JP7416136B2 - Colony detection device - Google Patents

Description

The present invention relates to a colony detection device .

In recent years, with the growth of fast food and other restaurant industries and the increase in conveniently processed foods, opportunities to eat food prepared by others have increased significantly. On the other hand, when dangerous substances such as poisonous substances or harmful microorganisms are mixed into foods such as processed foods, they directly affect the human body. Therefore, when handling processed foods, it is necessary not only to ensure the safety of raw materials, but also to strictly control the hygiene aspects of the manufacturing process, and to ensure food safety throughout the food production and processing.

Patent Document 1 discloses a colony detection system in which microorganisms collected from food are cultured on a petri dish or film and colonies are counted.

Japanese Patent Application Publication No. 2016-15944

However, there are various types of microorganisms contained in food, such as general viable bacteria, Escherichia coli, coliform bacteria, and Staphylococcus aureus. This requires complicated work such as selecting detection logic and specifying detection processing parameters to detect colonies according to various conditions. In particular, identifying detection processing parameters is an important task for increasing colony detection accuracy.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a colony detection device that can improve colony detection accuracy.

A colony detection device according to an embodiment of the present invention includes a culture medium image acquisition unit that acquires culture medium image data of a culture medium in which microorganisms are cultured, and a culture medium image data acquired by the culture medium image acquisition unit, and a culture medium image and a culture medium type. a specifying unit that specifies a detector for detecting a colony based on a medium type that is input to a discriminator generated using learning data that is identified by the discriminator; and a detector specified by the specifying unit. and a detection section that detects colonies based on the culture medium image data acquired by the culture medium image acquisition section using the medium image acquisition section.

A colony detection device according to an embodiment of the present invention includes a culture medium image acquisition unit that acquires culture medium image data of a culture medium in which microorganisms are cultured, and a culture medium image data acquired by the culture medium image acquisition unit, and a culture medium image and colonies. input to a discriminator generated using learning data including detector data and detection processing parameter data of a detector for detection, and generated based on the detector information and detection processing parameter information specified by the discriminator. and a detection section that detects colonies based on the culture medium image data acquired by the culture medium image acquisition section using a detector.

A computer program according to an embodiment of the present invention causes a computer to perform a process of acquiring medium image data of a medium in which microorganisms are cultured, and to use the acquired medium image data using learning data including the medium image data and the medium type. processing to identify a detection algorithm for detecting colonies based on the culture medium type identified by the learning model, and colony detection based on culture medium image data acquired using the identified detection algorithm. and the process of detecting it.

A colony detection method according to an embodiment of the present invention acquires medium image data of a medium in which microorganisms are cultured, and generates the acquired medium image data using learning data including the medium image data and the medium type. A detector for detecting colonies is specified based on the medium type identified by the discriminator, and colonies are detected based on the medium image data acquired using the specified detector.

A discriminator generation method according to an embodiment of the present invention acquires medium image data of a medium in which microorganisms are cultured, acquires a medium type of the medium image data, and includes the acquired medium image data and medium type. Generate a classifier using the training data.

The discriminator according to the embodiment of the present invention is generated using learning data including medium image data of a medium in which microorganisms are cultured and the medium type of the medium image data.

According to the present invention, the accuracy of colony detection can be improved.

FIG. 1 is a block diagram showing an example of the configuration of a colony detection device according to the present embodiment. It is a schematic diagram showing an example of a film type culture medium. FIG. 2 is a schematic diagram showing an example of the configuration of a learning model. FIG. 2 is a schematic diagram showing processing performed in a convolution layer. FIG. 2 is a schematic diagram showing processing performed in a pooling layer. FIG. 2 is a schematic diagram showing a first example of a learning method of a learning model by a learning processing unit. It is a schematic diagram which shows the 1st example of the colony detection method by a processing part. It is a schematic diagram which shows an example of a colony detection result. FIG. 7 is a schematic diagram showing a second example of a learning method of a learning model by a learning processing unit. It is a schematic diagram which shows the 2nd example of the colony detection method by a processing part. It is a schematic diagram which shows the 3rd example of the colony detection method by a processing part. FIG. 2 is a schematic diagram illustrating an example of a method for learning a learning model using a learning server. 7 is a flowchart illustrating an example of a learning process procedure of a learning model by a learning processing unit. It is a flow chart which shows an example of a colony detection processing procedure by a colony detection device.

