Fluorescence signal image enhancement method, equipment and program product
By performing channel enhancement processing on fluorescent signal images and inputting into deep learning models, the problem of poor detection accuracy under small samples is solved, the detection efficiency and accuracy are improved, and user operations are simplified.
Patent Information
- Application Number
- CN202411100493.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-12-03
AI Technical Summary
Under small sample conditions, when using deep learning models for image classification, the detection accuracy is poor. The existing technology requires users to invest a lot of energy and time in debugging and iterative training for feedback, and the feedback process is complicated and difficult to post to non-technical users.
By acquiring fluorescent signal sample images of different colors, dividing them into R, G, and B channels, performing single-channel signal enhancement processing, and combining them into enhanced images, inputting pre-trained deep learning models for object detection, and improving using EfficientNet and bidirectional feature pyramid networks Detection accuracy.
It improves the accuracy and efficiency of target detection, reduces the operation complexity of users, and enables more accurate detection under small sample size without affecting non-target areas.
Smart Images

Figure CN119067865A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of fluorescence signal image processing, and in particular to a fluorescence signal image enhancement method, device and program product. Background Art
[0002] At present, in the fields of scientific research, medicine, and industry, when using images for classification, such as cell classification, there is a problem of small sample size. Under small sample conditions, when using deep learning models to identify targets in sample images, there is a problem of poor detection accuracy. To solve this problem, the existing technology adopts a small sample learning method, which adds manual supervision to the process to provide feedback, superimposed training, feedback, and superimposed training on the results in real time until a more satisfactory result is obtained. The small sample learning method of the existing technology requires users to invest energy and time to perform debugging, that is, feedback → iterative training process; and the feedback and superimposed training process cannot be delegated to users who do not understand the technology, resulting in inconvenience in use. Summary of the invention
[0003] The embodiments of the present invention provide a method, device and program product for image enhancement of fluorescence signals, which enhance only the target signal without affecting other areas, so that the characteristics of the target are more prominent in the detection model, thereby achieving better detection results under small sample conditions.
[0004] In order to achieve the above object, on the one hand, a method for image enhancement of a fluorescence signal is provided, comprising:
[0005] Using different selected dyes, fluorescent signal sample images of different colors captured in the same field of view are respectively obtained, wherein the fluorescent signal sample images of different colors include: a red signal sample image, a green signal sample image and / or a blue signal sample image, wherein the following processing is performed on each of the sample images:
[0006] Dividing the sample image into single-channel signal point images of R, G, and B channels according to the channel information;
[0007] According to the color of the fluorescence signal of the sample image, single-channel signal point images of corresponding colors in the R, G, and B channels are subjected to single-channel enhancement processing to obtain a single-channel signal point enhanced image of the single-channel signal point image of the corresponding color;
[0008] The single-channel signal point enhanced image is merged with the unenhanced single-channel signal point images of the other two channels to obtain a merged first enhanced image.
[0009] Preferably, in the image enhancement method, the single-channel enhancement process comprises:
[0010] Histogram equalization is performed on the single-channel signal point image to be enhanced to obtain a corresponding enhanced single-channel signal point enhanced image.
[0011] Preferably, the image enhancement method further comprises using the first enhanced image corresponding to each sample image to perform target detection of the fluorescence signal, wherein the target detection of the fluorescence signal comprises:
[0012] Performing binarization processing on the sample image to obtain a binarized sample image;
[0013] Superimposing the binarized sample image with the first enhanced image corresponding to the sample image to obtain a second enhanced image corresponding to the sample image;
[0014] The second enhanced image is input into a pre-trained deep learning target detection model to identify the channel target of the color corresponding to the sample image.
[0015] Preferably, in the image enhancement method, the step of binarizing the sample image comprises:
[0016] Calculating the edge strength in the horizontal direction and the edge strength in the vertical direction of the sample image;
[0017] Calculating the gradient amplitude of the sample image according to the edge strength in the horizontal direction and the edge strength in the vertical direction;
[0018] Performing a linear transformation on the gradient amplitude to adjust the brightness and contrast of the sample image by scaling and translating to obtain a sample image after the linear transformation;
[0019] Binarization is performed on the linearly transformed sample image to obtain the binarized sample image.
