Self supervised learning based high precision bacteria analysis system

Abstract

The self-supervised learning-based high-precision bacteria analysis system of the present invention includes an analysis module that measures the number of bacteria in the air by analyzing image data of cultured bacteria captured in a time-series order, and the analysis module includes a reading unit that analyzes the image data using a reading model and reads the type and number of cultured bacteria; a prediction unit that predicts the growth of bacteria based on the type and number of bacteria through a prediction model; and a learning unit that trains the prediction model through self-supervised learning using the image data, thereby reducing errors and providing accurate measurements of the number of bacteria.

Description

Technology Field

[0001] The present invention relates to a self-supervised learning-based high-precision bacteria analysis system, and more specifically, to a self-supervised learning-based high-precision bacteria analysis system capable of reducing culture time and detecting bacteria by utilizing an AI model capable of self-supervised learning. Background Technology

[0002] As the global pharmaceutical and food markets expand, the number of sterile facilities for the production of pharmaceuticals and food is increasing. In accordance with regulations, sterile facilities must measure bacteria at least several times a day.

[0003] As sterile facilities increase, the demand for bacterial measurement is on the rise; however, the reality is that most measurement personnel perform these measurements manually, going through the processes of sampling, culture, and analysis.

[0004] When measuring bacteria manually, errors are likely to occur due to inconsistent skill levels and measurement standards.

[0005] Automatic bacteria measurement systems have been developed and are being distributed to overcome the problems of manual methods, but the measurement range remains at 2,000 to 20,000 CFU.

[0006] However, since sterile facilities and hospitals need to measure bacteria in the range of 1 to 800 CFU, there is a need to improve the measurement sensitivity and accuracy of existing automatic bacteria measurement systems. Prior art literature

[0007] Korean Patent Registration No. 10-2285089 (Registration Date: July 28, 2021) The problem to be solved

[0008] To solve the aforementioned problems, the present invention aims to provide a self-supervised learning-based high-precision bacterial analysis system capable of automatically measuring bacteria in the range of 1 to 800 CFU. means of solving the problem

[0009] The self-supervised learning-based high-precision bacteria analysis system of the present invention includes an analysis module that measures the number of bacteria in the air by analyzing image data of cultured bacteria captured in a time-series order, and the analysis module may include: a reading unit that analyzes the image data using a reading model and reads the type and number of cultured bacteria; a prediction unit that predicts the growth of bacteria based on the type and number of bacteria through a prediction model; and a learning unit that trains the prediction model through self-supervised learning using the image data.

[0010] The above reading unit performs image segmentation for bacterial colonies included in the image data using the above reading model and can output a segmentation map reflecting the characteristics of the bacterial colonies.

[0011] The above reading unit can calculate the rate of boundary change of the bacterial colony by comparing the segmentation map according to the time series order.

[0012] The above reading unit can detect a bacterial colony whose boundary change rate is less than a preset reference rate as a first colony, and detect a bacterial colony whose boundary change rate is greater than or equal to the reference rate as a second colony.

[0013] The above prediction unit can predict the size growth rate of the first colony by inputting the rate of boundary change of the first colony into the prediction model.

[0014] The prediction unit compares the size growth rate with a preset size growth rate standard to calculate a reach probability, which is the probability that the size growth rate exceeds the size growth rate standard, and can detect a first cluster with a reach probability greater than or equal to a preset reach probability standard as a third cluster.

[0015] The above reading unit can determine the number of bacteria by measuring the number of second colonies and the number of third colonies.

[0016] The learning unit may include: a target image selection unit that selects one or more target image data to be learned from the above image data; an error calculation unit that calculates an error between a first reading data output by inputting the target image data into the reading model and a second reading data output by inputting the first reading data into the prediction model; and a model update unit that minimizes the error by updating the weights of the prediction model.

[0017] The error calculation unit may input the first reading data into the prediction model to receive the first prediction data, and re-input the first prediction data into the prediction model to receive the second reading data after performing restoration.

