-range microscopic imaging image of the present invention, in the fifth step, the specific method for stitching the corrected accurate image is as follows:

[0048] S5-1. Based on step S3-3, after obtaining the displacement amounts Δ x and Δ y in the x and y directions of the overlapping region, perform a displacement transformation on the gray-scale distribution function I2(x, y) of the image to be calibrated: I‘2(x, y) = I2(x + Δ x , y + Δ y );

[0049] S5-2. Extract feature points from the image to be calibrated;

[0050] S5-3. After matching the feature points and performing image registration, copy the current image to be calibrated to a specific position in another image, and preprocess the overlapping boundary between the current image to be calibrated and the other image;

[0051] S5-4. Classify and store the images according to the stitching processing results of the current image to be calibrated.

[0052] As the second aspect of the present invention, a device for obtaining a large-range microscopic imaging image is proposed, including:

[0053] An image acquisition module for acquiring an original image, where the original image includes an initially acquired image to be calibrated and a base image adjacent to the previous image to be calibrated;

[0054] An image processing module for preprocessing the original image to perform displacement correction on the image to be calibrated and obtain an accurate image;

[0055] A storage module for classifying and storing the accurate image.

[0056] As an improvement to the device for obtaining a large-range microscopic imaging image of the present invention, the image processing module includes a training unit, a matching unit, a calculation unit, and a stitching unit, where

[0057] The training unit is used to input the acquired original image into a convolutional neural network for training to obtain a plurality of high-level feature maps representing the original image;

[0058] The matching unit is used to construct a gray-scale template matching algorithm and find the positional relationship of the overlapping region between two high-level feature maps based on a two-dimensional spatial sliding template to obtain the displacement amounts in the x and y directions of the overlapping region;

[0059] The calculation unit is used to calculate the actual displacement amount of the image to be calibrated according to the displacement amounts in the x and y directions of the acquired overlapping region to obtain an accurate image;

[0060] A splicing unit for splicing the accurate images to form a complete regional image

[0061] As an improvement to the apparatus for acquiring a large - range microscopic imaging image of the present invention, the training unit is trained based on the Alextnet neural network or the VGG16 neural network.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] To solve the problem of efficiently acquiring accurate spliced images during large - range microscopic imaging, on the one hand, the present invention dynamically adjusts the displacement amount of the stage of the imaging control system, so as to avoid inaccuracies of the target image caused by image distortion and the like resulting from complex processing of the acquired images during the image splicing process while ensuring real - time and efficient acquisition of the target image. On the other hand, dynamic step correction is performed during the acquisition process to ensure excellent acquisition effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Schematic diagram showing the step deviation during image acquisition due to equipment error in the prior art in an embodiment of the present invention;

[0065] Figure 2 Block diagram of the apparatus for acquiring a large - range microscopic imaging image proposed in an embodiment of the present invention;

[0066] Figure 3 Schematic flow diagram of the method for acquiring a large - range microscopic imaging image proposed in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. 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 making creative efforts shall fall within the protection scope of the present invention.

[0068] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention.

[0069] The present invention will be further described in detail below in conjunction with the accompanying drawings, but it is not intended to limit the present invention.

[0070] As Figures 1-3 , as an embodiment of the present invention, a method for obtaining a large-range microscopic imaging image is proposed, including the following steps:

[0071] In the first step, an initial acquisition of a calibration image to be calibrated and a base image adjacent to the previous calibration image to be calibrated is performed to obtain a base image gray-scale distribution function I1(x, y) and a calibration image gray-scale distribution function I2(x, y) respectively, where (x, y) represents the coordinates of any point. It should be noted that when performing the initial acquisition, the detection item containing the item to be imaged needs to be placed at the specified position on the stage. Then, the PC is used to control the microscopic imaging device to enter the image acquisition mode. Secondly, the stage controls the displacement amount and displacement direction according to the step size set by the control system of the PC, and the camera is used to perform image acquisition at the specified area position:

[0072] Based on the above technical concept, after obtaining the base image gray-scale distribution function I1(x, y) and the calibration image gray-scale distribution function I2(x, y) respectively, it is also necessary to perform filtering processing on them based on Gaussian filtering to reduce the noise of the acquired images.

