Laboratory strain AI identification method and system based on YOLOv8

Abstract

The invention provides a laboratory strain AI identification method and system based on YOLOv8. The method comprises the following steps: S1, carrying out laboratory detection on data acquisition of related strains; s2, cleaning, labeling, cutting, grouping and preprocessing the collected data set; s3, performing model training and algorithm adjustment on the processed data based on YOLOv8; s4, according to the trained data, establishing a basic strain identification reasoning model, and performing reasoning; s5, after reasoning is completed, the strain identification system automatically displays an identification result and pushes the identification result to a business system; by adopting a YOLOv8-based advanced visual model structure and combining a self-developed calculation algorithm, the problems of low automation degree, insufficient recognition accuracy and low classification and counting efficiency in a laboratory strain recognition process are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of bacterial strain identification and counting, and in particular to a YOLOv8-based laboratory bacterial strain AI identification method and system. Background Art

[0002] The identification and counting of laboratory bacterial species is a critical task. Accurate and rapid identification and counting of bacterial species are of great significance to laboratory testing. At present, traditional bacterial species identification methods mainly rely on manual microscope observation, culture medium testing and biochemical analysis, while bacterial species counting mostly relies on manual visual confirmation and manual counting. These methods are not only time-consuming and laborious, but also easily affected by subjective factors, with certain errors, which affect the accuracy of the results.

[0003] At present, some automated strain identification systems have begun to be gradually applied in laboratories, using image processing technology and machine learning algorithms to identify strains. However, most of these systems are based on traditional image processing algorithms, and the recognition effect is limited by image quality, background noise, and the complexity of strain morphology. Traditional image processing methods are difficult to cope with complex strain morphological changes and are highly dependent on the background, resulting in low recognition accuracy. In addition, the recognition speed of existing systems is also difficult to meet the needs of actual applications, especially when faced with a large number of samples, the recognition efficiency is low.

[0004] YOLOv8 is a more stable version of the YOLO object detection and image segmentation model developed by Ultralytics. It builds on the previous success of YOLO and introduces new features and improvements to further improve performance and flexibility. YOLOv8 can be trained on large datasets and can run on a variety of hardware platforms. However, early versions of YOLO still have shortcomings in detecting small targets, fine structures, and complex backgrounds. For the specific field of laboratory strain identification, the existing YOLO model still needs to improve its recognition accuracy and efficiency when processing highly complex strain images.

[0005] Patent document CN116433954A discloses a method for identifying multiple deep fungal species based on a neural network model. It uses various pictures of 5 common deep fungi, and after preliminary image feature processing, it is placed in a specific artificial intelligence deep neural network model for training, so that the model can quickly identify deep fungi and provide feedback in the form of a Web, providing auxiliary decision-making for early diagnosis and early treatment. However, the invention does not solve the problems of low automation, insufficient recognition accuracy, and low classification and counting efficiency in the process of laboratory strain identification by adopting an advanced visual model structure based on YOLOv8. Summary of the invention

[0006] In view of the defects in the prior art, the purpose of the present invention is to provide a laboratory strain AI identification method and system based on YOLOv8.

[0007] According to the present invention, a YOLOv8-based laboratory strain AI identification method is provided, comprising:

[0008] Step S1: Collecting data on relevant bacterial species for laboratory testing;

[0009] Step S2: cleaning, labeling, cutting, grouping and preprocessing the collected data set;

[0010] Step S3: Perform model training and algorithm adjustment on the processed data based on YOLOv8;

[0011] Step S4: Based on the trained data, a basic strain identification inference model is established and inference is performed;

[0012] Step S5: After the reasoning is completed, the strain identification system automatically displays the identification results and pushes them to the business system;

[0013] Repeat steps S2 to S5 until the model accuracy reaches a preset standard.

[0014] Preferably, in step S2:

[0015] Manually clean the collected data sets to filter out sterile data or data with shooting effects below the preset standards;

[0016] The cleaned data is annotated using an annotation tool; the annotated data is automatically cut using a cutting tool;

[0017] The data that has been cut is grouped using the data grouping tool, including training set, test set, and validation set; the ratio of grouped data volume is 7:2:1, and the configuration parameters are adjusted according to the training results;

[0018] Data preprocessing is performed on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc. Data expansion expands the data set by flipping the images in the data set at different angles to improve the robustness of the model; data brightness conversion and data contrast conversion are both implemented by the openCV framework to improve the robustness of the model.

