Self-adaptive brightness classification method and system for sparsely distributed microbeads on gene chip

Abstract

The invention provides a self-adaptive brightness classification method and system for sparsely distributed microbeads on a gene chip. The method comprises the following steps: converting gene chip fluorescence image data into a frequency spectrum space by using fast Fourier transform, calculating a rotation angle, and then rotating in a reverse direction to obtain a first image; defining a concerned microbead area on the first image to obtain a second image; performing inverse transformation on the second picture to obtain a third picture, then projecting the third picture and extracting row and column peak values, and finally generating a microbead positioning point array; extracting a foreground value and a background value from the second picture, and outputting a feature gray value at each microbead position; and drawing a histogram according to the feature gray values, determining a threshold value of brightness and darkness division, dividing the beads into bright beads and dark beads, determining the beads with the gray values greater than the threshold value as the bright beads, and determining the beads with the gray values smaller than the threshold value as the dark beads. According to the technical scheme, the method is suitable for microbead positioning and adaptive brightness classification when microbeads are distributed very sparsely, and the positioning success rate is high.

Description

Technical Field

[0001] The invention relates to the technical field of electrical digital data processing, and in particular to an adaptive light-dark classification method and system for sparsely distributed microbeads of a gene chip. Background Art

[0002] In the decoding process of high-density gene chips, it is necessary to classify the brightness of the beads based on the fluorescent signals of the beads on the gene chip, so as to decode the gene chip. There are two methods for locating and classifying the brightness of sparse beads on gene chips in the prior art:

[0003] (1) Projection method: Figure 1 As shown, the image is calculated along the x and y directions (or mean, variance, etc.) to obtain a one-dimensional curve (such as Figure 2 As shown in the figure), the peak point is calculated as the column / row coordinate position of the microbead array. However, this method has the following disadvantages: in sparse situations, some columns or entire columns have no bright beads, and the projection curve cannot reflect the corresponding peak, resulting in missed recognition of some rows and columns.

[0004] (2) Iterative method (illumina): Figure 3 As shown in the figure, ① find all bright beads and calculate the grid spacing; ② construct an incomplete Voronoi diagram to find the position of the dark beads; ③ calculate the affine transformation and the mapping transformation with the standard grid. However, this method has the following disadvantages: the amount of calculation is large, and in very sparse situations, there are fewer bright beads and fewer combinations of adjacent bright beads, which makes the first step prone to errors.

[0005] In summary, the above two methods in the prior art are only applicable to the case where the microbeads on the gene chip are sparsely distributed, for example, 15%≤number of bright beads / total number of wells≤30%. And the success rate is only about 20-30%. Moreover, for the case where the microbeads on the gene chip are even more sparsely distributed (i.e. number of bright beads / total number of wells≤15%), the above two methods in the prior art are no longer applicable, and the success rate is extremely low. Summary of the invention

[0006] In order to overcome the above technical problems, the first aspect of the present application provides an adaptive light-dark classification method for sparsely distributed microbeads on a gene chip, which comprises:

[0007] Step S1: Using fast Fourier transform, the gene chip fluorescence image data is converted into the spectrum space, the rotation angle is calculated, and then the first image is obtained by rotating in the opposite direction;

[0008] Step S2: limiting the microbead area of ​​interest on the first image, cutting out irrelevant areas, and obtaining a second image;

[0009] Step S3: performing an inverse transformation on the second image to obtain a third image, then projecting the third image and extracting row and column peaks to finally generate a microbead positioning point array;

[0010] Step S4: Based on the microbead positioning point, take it as the center point and extract pixels near the center point from the second image; sort / cluster the pixels, extract the foreground value representing the grayscale of the beads and the background value representing the grayscale of the area without beads from the second image, and output a characteristic grayscale value for each microbead position, and the characteristic grayscale value = foreground value - background value;

[0011] Step S5: Draw a histogram based on the characteristic grayscale value, determine the grayscale value threshold for light and dark division by finding the trough between two peaks, and divide the beads into two categories: bright beads and dark beads. Beads with grayscale values ​​greater than the threshold are bright beads, and beads with grayscale values ​​less than the threshold are dark beads.

[0012] Furthermore, in step S1, the calculation method of the rotation angle includes the following two cases:

[0013] (1) For the rectangular array of microbeads, find the five microbeads with the largest peaks in the local maximum in the spectrum space and with a specific distance between them in five times. The brightest point is always the middle point. Then find the remaining four points in turn according to their positions. Then calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles.