[0018] The self-supervised learning-based high-precision bacteria analysis system of the present invention may further include: a sampling module that captures airborne bacteria into a culture medium using a built-in impactor; a culture module that receives the culture medium and cultures it according to a preset temperature and time; a shooting module that captures the cultured bacteria to generate image data; and a transfer module that transfers the culture medium between the sampling module and the culture module. Effects of the invention

[0019] According to the present invention as described above, the following effects are achieved.

[0020] The present invention can reduce errors and provide accurate bacterial count measurements by capturing, culturing, and measuring bacteria in the air according to a standard method.

[0021] The present invention can ensure the reproducibility and accuracy of the measurement by performing the measurement process using a robot structure.

[0022] The present invention can eliminate deviations caused by the subjectivity or skill level of measurement personnel by using an AI model to identify bacteria.

[0023] The present invention can shorten analysis time and ensure measurement accuracy by predicting bacterial growth using an AI model.

[0024] The present invention can improve the accuracy of measurements by updating an AI model through self-supervised learning.

[0025] In addition, other features and advantages of the present invention may be newly identified through the embodiments of the present invention. Brief explanation of the drawing

[0026] FIG. 1 is a drawing showing the appearance of a self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention. FIG. 2 is a schematic diagram showing the configuration of a self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention. FIG. 3 is a diagram showing the configuration of an analysis module according to an embodiment of the present invention. FIG. 4 is a diagram showing the process of an analysis module (1100) performing analysis and learning according to an embodiment of the present invention. Figure 5 is a diagram to help understand the process of a reading unit reading a bacterial colony according to an embodiment of the present invention. Figure 6 is a diagram to help understand the process of a prediction unit predicting bacterial colonies according to an embodiment of the present invention. FIG. 7 is a diagram showing the configuration of a learning unit according to an embodiment of the present invention. FIG. 8 is a diagram illustrating the process of a self-supervised learning-based high-precision bacteria analysis system measuring bacteria according to an embodiment of the present invention. FIG. 9 is a diagram illustrating the process of a self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention performing self-supervised learning. Specific details for implementing the invention

[0027] It should be noted that in assigning reference numbers to the components of each drawing in this specification, identical components are assigned the same number whenever possible, even if they are shown in different drawings. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used in this specification, "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components in addition to the components mentioned.

[0028] The above objects, other objects, features, and advantages of the present invention will be easily understood through the following preferred embodiments associated with the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided to ensure that the disclosed content is thorough and complete and to ensure that the spirit of the invention is sufficiently conveyed to a person skilled in the art.

[0029] Hereinafter, preferred embodiments of the present invention designed to solve the above problems will be described in detail with reference to the attached drawings. The self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention will be referred to as the high-precision bacteria analysis system hereinafter.

[0030] FIG. 1 is a drawing showing the appearance of a self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention.

[0031] The high-precision bacterial analysis system illustrated in Fig. 1 can perform the sampling-culture-analysis process to measure bacteria in the air. The high-precision bacterial analysis system can be installed in sterile facilities such as clean rooms, pharmaceutical process lines, hospitals, and daycare centers, or in facilities where bacterial concentration management is required. The high-precision bacterial analysis system may be installed as a single unit, or multiple units may be installed at regular intervals for efficient measurement.

[0032] FIG. 2 is a schematic diagram showing the configuration of a self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention.

[0033] Referring to FIG. 2, a high-precision bacteria analysis system according to an embodiment of the present invention may include an analysis module (1100), a sampling module (1200), a culture module (1300), an imaging module (1400), and a transfer module (1500).

[0034] The sampling module (1200) can capture airborne bacteria in a culture medium using a built-in impactor. The impactor is a device for capturing airborne particles and can capture particles such as fine dust, viruses, and bacteria floating in the air.

[0035] Culture medium (TSA - Tryptic soyagar) can be prepared in a semi-solid form by adding agar to liquid medium (nutrient solution) in a Petri dish. For reference, solid culture media fix cells in place and allow them to grow into isolated clumps visible to the naked eye, known as colonies.

[0036] The impactor can be connected to an air inlet pipe through which indoor air is drawn in. A micro-flow sensor capable of measuring the volume of incoming air can be installed in the air inlet pipe.