[0019] Preferably, in step S3:

[0020] Specify the dataset path, train the model and adjust the algorithm for the visual model; use the YOLOv8 training instructions to train the model based on the preprocessed dataset; adjust the algorithm according to the training effect during the training process, and continuously tune the model; determine the training accuracy of the model by viewing the output data in the training result file, identify it with the actual data, and determine the effect of the model training.

[0021] Preferably, in step S4:

[0022] After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server.

[0023] After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference.

[0024] Preferably, in step S5:

[0025] After the inference is completed, the strain identification system automatically displays the identification results, and the identification results are manually confirmed. If there are any problems, they are corrected online;

[0026] The confirmed reasoning results are automatically pushed to the business system; the business system is the laboratory information management system LIMS, and the strain identification system is integrated with the LIMS system to automatically transmit the strain identification and counting results to the LIMS system;

[0027] After review and confirmation, the online manually corrected data is used as a new training data set. The model is retrained and the algorithm is adjusted to further improve the precision and accuracy of model recognition before it is put into operation again.

[0028] A laboratory strain AI identification system based on YOLOv8 is provided according to the present invention, comprising:

[0029] Module M1: Data collection for laboratory testing of related bacterial species;

[0030] Module M2: Cleaning, labeling, cutting, grouping and preprocessing the collected data sets;

[0031] Module M3: Perform model training and algorithm adjustment on the processed data based on YOLOv8;

[0032] Module M4: Based on the trained data, establish a basic strain identification inference model and perform inference;

[0033] Module M5: After the reasoning is completed, the strain identification system automatically displays the identification results and pushes them to the business system;

[0034] Modules M2 to M5 are triggered repeatedly until the model accuracy reaches the preset standard.

[0035] Preferably, in the module M2:

[0036] Manually clean the collected data sets to filter out sterile data or data with shooting effects below the preset standards;

[0037] The cleaned data is annotated using an annotation tool; the annotated data is automatically cut using a cutting tool;

[0038] The data that has been cut is grouped using the data grouping tool, including training set, test set, and validation set; the ratio of grouped data volume is 7:2:1, and the configuration parameters are adjusted according to the training results;

[0039] Data preprocessing is performed on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc. Data expansion expands the data set by flipping the images in the data set at different angles to improve the robustness of the model; data brightness conversion and data contrast conversion are both implemented by the openCV framework to improve the robustness of the model.

[0040] Preferably, in the module M3:

[0041] Specify the dataset path, train the model and adjust the algorithm for the visual model; use the YOLOv8 training instructions to train the model based on the preprocessed dataset; adjust the algorithm according to the training effect during the training process, and continuously tune the model; determine the training accuracy of the model by viewing the output data in the training result file, identify it with the actual data, and determine the effect of the model training.

[0042] Preferably, in the module M4:

[0043] After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server.

[0044] After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference.

[0045] Preferably, in the module M5:

[0046] After the inference is completed, the strain identification system automatically displays the identification results, and the identification results are manually confirmed. If there are any problems, they are corrected online;

[0047] The confirmed reasoning results are automatically pushed to the business system; the business system is the laboratory information management system LIMS, and the strain identification system is integrated with the LIMS system to automatically transmit the strain identification and counting results to the LIMS system;

[0048] After review and confirmation, the online manually corrected data is used as a new training data set. The model is retrained and the algorithm is adjusted to further improve the precision and accuracy of model recognition before it is put into operation again.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention adopts an advanced visual model structure based on YOLOv8 and combines it with a self-developed computing algorithm to solve the problems of low automation, insufficient recognition accuracy, and low classification and counting efficiency in the process of laboratory strain identification;

[0051] 2. The present invention utilizes the efficient target detection algorithm of YOLOv8 to achieve high-precision recognition and real-time classification of bacterial species, and combines online correction technology to further improve the accuracy and reliability of the recognition results, thereby significantly improving the automation level and overall operational efficiency of laboratory bacterial species management. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0053] Figure 1 This is a flow chart for model training and strain identification;

[0054] Figure 2 This is a hardware architecture diagram. DETAILED DESCRIPTION

[0055] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0056] Embodiment 1:

[0057] Based on the YOLOv8 model, the present invention makes specific optimization for the identification of laboratory strains and proposes an efficient AI identification method and system. The system can achieve rapid processing of large-scale samples while maintaining high recognition accuracy. Compared with the prior art, the present invention solves the problems of slow recognition speed, insufficient accuracy and strong dependence on operators of traditional strain identification methods, and provides a new solution for laboratory strain identification.