[0014] (2) For microbeads arranged in a honeycomb or regular hexagonal pattern, find the seven microbeads with the largest peak values ​​in the local maximum in the spectrum space and with a specific distance between them in seven steps. The brightest point is always the middle point. Then find the remaining six points in turn according to their positions. Then, calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles.

[0015] Further, step S3 includes steps S3.1 to S3.3:

[0016] Step S3.1: Calculate the frequency spectrum space of the second image, retain the high-frequency signal, and perform inverse transformation to obtain the third image covered with microbeads;

[0017] Step S3.2: Project the third image to obtain a projection image, calculate the local peak point of the projection image, and pair and combine the row and column peak coordinates to obtain the center point coordinates of all microbeads in the third image;

[0018] Step S3.3: Determine the boundary location, including S3.3.1-S3.3.2:

[0019] S3.3.1: Use the coordinates of the center point of the microbead to obtain the corresponding grayscale value, sum / variance these grayscale values ​​by row / column to obtain a sequence, and filter out values ​​below a certain threshold to determine the boundary;

[0020] S3.3.2: Based on the preset number of rows and columns, the boundaries are modified to obtain the extracted rows and columns and their peak positions;

[0021] Step S3.4: Based on the extracted row and column peak positions, pair and combine to generate a microbead positioning point array.

[0022] Furthermore, in step S3.1, the method for generating the third picture includes:

[0023] Step S3.1.1: Find the five brightest micro-bead points arranged in a rectangle, or the seven brightest micro-bead points arranged in a honeycomb or regular hexagon from the second image;

[0024] Step S3.1.2: prepare a blank image with the same size as the original image of the second image as a mask image;

[0025] Step S3.1.3: On the mask image, draw circles with a radius of r and a gray value of 1, with the brightest microbead point as the center;

[0026] Step S3.1.4: multiply the mask image and the data at the same position of the spectrum to retain the high-frequency signal;

[0027] Step S3.1.5: Perform inverse Fourier transform on the product to obtain the reconstructed third image filled with microbeads.

[0028] Furthermore, the step S5 includes steps S5.1 to S5.3:

[0029] Step S5.1: Logarithmically transform the feature grayscale values ​​to compress the long-tail data together, while the points closer to the zero point are less compressed;

[0030] Step S5.2: Count the histogram data and use moving average for smoothing;

[0031] Step S5.3: Draw a histogram. When the second-order derivative is less than 0 and the first-order derivative is equal to 0, it is determined as a peak. When the second-order derivative is greater than 0 and the first-order derivative is equal to 0, it is determined as a trough. The dark beads are gathered into one peak, while the bright beads are gathered into another peak. By finding the trough between the two peaks, the threshold for the light-dark division is determined, and the beads are divided into two categories: bright beads and dark beads.

[0032] The second aspect of the present application provides an adaptive light-dark classification system for sparsely distributed microbeads on a gene chip, comprising:

[0033] A rotation module, the rotation module is used to convert the gene chip fluorescence image data into a spectrum space using a fast Fourier transform, calculate a rotation angle, and then rotate in the opposite direction to obtain a first image;

[0034] A microbead block segmentation module, the microbead block segmentation module is used to define a microbead area of ​​interest on the first image, cut out irrelevant areas, and obtain a second image;

[0035] A positioning module, the positioning module is used to perform an inverse transformation on the second image to obtain a third image, then project the third image and extract row and column peaks, and finally generate a microbead positioning point array;

[0036] A feature extraction module is used to extract pixels near the center point from the second image based on the microbead positioning point, taking it as the center point; sorting / clustering the pixels, extracting a foreground value representing the grayscale of the beads and a background value representing the grayscale of the area without beads from the second image, and outputting a feature grayscale value for each microbead position, where the feature grayscale value = foreground value - background value;

[0037] The module for classifying bright and dark beads is used to draw a histogram according to the characteristic grayscale value, determine the grayscale value threshold for bright and dark division by finding the trough between two peaks, and classify the beads into two categories: bright beads and dark beads. Beads with grayscale values ​​greater than the threshold are bright beads, and beads with grayscale values ​​less than the threshold are dark beads.

[0038] Furthermore, the method for calculating the rotation angle by the rotation module includes the following two cases:

[0039] (1) For the rectangular array of microbeads, find the five microbeads with the largest peaks in the local maximum in the spectrum space and with a specific distance between them in five times. The brightest point is always the middle point. Then find the remaining four points in turn according to their positions. Then calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles.