[0037] The microflow sensor can transmit data measuring the air volume to the microflow controller. The microflow controller receives the data measuring the air volume and controls the valve to finely adjust the amount of air flowing in through the air inlet pipe.

[0038] The sampling module (1200) can precisely measure bacteria by finely adjusting the amount of incoming air to capture ultra-trace amounts of bacteria.

[0039] The culture module (1300) can accommodate a culture medium and culture according to a preset temperature and time. The culture module (1300) can hold the culture medium in a sealed state for 34 to 36 It can be grown to a size suitable for analysis or observation by culturing at a temperature of 12 to 48 hours.

[0040] Although not shown, the culture module (1300) may further be equipped with a cooling device and a ventilation device to maintain a constant temperature.

[0041] The shooting module (1400) can generate image data by shooting cultured bacteria. The shooting module (1400) may include a sensor unit, a light source unit, and a shooting unit.

[0042] The sensor unit can measure the internal temperature and humidity of the culture medium. The sensor unit may include a thermometer and a hygrometer for measuring temperature and humidity. In addition to temperature and humidity, the sensor unit may include other measuring devices necessary for culturing bacteria and collect data.

[0043] The light source can supply heat to maintain the internal temperature of the culture medium and provide lighting for photography when shooting.

[0044] The imaging unit can generate multiple image data by capturing cultured bacteria at regular time intervals. The time interval for capturing the bacteria can be appropriately selected. As will be described later, since it is necessary to measure the rate of change of the boundary formed by bacterial colonies, more precise prediction is possible as the time interval is shorter. However, since a large amount of image data can slow down processing speed and consume a lot of computing resources, a time interval that enables efficient measurement can be selected.

[0045] The transfer module (1500) can transfer culture medium between the sampling module (1200) and the culture module (1300).

[0046] Although not illustrated, the transfer module (1500) may include a rotary type robot structure and a flipper type robot structure. The rotary type robot structure can transfer the culture medium around a rotation axis or flip the culture medium. The flipper type robot structure can pick up or rotate the culture medium.

[0047] The transfer module (1500) can prevent errors or contamination that may occur during the transfer process by precisely controlling the process of transferring the culture medium.

[0048] The analysis module (1100) can measure the number of bacteria in the air by analyzing image data of cultured bacteria taken in chronological order.

[0049] FIG. 3 is a diagram showing the configuration of an analysis module (1100) according to an embodiment of the present invention. FIG. 4 is a diagram showing the process of an analysis module (1100) according to an embodiment of the present invention performing analysis and learning.

[0050] Referring to FIGS. 3 and 4, the analysis module (1100) may include a reading unit (1110), a prediction unit (1120), and a learning unit (1130). The analysis module (1100) can measure airborne bacteria by analyzing image data of cultured bacteria captured through an AI model.

[0051] The reading unit (1110) can analyze image data using a reading model and read the type and number of cultured bacteria.

[0052] The reading model according to an embodiment of the present invention may be a hybrid model combining a U-NET model and an LSTM model. However, it is not limited thereto, and other appropriate AI models may be selected. For example, a U-NET++ model improved over the U-NET model may be used.

[0053] The U-NET model is an image segmentation neural network used for image analysis. The U-NET model performs accurate segmentation by combining features extracted from images of various resolutions.

[0054] The LSTM model is a type of recurrent neural network (RNN) and is an AI model used to process time series data.

[0055] FIG. 5 is a diagram to help understand the process of a reading unit (1110) reading a colony of bacteria according to an embodiment of the present invention.

[0056] Referring to FIG. 5, the reading unit (1110) can perform image segmentation for bacterial colonies included in image data using a reading model and output a segmentation map that reflects the characteristics of the bacterial colonies.

[0057] For example, the reading model could be a U-NET model. When the U-NET model receives an original image as input, it can output a segment map containing the image's features through shrinkage and expansion paths.