[0058] In the present invention, YOLOv8 is used as the core target detection algorithm, which is optimized and applied to AI identification of laboratory strains. The model training, real-time detection and classification are all completed by YOLOv8, realizing automatic identification, automatic classification and automatic counting of laboratory strains. During the identification process, the self-developed algorithm is combined to achieve high-precision identification of partition blocks and improve the recognition rate of small target objects. At the same time, the self-developed online annotation platform is used to realize online correction of strain identification results, and based on the Baosight iplat4J development platform, the strain identification system and the laboratory information management system are seamlessly connected in real time.

[0059] According to the present invention, a YOLOv8-based laboratory strain AI identification method is provided, comprising:

[0060] Step S1: Collecting data on relevant bacterial species for laboratory testing;

[0061] Step S2: cleaning, labeling, cutting, grouping and preprocessing the collected data set;

[0062] Preferably, in step S2:

[0063] Manually clean the collected data sets to filter out sterile data or data with shooting effects below the preset standards;

[0064] The cleaned data is annotated using an annotation tool; the annotated data is automatically cut using a cutting tool;

[0065] The data that has been cut is grouped using the data grouping tool, including training set, test set, and validation set; the ratio of grouped data volume is 7:2:1, and the configuration parameters are adjusted according to the training results;

[0066] Data preprocessing is performed on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc. Data expansion expands the data set by flipping the images in the data set at different angles to improve the robustness of the model; data brightness conversion and data contrast conversion are both implemented by the openCV framework to improve the robustness of the model.

[0067] Step S3: Perform model training and algorithm adjustment on the processed data based on YOLOv8;

[0068] Preferably, in step S3:

[0069] Specify the dataset path, train the model and adjust the algorithm for the visual model; use the YOLOv8 training instructions to train the model based on the preprocessed dataset; adjust the algorithm according to the training effect during the training process, and continuously tune the model; determine the training accuracy of the model by viewing the output data in the training result file, identify it with the actual data, and determine the effect of the model training.

[0070] Step S4: Based on the trained data, a basic strain identification inference model is established and inference is performed;

[0071] Preferably, in step S4:

[0072] After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server.

[0073] After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference.

[0074] Step S5: After the reasoning is completed, the strain identification system automatically displays the identification results and pushes them to the business system;

[0075] Preferably, in step S5:

[0076] After the inference is completed, the strain identification system automatically displays the identification results, and the identification results are manually confirmed. If there are any problems, they are corrected online;

[0077] The confirmed reasoning results are automatically pushed to the business system; the business system is the laboratory information management system LIMS, and the strain identification system is integrated with the LIMS system to automatically transmit the strain identification and counting results to the LIMS system;

[0078] After review and confirmation, the online manually corrected data is used as a new training data set. The model is retrained and the algorithm is adjusted to further improve the precision and accuracy of model recognition before it is put into operation again.

[0079] Repeat steps S2 to S5 until the model accuracy reaches a preset standard.

[0080] Embodiment 2:

[0081] Embodiment 2 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0082] The present invention also provides a laboratory strain AI identification system based on YOLOv8. The laboratory strain AI identification system based on YOLOv8 can be realized by executing the process steps of the laboratory strain AI identification method based on YOLOv8, that is, those skilled in the art can understand the laboratory strain AI identification method based on YOLOv8 as a preferred implementation of the laboratory strain AI identification system based on YOLOv8.

[0083] According to a YOLOv8-based laboratory strain AI identification system provided by the present invention, Figure 1-Figure 2 As shown, including:

[0084] Module M1: Data collection for laboratory testing of related bacterial species;

[0085] Module M2: Cleaning, labeling, cutting, grouping and preprocessing the collected data sets;

[0086] Preferably, in the module M2:

[0087] Manually clean the collected data sets to filter out sterile data or data with shooting effects below the preset standards;

[0088] The cleaned data is annotated using an annotation tool; the annotated data is automatically cut using a cutting tool;

[0089] The data that has been cut is grouped using the data grouping tool, including training set, test set, and validation set; the ratio of grouped data volume is 7:2:1, and the configuration parameters are adjusted according to the training results;

[0090] Data preprocessing is performed on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc. Data expansion expands the data set by flipping the images in the data set at different angles to improve the robustness of the model; data brightness conversion and data contrast conversion are both implemented by the openCV framework to improve the robustness of the model.

