Machine learning based focused microplate high throughput data processing method and system

Abstract

The present application relates to the technical field of data processing, and provides a high-throughput data processing method and system for focused microplate based on machine learning. The method first acquires the microplate image to be processed and a plurality of predefined microplate quality detection viewpoints, which cover common microplate quality problems and discrimination standards. Then, by combining high-throughput image data and a plurality of quality detection viewpoints, initial quality inspection discrimination information is generated. The information is then sent to a specially designed microplate quality inspection discrimination network for processing. Through learning a large number of labeled microplate image data, the network can effectively analyze and decide the initial quality inspection discrimination information, and finally output accurate microplate quality detection viewpoints as the discrimination result. This process realizes automatic microplate quality evaluation, significantly improves work efficiency and discrimination accuracy, and provides strong technical support for experimental research in the biomedical field.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a high-throughput data processing method and system for focused microplates based on machine learning. Background Technology

[0002] Microplates are widely used as a common experimental tool in biology, medicine, and drug development. They can hold multiple independent microwells, each capable of culturing cells, storing reagents, or conducting various biochemical reactions. With technological advancements, high-throughput imaging technology has been introduced into microplate processing, enabling rapid simultaneous imaging of a large number of microwells and generating high-throughput image data containing a wealth of experimental data. However, assessing the quality of high-throughput image data, especially during automated processing, remains a significant challenge.

[0003] Traditional microplate quality inspection relies on manual visual inspection or simple image processing algorithms. These methods are not only inefficient, but also prone to omissions or misjudgments when processing high-throughput image data. Manual inspection is affected by factors such as human fatigue, experience, and subjective judgment, making it difficult to guarantee the stability and consistency of quality inspection; while simple image processing algorithms often lack effective recognition and discrimination capabilities when faced with complex image backgrounds and diverse microplate types. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a machine learning-based method and system for high-throughput data processing of focused microplates.

[0005] In a first aspect, embodiments of the present invention provide a high-throughput data processing method for focused microplates based on machine learning, applied to an image data processing system, the method comprising:

[0006] Acquire images of microplates to be processed and multiple microplate quality inspection points, wherein the images of microplates to be processed contain the type of microplate to be tested and high-throughput image data;

[0007] The initial quality inspection discrimination information is generated by combining the multiple microplate quality inspection viewpoints and the high-throughput image data of the microplate image to be processed.

[0008] The microplate quality inspection decision network is used to make a microplate quality inspection opinion decision on the initial quality inspection opinion information to determine the microplate quality inspection opinion result of the microplate image to be processed, wherein the microplate quality inspection opinion result is one of the multiple microplate quality inspection opinions;

[0009] The debugging steps for the microplate quality inspection discrimination network include:

[0010] Obtain X network training examples, where X is an integer greater than 0;

[0011] For each network training example, the type of microplate to be tested in the current network training example is obtained as the target microplate type, and the types of microplate to be tested in the remaining network training examples from X-1 network training examples are obtained as adversarial microplate types, where Y is an integer greater than 0 and less than a set number of viewpoints, and the set number of viewpoints is the maximum number of prior quality detection viewpoints in the microplate quality detection judgment result of the microplate quality detection viewpoint decision;

[0012] Based on the target microplate type, the adversarial microplate type, and high-throughput image data of each network training example, image combination is performed to generate quality inspection discrimination information examples, resulting in a quality inspection discrimination information example set. The target microplate type is the matching type corresponding to the quality inspection discrimination information example. The quality inspection discrimination information example set contains multiple quality inspection discrimination information examples, and each quality inspection discrimination information example is generated from a network training example.

[0013] The basic quality inspection discrimination network is debugged based on the aforementioned quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network.

[0014] Preferably, the step of combining the high-throughput image data based on the multiple microplate quality inspection viewpoints and the microplate image to be processed to generate initial quality inspection discrimination information includes:

[0015] The multiple microplate quality detection viewpoints are combined with the high-throughput image data of the microplate image to be processed to obtain joint discrimination information;

[0016] If the number of viewpoints for microplate quality inspection is less than the total number of viewpoints in the microplate quality inspection discrimination network, then training annotations are configured in the joint discrimination information based on the comparison result between the total number of viewpoints and the number of viewpoints for microplate quality inspection, to obtain initial quality inspection discrimination information.

[0017] Preferably, the step of combining the high-throughput image data of the plurality of microplate quality detection points with the microplate image to be processed to obtain joint discrimination information includes:

[0018] Based on the microplate reference image corresponding to each microplate quality inspection point, image mapping is performed on the multiple microplate quality inspection points to obtain multiple quality inspection mapping image data.

[0019] The multiple quality inspection mapping image data and the high-throughput image data of the microplate image to be processed are combined to obtain joint discrimination information.

[0020] Preferably, the step of using the microplate quality inspection discrimination network to make microplate quality inspection opinion decisions on the initial quality inspection discrimination information and determining the microplate quality inspection discrimination result of the microplate image to be processed includes:

[0021] The initial quality inspection discrimination information is used to make a quality inspection opinion decision on the microplate by a microplate quality inspection discrimination network, and the quality inspection opinion decision information is obtained.

[0022] If the number of decision sequences in the quality inspection opinion decision information is greater than the number of opinions in the multiple microplate quality inspection opinions, then the quality inspection opinion decision information is extracted based on the number of opinions to obtain decision query features;

[0023] Based on the priority of the multiple microplate quality inspection viewpoints in the initial quality inspection discrimination information, the microplate quality inspection viewpoint corresponding to the decision query feature is obtained as the microplate quality inspection discrimination result of the microplate image to be processed.

[0024] Preferably, for each network training example, obtaining the microplate type to be tested in the current network training example as the target microplate type and obtaining the microplate types to be tested from the remaining Y network training examples from X-1 network training examples as adversarial microplate types includes:

[0025] Select any one of the X network training examples as the current network training example;

[0026] Within the quantization constraint interval where the number of prior viewpoints generated by the basic quality inspection and discrimination network is the maximum value, an arbitrary value is determined as the number of adversarial viewpoints Y.

[0027] Based on the number Y of adversarial viewpoints, the remaining network training examples are obtained from the X-1 network training examples, excluding the Y examples.

[0028] Preferably, the step of generating quality inspection discrimination information examples based on the high-throughput image data of the target microplate type, the adversarial microplate type, and the current network training examples includes:

[0029] The target microplate type and the antagonistic microplate type are randomly combined to obtain joint discrimination information;

[0030] If the number of training examples for the target microplate type and the adversarial microplate type is less than the total number of views in the microplate quality inspection discriminant network, then training annotations are configured for the joint discriminant information based on the comparison result between the total number of views and the number of training examples.

[0031] The joint discrimination information is combined with the high-throughput image data of the current network training examples to obtain quality inspection discrimination information examples.

[0032] Preferably, before debugging the basic quality inspection discrimination network based on the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network, the method further includes: combining images based on the target microplate type, the adversarial microplate type, and the high-throughput image data of the current network training samples to generate adversarial training images and adding them to the quality inspection discrimination information sample set, wherein the adversarial microplate type is the matching type corresponding to the adversarial training image;

[0033] The step of debugging the basic quality inspection discrimination network based on the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes: debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images in the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network.

[0034] Preferably, the step of debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images from the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes:

[0035] Two quality inspection discrimination information samples and their corresponding matching types are obtained from the quality inspection discrimination information sample set;

[0036] Based on the set enhancement weights, feature enhancement is performed on the knowledge features corresponding to the two quality inspection discrimination information samples obtained, and feature enhancement is performed on the knowledge features corresponding to the two matching types. The results are used as debugging training examples.

[0037] Using the obtained debugging and training examples, the basic quality inspection discrimination network is debugged to obtain the microplate quality inspection discrimination network.

[0038] Preferably, the step of debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images from the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes:

[0039] The quality inspection discrimination information samples in the quality inspection discrimination information sample set are decomposed into multiple training sample groups;

[0040] For each training sample group, the training sample group is loaded into the first initial decision tree network and the second initial decision tree network for debugging, and the first debugging output information and the second debugging output information are obtained. The first initial decision tree network and the second initial decision tree network are both initialized through the basic quality inspection and discrimination network.

[0041] The quality inspection judgment information sample where the decision information in the first debug output information and the second debug output information are inconsistent is obtained as the quality inspection judgment information sample to be optimized;

[0042] Based on the training error of the quality inspection discrimination information samples to be optimized in the first debugging output information, a set proportion of quality inspection discrimination information samples is selected from the quality inspection discrimination information samples to be optimized as the first training set, and based on the training error of the quality inspection discrimination information samples to be optimized in the second debugging output information, the set proportion of quality inspection discrimination information samples is selected from the quality inspection discrimination information samples to be optimized as the second training set;

[0043] The network parameters of the second initial decision tree network are adjusted based on the first training set, and the network parameters of the first initial decision tree network are adjusted based on the second training set.

[0044] The first initial decision tree network and the second initial decision tree network are determined as the microplate quality inspection discrimination network.