[0020] Preferably, in the image enhancement method, the deep learning target detection model comprises:
[0021] An EfficientNet backbone network, configured to receive the second enhanced image and output feature maps with different semantic levels;
[0022] A bidirectional feature pyramid network, used to fuse the feature maps output by the EfficientNet backbone network and output a multi-scale feature map with enhanced semantic information;
[0023] A classification prediction network, used for performing classification prediction of a target according to a multi-scale feature map output by the bidirectional feature pyramid network;
[0024] The detection box prediction network is used to predict the position and size of the target detection box according to the multi-scale feature map output by the bidirectional feature pyramid network.
[0025] Preferably, in the image enhancement method, the deep learning target detection model verifies the prediction result by the following steps during training:
[0026] Cutting a field of view image in a predetermined manner to obtain a plurality of sub-field of view images;
[0027] One of the plurality of sub-view images is used as training data to verify other sub-view images of the plurality of sub-view images.
[0028] Preferably, in the image enhancement method, the step of cropping a field of view image in a predetermined manner comprises:
[0029] The one field of view image is cut according to the cross lines to obtain four equal parts, each of which corresponds to a sub-field of view image.
[0030] Preferably, in the image enhancement method, the red signal sample image, the green signal sample image and the blue signal sample image are obtained by photographing samples stained with TRITC, FITC and DAPI, respectively.
[0031] On the other hand, an electronic device is provided, including a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement the steps of any of the image enhancement methods described above.
[0032] On the other hand, a computer program product is provided, comprising a computer program, characterized in that when the computer program is executed by a processor, the steps of any of the image enhancement methods described above are implemented.
[0033] The above technical solution has the following technical effects:
[0034] The technical solution of one embodiment of the present invention obtains sample images of fluorescence signals of different colors through different dyes, divides the sample images into single-channel signal point images of R, G, and B channels, and performs single-channel channel enhancement on the single-channel signal point images of corresponding colors in the three channels, and then merges the enhanced single-channel signal point image with the other two unenhanced single-channel signal point images into an RGB image to obtain a first enhanced image of the sample image, in which only the target signal points are enhanced. This image enhancement method can prevent non-targets from being enhanced, thereby improving the detection accuracy of targets;
[0035] The technical solution of the embodiment of the present invention improves the strength of each channel, making each channel more obvious and easier to be found, thereby achieving more accurate detection with a small amount of sample;
[0036] Furthermore, because it is difficult to obtain data of fluorescent specimens, obtaining a field of view is very valuable in many cases. In order to perform effective target detection with a small sample size, in the technical solution of one embodiment of the present invention, in the verification of the deep learning target detection model, a field of view image is divided into multiple sub-fields of view or small fields of view, and one of the small fields of view is used as training data to verify other field of view images, thereby improving the accuracy of signal point positioning. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a schematic diagram of the process of the image enhancement method of the fluorescence signal according to one embodiment of the present invention;
[0038] Figure 2 This is a schematic diagram of the overall process of a method for enhancing an image of a fluorescent signal according to an embodiment of the present invention;
[0039] Figure 3 is a green signal sample image used in an embodiment of the present invention;
[0040] Figure 4 is a red signal sample image used in an embodiment of the present invention;
[0041] Figure 5 is a first enhanced image corresponding to the green signal sample image;
[0042] Figure 6 is the first enhanced image corresponding to the red signal sample image;
[0043] Figure 7 : is a comparison between the original image of the red signal sample image and the second enhanced image output after target enhancement; wherein the left side is the original image and the right side is the second enhanced image;
[0044] Figure 8 It is a comparison between the original image of the green signal sample image and the second enhanced image output after target enhancement; wherein, the left side is the original image and the right side is the second enhanced image. DETAILED DESCRIPTION
[0045] To further illustrate the various embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, a person of ordinary skill in the art should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.
