Microarray chip image identification method and system

Abstract

The invention provides a microarray chip image identification method and system, and the method comprises the steps: positioning a Block external rectangular frame of an array through a Block target detection model, and enabling the Block target detection model to be generated through the training of a predetermined target detection model through an image set which is formed by a plurality of arrays; performing semantic segmentation on the Block bounding rectangular frame area by using a Spot semantic segmentation model to generate a spot image block mask image, and adjusting the angle of the Block bounding rectangular frame based on the projection result of the spot image block mask image; and calculating by using the spot image block mask graph to obtain the center-of-mass coordinate and radius of each spot area in the adjusted Block circumscribed rectangular frame, and finally determining the circumscribed circle of each spot image block. According to the method, the position of the Block can be accurately recognized, the Spot can be accurately aligned, meanwhile, the analysis efficiency is greatly improved, and the analysis time is shortened.

Description

Technical Field

[0001] The present invention relates to the field of chip technology, and in particular to a method and system for recognizing microarray chip images. Background Art

[0002] A microarray chip uses methods such as photoconductive in situ synthesis or microspotting to systematically solidify a large number of biological macromolecules, such as nucleic acid fragments, peptide molecules, and even biological samples such as tissue sections and cells, on the surface of a support (such as a glass slide, nylon membrane, or other carrier), forming a dense two-dimensional molecular array. The array then reacts with labeled target molecules in the biological sample to be tested. Using specific instruments such as a laser confocal scanner or a charge-coupled device camera, the intensity of the reaction signal is rapidly, concurrently, and efficiently detected and analyzed, thereby determining the number of target molecules in the sample. Depending on the probes fixed on the chip, microarray chips include gene chips, protein chips, cell chips, tissue chips, and others.

[0003] Image recognition is a crucial step in gene chip data analysis, and its accuracy directly impacts subsequent bioinformatics interpretation and understanding of biological phenomena. However, due to the inherent complexity of gene chip images, existing image recognition technologies often face significant challenges when processing images with complex shapes, unclear boundaries, or severe background interference. For example, array arrangements have various configuration files. From an image perspective, array arrangement must first ensure the brightness of the calibration points. If the calibration points are not bright, automatic positioning will typically deviate. Traditional algorithms use projection to directly locate the block position, but this approach is often subject to various interferences, such as noise, fences, and contamination points, leading to inaccurate positioning. These issues not only reduce the accuracy of image recognition but also significantly reduce the efficiency of data processing, thereby impacting the quality and reliability of the entire gene expression analysis. Summary of the Invention

[0004] The present invention provides a method for identifying a microarray chip image, comprising:

[0005] Locating the array's block circumscribed rectangular frame using a block target detection model, where the block target detection model is generated by training a predetermined target detection model using image sets of multiple array arrangements, where each image set includes chip image data and corresponding target detection labels;

[0006] The Spot semantic segmentation model is used to perform semantic segmentation on the Block circumscribed rectangular frame area to generate a spot image block mask map, and the angle of the Block circumscribed rectangular frame is adjusted based on the projection result of the spot image block mask map; the Spot semantic segmentation model is generated by training a semantic segmentation network model using the cropped Spot image blocks in the Block circumscribed rectangular frame area and the corresponding semantic segmentation labels.

[0007] The spot image block mask is used to calculate the centroid coordinates and radius of each spot area within the adjusted Block circumscribed rectangular frame.

[0008] Optionally, using image sets arranged in multiple arrays to train a predetermined target detection model to generate the Block target detection model includes:

[0009] Adjust the brightness and contrast of the chip image in the predetermined array arrangement and generate JPG images of grayscale, red, green, and pseudo-color channels respectively;

[0010] Mark the target detection label on the JPG image.

[0011] Optionally, the predetermined target detection model is a target detection model based on YoloV5, the initial learning rate parameter of the target detection model based on YoloV5 is 0.01, and the loss function is the nn.BCEWithLogitsLoss function provided by Python.