[0040] (2) For microbeads arranged in a honeycomb or regular hexagonal pattern, find the seven microbeads with the largest peak values ​​in the local maximum in the spectrum space and with a specific distance between them in seven steps. The brightest point is always the middle point. Then find the remaining six points in turn according to their positions. Then, calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles.

[0041] Furthermore, the positioning module is used for:

[0042] Calculate the spectrum space of the second image, retain the high-frequency signal, and perform inverse transformation to obtain the third image covered with microbeads;

[0043] Projecting the third image to obtain a projection image, calculating the local peak point of the projection image, and pairing and combining the row and column peak coordinates to obtain the center point coordinates of all microbeads in the third image;

[0044] Determine the boundary position: use the coordinates of the center point of the microbead to obtain the corresponding grayscale value, sum / variance these grayscale values ​​by row / column to obtain a sequence, filter out values ​​below a specific threshold to determine the boundary; combine the preset number of rows and columns to correct the boundary, so as to obtain the extracted rows and columns and their peak positions;

[0045] According to the extracted row and column peak positions, the pairs are combined to generate a microbead positioning point array.

[0046] Furthermore, the method for generating the third picture includes:

[0047] Find the five brightest micro-beads arranged in a rectangle, or the seven brightest micro-beads arranged in a honeycomb or regular hexagon from the second picture;

[0048] Prepare a blank image with the same size as the original image of the second image as a mask image;

[0049] On the mask image, draw circles with a radius of r and a gray value of 1, with the brightest microbead as the center;

[0050] The mask image is multiplied by the data at the same position of the spectrum to retain the high-frequency signal;

[0051] The inverse Fourier transform of the product is performed to obtain the reconstructed third image filled with microbeads.

[0052] Furthermore, the bright and dark beads classification module is used to:

[0053] Take a logarithmic transformation on the feature grayscale value to compress the long-tail data together, while the points closer to the 0 point are less compressed;

[0054] Count the histogram data and use moving average for smoothing;

[0055] Draw a histogram. When the second-order derivative is less than 0 and the first-order derivative is 0, it is determined as a peak. When the second-order derivative is greater than 0 and the first-order derivative is 0, it is determined as a trough. The dark beads are gathered in one peak, and the bright beads are gathered in another peak. By finding the trough between the two peaks, the threshold for the light-dark division is determined, and the beads are divided into two categories: bright beads and dark beads.

[0056] After adopting the above technical solution, this application has the following technical effects:

[0057] The technical solution of the present application is suitable for microbead positioning and adaptive light-dark classification when the distribution of microbeads is very sparse (i.e., the number of bright beads / total number of holes ≤ 15%). The positioning success rate is high, and can reach 80%-90%, which facilitates subsequent accurate decoding. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the principle of the projection method in the prior art.

[0059] Figure 2 It is a one-dimensional curve obtained by the projection method in the prior art.

[0060] Figure 3 It is a schematic diagram of the principle of the iterative method in the prior art.

[0061] Figure 4 Schematic diagram of finding the five brightest beads in the spectrum space when targeting a rectangular arrangement of beads.

[0062] Figure 5 Schematic diagram of the actual angle and theoretical angle for beads arranged in a honeycomb / regular hexagonal pattern.

[0063] Figure 6 a in the figure is a local example of image B; b is the projection corresponding to a, and the horizontal axis is the x coordinate of the image (corresponding to Figure 6 The x-axis corresponds to the coordinate in a), and the vertical axis is the grayscale value (0-255).

[0064] Figure 7 In the figure, a is the original image after cropping of image B, b is the mask image, and c is the reconstructed image C.

[0065] Figure 8 The projection image of the reconstructed image C, the horizontal axis is the image x coordinate (with Figure 6 The x-axis coordinate of a corresponds to the normalized grayscale value. Assuming {vi} is the original grayscale value, is the normalized value.

[0066] Fig. 9 is the projection of the original image, and the horizontal axis is the image x coordinate (with Figure 6 The x-axis corresponds to the coordinate of a), and the vertical axis is the grayscale value (0-255).

[0067] Fig.10 To capture a sample image of a neighborhood pixel area, the horizontal and vertical coordinates are the x and y coordinates (integers) of the pixel matrix.