[0046] The present invention will now be further described with reference to the accompanying drawings and specific implementation methods.
[0047] Embodiment 1:
[0048] Figure 1 FIG. 1 is a flow chart of a method for enhancing an image of a fluorescent signal according to an embodiment of the present invention. Figure 1 The method for enhancing the image of the fluorescent signal of this embodiment comprises the following steps:
[0049] Using different selected dyes, fluorescent signal sample images of different colors captured in the same field of view are respectively obtained, and the fluorescent signal sample images of different colors include: a red signal sample image, a green signal sample image and / or a blue signal sample image, wherein each of the sample images is processed as follows:
[0050] The sample image is divided into single-channel signal point images of R, G, and B channels according to the channel information;
[0051] According to the color of the fluorescence signal of the sample image, single-channel signal point images of corresponding colors in the R, G, and B channels are subjected to single-channel enhancement processing to obtain single-channel signal point enhanced images of the single-channel signal point images of corresponding colors;
[0052] The single-channel signal point enhanced image is merged with the unenhanced single-channel signal point images of the other two channels to obtain a merged first enhanced image.
[0053] In a specific implementation, the red signal sample image, the green signal sample image and the blue signal sample image are obtained by photographing the samples stained with the dye TRITC, the dye FITC and the dye DAPI respectively. By using DAPI, TRITC and FITC to correspond to the characteristics of the blue, red and green channels, and by increasing the intensity of each channel, the characteristics are made more obvious and easier to be found, thereby achieving a method for detection with a small amount of sample.
[0054] Among them, DAPI (4',6-diamidino-2-phenylindole) is a DNA dye that can bind to the adenine-cytosine pair of DNA and emit blue fluorescence. TRITC (Tetramethylrhodamine Isothiocyanate) is a fluorescent dye that is usually bound to antibodies or other biological molecules for use in immunofluorescence staining experiments and emits red to orange fluorescence. FITC (Fluorescein Isothiocyanate) is a commonly used fluorescent dye that is also often bound to antibodies or other biological molecules for use in immunofluorescence staining experiments and emits green fluorescence.
[0055] Furthermore, the method further includes using the first enhanced images corresponding to the sample images to perform target detection of the fluorescence signal, wherein the target detection of the fluorescence signal includes:
[0056] Performing binarization processing on the sample image to obtain a binarized sample image;
[0057] Superimposing the binarized sample image with the first enhanced image corresponding to the sample image to obtain a second enhanced image corresponding to the sample image;
[0058] The second enhanced image is input into a pre-trained deep learning target detection model to identify the channel target with the color corresponding to the sample image.
[0059] Embodiment 2:
[0060] Figure 2 Schematic diagram of the overall process of a method for enhancing an image of a fluorescent signal according to an embodiment of the present invention. Figure 2 , the processing flow of the green signal sample image and the red signal sample image and the corresponding image effects are shown; wherein the left side is the corresponding steps and the right side is the corresponding image.
[0061] like Figure 2 The overall process includes: acquiring an image; performing single-channel image signal enhancement on the single-channel signal of the corresponding color; obtaining the result of signal enhancement; mapping signal points; merging; calling a deep learning detection model for target detection; and locating the target in the image.
[0062] The specific steps are described below.
[0063] Data acquisition: Use existing corresponding instruments to collect fluorescence signals. After obtaining image data, the image data can be annotated.
[0064] Image processing / enhancement algorithm: Enhance signal points; specifically includes:
[0065] Get the original image, that is, the original sample image In, such as Figure 3 The green signal in the sample image is the green fluorescence signal. Figure 4 The red signal sample image in ;
[0066] Use the following formula to calculate the edge value of the sample image In:
[0067] d x =[f(i-1,j-1)+2f(i-1,j)+f(i-1,j+1)]-[f(i+1,j)+f(i+1,j+1)]
[0068] d y=[f(i-1,j+1)+2f(i,j+1)+f(i+1,j+1)]-[f(i-1,j-1)+2f(i,j-1)+f(i+1,j-1)]
[0069] in:
[0070] d x is the edge strength in the horizontal direction;
[0071] d y is the edge strength in the vertical direction;
[0072] The function f() indicates that the operation in the brackets is the Sobel operator calculation.