Embodiments of the present invention will be described below based on the drawings. FIG. 1 is a block diagram showing an example of the configuration of a colony detection device 50 according to the present embodiment. One or more (in the example of FIG. 1, one) scanners 10 as optical reading devices are connected to the colony detection device 50 via a network 1 such as the Internet. The scanner 10 can read a film-type culture medium, which will be described later, from a hygiene management system 100 of a company or company, and can transmit the culture medium image data obtained by the reading to the colony detection device 50.

For example, the hygiene management system 100 extracts the food being processed or processed as a sample in a manufacturing process for manufacturing food that is being processed or after processing, or an inspection process for inspecting the food, and adds a sample liquid to a film-type medium described below. (specimen), add an appropriate sterile diluent, dilute the sample solution, and culture under the required culture conditions. The film-type medium after culturing can be read with the scanner 10, as described above.

FIG. 2 is a schematic diagram showing an example of the film-type culture medium 20. The film-type culture medium 20 includes a base sheet 21 formed of a film, a circular frame 22 formed in the center of the base sheet 21, a culture layer 23 for cultivating bacteria provided within the frame 22, and a culture layer 23. A cover sheet 24 and the like are provided to cover the base sheet 21 including the base sheet 21. In addition, FIG. 2 shows the film-type culture medium after culture|cultivation, and colonies can be visually observed. A colony is a collection of bacteria that is created by cultivating bacteria so small that they are invisible to the naked eye and growing until they are visible. The number of bacteria is determined by counting the colonies.

Although not shown, supplementary information can be written in the margin of the film-type culture medium 20 except for the culture layer 23. The supplementary information can include, for example, the type of specimen, the type or dilution rate of diluent, culture conditions, and the like. The supplementary information may be printed characters or symbols, or handwritten characters. The supplementary information can be read by the scanner 10.

Further, medium type information can be written in the margin of the film-type medium 20 excluding the culture layer 23. The medium type information includes, for example, the type of medium (medium type), and may be printed characters or symbols, handwritten characters, or the like. The medium type information can be read by the scanner 10.

As shown in FIG. 1, the colony detection device 50 includes a control section 51 that controls the entire device, a communication section 52, a storage section 53, an adjustment section 54, a determination section 55, an output processing section 56, and a processing section 60. The processing unit 60 includes a learning model 61 as a classifier, a learning processing unit 62, and a detection unit 63. The detection unit 63 includes a detection logic unit 631 and a detection processing parameter unit 632 as a detector. Note that if the learning model 61 has been trained, the learning processing unit 62 may not be provided and is not an essential configuration.

The control unit 51 can be configured with a CPU, ROM, RAM, and the like.

The communication unit 52 has a function of communicating with the scanner 10 via the network 1, and can receive (acquire) information read by the scanner 10.

More specifically, the communication unit 52 has a function as a culture medium image acquisition unit, and can acquire culture medium image data of a culture medium in which microorganisms (bacteria) are cultured from the scanner 10.

The communication unit 52 has a function as an additional information acquisition unit, and can acquire additional information related to the acquired culture medium image data from the scanner 10. Further, the communication unit 52 has a function as a type information acquisition unit, and can acquire the culture medium type information of the acquired culture medium image data from the scanner 10.

The processing unit 60 is, for example, hardware such as a CPU (for example, a multi-processor equipped with a plurality of processor cores), a GPU (Graphics Processing Units), a DSP (Digital Signal Processor), or an FPGA (Field-Programmable Gate Array). It can be configured by combining. It is also possible to combine quantum processors.

The learning model 61 has a function as a discriminator, and can use a multilayer neural network (deep learning), for example, a convolutional neural network, but can also use other machine learning. May be used.

FIG. 3 is a schematic diagram showing an example of the configuration of the learning model 61. The learning model 61 shown in FIG. 3 is a convolutional neural network, and an input layer 611, a convolution layer 612, a pooling layer 613, a convolution layer 614, a pooling layer 615, a fully connected layer 616, and an output layer 617 are connected in this order. ing. Note that the numbers of convolution layers, pooling layers, and fully connected layers are for convenience and are not limited to the numbers shown in FIG. 3. Furthermore, for convenience, the activation function layer is omitted. Culture medium image data is input to the input layer 611 . The learning model 61 can output identification results (medium type, detection processing parameters, etc. to be described later) for each culture medium image data obtained by reading one film-type culture medium 20 with the scanner 10.