[0045] Preferably, the step of debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images from the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes:

[0046] Based on the set of multiple quality inspection criteria, the quality inspection criteria information samples and adversarial training images in the set of quality inspection criteria information samples are derived to obtain multiple network training derived samples corresponding to each quality inspection criterion.

[0047] For each quality inspection judgment item, a set of network training derivative examples is obtained by selecting from the corresponding network training derivative examples;

[0048] Based on the network training derived sample set, the corresponding quality inspection and discrimination items are processed in the basic quality inspection and discrimination network to obtain the debugging output information of each quality inspection and discrimination item;

[0049] The network parameters of the basic quality inspection discrimination network are adjusted based on the debugging output information of each quality inspection discrimination item to obtain the microplate quality inspection discrimination network.

[0050] In a second aspect, embodiments of the present invention provide an image data processing system, including at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the method described in the first aspect.

[0051] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, the computer program implementing the method described in the first aspect when running.

[0052] In recent years, deep learning technology, especially convolutional neural networks, has achieved remarkable results in the field of image processing. Leveraging the powerful feature learning and classification capabilities of deep learning models can effectively improve the accuracy and efficiency of image processing tasks. In the field of microplate quality inspection, constructing a specially designed quality inspection discrimination network has become an innovative solution. By training deep neural network models to learn various features and patterns in microplate images, and then achieving automated quality discrimination, the accuracy and processing speed of microplate quality inspection can be greatly improved.

[0053] Against this backdrop, this invention proposes an automated microplate quality detection method based on deep learning. The method first acquires images of the microplates to be processed, along with multiple predefined microplate quality detection viewpoints covering common microplate quality problems and discrimination criteria. Then, by combining high-throughput image data and multiple quality detection viewpoints, initial quality inspection discrimination information is generated. This information is then fed into a specially designed microplate quality inspection discrimination network for processing. This network, by learning from a large amount of labeled microplate image data, can effectively perform deep analysis and decision-making on the initial quality inspection discrimination information, ultimately outputting accurate microplate quality detection viewpoints as the discrimination results. This process achieves automated microplate quality assessment, significantly improving work efficiency and discrimination accuracy, and providing strong technical support for experimental research in the biomedical field.

[0054] The beneficial effects of this invention are mainly reflected in the following aspects:

[0055] Efficiency and Accuracy: By acquiring images of the microplate to be processed and multiple microplate quality inspection points in step 110, this invention can fully utilize high-throughput image data to ensure comprehensive capture and analysis of the microplate type under test. Combining multiple quality inspection points allows for a preliminary assessment of microplate quality from multiple dimensions, providing a comprehensive and accurate information foundation for subsequent decision-making.

[0056] Information Fusion and Optimization: In step 120, this invention combines multiple microplate quality inspection perspectives and high-throughput image data to generate initial quality inspection discrimination information. This step achieves effective fusion of information from different sources, improves the richness and reliability of quality inspection discrimination information, and helps reduce the bias or misjudgment that may be caused by a single perspective.

[0057] Intelligent Decision Making: Through the microplate quality inspection discrimination network in step 130, this invention can make intelligent decisions on the initial quality inspection discrimination information and automatically determine the quality inspection discrimination result of the microplate image to be processed. This network is specially trained and can accurately identify and process complex quality inspection situations, greatly improving the automation and intelligence level of the quality inspection process;

[0058] Reliability and Practicality of Results: The final microplate quality control results are derived from a comprehensive evaluation of multiple quality inspection perspectives, ensuring the reliability and practicality of the results. These results can be directly applied to experimental or production processes, providing clear guidance for researchers and production personnel, and helping to improve the efficiency and quality of experiments or production.

[0059] In summary, this invention achieves automation, intelligence, and accuracy in microplate quality inspection by efficiently acquiring and processing microplate image data and combining multiple quality inspection perspectives for intelligent decision-making, providing strong support for research and production in related fields. Attached Figure Description

[0060] Figure 1 This is a flowchart of a machine learning-based high-throughput data processing method for focused microplates, provided in an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of the structure of an image data processing system 200 provided in an embodiment of the present invention. Detailed Implementation

[0062] To better understand the above technical solutions, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solutions of the present invention, rather than limitations on the technical solutions of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0063] Figure 1 A high-throughput data processing method for focusing microplates based on machine learning is shown, which is applied to an image data processing system. The method includes the following steps 110-130.

[0064] Step 110: Obtain the microplate image to be processed and multiple microplate quality inspection points. The microplate image to be processed contains the type of microplate to be tested and high-throughput image data.

[0065] Step 120: Combine the high-throughput image data of the multiple microplate quality inspection viewpoints and the microplate images to be processed to generate initial quality inspection discrimination information.

[0066] Step 130: The microplate quality inspection opinion decision is made on the initial quality inspection opinion information through the microplate quality inspection discrimination network to determine the microplate quality inspection discrimination result of the microplate image to be processed, wherein the microplate quality inspection discrimination result is one of the multiple microplate quality inspection opinions.

[0067] In the field of biomedical research, microplates are a commonly used experimental tool for storing and processing small amounts of biological samples. To ensure the accuracy of experimental results, quality control of microplates is crucial. The following is a specific application scenario that details how image data processing systems can be used to perform quality control of microplates.

[0068] First, the image data processing system acquires images of the microplate to be processed. This step typically involves capturing images of the microplate using a high-resolution camera or scanner. These images contain information about the type of microplate under test, such as the number, layout, and size of the microwells, as well as high-throughput image data, i.e., detailed visual information about the sample in each microwell.

[0069] Next, the image data processing system receives multiple microplate quality inspection views. These views may come from different quality inspection algorithms, empirical rules from previous experiments, or expert knowledge input. Each view provides an assessment of a certain aspect of the microplate quality, such as the clarity of the wells, the uniformity of the sample, and the presence of air bubbles or impurities.

[0070] Then, the image data processing system performs a combined analysis based on these quality inspection perspectives and high-throughput image data. The system may employ advanced image processing techniques, such as feature extraction, image segmentation, and pattern recognition, to integrate information from various perspectives with the raw image data. Through this combination, the system generates initial quality inspection criteria, serving as a preliminary assessment of the microplate's quality.

[0071] Finally, the image data processing system uses a specially trained microplate quality inspection discrimination network to make decisions based on the initial quality inspection information. This network, likely a deep learning model trained on extensive historical data and expert knowledge, is capable of accurately predicting the final quality inspection result based on the initial discrimination information. The system determines the microplate quality inspection discrimination result for the microplate image to be processed. This result is either the most representative viewpoint selected from multiple quality inspection perspectives, or a new conclusion derived by combining various viewpoints.

[0072] In this process, the image data processing system not only automates complex image processing and quality inspection tasks, but also improves the accuracy and efficiency of detection through machine learning. For biomedical research, this translates to more reliable experimental results and a more efficient research workflow.

[0073] Furthermore, the following is a more detailed description of how the aforementioned image data processing system performs microplate quality inspection.

[0074] The first step in an image data processing system is acquiring clear, high-resolution images of the microplate to be processed. This typically involves using specialized image acquisition equipment, such as a high-resolution camera or scanner, to precisely capture the microplate. To ensure image quality, the acquisition equipment may need to be calibrated, and appropriate parameters such as exposure time, focal length, and light source may need to be set. During image acquisition, the system also records microplate type information, including the size of the microplate, the number of wells, and their arrangement. This information is crucial for subsequent quality inspection, as it helps the system accurately identify and locate each well.

[0075] Once images of the microplate to be processed are acquired, the image data processing system receives multiple quality inspection perspectives. These perspectives may originate from different quality inspection algorithms, historical experimental data, expert knowledge bases, or user input. Each quality inspection perspective provides a unique insight into a specific aspect of the microplate's quality. For example, some perspectives may focus on the sharpness of the microwells, assessing their quality by analyzing the edge sharpness and contrast of the microwells in the image; others may focus on the uniformity of the samples, determining whether they are evenly distributed by measuring the color, brightness, or texture of the samples in each microwell; still others may focus on the presence of air bubbles, impurities, or other contaminants in the microplate.

[0076] After receiving multiple quality inspection viewpoints, the image data processing system uses advanced image processing and analysis techniques to combine this information. This may include algorithms and techniques such as feature extraction, image segmentation, pattern recognition, and machine learning. The system first processes and analyzes each quality inspection viewpoint individually, extracting key features and parameters related to the microplate's quality. Then, it fuses these features and parameters to generate an initial quality inspection judgment. This information is a preliminary judgment that integrates the evaluation results of various viewpoints; it may include the overall quality score of the microplate, the location of problematic microwells, and potential quality issues.

[0077] Finally, the image data processing system uses a specially trained microplate quality inspection discrimination network to make decisions based on the initial quality inspection information. This network is a deep learning model that has been trained and optimized based on a large amount of historical data and expert knowledge, and can accurately predict the final quality inspection result based on the initial discrimination information.

[0078] During the decision-making process, the microplate quality control network comprehensively considers various features and parameters in the initial quality control information, as well as their correlations and influences. Through complex calculations and reasoning, the network outputs a final microplate quality control result. This result is a concrete and interpretable conclusion that indicates the quality status of the microplate, existing problems, and possible solutions. Furthermore, this result can serve as a reference for subsequent experiments or research, helping researchers better understand and control experimental conditions.

