A dynamic scanning compensation method for inkjet-printed DNA synthesis equipment and the DNA synthesis equipment itself.

Abstract

This invention relates to a dynamic scanning compensation method and DNA synthesis device based on inkjet printing, belonging to the field of DNA synthesis. By applying threshold segmentation, connected component analysis, and circular Hough transform to a standard analysis image, a circular region for each reaction site is obtained. Based on the distance of all edge points of each reaction site from the center of the circular boundary, it is determined whether each reaction site meets the roundness requirements, thus obtaining information on missing reaction sites. Simultaneously, to address potential inaccuracies after the circular Hough transform, the results are calibrated using the YOLO target detection model based on a convolutional neural network, achieving high-precision detection of missing reaction sites suitable for industrial applications. Furthermore, by adding recognition weights for defective reaction sites such as those with "partially excessive distances from the center" during YOLOv6 model training, the compatibility between the two is further improved, enhancing detection accuracy and meeting the requirements of high-precision DNA synthesis.

Description

Technical Field

[0001] This invention relates to the field of synthetic biology, specifically to a dynamic scanning compensation method for inkjet printing-based DNA synthesis equipment and the DNA synthesis equipment itself. Background Technology

[0002] In the field of biology, automated DNA synthesis is an emerging DNA synthesis method in synthetic biology. Existing DNA inkjet printing equipment uses nozzles to spray ink droplets onto the reaction sites of a DNA chip, which has multiple circular reaction sites arrayed on it.

[0003] Commonly used DNA chips are in the form of standard glass slides, with a length of 75.30 mm and a width of 25.30 mm, making them easy to integrate into automated equipment. For example, the DNA chip is designed with circular reaction sites with a diameter of 120 μm, which are distributed on the DNA chip in a microarray manner.

[0004] DNA inkjet printing equipment utilizes inkjet printing technology to formulate inks containing the four bases (ATCG) required for gene synthesis, along with activators. This ink is then precisely printed onto different reaction sites on the DNA chip, ensuring thorough mixing of the base ink and activator ink for efficient reaction. Because DNA synthesis demands extremely precise precision in terms of droplet location and volume, the inkjet printing accuracy directly impacts the effectiveness of subsequent chemical reactions and the quality stability of the chip product.

[0005] During long-term operation, existing inkjet printers are susceptible to nozzle clogging, ink crystallization, and printhead corrosion, leading to issues such as nozzle blockage, jet direction deviation (skewed spraying), and insufficient jet volume. When the nozzle is malfunctioning, ink droplets cannot accurately land on the target reaction site, resulting in missing reaction sites. Similarly, misalignment of the DNA chip's installation position can also affect ink jet positioning, preventing droplets from accurately landing on the target reaction site.

[0006] If an ink droplet cannot accurately land on the target reaction site, the corresponding reaction site is considered a missing reaction site. To avoid missing reaction sites affecting the overall quality of DNA synthesis, current technologies typically use the following methods for detection:

[0007] Firstly, some devices rely on manual microscopic observation or offline imaging to manually determine missing reaction sites. However, this method is inefficient, highly subjective, and cannot meet the rapid detection requirements of high-density arrays.

[0008] Secondly, some technologies employ a fixed template comparison method. This involves comparing the grayscale of a sample image (a normally printed image) with the actual printed image, and then identifying missing reaction sites through simple threshold segmentation, region statistics, or grayscale judgment. However, this method requires the chip position to be perfectly aligned with the sample chip position. In reality, placement deviations are possible, typically reflecting micrometer-level deviations in the reaction sites. Due to chip placement misalignment and the unstable droplet morphology caused by nozzle changes, the fixed template method has low accuracy in complex scenarios and is prone to false positives or false negatives.

[0009] Third, some high-end equipment has nozzle ejection volume detection or ejection waveform monitoring functions, but such detection is mostly based on the printhead's own condition and cannot directly reflect whether the ink droplets accurately fall on the target reaction site, let alone achieve automatic compensation for missing reaction sites.