[0068] Fig.11 This is a schematic diagram of the peaks and troughs in the grayscale histogram. The horizontal axis is the logarithm of the grayscale value, and the vertical axis is the count in each interval. DETAILED DESCRIPTION

[0069] The advantages of the present invention are further described below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art should understand that the following specific description is illustrative rather than restrictive, and should not be used to limit the scope of protection of the present invention.

[0070] This embodiment provides an adaptive light-dark classification system for sparsely distributed microbeads on a gene chip, which includes: a rotation module, a microbead block segmentation module, a positioning module, a feature extraction module and a light-dark bead classification module.

[0071] The rotation module is used to convert the gene chip fluorescence image data into the spectrum space using fast Fourier transform, calculate the rotation angle, and then rotate in the opposite direction to obtain the first image.

[0072] The microbead block segmentation module is used to define the microbead area of ​​interest on the first image, and cut out irrelevant areas to obtain a second image.

[0073] The positioning module is used to perform an inverse transformation on the second image to obtain a third image, and then project the third image and extract row and column peaks to finally generate a microbead positioning point array.

[0074] The feature extraction module is used to extract pixels near the center point from the second image based on the microbead positioning point as the center point; sort / cluster the pixels, extract the foreground value representing the grayscale of the beads and the background value representing the grayscale of the area without beads from the second image, and output a feature grayscale value for each microbead position, and the feature grayscale value = foreground value - background value.

[0075] The module for classifying bright and dark beads is used to draw a histogram based on the characteristic grayscale value. By finding the trough between two peaks, the grayscale value threshold for bright and dark classification is determined, and beads are divided into two categories: bright beads and dark beads. Beads with grayscale values ​​greater than the threshold are bright beads, and beads with grayscale values ​​less than the threshold are dark beads.

[0076] The method for performing microbead adaptive light-dark classification using the adaptive light-dark classification system of sparsely distributed microbeads on the gene chip comprises the following steps S1 to S5:

[0077] Step S1: Rotation: The microbead array is regularly distributed (adjacent regular hexagonal honeycomb or square), calculate the angle of rotating the grid points to a horizontal and vertical state, and rotate the image to obtain image A.

[0078] The image data is converted into the spectrum space using the fast Fourier transform, the rotation angle is calculated, and the image A is obtained by rotating in the opposite direction.

[0079] (1) For the beads arranged in a rectangular shape, find the five brightest beads in the spectrum space, such as Figure 4 shown.

[0080] The brightest point is always the middle point O, and then A, B, C, and D are found in turn according to their positions. The rotation angle is then calculated through the included angle between the actual angles of these points and the theoretical angles.

[0081] Theoretical angle: AC is parallel to the y-axis, AB is at an angle of 45° to the x-axis, and the others are similar.

[0082] (2) For the beads arranged in a honeycomb / regular hexagonal pattern, find the seven brightest beads in the spectrum space (the seven with the largest peaks in the local maximum).

[0083] The brightest point is always the middle point O, and then find A, B, C, D, E, F in turn according to the position. Then calculate the rotation angle through the actual angle of these points and the theoretical angle, such as Figure 5 shown.

[0084] Theoretical angle: BC is parallel to the y-axis, the angle between AB and the x-axis is 30°, and the others are similar.

[0085] Step S2: Segment the microbead block: limit the microbead area of ​​interest on the image, crop out irrelevant areas, and obtain image B.

[0086] like Figure 6 As shown, find the left boundary of the region x-axis, calculate the mean by column, get the curve on the right, find the minimum value in a region on the left, and delete the region on the left of the minimum value. The right boundary of the x-axis is similar to the left boundary.

[0087] The boundary of the Y axis is similar to the boundary of the x axis. Projection is performed row by row to find the minimum value and then clip.

[0088] Step S3: Positioning:

[0089] S3.1: Calculate the spectrum space of the normalized image, retain the high-frequency signal, and perform inverse transformation to obtain the image C covered with microbeads;

[0090] like Figure 7 As shown, taking honeycomb as an example, the rotation steps are similar. Calculate the spectrum space of image B:

[0091] •Find the 7 brightest points;

[0092] •Prepare an all-0 image with the same size as the original image as the mask image;

[0093] •On the mask image, draw circles with a radius of r and a grayscale value of 1 with these 7 points as the center;

[0094] •The mask image is multiplied with the data at the same position of the spectrum (retaining high-frequency signals);

[0095] • Perform an inverse Fourier transform on the product to obtain the reconstructed image C filled with microbeads.