[0091] Module M3: Perform model training and algorithm adjustment on the processed data based on YOLOv8;

[0092] Preferably, in the module M3:

[0093] Specify the dataset path, train the model and adjust the algorithm for the visual model; use the YOLOv8 training instructions to train the model based on the preprocessed dataset; adjust the algorithm according to the training effect during the training process, and continuously tune the model; determine the training accuracy of the model by viewing the output data in the training result file, identify it with the actual data, and determine the effect of the model training.

[0094] Module M4: Based on the trained data, establish a basic strain identification inference model and perform inference;

[0095] Preferably, in the module M4:

[0096] After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server.

[0097] After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference.

[0098] Module M5: After the reasoning is completed, the strain identification system automatically displays the identification results and pushes them to the business system;

[0099] Modules M2 to M5 are triggered repeatedly until the model accuracy reaches the preset standard.

[0100] Preferably, in the module M5:

[0101] After the inference is completed, the strain identification system automatically displays the identification results, and the identification results are manually confirmed. If there are any problems, they are corrected online;

[0102] The confirmed reasoning results are automatically pushed to the business system; the business system is the laboratory information management system LIMS, and the strain identification system is integrated with the LIMS system to automatically transmit the strain identification and counting results to the LIMS system;

[0103] After review and confirmation, the online manually corrected data is used as a new training data set. The model is retrained and the algorithm is adjusted to further improve the precision and accuracy of model recognition before it is put into operation again.

[0104] Embodiment 3:

[0105] Embodiment 3 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0106] Based on the latest deep learning technology, the present invention studies high-performance computing methods and constructs a function that can automatically classify and count bacteria in real time and quickly. The specific method steps are as follows:

[0107] Step 1: Use a high-definition camera to collect data on laboratory testing of related bacterial species;

[0108] Step 2: Manually clean the collected data set to filter out sterile data or data with poor shooting effects;

[0109] Step 3: Use annotation tools to annotate the cleaned data;

[0110] Step 4: The labeled data is automatically cut using a self-developed cutting tool. In order to reduce the performance requirements of the GPU server for training, the image is automatically cut using the cutting tool, thereby reducing the resolution of a single image, but without affecting the resolution and recognition effect of the entire image.

[0111] Step 5: Use a self-developed data grouping tool to group the data that has been cut, including a training set, a test set, and a validation set; the data volume ratio of the three groups in the present invention is 7:2:1, and the configuration parameters can be adjusted according to the training results.

[0112] Step 6: Perform data preprocessing on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc.; data expansion is to expand the data set by flipping the images in the data set at different angles to improve the robustness of the model.

[0113] Data brightness conversion and data contrast conversion are both implemented by the openCV framework, and the purpose is to improve the robustness of the model.

[0114] Step 7: Specify the dataset path, train the model and adjust the algorithm for the visual model; Based on the preprocessed dataset, use the YOLOv8 training instructions to train the model; During the training process, adjust the algorithm according to the training effect, and continuously tune the model. The training accuracy of the model can be judged by viewing the output data in the training result file, and the recognition is combined with the actual data. The two aspects are combined to determine the effect of the model training.

[0115] Step 8: After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server.

[0116] Step 9: After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference;

[0117] Step 10: After the reasoning is completed, the strain identification system automatically displays the identification results, which can be manually confirmed. If there are any problems, they can be corrected online;

[0118] Step 11: Confirm that the completed reasoning results are automatically pushed to the business system; the business system refers to the laboratory information management system (LIMS), and the strain identification system can be integrated with the LIMS system to automatically transmit the results of strain identification and counting to the LIMS system.

[0119] Step 12: After review and confirmation, the online manually corrected data can be used as a new training data set to retrain the model and adjust the algorithm to further improve the precision and accuracy of model recognition before it is put into operation again;

[0120] Step 13: Repeat steps 4-12 until the model accuracy reaches the target.

[0121] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0122] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A laboratory strain AI identification method based on YOLOv8, characterized in that: include: Step S1: Collecting data on relevant bacterial species for laboratory testing; Step S2: cleaning, labeling, cutting, grouping and preprocessing the collected data set; Step S3: Perform model training and algorithm adjustment on the processed data based on YOLOv8; Step S4: Based on the trained data, a basic strain identification inference model is established and inference is performed; Step S5: After the reasoning is completed, the strain identification system automatically displays the identification results and pushes them to the business system; Repeat steps S2 to S5 until the model accuracy reaches a preset standard.