[0058] The contraction path is similar to the structure of a Convolutional Neural Network (CNN). The contraction path extracts high-level abstracted features from the input original image. Each step of the contraction path consists of two 3x3 convolution operations and a ReLU activation function. After the convolution operations, spatial resolution is reduced through 2x2 max pooling operations. As each step of the contraction path progresses, the number of channels doubles to learn more feature maps.

[0059] For example, if the input image is 572x572 in size, it is gradually reduced to 284x284, 142x142, 71x71, etc. through max pooling.

[0060] The expansion path gradually restores the feature map obtained from the contraction path to its original resolution. The expansion path performs fine-grained segmentation by restoring positional information. Each step of the expansion path consists of an upsampling operation and a convolution operation. The expansion path doubles the spatial resolution of the feature map through the upsampling operation.

[0061] After each upsampling step, the feature maps of corresponding sizes are concatenated and combined in the shrinkage path. The expansion path processes the combined feature maps with two 3x3 convolutions and a ReLU activation function.

[0062] The expansion path repeats the upsampling and combining processes to restore the image to the same resolution as the original input image.

[0063] The U-NET model connects the shrinking path and the expanding path using a structure called Skip Connections. Skip Connections directly combine high-resolution information obtained from the shrinking path with the upsampling process of the expanding path. Through Skip Connections, lost details can be restored, enabling more accurate segmentation results.

[0064] The reading unit (1110) can calculate the rate of boundary change of bacterial colonies by comparing segmentation maps according to time series order.

[0065] Specifically, the reading unit (1110) can arrange image data generated by the shooting unit at regular time intervals in chronological order.

[0066] The reading unit (1110) divides each image data listed in chronological order using a reading model. The reading unit (1110) can output a segmentation map that reflects the characteristics of bacterial colonies.

[0067] The process of the reading unit (1110) dividing the image data can be carried out along the aforementioned shrinkage path and expansion path.

[0068] The characteristics of the bacterial colonies reflected in the segmentation map output by the reading unit (1110) may be the boundaries of the colonies and the size of the colonies. In addition to the boundaries of the colonies and the size of the colonies, other characteristics may be included. For example, the segmentation map may include the shape of the colonies, the shape of the individual bacteria forming the colonies, the color of the colonies, etc.

[0069] The reading unit (1110) can calculate a velocity vector of cluster boundary change by comparing segmentation maps divided according to time series order. The boundary change velocity may be calculated in pixel units or in length units. For example, the boundary change velocity may be 2.5 pixels / min or 5.5 µm / min.

[0070] According to another embodiment of the present invention, the reading unit (1110) may calculate a velocity vector of the cluster area changing by comparing segmentation maps divided according to time series order. The velocity of area change may be calculated in pixel units or in area units. For example, 6.25 pixels 2 / min or 30.25um 2It can be / min.

[0071] According to an embodiment of the present invention, the reading unit (1110) can classify clusters by inputting a boundary change rate vector into a reading model. At this time, an LSTM model included in the reading model may be used.

[0072] The reading unit (1110) can detect a bacterial colony as a first colony in which the rate of change of the boundary of the bacterial colony is less than a preset reference rate. The first colony may be a bacterial colony in which the rate of change of the boundary of the colony is very low. For example, the first colony may be a colony of microbacteria in which the number of captured bacterial cells is very low.

[0073] The reading unit (1110) can detect a bacterial colony as a second colony if the rate of change in the boundary of the bacterial colony is greater than or equal to the reference rate.

[0074] For example, the second colony may be a colony of bacteria that can be easily observed because the number of captured bacterial cells is sufficient.

[0075] The boundary change rate criterion can serve as a standard for determining whether a colony of microbacteria exists. The boundary change rate criterion can be appropriately selected by considering factors such as the conditions of the location where the high-precision bacterial analysis system is installed and the inspection speed.

[0076] FIG. 6 is a diagram to help understand the process of a prediction unit (1120) according to an embodiment of the present invention predicting a colony of bacteria.

[0077] Referring to FIG. 6, the prediction unit (1120) can predict the growth of the bacteria based on the type and number of bacteria through a prediction model.

[0078] As shown in (A) of Fig. 6, conventionally, to measure bacteria, colonies were detected by culturing bacteria for 48 hours.