[0096] S3.2: Project the image C, calculate the local peak point of the projection image, pair and combine the row and column peak coordinates to obtain the coordinates PO of the center points of all beads in C;

[0097] like Figure 8As shown, the reconstructed image C is projected, and the local peak point of the projection image is calculated;

[0098] The peak point coordinate sequence {x i} i=1,…,n ;

[0099] The Y direction obtains the peak point coordinate sequence {y j} j=1,…,m .

[0100] S3.3: Determine the location of the boundary:

[0101] S3.3.1: Use the coordinates of the center point P0 of the microbead to obtain the corresponding grayscale value, and sum / variance these grayscale values ​​by row / column to obtain a sequence. The rows / columns with lower values ​​do not contain beads to determine the boundary;

[0102] S3.3.2: Based on the preset number of rows and columns, modify the boundary (if the calculated number is greater than the preset number, it will shrink inward; if the calculated number is less than the preset number, it will expand outward. The specific point to shrink or expand is based on the point "confidence", and the confidence is calculated as the similarity / distance with the set of neighboring points).

[0103] like Fig. 9 As shown, on the projection curve of the original image, take the sequence {x i} i=1,…,n , using percentiles (such as 20%) to filter out smaller values, starting from the left and counting the first retained value x k1 , which is the starting point, and the last point x k2 is the end point;

[0104] Finally, we get the peak point coordinate sequence {x i} i=k1,…,k2 ;

[0105] The same method can be used to obtain the peak point coordinate sequence {y j} j=l1,…,l2 .

[0106] S3.4: Based on the extracted row and column peak positions, pair and combine them to generate the microbead positioning point array P1.

[0107] Pair and combine the row and column peak coordinates to obtain the coordinates of all microbead center points in C.

[0108] P0={(x i ,y j )}, where i=k1,…,k2, and j=l1,…,l2.

[0109] Step S4: Feature extraction:

[0110] Step S4.1: Extract pixels near the center point based on the microbead positioning point, such as Fig.10 shown.

[0111] According to the center point of the bead (x i ,y j ), taking this as the center point, intercept the pixel area of ​​the neighborhood (rectangular / circular).

[0112] Step S4.2: Sort / cluster the pixels and extract the foreground value (grayscale of beads) / background (grayscale of the area without beads).

[0113] Sort the grayscale values ​​of the pixel area, take the average of the 5 smallest elements as the background value BG, and the average of the higher 30% pixel values ​​as the foreground value FG.

[0114] Each bead position (x i ,y j )output V=FG–BG.

[0115] Step S5: Bright and dark bead classification:

[0116] Step S5.1: Since there are many dark beads, the long-tail data are compressed together by taking a logarithmic transformation, while the points closer to the 0 point are less compressed.

[0117] Since there are many dark beads, the long-tail data are compressed together by taking a logarithmic transformation to make the peak obvious, while the points closer to point 0 are less compressed.

[0118] F(x)=log(x)

[0119] Where x is the grayscale value.

[0120] Step S5.2: Dark beads are gathered into one peak, while bright beads are gathered into another peak. By finding the valley between the two peaks, the threshold for dividing bright and dark is determined, and the beads are divided into two categories: bright beads and dark beads.

[0121] (1) Statistical histogram data

[0122] ① Divide the interval T into equal intervals i = [x i , x i+1 ];

[0123] ② For each bead value V, the corresponding interval T i The count is increased by one, and finally the count of each interval is obtained. ;

[0124] ③ Use moving average for smoothing, with a window size of 2k+1, where k is a positive integer, for example, the window size is 5;

[0125] .

[0126] in, is the count of each interval after smoothing.

[0127] (2) Dark beads gather in one peak, while bright beads gather in another peak. By finding the valley between the two peaks, we can determine the grayscale threshold for dividing light and dark, and divide the beads into two categories: bright beads and dark beads. Beads with grayscale values ​​greater than the threshold are bright beads, and beads with grayscale values ​​less than the threshold are dark beads. Fig.11 As shown, the wave peaks on the right are bright beads, and the wave peaks on the left are dark beads.

[0128] ①The peak is the local maximum, and the trough is the local minimum;

[0129] ②The trough to be found is the first trough on the right side of the first peak.