[0012] Optionally, the angle of the Block's circumscribed rectangular box is adjusted based on the projection result of the spot image block mask map, including: projecting the spot image block mask map from a predetermined angle range and simultaneously calculating the information entropy, and determining that the angle corresponding to the minimum information entropy is the offset angle of the Block; when the offset angle of the Block is greater than a threshold, rotating the Block's circumscribed rectangular box by the offset angle.

[0013] Optionally, the angle of the block's bounding rectangle is adjusted based on the projection result of the spot image block mask map, including matching a standard template constructed using the block's attributes with the spot image block mask map to locate the final position of the block's bounding rectangle.

[0014] Optionally, a standard template is constructed using Block attributes, including the number of rows and columns and the spacing between rows and columns of a Block.

[0015] Optionally, the method further includes: determining a circumscribed circle of each spot image block based on the centroid coordinates and radius.

[0016] The present invention provides a microarray chip image recognition system, the system comprising:

[0017] A block positioning unit is used to locate the block circumscribed rectangular frame of the array using a block target detection model. The block target detection model is generated by training a predetermined target detection model using image sets of multiple array arrangements, where each image set of array arrangement includes chip image data and corresponding target detection labels;

[0018] A Spot semantic segmentation unit is used to perform semantic segmentation on the Block circumscribed rectangular frame area using a Spot semantic segmentation model to generate a spot image block mask map. The Spot semantic segmentation model is generated by training a semantic segmentation network model using the Spot image blocks cropped within the Block circumscribed rectangular frame area and the corresponding semantic segmentation labels;

[0019] A Block positioning adjustment unit is used to adjust the angle of the Block circumscribed rectangular frame based on the projection result of the spot image block mask map;

[0020] The spot area recognition unit is used to calculate the centroid coordinates and radius of each spot area within the adjusted block circumscribed rectangular frame using the spot image block mask map.

[0021] The present invention provides an electronic device comprising a memory, a processor and a computer program stored in the memory and operable on the processor. When the processor executes the program, the aforementioned microarray chip image recognition method is implemented.

[0022] The present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the aforementioned microarray chip image recognition method is implemented.

[0023] The technical solution proposed by the present invention can accurately identify the location of the block and accurately align the spot, while significantly improving analysis efficiency and shortening analysis time. This solution introduces a block positioning model based on target detection and a block and spot recognition solution based on a combination of a semantic segmentation model and traditional image algorithms. After alignment, there is basically no need for manual adjustment, and the alignment effect is highly consistent with the manual alignment standard. This invention not only saves labor costs and improves efficiency, but also realizes the automation of the entire system from positioning the block to aligning the spot to extracting data. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 This is a flow chart of the main steps of the microarray chip image recognition method provided by the present invention.

[0026] Figure 2 It is a specific process flow chart of the microarray chip image recognition method provided by the present invention.

[0027] Figure 3a 、 3b They are schematic diagrams of a chip image with 1 column and 4 rows and a chip image with 2 columns and 7 rows that need to be identified in the present invention.

[0028] Figure 4a 、 4b They are the bounding rectangles recognized by the block target detection model.

[0029] Figure 5 It is the block recognition result after precise positioning.

[0030] Figure 6 This is the process of processing and changing the microarray chip image in the microarray chip image recognition method proposed by the present invention. DETAILED DESCRIPTION

[0031] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0032] This invention proposes an innovative technical solution that combines AI models with traditional image algorithms to identify blocks and spots in microarray chip images. This solution not only accurately identifies block locations but also precisely aligns spots, significantly improving analysis efficiency and shortening analysis time. The invention is described in detail below with reference to the accompanying figures.

[0033] Figure 1 The present invention provides a method for identifying microarray chip images, such as Figure 1 As shown, the method includes the following steps:

[0034] S1. Using the Block target detection model to locate the array's Block circumscribed rectangular frame, the Block target detection model is generated by training a predetermined target detection model using a variety of array arrangement image sets, each array arrangement image set includes chip image data and corresponding target detection labels;

[0035] S2. Using the Spot semantic segmentation model to perform semantic segmentation on the Block's circumscribed rectangular frame area, generating a spot image block mask map, and adjusting the angle of the Block's circumscribed rectangular frame based on the projection result of the spot image block mask map; the Spot semantic segmentation model is generated by training a semantic segmentation network model using the cropped Spot image blocks within the Block's circumscribed rectangular frame area and the corresponding semantic segmentation labels.