[0073] It should be noted that due to the control error of the control system of the PC, there is a certain difference between the theoretical error and the actual error of the acquired images. Therefore, in order to obtain accurate displacement parameters and acquire a precise overlapping area, it is necessary to use the method of image feature matching to obtain the magnitude of the actual displacement. At the same time, in order to avoid the influence of external factors on the images, the present invention proposes to use a convolutional neural network to extract high-level features and then perform matching to obtain the actual displacement of the images, that is, using the trained high-level features to represent the underlying images, and then registering through the position correspondence relationship between the high-level features and the underlying features to avoid differences in the images caused by interference due to changes in the acquisition conditions when acquiring the underlying features.

[0074] Based on this, the present invention proposes:

[0075] Second step, substitute the gray-scale distribution function I2(x,y) of the image to be calibrated and the gray-scale distribution function I1(x,y) of the base image into the convolutional neural network for training, to obtain a high-level feature map F2 representing the gray-scale distribution function I2(x,y) of the image to be calibrated and a high-level feature map F1 representing the gray-scale distribution function I1(x,y) of the base image. It should be noted that the convolutional neural network proposed in the present invention adopts an Alextnet neural network or a VGG16 neural network. At the same time, in specific implementation, only grayscale images are used as input images. Therefore, the dimension is 1, and the first layer input is the original image (the image to be calibrated and the base image collected initially). When the subsequent fully connected layers are removed, the feature maps F1 (base image) and F2 (image to be calibrated) can be extracted.

[0076] It can be understood that due to the invariance of the image position: the target position and the high-order feature position of the original image remain unchanged. Therefore, the obtained high-order features are less affected by external factors, and thus, better registration can be performed.

[0077] After obtaining the high-level feature map F2 representing the gray-scale distribution function I2(x,y) of the image to be calibrated and the high-level feature map F1 representing the gray-scale distribution function I1(x,y) of the base image through the convolutional neural network, it is necessary to:

[0078] Third step, construct a gray-scale template matching algorithm, and based on a two-dimensional spatial sliding template, find the positional relationship of the overlapping region of the high-level feature map F1 and the high-level feature map F2, and then calculate the displacement amounts of the overlapping region in the x and y coordinate directions respectively.

[0079] Based on the technical concept of the third step, it should be noted that gray-scale template matching is to find a sub-image similar to the template image in another image according to a known template image, and the matching algorithm based on gray scale is also called a correlation matching algorithm, which is matched with a two-dimensional spatial sliding template. Different matching algorithms are mainly reflected in the selection of the correlation criterion. Therefore, in order to achieve high matching accuracy and wide range of image matching, the present invention adopts the mean absolute difference (MAD) algorithm to calculate the displacement amounts of the overlapping region of the high-level feature map F1 and the high-level feature map F2 in the x and y coordinate directions. The specific implementation method is as follows:

[0080] S3-11, respectively obtain the image module size T in the displacement region of the image to be calibrated; the image module size s in the displacement region of the reference image;

[0081] S3-12, calculate the average difference D of the image in the displacement region of the image to be calibrated relative to the reference image:

[0082]

[0083] Wherein,

[0084] 1 ≤ i ≤ m - M + 1, 1 ≤ j ≤ n - N + 1

[0085] Wherein, M and N are respectively the length and width of the image of the region to be registered, and their values are related to the displacement distance plus the displacement step size in each direction; when D(i, j) is the smallest, the current (i, j) is the actual displacement value;

[0086] S3-2. Secondly, calculate the position of the maximum overlapping region between the high-level feature map F1 and the high-level feature map F2 in the gray distribution function I2(x, y) of the calibrated image, and obtain the positional relationship of the overlapping region between the high-level feature map F1 and the high-level feature map F2:

[0087] F1(x, y) = F2(x + Δ’ x , y + Δ’ y ) (2)

[0088] S3-3. Finally, obtain the displacement amount Δ’ of the overlapping region in the x direction of the coordinate x and the displacement amount Δ’ of the overlapping region in the y direction of the coordinate y .