FIG. 4 is a schematic diagram showing the processing performed in the convolution layer. The input/output data of the convolutional layer is also called a feature map, the input data of the convolutional layer is also called an input feature map, and the output data of the convolutional layer is also called an output feature map. The input feature map of the first convolutional layer is an input frame-by-frame image. The processing performed in the convolution layer (also referred to as "convolution operation") is a filter operation using a convolution filter (also referred to as "filter").

As shown in FIG. 4, for convenience, the input feature map is assumed to be 8×8 pixels. Furthermore, the size of the filter is assumed to be 3×3 pixels. In the convolution operation, while sliding the filter window at regular intervals on the input feature map, the elements of the filter are multiplied by the corresponding elements of the input feature map, the sum is calculated, and the calculated sum is used as the output feature map. to the corresponding pixel. In the example of FIG. 4, the calculation result of the region S corresponding to the filter F1 of the input feature map is stored in the pixel S1 of the output feature map. Further, the calculation result of the region S corresponding to the filter F2 of the input feature map is stored in the pixel S2 of the output feature map. Similarly, the calculation result of the region S corresponding to the filter F3 of the input feature map is stored in the pixel S3 of the output feature map. By moving the filter one pixel at a time and performing similar operations, the output feature map will have a size of 6×6 pixels. Here, by using three filters F1, F2, and F3, three output feature maps are obtained.

In learning the learning model 61, parameters related to filters include, for example, the values of filter elements, the number of filters (3 in the example of FIG. 4), the size of the filters (3×3 in the example of FIG. 4), and the filters. The movement width (also called "slide"; in the example of FIG. 4, 1 pixel), the padding of filling the periphery (edge area) of the input feature map with zero, etc. are optimized. Convolutional layers allow the extraction of spatial features of an image.

FIG. 5 is a schematic diagram showing the processing performed in the pooling layer. The pooling layer performs processing to reduce the size of the two-dimensional feature map output from the convolution layer. Specifically, processing is performed to aggregate local regions of the image into one element. For example, as shown in Figure 5, in a 6x6 pixel feature map (output feature map), the 2x2 local area (window W) is aggregated into "4", which is the maximum value of each element. ing. Note that the sliding of the window W is equal to the size of the window W, and in the example of FIG. 5, it slides by 2 pixels, so the feature map of 6×6 pixels is reduced to 3×3 pixels. With the pooling layer, even if a characteristic part is slightly deformed or displaced within the image, for example, the difference due to the deformation or displacement can be absorbed and the characteristic part can be extracted.

The learning processing unit 62 learns and updates (generates) the learning model 61 using the learning data. Details of the learning processing section 62 will be described later.

The detection unit 63 uses the optimal detection logic specified from among the plurality of detection logics (also referred to as detection algorithms or detectors) in the detection logic unit 631 according to the type of culture medium, and also uses the detection processing parameter unit 632. Colonies are detected by performing image processing on the culture medium image data using the optimal detection processing parameters specified from among the detection processing parameters within. Detection of a colony includes, for example, location information of the colony (X and Y coordinates on the image), reliability of colony detection, and the like. Reliability means detection accuracy, and for example, it can be assumed that the larger the difference or ratio between the color feature amount for detecting colonies and the threshold value of the color feature amount is, the higher the reliability is.

Detection processing parameters include, for example, color-related features (color features) when detecting colonies from culture medium images, thresholds for color features, pass/fail thresholds when making culture medium judgments, mask radius, detection logic parameters, etc. Contains various parameters.

Next, a method for generating (learning) the learning model 61 will be explained.

FIG. 6 is a schematic diagram showing a first example of a learning method of the learning model 61 by the learning processing unit 62. As shown in FIG. 6, the learning processing unit 62 provides learning input data and learning output data to the learning model 61 as learning data. The learning input data includes culture medium image data. In addition to the type of culture medium (medium type) and the detection processing parameters used by the detection unit 63, the learning output data includes location information of colonies output by the detection unit 63 based on culture medium image data, reliability of colony detection, etc. can be included. Note that colony location information is not essential.