[0073] (i, j) is the coordinate information of the pixel;
[0074] Use the following formula to calculate the gradient magnitude Magnirude of the edge value:
[0075]
[0076] Among them, the gradient amplitude makes the edge of the image more prominent.
[0077] Perform a linear transformation on the obtained gradient magnitude:
[0078] T=a×Magnitude+b
[0079] in:
[0080] T is the pixel value of the output image, ensuring that the result is between 0 and 255.
[0081] Magnitude is the gradient magnitude of the input image;
[0082] a is the scaling factor, which is used to control the brightness and contrast of the input image;
[0083] b is the translation amount, which is used to adjust the brightness of the input image.
[0084] In a specific implementation, the appropriate values of a and b are determined by observing the histogram of the image. If the histogram of the image is too concentrated in a certain range, the range of the histogram can be expanded by adjusting the values of a and b to make the image contrast better.
[0085] Edge features can be enhanced through linear transformation. Specifically, linear transformation can adjust the contrast and brightness of edge images, thereby enhancing edge features in the image and making it clearer and more obvious. In addition, linear transformation can reduce the impact of noise in the image on edge detection. By adjusting the contrast and brightness of the image, the edge can be made more prominent, making it easier to distinguish between the edge and noise. Furthermore, linear transformation can help improve the accuracy and stability of the edge detection algorithm. By adjusting the contrast and brightness of the image, the edge can be made clearer and more continuous, thereby improving the performance of the edge detection algorithm.
[0086] Then, the image obtained after the linear transformation is binarized; in a specific implementation, wherein:
[0087]
[0088] in,
[0089] T(x,y) is the pixel value of the input image; Threshold is the preset pixel value threshold, which is predetermined based on demand or experience; Binary(x,y) is the pixel value of the output image.
[0090] Perform channel enhancement on a single-channel image; take the green signal sample image as an example to illustrate, specifically including:
[0091] A) Split the original sample image In into three channels of single-channel signal point images according to the communication information, including: R channel single-channel signal point image In r , single channel signal point image of G channel In g And the single channel signal point image of the B channel In b ;
[0092] B) Perform histogram equalization on the signal point image of the green channel, perform single-channel signal enhancement on the signal point image of the green channel, and obtain an enhanced green channel signal point image: genhanced ;
[0093] C) The original single-channel image obtained in step A) is r and b and enhanced In genhanced Merge and reassemble into an RGB image with enhanced signal points of the green channel, i.e., a first enhanced image;
[0094] Using a similar method, the red signal sample image and the blue signal sample image can be processed to enhance the single-channel signal point intensity of the corresponding color respectively, to obtain the corresponding enhanced single-channel signal point image, and then re-merge it with the other two unenhanced single-channel signal point images into an RGB image, that is, to obtain the first enhanced image corresponding to the sample image. Figure 5 is a first enhanced image corresponding to the green signal sample image, in which the green channel is enhanced; Figure 6 It is the first enhanced image corresponding to the red signal sample image, in which the red channel has been enhanced.
[0095] Next, image mapping is performed to extract features and reduce noise; specifically:
[0096] The binarized sample image obtained by binarizing the original sample image above and the first enhanced image corresponding to the sample image are superimposed, so that the non-target is not enhanced. After superposition, the final image for inputting the deep learning target detection model for inference is obtained: Out predict , i.e. the second enhanced image.
[0097] Figure 7 The figure shows the comparison between the original image of the red signal sample image and the second enhanced image output after target enhancement; wherein the left side is the original image and the right side is the second enhanced image. Figure 8 The figure shows the comparison between the original image of the green signal sample image and the second enhanced image output after target enhancement; the left side is the original image and the right side is the second enhanced image. Figure 7 and Figure 8 It can be seen that the method of this embodiment of the present invention makes the signal point of the target channel more obvious and easier to identify by only enhancing the strength of the signal point of the target channel.