2. The YOLOv8-based laboratory strain AI identification method according to claim 1, characterized in that: In step S2: Manually clean the collected data sets to filter out sterile data or data with shooting effects below the preset standards; The cleaned data is annotated using an annotation tool; the annotated data is automatically cut using a cutting tool; The data that has been cut is grouped using the data grouping tool, including training set, test set, and validation set; the ratio of grouped data volume is 7:2:1, and the configuration parameters are adjusted according to the training results; Perform data preprocessing on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc.; Data expansion expands the data set by flipping the images in the data set at different angles to improve the robustness of the model; data brightness conversion and data contrast conversion are both implemented by the openCV framework to improve the robustness of the model.

3. The laboratory strain AI identification method based on YOLOv8 according to claim 1, characterized in that: In step S3: Specify the dataset path, train the model and adjust the algorithm for the visual model; based on the preprocessed dataset, use the YOLOv8 training instructions to train the model; During the training process, the algorithm is adjusted according to the training effect, and the model is continuously tuned; the training accuracy of the model is judged by viewing the output data in the training result file, and the actual data is combined for identification to determine the effect of the model training.

4. The laboratory strain AI identification method based on YOLOv8 according to claim 1, characterized in that: In step S4: After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server. After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference.

5. The laboratory strain AI identification method based on YOLOv8 according to claim 1, characterized in that: In step S5: After the inference is completed, the strain identification system automatically displays the identification results, and the identification results are manually confirmed. If there are any problems, they are corrected online; Confirm that the completed reasoning results are automatically pushed to the business system; The business system is the laboratory information management system LIMS. The strain identification system is integrated with the LIMS system to automatically transmit the results of strain identification and counting to the LIMS system. After review and confirmation, the online manually corrected data is used as a new training data set. The model is retrained and the algorithm is adjusted to further improve the precision and accuracy of model recognition before it is put into operation again.

6. A laboratory strain AI identification system based on YOLOv8, characterized in that: include: Module M1: Data collection for laboratory testing of related bacterial species; Module M2: Cleaning, labeling, cutting, grouping and preprocessing the collected data sets; Module M3: Perform model training and algorithm adjustment on the processed data based on YOLOv8; Module M4: Based on the trained data, establish a basic strain identification inference model and perform inference; Module M5: After the reasoning is completed, the strain identification system automatically displays the identification results and pushes them to the business system; Modules M2 to M5 are triggered repeatedly until the model accuracy reaches the preset standard.

7. The laboratory strain AI identification system based on YOLOv8 according to claim 6, characterized in that: In the module M2: Manually clean the collected data sets to filter out sterile data or data with shooting effects below the preset standards; The cleaned data is annotated using an annotation tool; the annotated data is automatically cut using a cutting tool; The data that has been cut is grouped using the data grouping tool, including training set, test set, and validation set; the ratio of grouped data volume is 7:2:1, and the configuration parameters are adjusted according to the training results; Perform data preprocessing on the grouped data, including data expansion, data brightness conversion, data contrast conversion, etc.; Data expansion expands the data set by flipping the images in the data set at different angles to improve the robustness of the model; data brightness conversion and data contrast conversion are both implemented by the openCV framework to improve the robustness of the model.

8. The laboratory strain AI identification system based on YOLOv8 according to claim 6, characterized in that: In the module M3: Specify the dataset path, train the model and adjust the algorithm for the visual model; based on the preprocessed dataset, use the YOLOv8 training instructions to train the model; During the training process, the algorithm is adjusted according to the training effect, and the model is continuously tuned; the training accuracy of the model is judged by viewing the output data in the training result file, and the actual data is combined for identification to determine the effect of the model training.

9. The laboratory strain AI identification system based on YOLOv8 according to claim 6, characterized in that: In the module M4: After the training is completed, a basic strain identification inference model is established and deployed. After the model training is completed, the training model with the best effect is defined as the basic strain identification inference model, and the model file is exported and deployed to the inference server. After the strain code is scanned, the strain identification inference model is automatically triggered to perform inference.

10. The laboratory strain AI identification system based on YOLOv8 according to claim 6, characterized in that: In the module M5: After the inference is completed, the strain identification system automatically displays the identification results, and the identification results are manually confirmed. If there are any problems, they are corrected online; Confirm that the completed reasoning results are automatically pushed to the business system; The business system is the laboratory information management system LIMS. The strain identification system is integrated with the LIMS system to automatically transmit the results of strain identification and counting to the LIMS system. After review and confirmation, the online manually corrected data is used as a new training data set. The model is retrained and the algorithm is adjusted to further improve the precision and accuracy of model recognition before it is put into operation again.

Images