[0079] In some other embodiments, the following is a more detailed description of how the image data processing system performs microplate quality inspection.

[0080] I. Image Acquisition and Preprocessing

[0081] 1. Image Acquisition

[0082] The primary task of an image data processing system is to acquire high-quality images of microplates. This typically involves specialized image acquisition equipment, such as high-resolution, high-sensitivity cameras, along with appropriate light sources and optical components, to ensure that every detail within the microplate is captured.

[0083] 2. Image Preprocessing

[0084] Before an image is further analyzed by the system, a series of preprocessing steps are usually required, including denoising, enhancement, and correction, in order to improve image quality and reduce the complexity of subsequent processing.

[0085] II. Integration of Multi-Source Quality Inspection Perspectives

[0086] 1. Source of viewpoints

[0087] Quality inspection perspectives can come from multiple sources, such as traditional algorithms based on image features, deep learning models, expert knowledge bases, historical data, and manual input from users.

[0088] 2. Integration of viewpoints

[0089] The system needs to effectively integrate these quality inspection perspectives from different sources. This may involve complex data processing and analysis processes such as weight allocation, feature fusion, and decision tree construction.

[0090] III. Generation of Initial Quality Inspection Judgment Information

[0091] 1. Feature Extraction

[0092] Key features related to the quality of the microplate, such as the shape, size, edge sharpness, and internal texture of the micropores, are extracted from the preprocessed images.

[0093] 2. Information fusion

[0094] The extracted features are fused with multi-source quality detection perspectives to generate an initial quality inspection discrimination information rich in information. This information not only includes the features of the image itself, but also incorporates quality assessment results from different sources.

[0095] IV. Decision Process of Microplate Quality Inspection Discriminant Network

[0096] 1. Network Structure

[0097] The quality control network for microplates is typically a complex deep learning model, such as a convolutional neural network (CNN) or a deep learning classifier. These models are capable of processing a large number of input features and outputting accurate prediction results.

[0098] 2. Training and Optimization

[0099] The network needs to be trained using a large amount of historical data, and its parameters need to be continuously adjusted through optimization algorithms (such as gradient descent) to improve the accuracy and stability of predictions. During the training process, common problems such as overfitting and underfitting also need to be considered, and corresponding measures should be taken to prevent and deal with them.

[0100] 3. Decision Output

[0101] Once the network is trained and reaches satisfactory performance, it can perform quality inspection on new microplate images. After inputting initial quality inspection information, the network undergoes a series of calculations and inferences, ultimately outputting a specific and interpretable quality inspection result. This result may include the overall quality score of the microplate, the specific location of the problematic microwell, the possible types of quality problems, and suggested follow-up actions.

[0102] V. Subsequent Applications and Feedback

[0103] 1. Application of Results

[0104] The quality inspection results can be directly applied to the experimental or production process, helping researchers or production personnel understand the quality status of the microplate and thus take corresponding measures for adjustment or improvement. For example, if the test results show that a microplate is clogged or contaminated, the microplate can be cleaned or replaced in a timely manner.

[0105] 2. Feedback and Improvement

[0106] In practical applications, it is also necessary to continuously collect user feedback on quality inspection results and evaluate the system's performance through actual performance during experiments or production processes. These feedback and evaluation results can be used to further improve and optimize the image data processing system, enhancing its accuracy and reliability. Simultaneously, new data and knowledge can be continuously added to the system, enabling it to adapt to more diverse application scenarios and needs.

[0107] Based on the above, the following is a glossary of relevant technical terms.

[0108] Images of microplates awaiting processing: These are images of microplates that require quality control via image data processing systems. These images typically contain information about the microplate and the samples on it, representing visual data captured during experiments or testing. For example, in biomedical research, researchers may use cameras or scanners to capture images of biological samples in microplates for subsequent analysis and processing. These captured images are the microplate images awaiting processing; they contain a wealth of information about the microplate and the samples, such as the shape, size, and location of the wells, as well as the color and texture of the samples.

[0109] Microplate quality assessment perspectives refer to the evaluations or judgments of microplate quality derived from different angles or based on different methods during microplate quality assessment. These perspectives may originate from various quality assessment algorithms, expert knowledge, historical data, or user input. For example, one perspective might focus on the clarity of the microwells, evaluating the sharpness and contrast of the microwell edges using image analysis algorithms; another perspective might focus on the uniformity of the samples, determining whether the distribution is uniform by measuring changes in the color and brightness of the samples. These perspectives collectively constitute a comprehensive assessment of microplate quality.

[0110] Microplate type to be tested: This refers to the type or specification of the microplate to be tested. A microplate is a commonly used laboratory tool for storing and processing small amounts of biological samples or other liquids. Different types of microplates have different characteristics such as the number of wells, pore size, and well spacing to suit different experimental needs. For example, 96-well plates and 384-well plates are two common types of microplates, with 96 and 384 wells respectively, for storing corresponding numbers of samples. When performing quality testing, it is necessary to identify the type of microplate to be tested in order to select the appropriate testing method and parameters.

[0111] High-throughput image data refers to large amounts of image data generated in a short period of time, typically used to describe scenarios requiring rapid and efficient image processing and analysis. In microplate quality inspection, high-throughput image data may originate from automated image acquisition systems capable of capturing and processing large numbers of microplate images in a short time. This image data contains rich information, such as the shape, size, and location of the microwells, as well as the color and texture of the samples, which can be used for subsequent image analysis and quality assessment. Processing high-throughput image data requires efficient algorithms and powerful computing capabilities to ensure accuracy and real-time performance.

[0112] Initial quality inspection discrimination information refers to the preliminary assessment of the quality status of microplates generated after initial analysis and processing based on the acquired microplate images and multiple quality inspection perspectives during the quality inspection of microplates by the image data processing system. This information may include key features extracted from the image, the evaluation results of each quality inspection perspective, and the correlation and influence between them. Initial quality inspection discrimination information provides important reference for subsequent quality inspection decisions.

[0113] Microplate quality control discriminant network: This is a deep learning model or network structure specifically designed for quality inspection of microplate images. It is typically a complex network composed of multiple neuron layers, capable of automatically extracting key information from images through learning and training based on initial quality control information or other relevant features, and outputting the final quality judgment result for the microplate. This network requires training and optimization with a large amount of historical data to improve its prediction accuracy and stability.

[0114] Microplate quality inspection opinion decision-making refers to the process in an image data processing system where, based on initial quality inspection information and the capabilities of the microplate quality inspection network, multiple quality inspection opinions are comprehensively evaluated and selected to ultimately determine the most representative or optimal opinion as the basis for judging microplate quality. This process may involve various methods and techniques such as weight allocation, decision tree construction, and voting mechanisms to ensure the accuracy and reliability of the output quality inspection results.

[0115] Microplate quality inspection results refer to the final conclusion or judgment output by the image data processing system after quality inspection of the microplate image. This result is typically a specific and interpretable quality assessment, which may include the overall quality score of the microplate, the specific location of the problematic microwell, the possible type of quality problem, and suggested follow-up actions. This result can be directly applied to experimental or production processes, helping researchers or production personnel understand the quality status of the microplate and take corresponding measures for adjustment or improvement. Simultaneously, this result can also serve as a reference for subsequent experiments or research, helping researchers better understand and control experimental conditions.

[0116] The beneficial effects of this invention are mainly reflected in the following aspects:

[0117] Efficiency and Accuracy: By acquiring images of the microplate to be processed and multiple microplate quality inspection points in step 110, this invention can fully utilize high-throughput image data to ensure comprehensive capture and analysis of the microplate type under test. Combining multiple quality inspection points allows for a preliminary assessment of microplate quality from multiple dimensions, providing a comprehensive and accurate information foundation for subsequent decision-making.

[0118] Information Fusion and Optimization: In step 120, this invention combines multiple microplate quality inspection perspectives and high-throughput image data to generate initial quality inspection discrimination information. This step achieves effective fusion of information from different sources, improves the richness and reliability of quality inspection discrimination information, and helps reduce the bias or misjudgment that may be caused by a single perspective.

[0119] Intelligent Decision Making: Through the microplate quality inspection discrimination network in step 130, this invention can make intelligent decisions on the initial quality inspection discrimination information and automatically determine the quality inspection discrimination result of the microplate image to be processed. This network is specially trained and can accurately identify and process complex quality inspection situations, greatly improving the automation and intelligence level of the quality inspection process;

[0120] Reliability and Practicality of Results: The final microplate quality control results are derived from a comprehensive evaluation of multiple quality inspection perspectives, ensuring the reliability and practicality of the results. These results can be directly applied to experimental or production processes, providing clear guidance for researchers and production personnel, and helping to improve the efficiency and quality of experiments or production.

[0121] In summary, this invention achieves automation, intelligence, and accuracy in microplate quality inspection by efficiently acquiring and processing microplate image data and combining multiple quality inspection perspectives for intelligent decision-making, providing strong support for research and production in related fields.

[0122] In the above embodiments, the debugging steps of the microplate quality inspection discrimination network include steps 210-240.