[0079] According to an embodiment of the present invention, the prediction unit (1120) can predict the size growth rate of the first colony by inputting the boundary change rate of the first colony into a prediction model.

[0080] For example, the size growth rate of the first colony may be the rate of increase in the area of ​​the bacterial colony. Alternatively, the size growth rate of the first colony may refer to the rate of increase in the diameter of the bacterial colony. Depending on the type of bacteria, it may be more efficient to predict the size growth rate based on area, or it may be more efficient to predict it based on diameter. Therefore, the criterion for the growth rate can be appropriately selected.

[0081] The growth rate of bacterial colony size can be an indicator showing how much the colony has grown over a certain period of time. By calculating the growth rate, it is possible to determine the size of the bacterial colony after a certain period.

[0082] The prediction unit (1120) can calculate the probability of reaching the size growth rate, which is the probability that the size growth rate exceeds the size growth rate standard, by comparing the size growth rate with a pre-set size growth rate standard.

[0083] If the size growth rate of the first colony exceeds the size growth rate threshold, the first colony can be predicted to grow to a detectable size after a certain period of time. The size growth rate threshold can be appropriately selected based on the type of bacteria, the environment in which the high-precision bacterial analysis system is installed, the season, etc.

[0084] If the probability of the first colony reaching the target is greater than or equal to the target probability threshold, the first colony can be predicted to survive for a certain period of time. The target probability threshold can be appropriately selected based on the type of bacteria, the environment in which the high-precision bacterial analysis system is installed, the season, etc. For example, the target probability threshold may be 95%.

[0085] The prediction unit (1120) can detect a first cluster with a reach probability greater than or equal to a preset reach probability standard as a third cluster.

[0086] The reading unit (1110) can determine the number of bacteria by measuring the number of second colonies and the number of third colonies.

[0087] As illustrated in FIG. 6 (B), the culture module (1300) can culture bacteria for 24 hours to measure bacteria. For example, at the time when the culture time reaches 24 hours, the reading unit (1110) can use a reading model to detect a colony whose boundary change rate is greater than or equal to a reference rate as a second colony. That is, at the time when the culture time reaches 24 hours, a colony of bacteria that has grown sufficiently to be detectable can be detected. For example, the second colony may be 8 CFU.

[0088] For example, the prediction unit (1120) can detect the third colony by using a prediction model to predict the growth rate of the first colony at the time when the culture time reaches 24 hours and calculating the probability of reaching a size growth rate criterion. That is, it can predict whether the bacterial colony will grow and be detectable when the culture time reaches 48 hours. The third colony may be a colony among the first colonies that can grow to a detectable size when 48 hours have passed. For example, the third colony may be predicted to be 4 CFU.

[0089] The reading unit (1110) can detect the number of bacteria in the air as 12 CFU by adding the number of the second colony, 8 CFU, and the number of the third colony, 4 CFU.

[0090] According to an embodiment of the present invention, a high-precision bacteria analysis system can culture and detect bacteria in a relatively short time, unlike conventional systems.

[0091] In addition, microscopic bacteria that require a relatively long period of culture to detect can also be detected by predicting their growth based on AI models.

[0092] FIG. 7 is a diagram showing the configuration of a learning unit (1130) according to an embodiment of the present invention.

[0093] Referring to FIG. 7, the learning unit (1130) may include a target image selection unit (1131), an error calculation unit (1132), and a model update unit (1133).

[0094] The learning unit (1130) can train a prediction model through self-supervised learning using image data.

[0095] According to one embodiment of the present invention, the learning unit (1130) can perform time series prediction-based self-supervised learning.

[0096] Specifically, the reading model can analyze data at the current time point (t) and output a reading result. The prediction model can output a prediction result for a future time point (t + Δt) based on data at the current time point (t). The prediction model can restore the prediction result for the future time point (t + Δt) in reverse order and output the data at the current time point (t) again.

[0097] The reading model can output a reading result by analyzing the data at the current time point (t) restored by the prediction model.