[0010] In summary, existing technologies generally suffer from the following shortcomings:

[0011] (1) It lacks the ability to accurately identify the reaction site image after printing, and cannot stably and quickly detect missing reaction sites when the chip posture changes or the jetting is offset.

[0012] (2) Lack of subsequent compensation printing capability, unable to perform subsequent compensation printing;

[0013] (3) The lack of an automatic verification mechanism after compensation printing makes it impossible to form a closed-loop control, and the success rate of compensation is difficult to guarantee;

[0014] (4) Existing technologies do not fully utilize machine learning or artificial intelligence methods to improve the accuracy of recognition, compensation strategies and the intelligence of cleaning judgment.

[0015] Therefore, there is an urgent need for a dynamic scanning compensation method and a DNA synthesis device based on inkjet printing to solve at least one of the above problems. Summary of the Invention

[0016] To alleviate or solve at least one aspect or point of the above-mentioned problems, the present invention is proposed. A dynamic scan compensation method for a DNA synthesis device based on inkjet printing according to the present invention includes the following steps:

[0017] After completing one DNA reagent jet printing, the inkjet-printed area is photographed to obtain a grayscale image of the inkjet-printed area. This grayscale image is called the initial image.

[0018] The initial image is preprocessed to obtain a standard analysis image;

[0019] Thresholding segmentation, connected component analysis, and circular Hough transform are performed on the standard analysis image. Thresholding segmentation distinguishes the reaction sites from the background. Connected component analysis extracts each reaction site and obtains the region of each reaction site. Circular Hough transform is used to determine the circular region of each reaction site, locate the circular boundary of each reaction site, and obtain the center coordinates (a, b) and radius r.

[0020] Based on the distances from all edge points of each reaction site to the center of the circular boundary of each reaction site, determine whether each reaction site meets the roundness requirements;

[0021] The YOLO target detection model based on convolutional neural networks was used to analyze the above standard analysis images to obtain the probability that each reaction site is a missing reaction site;

[0022] Based on whether each reaction site meets the roundness requirement and the probability that each reaction site is a missing reaction site, determine whether each reaction site is a missing reaction site;

[0023] Inkjet compensation printing was used to print missing reaction sites.

[0024] Preferably, determining whether each reaction site meets the roundness requirement based on the distances from all edge points of each reaction site to the center of the circular boundary of each reaction site includes:

[0025] Calculate the distance from all edge points of each reaction site to the center of its circular boundary. , , in, , (a, b) represents the coordinates of the i-th edge point of the reaction site, and (a, b) represents the coordinates of the center of the circular boundary of the reaction site.

[0026] Calculate the average distance deviation Dr, Dr= Where n is the number of edge points;

[0027] Dr is compared with a preset deviation value. If it is less than the preset deviation, it is considered to meet the roundness requirement; otherwise, it is considered not to meet the roundness requirement.

[0028] Preferably, the YOLO target detection model based on convolutional neural networks is used to analyze the above standard analysis images, and the probability of each reaction site being a missing reaction site is determined as follows:

[0029] Multiple grayscale images of the printed area under different inkjet printing conditions are acquired. The grayscale images are preprocessed to obtain standard analysis images. The standard analysis images are used as training images to ensure that the training images cover samples of normal reaction sites and various missing reaction sites.

[0030] The training images were manually annotated to identify missing and normal response sites.

[0031] The labeled image dataset is input into the YOLO model for training;

[0032] The trained model is evaluated and optimized using a test set.

[0033] Preferably, in YOLO model training, the recognition weight of defect response sites such as "some edge points are too far from the center of the circle" is increased.

[0034] Preferably, in YOLO model training: all defect samples with "edge points too far from the center" are selected from the dataset and denoted as S-class samples; S-class samples are replicated multiple times in the training set to increase their proportion in the training set; the validation set and test set remain unchanged.