[0036] S3. Use the spot image block mask to calculate the centroid coordinates and radius of each spot area within the adjusted block's circumscribed rectangular frame.

[0037] The Block target detection model and Spot semantic segmentation model are both AI models, and the training adopts the pytorch framework. The present invention adopts the innovative idea of ​​combining AI models with traditional image algorithms.

[0038] The technical solution proposed by the present invention can accurately identify the location of the block and accurately align the spot, while significantly improving analysis efficiency and shortening analysis time. This solution introduces a block positioning model based on target detection and a block and spot recognition solution based on a combination of a semantic segmentation model and traditional image algorithms. After alignment, there is basically no need for manual adjustment, and the alignment effect is highly consistent with the manual alignment standard. This invention not only saves labor costs and improves efficiency, but also realizes the automation of the entire system from positioning the block to aligning the spot to extracting data.

[0039] The microarray chip image recognition method proposed in the present invention includes two major stages: Block positioning and Spot recognition. Figure 2 As shown, the specific process includes the following steps:

[0040] S01. Use the YoloV5 network structure to train the Block target detection model;

[0041] S02. Use the UNet++ network structure to train the Spot semantic segmentation model;

[0042] S03. The Tif image of the microarray chip is input into the Block target detection model, and the coordinates of the outer rectangle of the Block are output;

[0043] S04. Capture the regional image of each Block, input it into the Spot semantic segmentation model, and segment it to obtain the region and category of each Spot;

[0044] S05. Determine the deflection angle of the Block based on the binary image of the Spot: that is, for the binary image of the good and bad spots in the Block area, use the projection method to calculate the angle, that is, rotate within the range of 0-180 degrees, and calculate the information entropy every 0.2 degrees. After traversal, calculate the angle corresponding to the minimum information entropy and determine it as the deflection angle of the Block.

[0045] S06. If the deflection angle of the Block is greater than the threshold, the Block is rotated according to the deflection angle, otherwise it is not rotated;

[0046] S07. Using the properties of the Block, a standard template is established, and the binary image is matched with the standard template to calculate the precise position of the Block.

[0047] S08. After the Block is accurately positioned, calculate the distance between the center coordinates of each Circle in the Block and the X and Y values ​​of the spot centroid calculated by the Spot semantic segmentation model. If the distance is within the threshold range, assign the X and Y values ​​calculated by the Spot semantic segmentation model to the X and Y values ​​of the Circle.

[0048] S09. Traverse and calculate the threshold value in the Local area of ​​each Spot, and calculate the center coordinates and radius of the Cricle.

[0049] To recognize the image of a microarray chip, we must first accurately locate the positions of the various dot arrays in the microarray chip, and the array arrangement has various array arrangement configuration files. Because the traditional algorithm directly locates the position of the block through the projection method, and the array arrangement needs to ensure the brightness of the calibration point, if the calibration point is not bright, it will usually cause deviations in the automatic positioning, and this solution is often affected by various interferences, such as noise, fences, pollution points, etc., resulting in inaccurate positioning. In order to allow the system to accurately and precisely locate the positions of various dot matrix areas (Blocks) in the microarray chip image, the present invention, as a specific embodiment, adopts Yolov5 as a target detection deep learning network, and generates a Block target detection model through training of image data sets with multiple arrangements. This model makes it no longer necessary for all four corners to be bright, and mainly allows the model to recognize models of various arrays.