[0130] It should be noted that the embodiments of the present invention have better practicability and do not impose any form of limitation on the present invention. Any technician familiar with the field may use the technical content disclosed above to change or modify it into an equivalent effective embodiment. However, any modification or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An adaptive light-dark classification method for sparsely distributed microbeads on a gene chip, characterized in that: include: Step S1: Using fast Fourier transform, the gene chip fluorescence image data is converted into the spectrum space, the rotation angle is calculated, and then the first image is obtained by rotating in the opposite direction; Step S2: limiting the microbead area of ​​interest on the first image, cutting out irrelevant areas, and obtaining a second image; Step S3: performing an inverse transformation on the second image to obtain a third image, then projecting the third image and extracting row and column peaks to finally generate a microbead positioning point array; Step S4: Based on the microbead positioning point, taking it as the center point, extracting pixels near the center point from the second image; Sort / cluster the pixels, extract the foreground value representing the grayscale of the beads and the background value representing the grayscale of the area without beads from the second image, and output a characteristic grayscale value for each microbead position, where the characteristic grayscale value = foreground value - background value; Step S5: Draw a histogram based on the characteristic grayscale value, determine the grayscale value threshold for light and dark division by finding the trough between two peaks, and divide the beads into two categories: bright beads and dark beads. Beads with grayscale values ​​greater than the threshold are bright beads, and beads with grayscale values ​​less than the threshold are dark beads.

2. The adaptive light-dark classification method for sparsely distributed microbeads on a gene chip as claimed in claim 1, characterized in that: In step S1, the calculation method of the rotation angle includes the following two cases: (1) For the rectangular array of microbeads, find the five microbeads with the largest peaks in the local maximum in the spectrum space and with a specific distance between them in five times. The brightest point is always the middle point. Then find the remaining four points in turn according to their positions. Then calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles. (2) For microbeads arranged in a honeycomb or regular hexagonal pattern, find the seven microbeads with the largest peak values ​​in the local maximum in the spectrum space and with a specific distance between them in seven steps. The brightest point is always the middle point. Then find the remaining six points in turn according to their positions. Then, calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles.

3. The adaptive light-dark classification method for sparsely distributed microbeads on a gene chip as claimed in claim 1, characterized in that: Step S3 includes steps S3.1 to S3.3: Step S3.1: Calculate the frequency spectrum space of the second image, retain the high-frequency signal, and perform inverse transformation to obtain the third image covered with microbeads; Step S3.2: Project the third image to obtain a projection image, calculate the local peak point of the projection image, and pair and combine the row and column peak coordinates to obtain the center point coordinates of all microbeads in the third image; Step S3.3: Determine the boundary location, including S3.3.1-S3.3.2: S3.3.1: Use the coordinates of the center point of the microbead to obtain the corresponding grayscale value, sum / variance these grayscale values ​​by row / column to obtain a sequence, and filter out values ​​below a certain threshold to determine the boundary; S3.3.2: Based on the preset number of rows and columns, the boundaries are modified to obtain the extracted rows and columns and their peak positions; Step S3.4: Based on the extracted row and column peak positions, pair and combine to generate a microbead positioning point array.

4. The method for adaptive light-dark classification of sparsely distributed microbeads on a gene chip as claimed in claim 3, characterized in that: In step S3.1, the method for generating the third picture includes: Step S3.1.1: Find the five brightest micro-bead points arranged in a rectangle, or the seven brightest micro-bead points arranged in a honeycomb or regular hexagon from the second image; Step S3.1.2: prepare a blank image with the same size as the original image of the second image as a mask image; Step S3.1.3: On the mask image, draw circles with a radius of r and a gray value of 1, with the brightest microbead point as the center; Step S3.1.4: multiply the mask image and the data at the same position of the spectrum to retain the high-frequency signal; Step S3.1.5: Perform inverse Fourier transform on the product to obtain the reconstructed third image filled with microbeads.

5. The adaptive light-dark classification method for sparsely distributed microbeads on a gene chip as claimed in claim 1, characterized in that: The step S5 includes steps S5.1 to S5.3: Step S5.1: Logarithmically transform the feature grayscale values ​​to compress the long-tail data together, while the points closer to the zero point are less compressed; Step S5.2: Count the histogram data and use moving average for smoothing; Step S5.3: Draw a histogram. When the second-order derivative is less than 0 and the first-order derivative is equal to 0, it is determined as a peak. When the second-order derivative is greater than 0 and the first-order derivative is equal to 0, it is determined as a trough. The dark beads are gathered into one peak, while the bright beads are gathered into another peak. By finding the trough between the two peaks, the threshold for the light-dark division is determined, and the beads are divided into two categories: bright beads and dark beads.