[0035] Preferably, during YOLO model training, the mAP@0.5 and mAP@0.5:0.95 metrics are monitored on the validation set, and the weight file with the best performance on the validation set is saved to obtain the trained model.

[0036] Preferably, in YOLO model training, by assigning higher weights to the category of "edge points are too far from the center", the model will penalize the misclassification of this type of sample more severely.

[0037] Preferably, when photographing the inkjet-printed area, a grayscale camera is used to directly acquire a grayscale image.

[0038] Preferably, image preprocessing includes: pose transformation and region localization.

[0039] In addition, the present invention also provides a DNA synthesis device that employs the dynamic scanning compensation method described in any of the preceding claims.

[0040] This invention combines a missing reaction site detection algorithm with the YOLO model to achieve automated identification of the state of reaction sites after printing. It can accurately detect missing, offset, or abnormally shaped ink droplets even when the chip's orientation is off or the ejection state is unstable. Compared to existing technologies that rely on manual observation or fixed threshold judgments, this invention significantly improves the accuracy and robustness of missing detection, and reduces false positives and false negatives.

[0041] This invention employs threshold segmentation, connected component analysis, and circular Hough transform on standard analytical images. It then determines whether each reaction site meets roundness requirements based on the distance of all edge points from the center of the circular boundary of each reaction site, thus obtaining information on missing reaction sites. Simultaneously, to address potential inaccuracies after the circular Hough transform, the results are calibrated using the YOLO target detection model based on a convolutional neural network, achieving high-precision detection of missing reaction sites suitable for industrial applications. Furthermore, by adding recognition weights for defective reaction sites such as those with "partially excessive distances from the center" during YOLOv6 model training, the compatibility between the two is further improved, enhancing detection accuracy and meeting the requirements of high-precision DNA synthesis.

[0042] This invention, through high-precision compensation printing, can automatically select a compensation strategy based on the number and distribution of missing reaction sites, avoiding the secondary loss problem caused by inaccurate manual judgment or compensation position deviation in traditional re-spraying methods, thereby improving the success rate of compensation printing.

[0043] This invention automatically decides and executes compensation printing, flash print recovery, or cleaning processes, enabling the entire printing process to be adaptive. This intelligent decision-making mechanism reduces manual intervention, improves equipment operating efficiency, and lowers the risk of printing failures due to abnormal printhead conditions.

[0044] The overall closed-loop control system of this invention can continuously monitor print quality in the cycle of printing-detection-compensation-verification-cleaning, upgrading the DNA inkjet printing process from a traditional "open process" to a "monitorable, feedback-able, and correctable closed-loop process", significantly improving print quality consistency, system stability, and production yield.

[0045] This invention solves the problems of inaccurate detection, imprecise compensation, and unverifiable cleaning in existing technologies by using image recognition, intelligent compensation, cleaning detection, and closed-loop control. It has significant technological advancements and practical value. Attached Figure Description

[0046] Figure 1 A flowchart illustrating a dynamic scanning compensation method for a DNA synthesis device based on inkjet printing, as an exemplary embodiment of the present invention.

[0047] Figure 2 This is a grayscale schematic diagram of an inkjet-printed area, which is an exemplary embodiment of the present invention.

[0048] Figure 3 This is a schematic diagram of a standard analysis image after threshold segmentation and connected component analysis, which is an exemplary embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram illustrating the manual annotation of training images, as an exemplary embodiment of the present invention, to indicate missing and normal reaction sites.

[0050] Figure 5 This is a schematic diagram illustrating the identification of missing reaction sites, as an exemplary embodiment of the present invention. Detailed Implementation

[0051] The following description of embodiments of the present invention with reference to the accompanying drawings is intended to explain the overall inventive concept of the invention and should not be construed as a limitation thereof. In this invention, the same reference numerals denote the same or similar parts.

[0052] The features described herein may be implemented in various forms and should not be construed as limited to the examples described herein. Rather, the examples described herein are provided only to illustrate some of the many feasible ways in which the methods, apparatuses, and / or systems described herein will become clear upon understanding the disclosure of the invention.