[0050] Prepare a dataset for training the Block target detection model: collect and organize chip image data and annotate target detection labels. Here, we use a 1*4 microarray chip image (Figure 3A) to train the Block positioning model as an example. A total of 40 chip images with 1 column and 4 rows in TIFF format are selected. The brightness and contrast of each TIFF image are automatically adjusted. After adjustment, jpg images with grayscale, red, green, and pseudo-color channels are generated respectively. Target detection labels are annotated on the jpg images, resulting in a total of 160 labeled data. These data are divided into training and validation sets at an 8:2 ratio. After 100 rounds of iterative training, the Block target detection model is obtained. The model is trained and optimized through multiple rounds of iteration until a detection model that can accurately locate the Block area is obtained. Another 50 1*4 microarray chip images are taken to verify the obtained Block target detection model. The results show that the Block detection rate is 100%.

[0051] To enable the model to locate block regions on 2x7 microarray chip images (Figure 3B), a certain amount of high-quality 2x7 microarray images was added to the training set to fine-tune the existing Yolov5-based block target detection model. This resulted in a target detection model capable of locating block regions on both types of microarray images. To enable the model to recognize block regions on a wider variety of array patterns, microarray images with other layouts, such as 4x2 and 2x5, were further added. Finally, a total of 400 labeled data, 100 TIFF images, were trained as a sample. These images were divided into a training set and a validation set at an 8:2 ratio. After 100 rounds of iterative training, a block target detection model was obtained. After target detection is complete, the given box position is the initial position of the block location. The block target detection model's recognition result is the circumscribed rectangular box that locates the array.

[0052] Taking 1×4 and 2×7 microarray chip images as examples, the Block target detection model obtained by the above model can locate the external rectangular frame of the microarray dot area (Block). The results are as follows: Figure 4a 、 Figure 4b shown.

[0053] It is worth noting that when training the Block target detection model based on YoloV5, the input image size is set to 1024*1024, the initial learning rate is 0.01, and the loss function is the nn.BCEWithLogitsLoss function that comes with Python. Experimental verification shows that this setting has the best training effect.

[0054] After the target detection is completed, the given Box position is the initial position of the Block positioning. The angle calculation of the Block is crucial before the microarray Spot is aligned, because if the angle of the Block deviates and is not corrected, the rows and columns of the Spot may be completely messed up. This situation is very obvious on high-density large arrays. After the target is detected and positioned, the method for calculating the Block angle is as follows: Use the Spot semantic segmentation model in this area to perform semantic segmentation on the image blocks in the area to obtain a binary image, project based on the binary image, and calculate the information entropy from 0-180 degrees. The angle corresponding to the minimum entropy is the offset angle of the Block. After completing the angle calculation, if the Block angle offset is greater than the threshold, the Block graphic is rotated by that angle, and then the template constructed by the number of Block rows and columns and the row and column spacing is matched with the binary image to find the final precise position of the block, such as Figure 5 shown.

[0055] Adjusting the angle of the block's circumscribed rectangular frame based on the projection result of the spot image block mask includes: projecting the spot image block mask from a predetermined angle range and simultaneously calculating information entropy, determining the angle corresponding to the minimum information entropy as the block's offset angle; and rotating the block's circumscribed rectangular frame by the offset angle when the block's offset angle is greater than a threshold. Adjusting the angle of the block's circumscribed rectangular frame based on the projection result of the spot image block mask includes matching a standard template constructed using the block's attributes with the spot image block mask to locate the final position of the block's circumscribed rectangular frame.

[0056] In the present invention, as a specific embodiment, Figure 6 As shown, the UNet++ semantic segmentation model is used to semantically segment the image blocks within a block, generating a binary image. Projection is then performed on this binary image, with information entropy calculated every 0.2° from 0-180°. The angle corresponding to the minimum entropy is the block's offset angle. After calculating the offset angle, the block image is rotated by that angle. A standard template is then created using the block's properties, and template matching is used to precisely locate the block. Using 1×4 and 2×7 microarray chips as examples, the rotated blocks are shown in Figure 4.

[0057] After the block is accurately positioned, the accuracy of spot recognition directly affects the final numerical extraction of each signal point. The advantage of the semantic segmentation-based spot recognition in this invention is that it can well segment the approximate spot area for various signal points and give the signal category, whether it is a good point, a bad point, or an impurity point. In the entire Tiff image, impurity points include two categories: dust and fence marks.