[0123] Step 210: Obtain X network training examples, where X is an integer greater than 0.

[0124] Step 220: For each network training example, obtain the microplate type to be tested in the current network training example as the target microplate type, and obtain the microplate types to be tested from the remaining Y network training examples from X-1 network training examples as adversarial microplate types, where Y is an integer greater than 0 and less than the set number of viewpoints, where the set number of viewpoints is the maximum number of prior quality detection viewpoints in the microplate quality detection judgment result of the microplate quality detection viewpoint decision.

[0125] Step 230: Generate quality inspection discrimination information examples by combining images based on the target microplate type, the adversarial microplate type, and high-throughput image data of each network training example, thereby obtaining a quality inspection discrimination information example set. The target microplate type is the matching type corresponding to the quality inspection discrimination information example. The quality inspection discrimination information example set contains multiple quality inspection discrimination information examples, and each quality inspection discrimination information example is generated from a network training example.

[0126] Step 240: Debug the basic quality inspection discrimination network based on the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network.

[0127] In biomedical laboratories, researchers frequently use microplates to store and process biological samples. To ensure the accuracy and reliability of experimental results, quality control of the microplates is crucial. Image data processing systems play a key role in this process.

[0128] First, the image data processing system acquires X network training examples. These examples are selected from historical laboratory data and contain images of various types and qualities of microplates. Each example includes image data of the microplate, the corresponding type of microplate to be tested, and high-throughput image data.

[0129] For each network training example, the system uses its target microplate type as the primary microplate type. Simultaneously, the system randomly selects Y microplate types from the remaining X-1 network training examples as adversarial microplate types. These adversarial types are used to enhance the network's discriminative ability during training, enabling it to more accurately distinguish between different types of microplates.

[0130] Next, the system performs image combination processing based on the target microplate type, adversarial microplate type, and high-throughput image data for each network training example. Through this step, the system generates quality control discriminant information examples and combines these examples into a quality control discriminant information example set. Each quality control discriminant information example contains the image data used to train the network and the corresponding label information.

[0131] Finally, the system uses a set of quality inspection discrimination information samples to debug and train the basic quality inspection discrimination network. During training, the network continuously learns and adjusts its parameters to more accurately identify and process various microplate images. Through multiple iterations and optimizations, a trained microplate quality inspection discrimination network is finally obtained. This network can automatically perform quality inspection on input microplate images and output the corresponding discrimination results.

[0132] Through such application scenario examples, it is clear how the above technical solutions are implemented in image data processing systems. By acquiring network training examples, determining the target microplate type and the adversarial microplate type, generating quality inspection and discrimination information examples, and debugging the basic quality inspection and discrimination network, the system can achieve efficient and accurate detection of microplate quality, providing strong support for research in biomedical laboratories.

[0133] Furthermore, in image data processing systems, the debugging of the microplate quality inspection discrimination network is a crucial step, determining the accuracy and reliability of the final network model. The debugging process mainly involves steps such as setting network parameters, preparing training data, monitoring the training process, and evaluating model performance.

[0134] 1. Network parameter settings

[0135] Before debugging, some basic network parameters need to be set, such as the learning rate, batch size, and number of iterations. These parameters directly affect the network's training speed and convergence. The learning rate determines the step size of weight updates in each iteration, the batch size determines the number of samples used in each training iteration, and the number of iterations determines the length of the entire training process. These parameters need to be adjusted based on the specific training data and hardware resources to find the optimal training configuration.

[0136] 2. Training Data Preparation

[0137] Training data is the foundation of network training, and its quality and quantity directly determine the network's performance. Preparing training data requires selecting representative samples from a large number of raw images and performing necessary preprocessing operations, such as image cropping and normalization. Furthermore, each sample needs to be labeled with appropriate information based on its specific characteristics so that the network can learn the correct classification rules. In this scenario, the training data is a set of quality control discrimination information samples, containing multiple quality control discrimination information samples, each corresponding to a network training sample.

[0138] 3. Training process monitoring

[0139] During training, it's necessary to monitor the network's training status in real time, including changes in the loss function and improvements in accuracy. These metrics help determine if the network is in a normal training state and whether adjustments to the training strategy are needed. For example, if the loss function doesn't decrease significantly over a period of time, or if the accuracy starts to fluctuate after reaching a certain level, then measures such as reducing the learning rate or increasing the number of iterations should be considered to improve training effectiveness.

[0140] 4. Model Performance Evaluation

[0141] Once the network is trained, its performance needs to be evaluated to verify its ability to accurately identify and process various microplate images. Evaluation typically uses a set of independent test data that was not used during training. By comparing the network's performance on the test data with the true labels, metrics such as precision and recall can be calculated to assess its performance. If the network is found to perform poorly in certain aspects, it is necessary to return to the training phase and further improve its performance by adjusting the network structure or optimizing the training strategy.

[0142] In summary, the debugging of a microplate quality inspection discrimination network is a complex and delicate process that requires comprehensive consideration of multiple factors. Through reasonable parameter settings, sufficient training data preparation, real-time monitoring of the training process, and objective model performance evaluation, the network's performance can be gradually optimized and improved, ultimately achieving efficient and accurate detection of microplate quality.

[0143] In some possible embodiments, the high-throughput image data described in step 120, based on the multiple microplate quality inspection viewpoints and the microplate images to be processed, is combined to generate initial quality inspection discrimination information, including steps 121-122.

[0144] Step 121: Combine the high-throughput image data of the multiple microplate quality detection points with the microplate image to be processed to obtain joint discrimination information.

[0145] Step 122: If the number of viewpoints for microplate quality inspection is less than the total number of viewpoints in the microplate quality inspection discrimination network, then based on the comparison result between the total number of viewpoints and the number of viewpoints for microplate quality inspection, training annotations are configured in the joint discrimination information to obtain initial quality inspection discrimination information.

[0146] In some embodiments, during the microplate quality inspection process, the image data processing system pays particular attention to effectively combining multiple microplate quality inspection perspectives with high-throughput image data of the microplate image to be processed, in order to generate more valuable initial quality inspection discrimination information. This process is specifically divided into steps 121 and 122.

[0147] Step 121: Image combination to generate joint discriminant information

[0148] In this step, the system first performs image combination with high-throughput image data of the microplate quality inspection points and the images of the microplate to be processed. This "image combination" is not a simple image overlay or stitching, but rather a fusion of image features (such as brightness, contrast, and edge sharpness) of interest to each inspection point with corresponding regions in the high-throughput image data using specific algorithms and techniques. This results in a joint discriminant information that incorporates multiple aspects of quality information. This joint discriminant information not only preserves the detailed information in the original image data but also incorporates the professional judgments of multiple quality inspection points, providing a more comprehensive and accurate basis for subsequent quality inspection.

[0149] Step 122: Configure training annotations to obtain initial quality control discrimination information.

[0150] After image assemblage, the system further checks whether the number of viewpoints currently used for microplate quality detection is less than the total number of viewpoints supported by the microplate quality control discriminant network. If this is the case, the system configures corresponding training annotations into the joint discriminant information based on the comparison between the total number of viewpoints and the number of currently used viewpoints. These training annotations can be additional annotations on the quality status of certain specific regions or supplementary explanations of the overall quality level. Their purpose is to help the quality control discriminant network better understand and utilize this joint discriminant information during training. After configuring the training annotations, the joint discriminant information is transformed into initial quality control discriminant information, which will serve as an important basis for subsequent quality control decisions.

[0151] It's important to note that "total number of viewpoints" refers to the total number of possible quality inspection viewpoints considered during the design of the microplate quality inspection discriminant network, while "number of currently used viewpoints" refers to the number of viewpoints actually used in this quality inspection. This difference may stem from various reasons, such as different experimental requirements, data acquisition limitations, or adjustments to the quality inspection strategy. By configuring training annotations into the joint discriminant information, the system can effectively compensate for information gaps or biases caused by this difference, thereby improving the accuracy and reliability of the initial quality inspection discriminant information.

[0152] In summary, through the meticulous processing in steps 121 and 122, the image data processing system can fully leverage the information advantages of multiple microplate quality inspection perspectives and high-throughput image data to generate more valuable initial quality inspection judgment information. This information not only provides strong support for subsequent quality inspection decisions but also lays the foundation for improving the accuracy and efficiency of microplate quality inspection.

[0153] In some other possible embodiments, the step 121, which describes combining the high-throughput image data of the plurality of microplate quality inspection viewpoints with the microplate image to be processed to obtain joint discrimination information, includes: mapping the plurality of microplate quality inspection viewpoints to obtain a plurality of quality inspection mapping image data based on the microplate reference image corresponding to each microplate quality inspection viewpoint; and combining the plurality of quality inspection mapping image data with the high-throughput image data of the microplate image to be processed to obtain joint discrimination information.

[0154] In other embodiments, the image data processing system employs a more refined image combination method when performing step 121 to generate joint discriminant information. This method involves image mapping for each microplate quality inspection viewpoint and combining it with high-throughput image data of the microplate images to be processed. Detailed examples are provided below.