[0089] Based on the above technical concept, after obtaining the displacement amount Δ’ of the overlapping region in the x direction of the coordinate x and the displacement amount Δ’ of the overlapping region in the y direction of the coordinate y , it is necessary to evaluate the effect of the collected image, and then dynamically adjust the moving step size and moving speed of the stage. The specific implementation method for establishing the collection effect evaluation is as follows:

[0090] S4-1. Customize the minimum step size Z of the overlapping region in the x and y directions of the coordinate, and determine the impact on the next image acquisition when adjusting the step size. If adjustment is required, obtain the adjustment amplitude size for adjustment and execute step S4-2; otherwise, directly calculate the actual displacement amount of the image to be calibrated according to step S4-3;

[0091] S4-2. Calculate the overlapping region f(I1(x, y), I2(x, y)) between the base image and the image to be calibrated:

[0092] S4-21. First, set the threshold value δ as the threshold value of the minimum spliceable overlapping region between the base image and the image to be calibrated. It should be noted that the threshold value δ is an empirical value, which is strongly related to the complexity of the currently processed image and the moving step size. Generally, it is set to 0.2 during initialization and is adjusted through experiments later;

[0093] S4-22. Judge the size relationship between the overlapping region S between the base image and the image to be calibrated and the threshold value δ. If it is greater than, return to the fifth step to splice the accurate image. Otherwise, adjust the movement control, that is, calculate the parameters that need to be adjusted:

[0094] In the X direction: n1 = Δ x / z (rounded down), n2 = Δ x / z + 1 (rounded up);

[0095] In the Y direction: m1 = Δ x / z (rounded down, displacement value), m2 = Δ x / z + 1; where n1 and n2 are the upper and lower adjustment movement values in the x direction respectively; m1 and m2 are the upper and lower adjustment movement values in the y direction respectively;

[0096] Calculate the overlapping area between the reference image and the calibrated image after correction respectively:

[0097] S1 = f(I1(x, y), I2(x + n1*z, y + m1*z))

[0098] S2 = f(I1(x, y), I2(x + n2*z, y + m1*z))

[0099] S3 = f(I1(x, y), I2(x + n1*z, y + m2*z))

[0100] S4 = f(I1(x, y), I2(x + n2*z, y + m2*z))

[0101] Among them, compare the magnitudes of these four values, and select the largest one as the displacement correction direction parameter for the next step. For example: assuming S3 is the largest, the displacement correction parameter is a step size of n1 in the X direction and a step size of m2 in the Y direction.

[0102] Based on the above technical concept, it can be understood that after evaluating the effect of the collected image and dynamically adjusting the movement step size and movement speed of the stage, the acquired image ensures its acquisition effect.

[0103] Fourth step, calculate the actual displacement amount of the image to be calibrated according to the displacement amounts in the x and y directions of the obtained overlapping area, and obtain the precise image:

[0104] S4 - 3, calculate the actual displacement amount of the image to be calibrated:

[0105] Δx = ε·Δ’x and Δy = ε·Δ’y

[0106] Where ε represents the proportional relationship between the sizes of the high-level feature map F1, high-level feature map F2 and the size of the image to be calibrated.

[0107] It should be noted that during the image acquisition process, there are two parallel modes in total. That is, first, the movement parameters of the control system are dynamically adjusted according to the evaluation of the newly acquired image, and the other is to splice the corrected image using the stitching algorithm. That is,

[0108] Step 5: Splice the corrected accurate images to form a complete regional image. It should be noted that the specific method for splicing the corrected accurate images is as follows:

[0109] S5-1: Based on step S3-3, after obtaining the displacement amounts Δ x and Δ y in the x and y directions of the overlapping region, perform a displacement transformation on the gray distribution function I2(x, y) of the image to be calibrated: I‘2(x, y) = I2(x + Δ x , y + Δ y ); It can be understood that I‘2(x, y) is the actual deviation value of the image to be calibrated. After obtaining the actual displacement deviation, directly use the deviation value for calculation. It can be understood that the calculation steps include:

[0110] S5-2: Extract feature points from the image to be calibrated;

[0111] S5-3: After matching the feature points and performing image registration, copy the current image to be calibrated to a specific position in another image, and perform special preprocessing on the overlapping boundary between the current image to be calibrated and the other image. It should be noted that the method for performing special preprocessing on the overlapping boundary is: increment or decrement the overlapping region;

[0112] S5-4: Classify and store the images according to the stitching processing results of the current image to be calibrated.