[0058] To build a Spot semantic segmentation model, as a specific implementation method, UNet++ is first selected as the semantic segmentation network, and then the spot image blocks are cropped from the block area after the previous angle adjustment in a sliding window manner. The spot image blocks are selected to construct the training set data and annotated with semantic segmentation labels. A total of 52 TIFF format chip data with 1 column and 4 rows and 2 columns and 7 rows of chip data are selected. Similarly, jpg images with grayscale, red, green, and pseudo-color channels are generated. Each jpg image is cropped with a sliding window size of 500*500 pixels and a step size of 380 pixels. Considering the diversity and balance of the samples, 3595 image blocks are selected from the cropped image blocks for semantic segmentation label annotation, and a total of 3595 labeled data are obtained. During the training process, these labeled data are divided into training and validation sets at an 8:2 ratio. After 300 rounds of iterative training and multiple rounds of iterative optimization, the spot semantic segmentation model is obtained and tested on a test set of 50 TIFF images. The model pixel accuracy is about 98.74% and the DICE coefficient is 86.01%.

[0059] After building the Spot semantic segmentation model, the center of mass and radius of each spot area are calculated using the mask image of each spot. The spot center of mass and radius are used to accurately calculate parameters such as the spot center, radius, and circularity on the original 16-bit TIFF data. Image morphology is used to calculate the center radius of the signal point area segmented by the Spot semantic segmentation model. The next step is determined based on the signal category. Only good and bad points are calculated for the next step; impurity points are not considered for the next step. The distance between the initial X and Y positions of the circle center of mass in the block and the X and Y positions calculated by the Spot semantic segmentation model is calculated. If the distance is less than the threshold, the X and Y coordinates calculated by the Spot semantic segmentation model are assigned to the initial X and Y values ​​of the circle center of mass. Because not all circles within the block can be detected using the Spot semantic segmentation model, some positions are exactly the same as the background. For weak points and background points, the initial X and Y positions of the circle center of mass are retained. The above threshold range is calculated using the traditional Otsu algorithm. As a specific implementation method, the segmentation threshold range of a 16-bit Tiff image is calculated.

[0060] Because the final spot recognition result is to calculate the circumscribed circle of the spot, when the spot semantic segmentation model is trained and can be used for spot recognition, the circumscribed circle of the spot is obtained according to the X and Y coordinates of the spot centroid obtained by the spot semantic segmentation model.

[0061] Through the above specific description, the present invention first uses its reasoning results to perform image projection algorithm and template matching algorithm to calculate the precise position and angle of the block; then uses the semantic segmentation reasoning results to perform morphological calculation to obtain the center, radius and category of the spot, and combines it with the Otsu threshold algorithm of the traditional image algorithm to calculate the final spot's precise center and radius.

[0062] The present invention provides a microarray chip image recognition system, the system comprising:

[0063] A block positioning unit is used to locate the block circumscribed rectangular frame of the array using a block target detection model. The block target detection model is generated by training a predetermined target detection model using image sets of multiple array arrangements, where each image set of array arrangement includes chip image data and corresponding target detection labels;

[0064] A Spot semantic segmentation unit is used to perform semantic segmentation on the Block circumscribed rectangular frame area using a Spot semantic segmentation model to generate a spot image block mask map. The Spot semantic segmentation model is generated by training a semantic segmentation network model using the Spot image blocks cropped within the Block circumscribed rectangular frame area and the corresponding semantic segmentation labels;

[0065] A Block positioning adjustment unit is used to adjust the angle of the Block circumscribed rectangular frame based on the projection result of the spot image block mask map;

[0066] The spot area recognition unit is used to calculate the centroid coordinates and radius of each spot area within the adjusted block circumscribed rectangular frame using the spot image block mask map.

[0067] Preferably, the system also includes an assignment determination unit for calculating the distance between the center coordinates of each Circle in the Block and the center of mass X, Y of the spot calculated by the Spot semantic segmentation model. If the distance is within a threshold range, the center of mass X, Y of the spot calculated by the Spot semantic segmentation model is assigned to the X, Y values ​​of the Circle center.