[0053] The terminology used herein is for the purpose of describing various examples only and is not intended to limit disclosure. Unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. The terms “comprising,” “including,” and “having” indicate the presence of the described features, quantities, operations, components, elements, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, quantities, operations, components, elements, and / or combinations thereof.

[0054] To enable those skilled in the art to utilize the content of this invention, the following exemplary embodiments may be provided in conjunction with specific application scenarios, specific systems, device and component parameters, and specific connection methods. However, these embodiments are merely examples for those skilled in the art, and the general principles defined herein can be applied to other embodiments and application scenarios without departing from the spirit and scope of this invention.

[0055] According to an exemplary embodiment of the present invention: Figures 1-5 As shown, this invention provides a dynamic scan compensation method for a DNA synthesis device based on inkjet printing. After completing one DNA reagent jet printing cycle, an image is taken of the inkjet-printed area to obtain a grayscale image of the inkjet-printed area; this grayscale image is referred to as the initial image. Figure 2As shown, a grayscale image of an inkjet-printed area is illustrated, revealing missing reaction sites. This invention involves photographing the inkjet-printed area to be detected to obtain its grayscale image. For example, a grayscale camera is used to directly acquire a black-and-white image. Compared to using a color camera to acquire a color image and then converting it to grayscale, this method requires no conversion, has less noise, and higher accuracy.

[0056] The initial image is preprocessed to obtain a standard analysis image. For example, image preprocessing includes pose transformation and region localization. Region localization includes selecting a predetermined analysis region, such as choosing a rectangular region as the analysis region on the grayscale image acquired by photography. Pose transformation refers to correcting the image, such as rotating the image so that the longer side of the aforementioned rectangular region is parallel to or coincides with a preset straight line.

[0057] According to an exemplary embodiment of the present invention: Figure 3 As shown, threshold segmentation, connected component analysis, and circular Hough transform are applied to the standard analysis image. Threshold segmentation distinguishes the reaction sites from the background; connected component analysis extracts each reaction site and obtains the region of each reaction site; circular Hough transform determines the circular region of each reaction site, locates the circular boundary of each reaction site, and obtains the center coordinates and radius.

[0058] For example, thresholding is used to segment a standard analysis image to distinguish between reaction sites and the background. Thresholding divides the standard analysis image into foreground and background, where the foreground represents reaction sites and the background represents non-reaction site regions. If the grayscale image contains noise or uneven illumination, image preprocessing can be performed before thresholding. That is, if the image contains noise or uneven illumination, smoothing filtering (such as Gaussian filtering) can be applied first to improve segmentation accuracy. Threshold selection: Common methods include global thresholding (such as the Otsu method) or local adaptive thresholding. The Otsu method automatically determines the optimal threshold by maximizing the inter-class variance, and is suitable for images with significant grayscale differences between the foreground and background. Since the foreground and background grayscale differences are significant after inkjet printing on a DNA chip, the Otsu method is preferred. Each pixel in the image is compared with a threshold; pixels greater than the threshold are designated as background, and pixels less than the threshold are designated as foreground. A binary image is obtained, where the foreground region corresponds to the possible reaction sites.

[0059] For example, such as Figure 3As shown, connected component analysis is performed to extract each reaction site. The aforementioned threshold segmentation distinguishes the reaction sites from the background. However, the foreground, i.e., multiple array reaction sites, is a whole; therefore, it is necessary to extract each reaction site individually. Connected component analysis can be performed to identify each independent black region in the binary image and extract its geometric features. Connected component analysis refers to grouping adjacent foreground pixels (with the same gray value) in an image into a connected region. Through this analysis, each independent reaction site can be extracted, isolated noise points can be filtered out, and information such as the position, area, and edges of each reaction site can be obtained.

[0060] A circular Hough transform is performed to determine the circular region of each reaction site. The circular Hough transform precisely locates the circular boundary of each reaction site, obtaining the center coordinates and radius. The circular Hough transform yields the following information: the circular region is (a, b, r), where a and b represent the coordinates of the center, and r represents the radius of the circle.