[0155] First, the system acquires a microplate reference image corresponding to each microplate quality inspection viewpoint. These reference images are predefined and used to guide the mapping operation of each inspection viewpoint during the image combination process. Each reference image contains image features and information related to a specific quality inspection viewpoint, such as specific brightness distribution, color mode, or texture structure.

[0156] Next, the system will perform image mapping on multiple microplate quality inspection viewpoints based on these microplate reference images. "Image mapping" refers to mapping the image features or information relevant to each quality inspection viewpoint onto the high-throughput image data of the microplate images to be processed, guided by the reference images. During this process, the system may employ various image processing techniques, such as feature extraction, image transformation, and filters, to ensure the accuracy and effectiveness of the mapping.

[0157] Through image mapping, the system generates multiple quality inspection mapping image data sets. Each quality inspection mapping image data set reflects the specific performance or impact of the corresponding quality inspection perspective on the microplate image to be processed. These mapped image data sets not only retain the information of the original inspection perspective but also effectively integrate it with the high-throughput data of the image to be processed.

[0158] Finally, the system combines these multiple quality inspection mapping image data with the high-throughput image data of the microplate images to be processed. In this step, the system may use weighted averaging, image overlay, or other complex image fusion algorithms to ensure that the combined joint discrimination information can comprehensively reflect the influence of each quality inspection perspective and highlight key features related to microplate quality.

[0159] In summary, through these detailed image processing and combination steps, the image data processing system can generate joint discrimination information containing multifaceted quality information. This information not only provides a comprehensive and accurate basis for subsequent quality inspection decisions but also helps improve the accuracy and efficiency of microplate quality inspection.

[0160] Under some preferred design approaches, step 130, which describes using a microplate quality inspection discrimination network to make microplate quality inspection viewpoint decisions on the initial quality inspection discrimination information and determine the microplate quality inspection discrimination result of the microplate image to be processed, includes: making microplate quality inspection viewpoint decisions on the initial quality inspection discrimination information through a microplate quality inspection discrimination network to obtain quality inspection viewpoint decision information; if the number of decision sequences in the quality inspection viewpoint decision information is greater than the number of viewpoints of the multiple microplate quality inspection viewpoints, then extracting the quality inspection viewpoint decision information based on the number of viewpoints to obtain decision query features; and obtaining the microplate quality inspection viewpoint corresponding to the decision query features as the microplate quality inspection discrimination result of the microplate image to be processed, according to the priority of the multiple microplate quality inspection viewpoints in the initial quality inspection discrimination information.

[0161] Under certain preferred design approaches, image data processing systems employ a specific decision-making process to determine the quality inspection result of the microplate image during microplate quality inspection. This process involves processing the initial quality inspection information through a microplate quality inspection discrimination network and deriving the final discrimination result based on a series of decision rules. The following is a detailed explanation with examples.

[0162] First, the system utilizes a microplate quality inspection discrimination network to make decisions on microplate quality inspection viewpoints based on the initial quality inspection discrimination information. In this step, the network analyzes and evaluates various quality characteristics contained in the initial discrimination information according to its internal preset algorithms and models, thereby generating quality inspection viewpoint decision information. This decision information reflects the network's judgment and preference for various quality inspection viewpoints.

[0163] Next, the system checks whether the number of decision sequences in the generated quality inspection viewpoint decision information is greater than the actual number of viewpoints used for multiple microplate quality inspections. The number of decision sequences refers to the number of judgments or choices generated by the network during the decision-making process, while the number of viewpoints refers to the number of quality inspection viewpoints actually involved in the decision-making process. If the number of decision sequences is greater than the number of viewpoints, it indicates that the network has generated additional decision outputs, which may be due to the complexity of the network design or the redundancy of data processing.

[0164] In this scenario, the system extracts decision information from quality inspection viewpoints based on the number of viewpoints actually used, resulting in decision query features. The purpose of this step is to filter out the portions of decision outputs that correspond to the actual quality inspection viewpoints for subsequent processing and judgment. The extraction method can be determined based on specific implementation requirements and network design; for example, rule-based filtering or feature selection algorithms can be used.

[0165] Finally, the system will determine the microplate quality inspection result corresponding to the decision query feature based on the priority of multiple microplate quality inspection viewpoints in the initial quality inspection discrimination information. Priority is the importance ranking assigned to each quality inspection viewpoint during the quality inspection process; it reflects the weight and influence of different viewpoints in the decision. The system will select the most important and representative quality inspection viewpoint from the decision query feature according to this priority order as the final quality inspection discrimination result.

[0166] In summary, through this optimized design approach and implementation steps, the image data processing system can more accurately and efficiently determine the quality inspection results of the microplate images to be processed. This not only improves the accuracy and reliability of microplate quality inspection but also provides strong support for subsequent experimental operations and data analysis.

[0167] Furthermore, regarding the core inventive point of determining the microplate quality inspection judgment result of the microplate image to be processed, the previous explanation can be further refined to provide a more detailed and in-depth understanding.

[0168] When determining the microplate quality inspection judgment result of the microplate image to be processed, the system execution process can be broken down into the following key steps.

[0169] Initializing the microplate quality inspection discriminant network: The system first initializes a pre-trained microplate quality inspection discriminant network. This network is trained based on a large number of microplate images and their corresponding quality labels, and it has the ability to perform quality detection on microplate images. During initialization, the network loads predefined network structures, weight parameters, and related configuration information.

[0170] Input Initial Quality Inspection Discriminant Information: Next, the system will pass the previously generated initial quality inspection discriminant information as input data to the microplate quality inspection discriminant network. This initial quality inspection discriminant information contains image features extracted from multiple quality inspection perspectives and related training annotations, providing necessary information for the network's decision-making.

[0171] Quality Inspection Perspective Decision-Making: Within the microplate quality inspection discrimination network, each quality inspection perspective is treated as an independent decision factor. The network analyzes the initial quality inspection discrimination information point by point based on the weights and interrelationships of these decision factors. Specifically, the network extracts feature maps related to each quality inspection perspective through a series of operations such as convolution, pooling, and fully connected layers, and makes decisions based on these feature maps. These decisions reflect the network's support level or confidence level for different quality inspection perspectives.

[0172] Post-processing of decision information: After obtaining the quality inspection viewpoint decision information, the system further processes this information. If the number of decision information sequences exceeds the number of quality inspection viewpoints actually used, the system will employ certain strategies for extraction or filtering. This may be because some decision information is generated due to network redundancy design or errors in the data processing process and needs to be filtered out. Extraction strategies may include threshold-based filtering, ranking selection based on feature importance, etc. Finally, the system will retain the portion of decision information corresponding to the actual quality inspection viewpoints used, forming decision query features.

[0173] Priority-based decision determination: After determining the decision query features, the system determines the final decision based on the priorities of multiple microplate quality inspection viewpoints set in the initial quality inspection judgment information. These priorities may be based on experience, experimental requirements, or statistical data analysis, reflecting the importance of different quality inspection viewpoints in the judgment process. The system weights and combines or selectively adopts the decision query features according to their priority order to determine the final microplate quality inspection judgment result. This process can be compared to obtaining a final decision based on the advice of multiple experts, taking into account each expert's authority and area of ​​expertise.

[0174] Through the detailed execution process and explanation described above, it becomes clearer how the system determines the microplate quality inspection result of the microplate image to be processed, revealing the implementation details and internal logic of this core invention. This not only provides strong assurance for the accuracy and reliability of microplate quality inspection but also offers new ideas and inspiration for technological innovation and development in related fields.

[0175] In an alternative embodiment, step 220, which describes obtaining the microplate type to be tested for each network training example as the target microplate type and obtaining the microplate types to be tested from the remaining Y network training examples from the X-1 network training examples as adversarial microplate types, includes: obtaining any one of the X network training examples as the current network training example; determining any value as the number of adversarial viewpoints Y within the quantization constraint interval where the number of prior viewpoints generated by the basic quality control discriminant network is the maximum value; and based on the number of adversarial viewpoints Y, obtaining the remaining network training examples (excluding Y) from the X-1 network training examples.

[0176] In an alternative embodiment, when the system performs step 220, it selects and processes network training examples in a specific manner to provide effective training data for subsequent microplate type detection. The following is a detailed explanation with examples.

[0177] First, the system randomly selects one of X network training examples as the current network training example. This selection is random or based on some preset rules to ensure that each example has a chance to be selected as the current training object.

[0178] Next, the system considers the number of prior viewpoints generated by the basic quality control network and uses this as the maximum value to define a quantization constraint interval. The number of prior viewpoints is derived from the network's previous training experience and knowledge, reflecting the network's ability and accuracy in detecting microplate types. Within this quantization constraint interval, the system determines an arbitrary value as the number of adversarial viewpoints, Y. The determination of the number of adversarial viewpoints Y can be random or calculated according to a certain algorithm or strategy; its purpose is to provide a certain number of negative examples for subsequent adversarial training.

[0179] Then, based on the number Y adversarial viewpoints, the system selects Y examples from the remaining X-1 network training examples as adversarial network training examples. The selection of these adversarial network training examples can be random or based on some similarity or difference metric to ensure that they are challenging and adversarial on the target microplate type of the current network training examples.