[0113] As the second aspect of the present invention, a large-range microscopic imaging image acquisition device is proposed, including:

[0114] An image acquisition module for acquiring an original image, where the original image includes an initially acquired image to be calibrated and a base image adjacent to the previous image to be calibrated;

[0115] An image processing module for preprocessing the original image to perform displacement correction on the image to be calibrated and obtain an accurate image. It should be noted that the image processing module includes a training unit, a matching unit, a calculation unit, and a stitching unit, where

[0116] The training unit is used to input the acquired original image into a convolutional neural network for training to obtain multiple high-level feature maps representing the original image;

[0117] A matching unit for constructing a grayscale template matching algorithm and finding the positional relationship of the overlapping area between two high-level feature maps based on a two-dimensional spatial sliding template, and obtaining the displacement amounts of the overlapping area in the x and y coordinate directions;

[0118] A calculation unit for calculating the actual displacement amount of the image to be calibrated according to the obtained displacement amounts of the overlapping area in the x and y coordinate directions to obtain a precise image;

[0119] A splicing unit for splicing the precise images to form a complete regional image

[0120] A storage module for classifying and storing the precise images.

[0121] The above has shown and described the basic principles, main features and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention, and any reference signs in the claims should not be regarded as limiting the claimed rights involved.

[0122] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0123] ​

Claims

1. A method for obtaining a large-range microscopic imaging image, characterized in that, It includes the following steps: In the first step, initially collect the image to be calibrated and the base image adjacent to the previous image to be calibrated, and obtain the base image gray distribution function I1(x, y) and the image to be calibrated gray distribution function I2(x, y) respectively, where (x, y) represents the coordinates of any point; In the second step, substitute the gray distribution function I2(x, y) of the image to be calibrated and the gray distribution function I1(x, y) of the base image into the convolutional neural network for training to obtain the high-level feature map F2 representing the gray distribution function I2(x, y) of the image to be calibrated and the high-level feature map F1 representing the gray distribution function I1(x, y) of the base image; In the third step, construct a gray template matching algorithm, and based on the spatial two-dimensional sliding template, find the positional relationship of the overlapping region between the high-level feature map F1 and the high-level feature map F2, and then calculate the displacement amounts of the overlapping region in the x and y coordinate directions respectively; In the fourth step, calculate the actual displacement amount of the image to be calibrated according to the displacement amounts of the overlapping region in the x and y coordinate directions to obtain the accurate image; In the fifth step, splice the corrected accurate images to form a complete regional image, and then perform classification storage to complete image acquisition and classification storage; In the third step, after obtaining the high-level feature map F2 and the high-level feature map F1 based on the convolutional neural network, when calculating the displacement amounts of the overlapping region in the x and y coordinate directions: S3-1, First, it is necessary to construct a gray template matching algorithm in the way of the mean absolute difference (MAD) algorithm: S3-11, respectively obtain the image module size T of the image to be calibrated in the displacement region; the image module size S of the base image in the displacement region; S3-12, calculate the average difference D of the image to be calibrated relative to the base image in the displacement region image: where 1 ≤ i ≤ m - M + 1, 1 ≤ j ≤ n - N + 1 In the formula, M and N are the length and width of the module image of the region to be registered respectively, and their values are related to the displacement distance plus the displacement step in each direction; when D(i, j) is the smallest, then the current (i, j) is the actual displacement value, and m and n are the length and width of the original image collected each time; S3-2, Secondly, calculate the position of the maximum overlapping region between the high-level feature map F1 and the high-level feature map F2 in the gray distribution function I2(x, y) of the calibration image to obtain the positional relationship of the overlapping region between the high-level feature map F1 and the high-level feature map F2; F1(x,y) = F2(x + Δ x , y + Δ y ) S3-3. Finally, obtain the displacement Δ of the overlapping region in the x coordinate direction x and the displacement Δ of the overlapping region in the y coordinate direction y ; In the fifth step, the specific method of splicing the corrected accurate images is: S5-1, based on step S3-3, after obtaining the displacement amounts Δ x and Δ y in the x and y directions of the overlapping region, perform a displacement transformation on the grayscale distribution function I2(x, y) of the image to be calibrated: I2(x, y) = I2(x + Δ x , y + Δ y ); S5-2, extract feature points from the image to be calibrated; S5-3, after matching the feature points and performing image registration, copy the current image to be calibrated to a specific position of another image, and preprocess the overlapping boundary between the current image to be calibrated and another image; S5-4, perform image classification storage according to the splicing processing result of the current image to be calibrated.