[0061] For example, a coordinate system is established with the lower left corner of the rectangular region of the standard analysis image as the origin, the longer side of the rectangle as the X-axis, and the shorter side as the Y-axis. The Circular Hough Transform (CHT) utilizes the mathematical expression for a circle (xa). 2 +(yb) 2 =r 2 , where (a,b) is the center of the circle and r is the radius. The position (a,b) and radius of the circle are determined by mapping points on the edge of the image space to the parameter space (a,b,r) and statistically analyzing the peak values ​​in the parameter space.

[0062] Determine if the reaction site meets the roundness requirements. For example, calculate the distance from all edge points of the reaction site to the center of its circular region. , in, , Represents the coordinates of the i-th edge point of the reaction site, and calculates the average distance deviation Dr= Where n is the number of edge points. Dr is compared with a preset deviation value. If it is less than the preset deviation, it is considered to meet the roundness requirement; otherwise, it is considered not to meet the roundness requirement. The preset deviation value can be set according to the detection accuracy requirements.

[0063] This invention compares Dr with a preset deviation value. If Dr is less than the preset deviation, the reaction site is considered to meet the roundness requirement, and the inkjet printing of the reaction site is considered to meet the requirements. Conversely, if Dr is greater than the preset deviation, the reaction site is considered to not meet the inkjet printing requirements. In practical applications, the above judgment has some inaccuracies. For example, although the average distance deviation Dr meets the requirements, there may be some individual points that deviate too far, i.e., some edge points are significantly offset, while other edge points are basically not offset. In this case, the average distance deviation Dr meets the requirements, but in reality, the inkjet printing of the above reaction site does not meet the requirements.

[0064] To address the above issues, this invention employs the YOLO target detection model based on convolutional neural networks to calibrate the results, thereby achieving a dynamic scanning compensation method that meets the accuracy requirements for industrial applications.

[0065] The YOLO target detection model, based on a convolutional neural network, is used to analyze the aforementioned standard analysis images to determine the probability that each reaction site is a missing reaction site. For example, multiple grayscale images of the printed area under different inkjet printing conditions are acquired as training images, ensuring that the images cover samples of missing reaction sites and various common defects (such as uneven inkjet printing, missing parts, and tailing), and that the image resolution is sufficient to clearly distinguish the reaction sites. Optionally, multiple grayscale images of the printed area under different inkjet printing conditions are acquired, and the grayscale images are preprocessed to obtain standard analysis images, which are then used as training images.

[0066] like Figure 4 As shown, the labeled standard analysis images are used as training images, with missing response sites annotated. The training images are manually annotated to create images of missing response sites and images of missing response sites, thus obtaining the sample dataset. For example, an annotation tool (such as LabelMe) is used to annotate the bounding box of each response site in each image and assign a category label: normal response sites (complete shape, uniform grayscale, clear edges), and missing response sites (with defects, such as missing, overspray, underspray, tailing, etc.). Figure 4 The green boxes in the annotations represent labeled normal response sites, and the red boxes represent labeled missing response sites. The annotation files are saved in YOLO format. The dataset is randomly divided into training (70%), validation (15%), and test (15%) sets according to the specified proportions.

[0067] The labeled images are input into the YOLO model. This invention preferably uses the YOLOv6 model for training, which is version 6 of the YOLO model. For example, YOLOv6 is selected as the base model, with EfficientRep as the backbone network, Rep-PAN as the neck network, and a decoupled detection head as the head. In the training configuration file, the number of classes is set to 2 (missing response sites and non-missing response sites), the input image size is consistent, the initial learning rate is set to 0.01, and the number of training epochs is at least 100. During training, the mAP@0.5 and mAP@0.5:0.95 metrics are monitored on the validation set. The weight file with the best performance on the validation set is saved to obtain the trained model.