6. An adaptive light-dark classification system for sparsely distributed microbeads on a gene chip, characterized in that: include: A rotation module, the rotation module is used to convert the gene chip fluorescence image data into a spectrum space using a fast Fourier transform, calculate a rotation angle, and then rotate in the opposite direction to obtain a first image; A microbead block segmentation module, the microbead block segmentation module is used to define a microbead area of ​​interest on the first image, cut out irrelevant areas, and obtain a second image; A positioning module, the positioning module is used to perform an inverse transformation on the second image to obtain a third image, then project the third image and extract row and column peaks, and finally generate a microbead positioning point array; A feature extraction module, wherein the feature extraction module is used to extract pixels near the center point from the second image based on the microbead positioning point as the center point; Sort / cluster the pixels, extract the foreground value representing the grayscale of the beads and the background value representing the grayscale of the area without beads from the second image, and output a characteristic grayscale value for each microbead position, where the characteristic grayscale value = foreground value - background value; The module for classifying bright and dark beads is used to draw a histogram according to the characteristic grayscale value, determine the grayscale value threshold for bright and dark division by finding the trough between two peaks, and classify the beads into two categories: bright beads and dark beads. Beads with grayscale values ​​greater than the threshold are bright beads, and beads with grayscale values ​​less than the threshold are dark beads.

7. The adaptive light-dark classification system for sparsely distributed microbeads on a gene chip as claimed in claim 6, characterized in that: The method for calculating the rotation angle by the rotation module includes the following two cases: (1) For the rectangular array of microbeads, find the five microbeads with the largest peaks in the local maximum in the spectrum space and with a specific distance between them in five times. The brightest point is always the middle point. Then find the remaining four points in turn according to their positions. Then calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles. (2) For microbeads arranged in a honeycomb or regular hexagonal pattern, find the seven microbeads with the largest peak values ​​in the local maximum in the spectrum space and with a specific distance between them in seven steps. The brightest point is always the middle point. Then find the remaining six points in turn according to their positions. Then, calculate the rotation angle through the angle between the actual angles of these points and the theoretical angles.

8. The adaptive light-dark classification system for sparsely distributed microbeads on a gene chip as claimed in claim 6, characterized in that: The positioning module is used for: Calculate the spectrum space of the second image, retain the high-frequency signal, and perform inverse transformation to obtain the third image covered with microbeads; Projecting the third image to obtain a projection image, calculating the local peak point of the projection image, and pairing and combining the row and column peak coordinates to obtain the center point coordinates of all microbeads in the third image; Determine the boundary position: use the coordinates of the center point of the microbead to obtain the corresponding grayscale value, sum / variance these grayscale values ​​by row / column to obtain a sequence, filter out values ​​below a specific threshold to determine the boundary; combine the preset number of rows and columns to correct the boundary, so as to obtain the extracted rows and columns and their peak positions; According to the extracted row and column peak positions, the pairs are combined to generate a microbead positioning point array.

9. The adaptive light-dark classification system for sparsely distributed microbeads on a gene chip as claimed in claim 8, characterized in that: The method for generating the third picture includes: Find the five brightest micro-beads arranged in a rectangle, or the seven brightest micro-beads arranged in a honeycomb or regular hexagon from the second picture; Prepare a blank image with the same size as the original image of the second image as a mask image; On the mask image, draw circles with a radius of r and a gray value of 1, with the brightest microbead as the center; The mask image is multiplied by the data at the same position of the spectrum to retain the high-frequency signal; The inverse Fourier transform of the product is performed to obtain the reconstructed third image filled with microbeads.

10. The adaptive light-dark classification system for sparsely distributed microbeads on a gene chip as claimed in claim 6, characterized in that: The bright and dark beads classification module is used for: Take a logarithmic transformation on the feature grayscale value to compress the long-tail data together, while the points closer to the 0 point are less compressed; Count the histogram data and use moving average for smoothing; Draw a histogram. When the second-order derivative is less than 0 and the first-order derivative is 0, it is determined as a peak. When the second-order derivative is greater than 0 and the first-order derivative is 0, it is determined as a trough. The dark beads are gathered in one peak, and the bright beads are gathered in another peak. By finding the trough between the two peaks, the threshold for the light-dark division is determined, and the beads are divided into two categories: bright beads and dark beads.

Images