[0180] It's important to note that the "adversarial" aspect here doesn't refer to the examples themselves being incorrect or of poor quality. Rather, it refers to their differences or challenges in microplate type compared to the current network training examples, which can be used to enhance the network's generalization ability and robustness. Through adversarial training, the network can better learn and distinguish different types of microplates, thereby improving detection accuracy and reliability in practical applications.

[0181] Finally, the remaining X-1-Y network training examples are considered as the remaining network training examples. They may not be directly used for adversarial training in the current training iteration, but can be used for other training purposes, such as supplementary training and validation.

[0182] In summary, through this alternative implementation method, the system can more flexibly and effectively select and process network training examples, providing richer and more diverse training data support for microplate type detection.

[0183] Furthermore, the step of generating quality inspection discrimination information examples based on the high-throughput image data of the target microplate type, the adversarial microplate type, and the current network training examples includes: randomly combining the target microplate type and the adversarial microplate type to obtain joint discrimination information; if the number of training examples of the target microplate type and the adversarial microplate type is less than the total number of views in the microplate quality inspection discrimination network, then configuring training annotations to the joint discrimination information based on the comparison result of the total number of views and the number of training examples; and combining the joint discrimination information with the high-throughput image data of the current network training examples to obtain quality inspection discrimination information examples.

[0184] Furthermore, the system executes a series of detailed steps during the generation of quality inspection and judgment information samples. The following is a detailed explanation of these steps.

[0185] First, the system randomly combines target microplate types with adversarial microplate types. This random combination aims to create a hybrid and challenging joint discriminative information. This joint discriminative information integrates features from both target and adversarial types, enabling the network to learn more complex and subtle differences during training.

[0186] Next, the system will check whether the number of training examples for the target microplate type and the adversarial microplate type is less than the total number of viewpoints in the microplate quality inspection discrimination network. The total number of viewpoints refers to the total number of different quality inspection viewpoints that the network is designed to handle. If the number of training examples is less than the total number of viewpoints, it means that the currently provided examples are insufficient to cover all possible judgment angles of the network.

[0187] In this scenario, the system configures training annotations on the joint discriminative information based on the comparison between the total number of viewpoints and the number of training examples. Training annotations are additional descriptions or labels for the joint discriminative information, used to guide the network to focus on specific details or features during training. These annotations may include indicating which viewpoints are missing, which viewpoints are easily confused, and which viewpoints require special attention.

[0188] Finally, the system combines the joint discriminant information with the high-throughput image data of the current network training examples. This combination integrates the joint discriminant information (containing features of target and adversarial types as well as training annotations) with the high-throughput image data (containing detailed visual information of the microplates) to form a complete quality inspection discriminant information example. This example not only includes image information of the microplates but also related quality inspection viewpoints and annotation information, providing comprehensive and detailed data support for network training.

[0189] Through these steps, the system can generate challenging and diverse quality inspection discrimination information samples, thereby improving the training effect and accuracy of the microplate quality inspection discrimination network.

[0190] In some alternative embodiments, before debugging the basic quality inspection discriminant network based on the quality inspection discriminant information sample set to obtain the microplate quality inspection discriminant network, the method further includes: combining images based on the target microplate type, the adversarial microplate type, and the high-throughput image data of the current network training samples to generate adversarial training images and adding them to the quality inspection discriminant information sample set, wherein the adversarial microplate type is the matching type corresponding to the adversarial training image. Then, debugging the basic quality inspection discriminant network based on the quality inspection discriminant information sample set to obtain the microplate quality inspection discriminant network includes: debugging the basic quality inspection discriminant network using the quality inspection discriminant information samples and adversarial training images in the quality inspection discriminant information sample set to obtain the microplate quality inspection discriminant network.

[0191] In some alternative embodiments, the system's method adds an important step—generating adversarial training images—before debugging the basic quality control network based on a set of quality control discrimination information samples. The following is a detailed explanation with examples.

[0192] First, the system performs image combination based on high-throughput image data of the target microplate type, the adversarial microplate type, and the current network training examples. The purpose of this step is to create a special type of training image, namely, an adversarial training image. In this process, the system may employ techniques such as image stitching, fusion, and overlay to combine the image features of the target and adversarial microplate types, forming a challenging new image.

[0193] The adversarial training images are generated to simulate complex situations that may occur in real-world applications, such as blurring, distortion, and occlusion in microplate images. By introducing adversarial training images, the system can enhance the basic quality inspection discrimination network's ability to handle such complex situations.

[0194] After generating adversarial training images, the system adds these images to the quality control and discrimination information sample set. Thus, the quality control and discrimination information sample set includes not only ordinary quality control and discrimination information samples but also adversarial training images. The label or matching type of the adversarial training images in the set is set to the adversarial microplate type to indicate their special role in network training.

[0195] Next, when the system reaches the step of debugging the basic quality inspection discrimination network based on the quality inspection discrimination information sample set, it will simultaneously utilize the quality inspection discrimination information samples and adversarial training images in the set. The debugging process may include adjusting the network weights, optimizing the structure, and setting hyperparameters, aiming to enable the network to better adapt to and handle various quality inspection tasks.

[0196] By introducing adversarial training images and incorporating them into the network's debugging process, the resulting microplate quality inspection discriminant network exhibits stronger generalization ability and robustness. In practical applications, this network can more accurately identify and judge the quality status of microplates, providing reliable support for subsequent experiments and data analysis.

[0197] In some exemplary embodiments, the step of debugging the basic quality inspection discriminant network using quality inspection discriminant information samples and adversarial training images from the quality inspection discriminant information sample set to obtain the microplate quality inspection discriminant network includes: obtaining two quality inspection discriminant information samples and their corresponding matching types from the quality inspection discriminant information sample set; performing feature enhancement on the knowledge features corresponding to the two obtained quality inspection discriminant information samples and on the knowledge features corresponding to the two matching types according to the set enhancement weights, and using the obtained result as a debugging training example; and using the obtained debugging training example to debug the basic quality inspection discriminant network to obtain the microplate quality inspection discriminant network.

[0198] In some exemplary embodiments, the system debugs the basic quality inspection discriminant network using quality inspection discriminant information samples from the quality inspection discriminant information sample set and adversarial training images to obtain the microplate quality inspection discriminant network. This process can be explained in detail below.

[0199] First, the system randomly selects two quality inspection discrimination information samples and their corresponding matching types from the quality inspection discrimination information sample set, or according to a certain strategy. These two samples may come from different microplate images or different regions of the same image, but they both contain important information about the quality of the microplate.

[0200] Next, the system performs feature enhancement on the knowledge features corresponding to the two quality inspection discrimination information examples according to preset enhancement weights. Knowledge features refer to those features extracted during network training and used to represent the quality characteristics of the microplate. Feature enhancement is a technique to increase the influence of these features during training; it can be achieved by increasing feature weights, transforming or combining features, etc. Similarly, the system also performs feature enhancement on the knowledge features corresponding to the two matching types. The purpose of this is to make the network pay more attention to features closely related to the matching type, thereby improving the network's ability to discriminate between different matching types.

[0201] After feature enhancement, the system will obtain a new debugging training example. This example contains enhanced quality inspection and discrimination information samples and matching types, as well as the correspondence between them. This example will be used to debug the basic quality inspection and discrimination network.

[0202] Finally, the system uses the obtained debugging training examples to debug the basic quality inspection discrimination network. The debugging process may include adjusting the network parameters, optimizing the network structure, and updating the network weights, so that the network can better adapt to and handle the quality inspection discrimination task. Through multiple iterations and debugging, the system will eventually obtain a high-performance microplate quality inspection discrimination network that can accurately identify and judge the quality status of microplates, providing strong support for subsequent experiments and data analysis.

[0203] In a preferred embodiment, the step of debugging the basic quality inspection discrimination network by using the quality inspection discrimination information samples and adversarial training images in the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes steps 310-360.

[0204] Step 310: Decompose the quality inspection discrimination information samples in the quality inspection discrimination information sample set into multiple training sample groups.

[0205] Step 320: For each training sample group, load the training sample group into the first initial decision tree network and the second initial decision tree network respectively for debugging, and obtain the first debugging output information and the second debugging output information. The first initial decision tree network and the second initial decision tree network are both obtained by initializing through the basic quality inspection discriminant network.

[0206] Step 330: Obtain quality inspection discrimination information samples where the decision information in the first debug output information and the second debug output information are inconsistent as quality inspection discrimination information samples to be optimized.

[0207] Step 340: Based on the training error of the quality inspection discrimination information samples to be optimized in the first debugging output information, select a set proportion of quality inspection discrimination information samples from the quality inspection discrimination information samples to be optimized as the first training set, and based on the training error of the quality inspection discrimination information samples to be optimized in the second debugging output information, select the set proportion of quality inspection discrimination information samples from the quality inspection discrimination information samples to be optimized as the second training set.

[0208] Step 350: Adjust the network parameters of the second initial decision tree network based on the first training set, and adjust the network parameters of the first initial decision tree network according to the second training set.

[0209] Step 360: Determine the first initial decision tree network and the second initial decision tree network as the microplate quality inspection discrimination network.