[0068] Model evaluation and tuning. The trained model is evaluated and optimized using a test set. If the model's performance on the test results is insufficient, more relevant samples are collected for incremental training or the class weights are adjusted until the application requirements are met.

[0069] Preferably, when evaluating and optimizing the trained model using the test set, the ability to identify... Weighting of excessively large cases. In YOLOv6 model training, the recognition weight of defect reaction sites such as "some edge points are too far from the center" is increased, so that the model pays more attention to these abnormal cases that are easily missed by the average distance deviation Dr, thereby accurately calibrating the original judgment results.

[0070] For example, select all defect samples with "edge points too far from the center" from the dataset, denoted as class S samples; duplicate class S samples in the training set (e.g., duplicate 2-3 times) to increase their proportion in the training set; keep the validation / test set unchanged (to avoid evaluation bias). With these settings, the model will encounter these samples more frequently during training, forcing it to learn their features and improving its recognition sensitivity. For example, modify the class weights of the loss function. YOLOv6's loss includes a "classification loss." By assigning a higher weight to the "edge points too far from the center" class, the model will penalize misclassification of these samples more severely. This approach allows for better calibration.

[0071] Based on whether each reaction site meets the roundness requirement and the probability that each reaction site is a missing reaction site, it is determined whether the reaction site is a missing reaction site. If a reaction site meets the roundness requirement and the probability of it being a normal reaction site analyzed by the YOLO model is greater than 90%, it is considered a normal reaction site; otherwise, it is marked as an abnormal reaction site, which can also be called a missing reaction site. Figure 5 The diagram illustrates the missing reaction sites obtained using the method of the present invention. For example, Figure 5The reaction sites marked "missing" are missing reaction sites. Therefore, it has high recognition accuracy.

[0072] When missing reaction sites are identified, existing re-printing methods can be used for repair. Preferably, different inkjet compensation methods are employed for inkjet compensation printing based on the number and distribution of missing reaction sites. When the number of missing reaction sites is small, a flash printhead recovery is first performed to clear any possible minor blockages or jet instability factors. Then, compensating inkjet printing is performed on the missing reaction sites. When the number of missing reaction sites is large, a reaction site area cleaning process is prioritized. After cleaning, a printhead flash printhead recovery is performed, and all reaction sites are re-printed.

[0073] For example, based on the image coordinates of the missing reaction sites, a coordinate mapping algorithm is used to convert them into compensated printing coordinates in the nozzle coordinate system, generating a compensated printing file, which is then executed by the nozzle for compensated spraying. The image coordinates of the missing reaction sites can be obtained from the information of (a, b) within the circular region (a, b, r) in the circular Hough transform described above.

[0074] This invention combines a missing reaction site detection algorithm with the YOLO model to achieve automated identification of the state of reaction sites after printing. It can accurately detect missing, offset, or abnormally shaped ink droplets even when the chip's orientation is off or the ejection state is unstable. Compared to existing technologies that rely on manual observation or fixed threshold judgments, this invention significantly improves the accuracy and robustness of missing detection, and reduces false positives and false negatives.

[0075] The high-precision compensation printing method of this invention can automatically select a compensation strategy based on the number and distribution of missing reaction sites, avoiding the secondary loss problem caused by inaccurate manual judgment or compensation position deviation in traditional re-spraying methods, thereby improving the success rate of compensation printing. This invention automatically decides and executes compensation printing, flash spray recovery, or cleaning processes, making the entire printing process adaptive. This intelligent decision-making mechanism reduces manual intervention, improves equipment operating efficiency, and reduces the risk of printing failure due to abnormal printhead conditions.

[0076] The overall closed-loop control system of this invention can continuously monitor print quality in the cycle of printing-detection-compensation-verification-cleaning, upgrading the DNA inkjet printing process from a traditional "open process" to a "monitorable, feedback-able, and correctable closed-loop process", significantly improving print quality consistency, system stability, and production yield.