[0098] The learning unit (1130) can compare the reading result of the reading model analyzing the data at the current time point (t) with the reading result of the reading model analyzing the data at the current time point (t) restored by the prediction model, and calculate the error.

[0099] The learning unit (1130) can define the calculated error as a loss function and train the prediction model so that the loss function is minimized.

[0100] The target image selection unit (1131) can select one or more target image data to be trained from among the image data. The criteria for selecting the image data can be appropriately selected and are not limited.

[0101] The error calculation unit (1132) can calculate the error between the first reading data and the second reading data.

[0102] According to one embodiment, the first reading data may be output by the error calculation unit (1132) inputting target image data into the reading model. For example, the first reading data may be the boundary change rate of a bacterial colony.

[0103] The second reading data may be output by the error calculation unit (1132) inputting the first reading data into the prediction model.

[0104] More specifically, the error calculation unit (1132) can input the first reading data into the prediction model and receive the first prediction data. For example, the first prediction data may be the size growth rate of a bacterial colony.

[0105] The error calculation unit (1132) can receive output of second reading data in which restoration is performed by re-inputting the first prediction data into the prediction model. For example, the second reading data may be the boundary change rate of a bacterial colony restored from the size growth rate of a bacterial colony.

[0106] The model update unit (1133) can minimize the error by updating the weights of the prediction model. The model update unit (1133) defines the error between the first reading data and the second reading data as a loss function and can update the weights of the prediction model so that the loss function is minimized.

[0107] According to one embodiment, the learning unit (1130) can further improve the prediction accuracy of the prediction model by performing self-supervised learning using a reading model and a prediction model.

[0108] FIG. 8 is a diagram illustrating the process of a self-supervised learning-based high-precision bacteria analysis system measuring bacteria according to an embodiment of the present invention.

[0109] Referring to FIG. 8, the process of the high-precision bacteria analysis system of the present invention measuring bacteria is explained.

[0110] A high-precision bacteria analysis system according to an embodiment of the present invention can capture bacteria in the air in a culture medium (S10).

[0111] A high-precision bacterial analysis system can accommodate a culture medium and culture bacteria according to a preset temperature and time (S20).

[0112] A high-precision bacterial analysis system can generate image data by photographing bacteria in a culture medium during the time bacteria are cultured (S30).

[0113] A high-precision bacterial analysis system can arrange the image data included in the generated image data in chronological order and perform image segmentation of bacterial colonies included in the image data using a reading model (S40).

[0114] A high-precision bacterial analysis system can output a segmentation map containing the characteristics of bacterial colonies (S50).

[0115] A high-precision bacterial analysis system can calculate the boundary change rate vector of bacterial colonies by comparing segmentation maps according to time series order (S60).

[0116] A high-precision bacterial analysis system inputs a boundary change rate vector into a reading model, detects bacterial colonies whose boundary change rate is less than a preset reference rate as first colonies, and detects bacterial colonies whose boundary change rate is greater than or equal to the reference rate as second colonies (S70).

[0117] A high-precision bacterial analysis system can input the boundary change rate vector of the first colony into a prediction model and predict the size growth rate of the first colony after a certain period of time (S80).

[0118] A high-precision bacterial analysis system can calculate a probability of reaching the size growth rate, which is the probability that the size growth rate exceeds the size growth rate standard, by comparing the size growth rate with a preset size growth rate standard, and can detect a first colony with a probability of reaching the size growth rate standard as a third colony (S90).

[0119] A high-precision bacterial analysis system can measure the second and third colonies and output the number of colonies per unit area (CFU / m3) (S100).

[0120] FIG. 9 is a diagram illustrating the process of a self-supervised learning-based high-precision bacteria analysis system according to an embodiment of the present invention performing self-supervised learning.

[0121] Referring to FIG. 9, the process of the high-precision bacteria analysis system of the present invention performing self-supervised learning is explained.

[0122] A high-precision bacteria analysis system can select one or more target image data from the image data to be used for training (S210).

[0123] A high-precision bacteria analysis system can input target image data into a reading model and output first reading data (S220).

[0124] A high-precision bacteria analysis system can input first reading data into a prediction model and output first prediction data (S230).