[0210] In a preferred technical solution, the system debugs the basic quality inspection discriminant network through a series of detailed steps, and finally obtains the microplate quality inspection discriminant network. The following is a detailed explanation of these steps.

[0211] First, the system decomposes the quality inspection discrimination information samples in the quality inspection discrimination information sample set into multiple training sample groups. The purpose of this is to divide a large number of samples into smaller, more easily processed units for subsequent network debugging. Each training sample group contains a certain number of quality inspection discrimination information samples, which may come from different microplate images and may contain different quality features and matching types.

[0212] Next, for each training example group, the system loads it into a first initial decision tree network and a second initial decision tree network for debugging. Both initial decision tree networks are initialized using a basic quality control discriminant network, but they may differ in structure and parameters. By debugging these two networks, the system obtains first and second debugging output information, which reflects the network's processing results and performance on the training example groups.

[0213] Then, the system acquires quality inspection discrimination information samples where the decision information in the first and second debug output information is inconsistent as quality inspection discrimination information samples to be optimized. The differences in processing results for these samples in the two networks indicate that they may be complex or boundary cases that the networks struggle to accurately discriminate. Selecting these samples for further optimization helps improve the network's discrimination ability and generalization performance.

[0214] Next, based on the training error of the quality inspection discrimination information samples to be optimized in the first debugging output information, the system selects a predetermined proportion of quality inspection discrimination information samples as the first training set. Similarly, based on the training error of the quality inspection discrimination information samples to be optimized in the second debugging output information, the system also selects a predetermined proportion of quality inspection discrimination information samples as the second training set. These two training sets are used to further debug and optimize the two initial decision tree networks.

[0215] Subsequently, the system uses the first training set to adjust the network parameters of the second initial decision tree network to reduce its error when processing the quality inspection discrimination information samples to be optimized. Simultaneously, the system also uses the second training set to adjust the network parameters of the first initial decision tree network to achieve a similar purpose. This cross-tuning strategy helps the two networks learn from and complement each other, thereby improving the overall discrimination performance.

[0216] Finally, the system determined the first and second initial decision tree networks, after debugging and optimization, as the microplate quality control discriminant networks. These two networks underwent fine-tuning and optimization in structure and parameters, enabling them to more accurately identify and judge the quality status of microplates. Combining them as the final microplate quality control discriminant network provides strong support for subsequent experiments and data analysis.

[0217] In other application scenarios, the system first decomposes the quality inspection discrimination information sample set into multiple smaller training sample groups. Each training sample group contains a set of quality inspection discrimination information samples, covering different microplate types, quality features, and matching types. By grouping, the system can more effectively manage and process a large number of quality inspection discrimination information samples, facilitating subsequent network debugging. Next, the system processes each training sample group. It inputs each training sample group into two initial decision tree networks—a first initial decision tree network and a second initial decision tree network. Both networks are initialized based on the basic quality inspection discrimination network, but may differ in network structure and parameters. By debugging in these two networks, the system can obtain two sets of debugging output information for each training sample group: the first debugging output information and the second debugging output information. This information reflects the network's processing results and performance on the training samples. After obtaining the two sets of debugging output information, the system compares and analyzes them. It pays particular attention to quality inspection discrimination information samples that produce inconsistent decision information in the two networks. These examples are called quality control discriminant information examples to be optimized because they reveal potential difficulties or uncertainties in the network's handling of complex or boundary cases. Selecting these examples for further optimization is crucial for improving the network's discriminative ability and generalization performance. To further optimize the network, the system needs to select a training set based on the training error of the quality control discriminant information examples to be optimized. It first calculates the training error of each example in the first debug output and selects a certain proportion of examples as the first training set based on the error magnitude. Similarly, the system also calculates the training error of each example in the second debug output and selects the same proportion of examples as the second training set. These two training sets will be used to further debug and optimize the two initial decision tree networks, respectively. Using the selected training sets, the system begins to adjust the network parameters of the two initial decision tree networks. It uses the first training set to debug the second initial decision tree network, reducing its error in processing the quality control discriminant information examples by adjusting the network parameters and structure. Simultaneously, the system also used a second training set to perform a similar debugging and optimization process on the first initial decision tree network. This cross-debugging strategy helps the two networks learn from and complement each other, thereby improving the overall discrimination performance. After a series of detailed debugging and optimization steps, the system finally determined two high-performance decision tree networks: the first initial decision tree network and the second initial decision tree network. Both networks underwent fine-tuning and optimization in structure and parameters, enabling them to more accurately identify and judge the quality status of microplates. Combining them as the final microplate quality inspection discrimination network can provide strong support for subsequent experiments and data analysis.This network not only has high accuracy and reliability, but also good generalization ability and robustness, and can adapt to various complex quality inspection scenarios and needs.

[0218] In step 340, the training error refers to the difference between the model's performance on the training data and the actual labels in machine learning. Specifically, in step 340 of this technical solution, the training error involves the errors generated by the first initial decision tree network and the second initial decision tree network when processing quality inspection and discrimination information samples.

[0219] In detail, training error can be measured by various metrics, such as Mean Squared Error (MSE) and Cross-Entropy Loss. These metrics quantify the degree of inconsistency between the model's predictions and the actual labels. In this step, the system calculates the difference between the predicted result and the actual label in the network output for each quality control information sample to be optimized; this difference is the training error.

[0220] For classification tasks, a common method for calculating training error is cross-entropy loss. Suppose a sample of quality control information to be optimized has the actual label "qualified," but the output of the first initial decision tree network is "unqualified" with a higher probability. In this case, the first initial decision tree network will generate a large training error for this sample. The system will use the magnitude of this error to determine whether this sample should be selected into the first training set for further debugging and optimization of the network.

[0221] In step 340, the system calculates not only the training error of each quality inspection discrimination information sample to be optimized in the first initial decision tree network, but also their training error in the second initial decision tree network. This is done to select the most representative samples from the two networks to form the training set, allowing for more effective adjustment and optimization of network parameters.

[0222] In summary, the training error in step 340 is an important indicator of how well the model performs on the training data. By calculating and analyzing the training error, the system can select the quality control information samples that most need optimization, thereby enabling targeted network debugging and optimization.

[0223] In another preferred embodiment, the step of debugging the basic quality inspection network using the quality inspection discrimination information samples and adversarial training images in the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes: deriving from the quality inspection discrimination information samples and adversarial training images in the quality inspection discrimination information sample set according to multiple set quality inspection discrimination items, obtaining multiple network training derived samples corresponding to each quality inspection discrimination item; for each quality inspection discrimination item, selecting from the corresponding network training derived samples to obtain a network training derived sample set; performing the corresponding quality inspection discrimination item processing in the basic quality inspection discrimination network based on the network training derived sample set, obtaining debugging output information for each quality inspection discrimination item; and adjusting the network parameters of the basic quality inspection discrimination network according to the debugging output information for each quality inspection discrimination item to obtain the microplate quality inspection discrimination network.

[0224] In another preferred embodiment, the system uses a series of detailed steps to debug the basic quality inspection discriminant network using quality inspection discriminant information samples from the quality inspection discriminant information sample set and adversarial training images, in order to obtain the microplate quality inspection discriminant network. The following is a detailed explanation of these steps.

[0225] First, based on multiple predefined quality control criteria, the system performs derivation processing on the quality control criteria samples and adversarial training images in the quality control criteria sample set. This means that for each quality control criterion, the system generates multiple corresponding network training derivation samples. These derivation samples are generated by transforming, enhancing, or combining the original samples, aiming to increase the diversity and generalization ability of the data.

[0226] Next, for each quality inspection judgment item, the system selects a set of network training derived samples from the corresponding network training derived samples. The selection process may include criteria such as the quality, representativeness, or difference from other samples to ensure that the selected samples can fully reflect the characteristics and requirements of the quality inspection judgment item.

[0227] Then, the system uses the network training to derive a set of sample examples to perform the corresponding quality inspection and discrimination tasks in the basic quality inspection and discrimination network. This means that the system inputs the selected samples into the basic quality inspection and discrimination network and observes the network's performance when processing these samples. Through this process, the system can obtain debugging output information for each quality inspection and discrimination task, which reflects the network's performance and accuracy in processing specific quality inspection and discrimination tasks.

[0228] Finally, the system adjusts the network parameters of the basic quality inspection discrimination network based on the debugging output information for each quality inspection discrimination item. This may include modifying the network weights, adjusting the network structure, or optimizing the network training strategy. By continuously adjusting the network parameters, the system can gradually improve the network's performance and accuracy in handling various quality inspection discrimination items, thereby obtaining the final microplate quality inspection discrimination network.

[0229] The advantage of this technical solution lies in its full utilization of the quality inspection discrimination information sample set and adversarial training images to debug and optimize the basic quality inspection discrimination network. Through derivation processing and filtering of network training derivative samples, the system can generate more representative and diverse data, thereby improving the network's generalization ability. Simultaneously, by independently processing and debugging each quality inspection discrimination item, the system can more accurately optimize network parameters, improving the network's performance and accuracy in handling specific quality inspection discrimination items. The resulting microplate quality inspection discrimination network will better meet the needs of practical applications, providing more reliable and accurate quality inspection discrimination results.