[0098] The following is an explanation of the deep learning target detection model used in an embodiment of the present invention. In a specific implementation, the detection model EfficientDet is used. The detection model EfficientDet includes the following four parts:
[0099] Part 1: EfficientNet Backbone:
[0100] Purpose: The EfficientNet backbone network is used to extract features from an input image such as the second enhanced image;
[0101] Input: The input of the EfficientNet backbone network is the image to be inferred, which is the second enhanced image in the embodiment of the present invention, and is usually preprocessed to meet the input requirements of the network;
[0102] Output: The output of the EfficientNet backbone network is a series of feature maps with different semantic levels;
[0103] Function: The backbone network converts the input image into high-level semantic feature representations through multi-layer convolution and pooling operations. These feature representations are very useful for target detection tasks.
[0104] Part 2: BiFPN Layer (Bidirectional Feature Pyramid Network):
[0105] Purpose: BiFPN is used to fuse multi-scale features from the EfficientNet backbone network to enhance target detection performance;
[0106] Input: The input of BiFPN is multiple feature maps output by the EfficientNet backbone network;
[0107] Output: The output of BiFPN is a multi-scale feature map with enhanced semantic information;
[0108] Function: BiFPN effectively enhances the semantic expression ability of features and improves the accuracy and robustness of target detection through bidirectional connection and feature fusion operations.
[0109] Part 3: Class Prediction Network:
[0110] Purpose: Class Prediction Network is used to predict the category of each target detection box;
[0111] Input: The input of Class Prediction Network is the feature map output by BiFPN;
[0112] Output: The output of the Class Prediction Network is the predicted probability of the category to which each detection box belongs;
[0113] Function: Class Prediction Network classifies the detection box according to the context information in the feature map to determine which category the detected object belongs to.
[0114] Part 4: Box Prediction Network:
[0115] Purpose: Box Prediction Network is used to predict the position and size of each target detection box;
[0116] Input: The input of Box Prediction Network is the feature map output by BiFPN.
[0117] Output: The output of the Box Prediction Network is the predicted result of the position and size of each detection box.
[0118] Function: Box Prediction Network regresses the position and size of the detection box based on the contextual information in the feature map, thereby accurately locating the detected object.
[0119] In a specific implementation, EfficientDet is deployed as follows:
[0120] Considering the desktop compiler VS (Visual Studio) C++ language, the deep learning model is deployed using the dnn module in OPENCV. The model can eventually be applied to Anaysis Pro and Analysis BIO software, and can be used in conjunction with existing corresponding upright and inverted microscopes.
[0121] In addition to the EfficientDet model mentioned above, other implementations can also use multimodal large models such as Google Gemini to achieve target detection. The cost is the computing power requirement. When there is a network, it can be used online in the cloud. When there is no network, it needs to be combined with edge devices for calculation.
[0122] In the specific implementation of the embodiment of the present invention, it is necessary to perform effective target detection with a small sample size, because it is difficult to obtain data of fluorescent specimens, so in many cases, obtaining a field of view is very valuable; therefore, in the embodiment of the present invention, in the verification of the target detection model, a field of view image is cropped in a predetermined manner to obtain multiple sub-field of view images; one sub-field of view image among the multiple sub-field of view images is used as training data to verify the other sub-field of view images among the multiple sub-field of view images. A specific implementation method is: crop a field of view image according to the crosshairs to obtain four equal parts, and use one of the small fields of view as training data to verify the other field of view images.
[0123] The image enhancement method of the embodiment of the present invention only enhances the target signal and does not affect other areas. Through this method, the data volume requirement is greatly reduced when using deep learning, so that the model can obtain a better detection result with very small amounts of data. Moreover, a detection model is adopted instead of a small sample learning method, which greatly reduces the energy and time invested by users.
[0124] Embodiment three:
[0125] The present invention also provides an electronic device, which includes a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement the steps of any of the image enhancement methods described above.