[0077] This invention solves the problems of inaccurate detection, imprecise compensation, and unverifiable cleaning in existing technologies by using image recognition, intelligent compensation, cleaning detection, and closed-loop control. It has significant technological advancements and practical value.

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that variations and combinations of elements may be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic scan compensation method for an inkjet printing-based DNA synthesizer, characterized by: Includes the following steps: After completing one DNA reagent jet printing, the inkjet-printed area is photographed to obtain a grayscale image of the inkjet-printed area. This grayscale image is called the initial image. The initial image is preprocessed to obtain a standard analysis image; Thresholding segmentation, connected component analysis, and circular Hough transform are performed on the standard analysis image. Thresholding segmentation distinguishes the reaction sites from the background. Connected component analysis extracts each reaction site and obtains the region of each reaction site. Circular Hough transform is used to determine the circular region of each reaction site, locate the circular boundary of each reaction site, and obtain the center coordinates (a, b) and radius r. Based on the distances from all edge points of each reaction site to the center of the circular boundary of each reaction site, determine whether each reaction site meets the roundness requirements; The YOLO target detection model based on convolutional neural networks was used to analyze the above standard analysis images to obtain the probability that each reaction site is a missing reaction site; Based on whether each reaction site meets the roundness requirement and the probability that each reaction site is a missing reaction site, determine whether each reaction site is a missing reaction site; Inkjet compensation printing was used to print missing reaction sites.

2. The dynamic scan compensation method of claim 1, wherein: Determining whether each reaction site meets the roundness requirement based on the distances from all edge points of each reaction site to the center of the circular boundary of each reaction site includes: Calculate the distance from all edge points of each reaction site to the center of its circular boundary. , , in, , (a, b) represents the coordinates of the i-th edge point of the reaction site, and (a, b) represents the coordinates of the center of the circular boundary of the reaction site. Calculate the average distance deviation Dr, Dr= Where n is the number of edge points; Dr is compared with a preset deviation value. If it is less than the preset deviation, it is considered to meet the roundness requirement; otherwise, it is considered not to meet the roundness requirement.

3. The dynamic scanning compensation method according to claim 1, characterized in that: The YOLO target detection model, based on a convolutional neural network, was used to analyze the above standard analysis images. The probability that each reaction site is a missing reaction site was determined as follows: Multiple grayscale images of the printed area under different inkjet printing conditions are acquired. The grayscale images are preprocessed to obtain standard analysis images. The standard analysis images are used as training images to ensure that the training images cover samples of normal reaction sites and various missing reaction sites. The training images were manually annotated to identify missing and normal response sites. The labeled images are divided into training, testing, and validation sets. The training dataset is then input into the YOLO model for training. The trained model is evaluated and optimized using a test set.

4. The dynamic scanning compensation method according to claim 3, characterized in that: In YOLO model training, the recognition weight of defect response sites such as "some edge points are too far from the center of the circle" is increased.

5. The dynamic scanning compensation method according to claim 4, characterized in that: In YOLO model training: all defective samples with "edge points too far from the center" are selected from the dataset and denoted as S-class samples; S-class samples are replicated multiple times in the training set to increase their proportion in the training set; the validation set and test set remain unchanged.

6. The dynamic scanning compensation method according to claim 3, characterized in that: During YOLO model training, the mAP@0.5 and mAP@0.5:0.95 metrics are monitored on the validation set. The weight file with the best performance on the validation set is saved to obtain the trained model.

7. The dynamic scanning compensation method according to claim 3, characterized in that: In YOLO model training, by assigning higher weights to the category "edge points are too far from the center", the model is penalized more severely for misclassification of this type of sample.

8. The dynamic scanning compensation method according to claim 1, characterized in that: When photographing the inkjet-printed area, use a grayscale camera to directly acquire grayscale images.

9. The dynamic scanning compensation method according to claim 1, characterized in that: Image preprocessing includes pose transformation and region localization.

10. A DNA synthesis apparatus, characterized in that: The dynamic scanning compensation method according to any one of claims 1-9 is adopted.

Images