[0125] The high-precision bacteria analysis system can output second reading data that has been restored by re-entering first prediction data into a prediction model (S240).

[0126] A high-precision bacteria analysis system can calculate the error between the first reading data and the second reading data (S250).

[0127] A high-precision bacteria analysis system can define the error as a loss function and update the weights of the prediction model to minimize the loss function (S260).

[0128] A high-precision bacteria analysis system according to an embodiment of the present invention can measure ultra-trace amounts of bacteria by predicting and analyzing bacterial growth using an AI model, and can improve the prediction accuracy of the prediction model through self-supervised learning.

[0129] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various modifications and variations from the description above. For example, suitable results can be achieved even if the described techniques are performed in a different order than described, and / or components such as the described structures and devices are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0130] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below. Explanation of the symbols

[0131] 1100: Analysis Module 1200: Sampling Module 1300: Culture Module 1400: Imaging Module 1500: Transfer module 1110: Reader 1120: Prediction section 1130: Learning section 1131: Target Image Selection Unit 1132: Error Calculation Unit 1133: Model Update Section

Claims

Claim 1 A self-supervised learning-based high-precision bacteria analysis system comprising: an analysis module that measures the number of bacteria in the air by analyzing image data captured of cultured bacteria in a time-series order; a reading unit that analyzes the image data using a reading model and reads the type and number of cultured bacteria; a prediction unit that predicts the growth of bacteria based on the type and number of bacteria through a prediction model; and a learning unit that trains the prediction model through self-supervised learning using the image data. Claim 2 A self-supervised learning-based high-precision bacterial analysis system according to claim 1, wherein the reading unit performs image segmentation for bacterial colonies included in the image data using the reading model and outputs a segmentation map reflecting the characteristics of the bacterial colonies. Claim 3 A self-supervised learning-based high-precision bacterial analysis system, wherein, in paragraph 2, the reading unit compares the segmentation map according to time series order to calculate the rate of boundary change of the bacterial colony. Claim 4 A self-supervised learning-based high-precision bacterial analysis system according to claim 3, wherein the reading unit detects a bacterial colony whose boundary change rate is less than a preset reference rate as a first colony, and detects a bacterial colony whose boundary change rate is greater than or equal to the reference rate as a second colony. Claim 5 A self-supervised learning-based high-precision bacteria analysis system according to claim 1, wherein the prediction unit inputs the rate of change in the boundary of the first colony into the prediction model to predict the size growth rate of the first colony. Claim 6 A self-supervised learning-based high-precision bacteria analysis system, wherein, in claim 5, the prediction unit compares the size growth rate with a preset size growth rate standard to calculate a probability of reaching the size growth rate, which is the probability that the size growth rate exceeds the size growth rate standard, and detects a first colony with a probability of reaching the size growth rate greater than or equal to the preset probability of reaching the size growth rate standard as a third colony. Claim 7 A self-supervised learning-based high-precision bacterial analysis system, wherein, in claim 6, the reading unit measures the number of second colonies and the number of third colonies to derive the number of bacteria. Claim 8 A self-supervised learning-based high-precision bacteria analysis system according to claim 1, wherein the learning unit comprises: a target image selection unit that selects one or more target image data to be learned from the image data; an error calculation unit that calculates an error between a first reading data output by inputting the target image data into the reading model and a second reading data output by inputting the first reading data into the prediction model; and a model update unit that minimizes the error by updating the weights of the prediction model. Claim 9 A self-supervised learning-based high-precision bacteria analysis system, wherein, in claim 8, the error calculation unit inputs the first reading data into the prediction model to receive the first prediction data, and re-inputs the first prediction data into the prediction model to receive the second reading data after performing restoration. Claim 10 A self-supervised learning-based high-precision bacteria analysis system according to claim 1, further comprising: a sampling module that captures airborne bacteria in a culture medium using a built-in impactor; a culture module that receives the culture medium and cultures it according to a preset temperature and time; a shooting module that captures the cultured bacteria to generate image data; and a transfer module that transfers the culture medium between the sampling module and the culture module.

Images