[0068] Preferably, the system further comprises a circumscribed circle finalization unit, which is used to traverse and calculate the threshold range within the Local area of ​​each Spot, and calculate the center coordinates and radius of the Circle.

[0069] The present invention provides an electronic device comprising a memory, a processor and a computer program stored in the memory and operable on the processor. When the processor executes the program, the aforementioned microarray chip image recognition method is implemented.

[0070] The present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the aforementioned microarray chip image recognition method is implemented.

[0071] The present invention can realize the automation of the entire system from positioning the Block, aligning the Spot, to extracting the microarray chip image data.

[0072] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying a microarray chip image, characterized in that: include: Locating the array's block circumscribed rectangular frame using a block target detection model, where the block target detection model is generated by training a predetermined target detection model using image sets of multiple array arrangements, where each image set includes chip image data and corresponding target detection labels; The Spot semantic segmentation model is used to perform semantic segmentation on the Block circumscribed rectangular frame area to generate a spot image block mask map, and the angle of the Block circumscribed rectangular frame is adjusted based on the projection result of the spot image block mask map; the Spot semantic segmentation model is generated by training a semantic segmentation network model using the cropped Spot image blocks in the Block circumscribed rectangular frame area and the corresponding semantic segmentation labels. The spot image block mask is used to calculate the centroid coordinates and radius of each spot area within the adjusted Block circumscribed rectangular frame.

2. The method for recognizing microarray chip images according to claim 1, wherein The block target detection model is generated by training a predetermined target detection model using image sets arranged in a variety of arrays, including: Adjust the brightness and contrast of the chip image in the predetermined array arrangement and generate JPG images of grayscale, red, green, and pseudo-color channels respectively; Mark the target detection label on the JPG image.

3. The method for recognizing microarray chip images according to claim 1, wherein The predetermined target detection model is a target detection model based on YoloV5, the initial learning rate parameter of the target detection model based on YoloV5 is 0.01, and the loss function is the nn.BCEWithLogitsLoss function provided by Python.

4. The method for recognizing microarray chip images according to claim 1, wherein Adjust the angle of the Block's circumscribed rectangular frame based on the projection result of the spot image block mask map, including: projecting the spot image block mask map from a predetermined angle range and simultaneously calculating the information entropy, determining the angle corresponding to the minimum information entropy as the Block's offset angle; when the Block's offset angle is greater than a threshold, rotating the Block's circumscribed rectangular frame by the offset angle.

5. The method for recognizing microarray chip images according to claim 4, wherein The angle of the block's bounding rectangle is adjusted based on the projection result of the spot image block mask map, including matching the standard template constructed using the block's attributes with the spot image block mask map to locate the final position of the block's bounding rectangle.

6. The method for recognizing microarray chip images according to claim 5, wherein Use the properties of Block to build a standard template, including the number of rows and columns and the spacing between rows and columns.

7. The method for recognizing microarray chip images according to claim 1, wherein The method also includes: determining a circumscribed circle of each spot image block according to the centroid coordinates and the radius.

8. A microarray chip image recognition system, characterized in that: The system includes: A block positioning unit is used to locate the block circumscribed rectangular frame of the array using a block target detection model. The block target detection model is generated by training a predetermined target detection model using image sets of multiple array arrangements, where each image set of array arrangement includes chip image data and corresponding target detection labels; A Spot semantic segmentation unit is used to perform semantic segmentation on the Block circumscribed rectangular frame area using a Spot semantic segmentation model to generate a spot image block mask map. The Spot semantic segmentation model is generated by training a semantic segmentation network model using the Spot image blocks cropped within the Block circumscribed rectangular frame area and the corresponding semantic segmentation labels; A Block positioning adjustment unit is used to adjust the angle of the Block circumscribed rectangular frame based on the projection result of the spot image block mask map; The spot area recognition unit is used to calculate the centroid coordinates and radius of each spot area within the adjusted block circumscribed rectangular frame using the spot image block mask map.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the microarray chip image recognition method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for recognizing a microarray chip image according to any one of claims 1 to 7 is implemented.

Images