[0230] Figure 2 This is a schematic diagram of the structure of an image data processing system 200 provided in an embodiment of the present invention. Figure 2 The image data processing system 200 shown includes a processor 210, which can call and run computer programs from memory to implement the methods in the embodiments of the present invention.

[0231] Optionally, such as Figure 2 As shown, the image data processing system 200 may further include a memory 230. The processor 210 can retrieve and run computer programs from the memory 230 to implement the methods described in this embodiment of the invention.

[0232] The memory 230 can be a separate device independent of the processor 210, or it can be integrated into the processor 210.

[0233] Optionally, such as Figure 2 As shown, the image data processing system 200 may also include a transceiver 220, which the processor 210 can control to interact with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0234] Optionally, the image data processing system 200 can implement the corresponding processes of the storage engine or components (such as processing modules) in the storage engine or the device with the storage engine deployed in the various methods of the embodiments of the present invention. For the sake of brevity, these will not be described in detail here.

[0235] It should be understood that the processor in the embodiments of the present invention may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0236] It is understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0237] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of the present invention may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0238] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, the computer program implementing the above method when running.

[0239] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art.

Claims

1. A machine learning-based method for high-throughput data processing of focused microplates, characterized in that, The method, applied to an image data processing system, includes: Acquire images of microplates to be processed and multiple microplate quality inspection points, wherein the images of microplates to be processed contain the type of microplate to be tested and high-throughput image data; The initial quality inspection discrimination information is generated by combining the multiple microplate quality inspection viewpoints and the high-throughput image data of the microplate image to be processed. The microplate quality inspection decision network is used to make a microplate quality inspection opinion decision on the initial quality inspection opinion information to determine the microplate quality inspection opinion result of the microplate image to be processed, wherein the microplate quality inspection opinion result is one of the multiple microplate quality inspection opinions; The debugging steps for the microplate quality inspection discrimination network include: Obtain X network training examples, where X is an integer greater than 0; For each network training example, the type of microplate to be tested in the current network training example is obtained as the target microplate type, and the types of microplate to be tested in the remaining network training examples from X-1 network training examples are obtained as adversarial microplate types, where Y is an integer greater than 0 and less than a set number of viewpoints, and the set number of viewpoints is the maximum number of prior quality detection viewpoints in the microplate quality detection judgment result of the microplate quality detection viewpoint decision; Based on the target microplate type, the adversarial microplate type, and high-throughput image data of each network training example, image combination is performed to generate quality inspection discrimination information examples, resulting in a quality inspection discrimination information example set. The target microplate type is the matching type corresponding to the quality inspection discrimination information example. The quality inspection discrimination information example set contains multiple quality inspection discrimination information examples, and each quality inspection discrimination information example is generated from a network training example. The basic quality inspection discrimination network is debugged based on the aforementioned quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network.

2. The method according to claim 1, characterized in that, The high-throughput image data based on the multiple microplate quality inspection viewpoints and the microplate images to be processed is combined to generate initial quality inspection discrimination information, including: The multiple microplate quality detection viewpoints are combined with the high-throughput image data of the microplate image to be processed to obtain joint discrimination information; If the number of viewpoints for microplate quality inspection is less than the total number of viewpoints in the microplate quality inspection discrimination network, then training annotations are configured in the joint discrimination information based on the comparison result between the total number of viewpoints and the number of viewpoints for microplate quality inspection, to obtain initial quality inspection discrimination information.

3. The method according to claim 2, characterized in that, The step of combining the high-throughput image data of the multiple microplate quality detection points with the microplate image to be processed to obtain joint discrimination information includes: Based on the microplate reference image corresponding to each microplate quality inspection point, image mapping is performed on the multiple microplate quality inspection points to obtain multiple quality inspection mapping image data. The multiple quality inspection mapping image data and the high-throughput image data of the microplate image to be processed are combined to obtain joint discrimination information.

4. The method according to claim 1, characterized in that, The step of using the microplate quality inspection discrimination network to make microplate quality inspection opinion decisions on the initial quality inspection discrimination information, and determining the microplate quality inspection discrimination result of the microplate image to be processed, includes: The initial quality inspection discrimination information is used to make a quality inspection opinion decision on the microplate by a microplate quality inspection discrimination network, and the quality inspection opinion decision information is obtained. If the number of decision sequences in the quality inspection opinion decision information is greater than the number of opinions in the multiple microplate quality inspection opinions, then the quality inspection opinion decision information is extracted based on the number of opinions to obtain decision query features; Based on the priority of the multiple microplate quality inspection viewpoints in the initial quality inspection discrimination information, the microplate quality inspection viewpoint corresponding to the decision query feature is obtained as the microplate quality inspection discrimination result of the microplate image to be processed.

5. The method according to claim 1, characterized in that, For each network training example, obtaining the microplate type to be tested in the current network training example as the target microplate type and obtaining the microplate types to be tested from the remaining Y network training examples from X-1 network training examples as adversarial microplate types includes: Select any one of the X network training examples as the current network training example; Within the quantization constraint interval where the number of prior viewpoints generated by the basic quality inspection and discrimination network is the maximum value, an arbitrary value is determined as the number of adversarial viewpoints Y. Based on the number Y of adversarial viewpoints, X-1-Y network training examples other than Y network training examples are selected from the X-1 network training examples as the remaining network training examples.

6. The method according to claim 5, characterized in that, The generation of quality inspection discrimination information examples based on the high-throughput image data of the target microplate type, the adversarial microplate type, and the current network training examples includes: The target microplate type and the antagonistic microplate type are randomly combined to obtain joint discrimination information; If the number of training examples for the target microplate type and the adversarial microplate type is less than the total number of views in the microplate quality inspection discriminant network, then training annotations are configured for the joint discriminant information based on the comparison result between the total number of views and the number of training examples. The joint discrimination information is combined with the high-throughput image data of the current network training examples to obtain quality inspection discrimination information examples.

7. The method according to claim 1, characterized in that, Before debugging the basic quality inspection discrimination network based on the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network, the method further includes: combining images based on the target microplate type, the adversarial microplate type, and the high-throughput image data of the current network training samples to generate adversarial training images and adding them to the quality inspection discrimination information sample set, wherein the adversarial microplate type is the matching type corresponding to the adversarial training image; The step of debugging the basic quality inspection discrimination network based on the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes: debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images in the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network.

8. The method according to claim 7, characterized in that, The step of debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images from the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes: Two quality inspection discrimination information samples and their corresponding matching types are obtained from the quality inspection discrimination information sample set; Based on the set enhancement weights, feature enhancement is performed on the knowledge features corresponding to the two quality inspection discrimination information samples obtained, and feature enhancement is performed on the knowledge features corresponding to the two matching types. The results are used as debugging training examples. Using the obtained debugging training example, the basic quality inspection discrimination network is debugged to obtain the microplate quality inspection discrimination network; Alternatively, the step of debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images from the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes: The quality inspection discrimination information samples in the quality inspection discrimination information sample set are decomposed into multiple training sample groups; For each training sample group, the training sample group is loaded into the first initial decision tree network and the second initial decision tree network for debugging, and the first debugging output information and the second debugging output information are obtained. The first initial decision tree network and the second initial decision tree network are both initialized through the basic quality inspection and discrimination network. The quality inspection judgment information sample where the decision information in the first debug output information and the second debug output information are inconsistent is obtained as the quality inspection judgment information sample to be optimized; Based on the training error of the quality inspection discrimination information samples to be optimized in the first debugging output information, a set proportion of quality inspection discrimination information samples is selected from the quality inspection discrimination information samples to be optimized as the first training set, and based on the training error of the quality inspection discrimination information samples to be optimized in the second debugging output information, the set proportion of quality inspection discrimination information samples is selected from the quality inspection discrimination information samples to be optimized as the second training set; The network parameters of the second initial decision tree network are adjusted based on the first training set, and the network parameters of the first initial decision tree network are adjusted based on the second training set. The first initial decision tree network and the second initial decision tree network are determined as the microplate quality inspection discrimination network.

9. The method according to claim 7, characterized in that, The step of debugging the basic quality inspection discrimination network using quality inspection discrimination information samples and adversarial training images from the quality inspection discrimination information sample set to obtain the microplate quality inspection discrimination network includes: Based on the set of multiple quality inspection criteria, the quality inspection criteria information samples and adversarial training images in the set of quality inspection criteria information samples are derived to obtain multiple network training derived samples corresponding to each quality inspection criterion. For each quality inspection judgment item, a set of network training derivative examples is obtained by selecting from the corresponding network training derivative examples; Based on the network training derived sample set, the corresponding quality inspection and discrimination items are processed in the basic quality inspection and discrimination network to obtain the debugging output information of each quality inspection and discrimination item; The network parameters of the basic quality inspection discrimination network are adjusted based on the debugging output information of each quality inspection discrimination item to obtain the microplate quality inspection discrimination network.

10. An image data processing system, characterized in that, The method includes at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, causing the at least one processor to perform the method according to any one of claims 1-9.

Images