[0126] Further, as an executable solution, the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, and various interfaces and lines are used to connect the various parts of the entire computer unit.
[0127] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer unit by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required for a function; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0128] Embodiment 4:
[0129] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.
[0130] If the module / unit integrated in the computer unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory) and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.
[0131] Embodiment five:
[0132] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the method described above are implemented.
[0133] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.
Claims
1. A method for enhancing an image of a fluorescence signal, characterized in that: include: Using different selected dyes, fluorescent signal sample images of different colors captured in the same field of view are respectively obtained, wherein the fluorescent signal sample images of different colors include: a red signal sample image, a green signal sample image and / or a blue signal sample image, wherein the following processing is performed on each of the sample images: Dividing the sample image into single-channel signal point images of R, G, and B channels according to the channel information; According to the color of the fluorescence signal of the sample image, single-channel signal point images of corresponding colors in the R, G, and B channels are subjected to single-channel enhancement processing to obtain a single-channel signal point enhanced image of the single-channel signal point image of the corresponding color; The single-channel signal point enhanced image is merged with the unenhanced single-channel signal point images of the other two channels to obtain a merged first enhanced image.
2. The image enhancement method according to claim 1, characterized in that: The single-channel enhancement process includes: Histogram equalization is performed on the single-channel signal point image to be enhanced to obtain a corresponding enhanced single-channel signal point enhanced image.
3. The image enhancement method according to claim 1, characterized in that: The method further includes using the first enhanced images corresponding to the sample images to perform target detection of the fluorescent signal, wherein the target detection of the fluorescent signal includes: Performing binarization processing on the sample image to obtain a binarized sample image; Superimposing the binarized sample image with the first enhanced image corresponding to the sample image to obtain a second enhanced image corresponding to the sample image; The second enhanced image is input into a pre-trained deep learning target detection model to identify the channel target of the color corresponding to the sample image.
4. The image enhancement method according to claim 3, characterized in that: The binarization process of the sample image includes: Calculating the edge strength in the horizontal direction and the edge strength in the vertical direction of the sample image; Calculating the gradient amplitude of the sample image according to the edge strength in the horizontal direction and the edge strength in the vertical direction; Performing a linear transformation on the gradient amplitude to adjust the brightness and contrast of the sample image by scaling and translating to obtain a sample image after the linear transformation; Binarization is performed on the linearly transformed sample image to obtain the binarized sample image.
5. The image enhancement method according to claim 3, characterized in that: The deep learning target detection model includes: An EfficientNet backbone network, configured to receive the second enhanced image and output feature maps with different semantic levels; A bidirectional feature pyramid network, used to fuse the feature maps output by the EfficientNet backbone network and output a multi-scale feature map with enhanced semantic information; A classification prediction network, used for performing classification prediction of a target according to a multi-scale feature map output by the bidirectional feature pyramid network; The detection box prediction network is used to predict the position and size of the target detection box according to the multi-scale feature map output by the bidirectional feature pyramid network.
6. The image enhancement method according to claim 3, characterized in that: The deep learning target detection model is trained by verifying the prediction results through the following steps: Cutting a field of view image in a predetermined manner to obtain a plurality of sub-field of view images; One of the plurality of sub-view images is used as training data to verify other sub-view images of the plurality of sub-view images.
7. The image enhancement method according to claim 6, characterized in that: The step of cropping a field of view image in a predetermined manner comprises: The one field of view image is cut according to the cross lines to obtain four equal parts, each of which corresponds to a sub-field of view image.
8. The image enhancement method according to claim 1, characterized in that: The red signal sample image, the green signal sample image and the blue signal sample image are obtained by photographing samples stained with TRITC, FITC and DAPI, respectively.
9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement the steps of the image enhancement method according to any one of claims 1 to 8.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the image enhancement method according to any one of claims 1 to 8 are implemented.
Citation Information
Patent Citations
Pathological image color quality comprehensive evaluation method based on color separation
CN104408717A
Image processing method and device for fluorescence reaction region of teeth
CN110874857A
OCTA image analysis method based on classification network
CN113421259A