2. The method for obtaining a large-range microscopic imaging image according to claim 1, wherein In the first step, after respectively obtaining the base image gray distribution function I1(x, y) and the image to be calibrated gray distribution function I2(x, y), and before substituting them into the convolutional neural network for training, Perform filtering processing on it based on Gaussian filtering to reduce the noise of the acquired image.

3. A method for obtaining a wide-range microscopic imaging image according to claim 1, characterized in that, In the second step, the convolutional neural network adopts an Alextnet neural network or a VGG16 neural network.

4. A method for obtaining a wide-range microscopic imaging image according to claim 1, characterized in that, In the fourth step, before calculating the actual displacement of the image to be calibrated, the acquisition effect of the displacement in the x and y directions of the acquired overlapping region is evaluated to dynamically correct the displacement. Among them, The steps for establishing the acquisition effect evaluation include: S4-1, Customize the minimum step size of the overlapping region in the x and y directions as z, and determine the impact on the next image acquisition when adjusting the step size. If adjustment is required, obtain the adjustment amplitude size and continue to execute step S4-2; otherwise, directly calculate the actual displacement of the image to be calibrated according to step S4-3; S4-2, Calculate the overlapping region f(I1(x,y), I2(x,y)) of the base image and the image to be calibrated; S4-21, first set a threshold value as the threshold for the minimum splicable coincidence area between the base image and the image to be calibrated; S4-22, determine the size relationship between the overlapping region f(I1(x, y), I2(x, y)) of the base image and the image to be calibrated and the threshold value. If it is greater, return to the fifth step to splice the precise images. Otherwise, calculate the adjusted parameters: If it is greater, return to the fifth step to splice the precise images. Otherwise, calculate the adjusted parameters: In the X direction: n1 = Δ x / z, n2 = Δ x / z + 1; In the Y direction: m1 = Δ y / z, m2 = Δ y / z + 1; In the formula, n1 and n2 are the upper and lower adjustment movement values in the x direction respectively; m1 and m2 are the upper and lower adjustment movement values in the y direction respectively; Calculate the overlapping areas of the base image and the corrected image to be calibrated respectively: f(I1(x,y), I2(x,y))1 = f(I1(x,y), I2(x + n1*z, y + m1*z)) f(I1(x,y), I2(x,y))2 = f(I1(x,y), I2(x + n2*z, y + m1*z)) f(I1(x,y), I2(x,y))3 = f(I1(x,y), I2(x + n1*z, y + m2*z)) f(I1(x,y), I2(x,y))4 = f(I1(x,y), I2(x + n2*z, y + m2*z)) Among them, compare the sizes of these four values and select the largest one as the next displacement correction direction parameter; S4-3, Calculate the actual displacement of the image to be calibrated: Δ‘ x = ε·Δ x and Δ‘ y = ε·Δ y In the formula, ε represents the proportional relationship between the sizes of the high-level feature map F1, the high-level feature map F2 and the size of the image to be calibrated.

5. An apparatus for obtaining a wide-range microscopic imaging image, based on the obtaining method according to any one of claims 1-4, characterized in that, Include: An image acquisition module that acquires an original image, where the original image includes an initially acquired image to be calibrated and a base image adjacent to the previous image to be calibrated; An image processing module that preprocesses the original image to correct the displacement of the image to be calibrated and obtain a precise image; A storage module that classifies and stores the precise image.

6. The wide-range microscopic imaging image acquisition device according to claim 5, characterized in that, The image processing module includes a training unit, a matching unit, a calculation unit, and a splicing unit. Among them, The training unit is used to substitute the acquired original image into a convolutional neural network for training to obtain multiple high-level feature maps representing the original image; The matching unit is used to construct a gray template matching algorithm and find the positional relationship of the overlapping region between two high-level feature maps based on a two-dimensional spatial sliding template to obtain the displacement in the x and y directions of the overlapping region; The calculation unit is used to calculate the actual displacement of the image to be calibrated according to the displacement in the x and y directions of the acquired overlapping region to obtain a precise image; The splicing unit is used to splice the precise image to form a complete regional image.

7. The wide-range microscopic imaging image acquisition device according to claim 6, wherein The training unit is trained based on the Alexnet neural network or the VGG16 neural network.

Images