The learning processing unit 62 can learn and update (generate) the learning model 61 using learning data including culture medium image data and the type of culture medium. Types of media include, for example, a general medium for measuring the number of viable bacteria (also referred to as AC), a medium for measuring the number of coliform bacteria (also referred to as CC), a medium for measuring the number of E. coli/coliform bacteria (also referred to as EC), and a medium for measuring the number of coliform bacteria (also referred to as EC). Contains a medium for measuring the number of staphylococci (also referred to as SA). Thereby, the learning model 61 can identify (output) the type of culture medium based on the culture medium image data.

Further, the learning processing unit 62 can learn and update (generate) the learning model 61 using learning data including culture medium image data, detection processing parameters for detecting colonies, and colony detection result information. The detection result information can include colony position information and the reliability of colony detection (for example, the probability of how accurate the colony detection result is). For example, medium image data is given to the learning model 61 as learning data, and each time the detection processing parameters are updated, the updated detection processing parameters and colony detection result information obtained as a result of the update are given to the learning model 61 as learning data, The learning model 61 can be trained. Thereby, the learning model 61 can identify (output) detection processing parameters necessary to make the reliability of colony detection equal to or higher than a predetermined threshold.

Next, a method for detecting colonies will be explained.

FIG. 7 is a schematic diagram showing a first example of a colony detection method by the processing unit 60. Since the learning model 61 is trained using learning data including culture medium image data and the type of culture medium, for example, when culture medium image data with unknown culture medium type is input to the learning model 61, the learning model 61 The type of culture medium can be identified based on the image data of the culture medium. As for the type of culture medium outputted (identified) by the learning model 61, for example, the probability thereof can be outputted for each of the above-mentioned AC, CC, EC, and SA. The type of culture medium with the highest probability can be used as the identification result.

The detection unit 63 has a function as a specifying unit, and based on the type of culture medium identified by the learning model 61, the type of culture medium identified from among the detection logics A, B, C, and D possessed by the detection logic unit 631. The optimal detection logic can be identified. For example, detection logics A, B, C, and D can be specified depending on whether the type of medium is AC, CC, EC, or SA. This makes it possible to specify the optimal detection logic depending on the type of culture medium. Although FIG. 7 shows detection logics A to D, the number of detection logics is not limited to four. Note that a detection logic that is a combination of at least two parts of detection logics A, B, C, and D may be specified depending on the magnitude of the probability for each of AC, CC, EC, and SA output by the learning model 61. good.

The detection unit 63 can detect colonies by performing image processing on the culture medium image data using the specified detection logic.

With the above-described configuration, by simply inputting culture medium image data to the processing unit 60 (learning model 61), the type of culture medium is identified, the detection logic optimal for the identified type of culture medium is specified, and colonies are detected. This eliminates the need for complicated work such as selecting detection logic, and improves the accuracy of colony detection.

In addition, since the learning model 61 is trained using learning data including detection processing parameters for detecting colonies and colony detection result information, for example, when culture medium image data is input to the learning model 61, the learning model 61 can specify detection processing parameters (detection processing parameter information) that are optimal for the detection logic that performs image processing on culture medium image data.

The detection unit 63 can detect colonies by performing image processing on the culture medium image data using the specified detection logic and detection processing parameters optimal for the detection logic.

With the above configuration, by simply inputting culture medium image data to the processing unit 60 (learning model 61), the optimum detection processing parameters for the detection logic can be specified and colonies can be detected. Complicated work such as identification becomes unnecessary, and the accuracy of colony detection can be improved.

The output processing unit 56 outputs a colony detection result based on the type of culture medium (medium type) identified by the learning model 61, the colony position information output by the detection unit 63, the reliability of colony detection, and the like.

FIG. 8 is a schematic diagram showing an example of colony detection results. As shown in FIG. 8, in the culture layer, a rectangular mark indicated by the symbol M1 indicates a colony counted as one. If the colonies are multicolored, the number of colonies can be counted by color. A mark indicated by the symbol M2 indicates a state in which a plurality of colonies are connected. In the example of FIG. 8, the number of colonies is four. The mark indicated by the symbol M3 indicates the boundary line of the colony. A border line is displayed between the colony and the medium. This makes it possible to clearly determine the boundaries between colonies. The mark indicated by the symbol M4 is a marker display for a colored specimen that may be mistaken for a colony, or a display for calling attention to a specimen that is likely to be falsely detected. This makes it possible to call attention to avoid misidentifying colonies.

The adjustment unit 54 can adjust the detection processing parameters specified by the learning model 61 based on the additional information acquired by the communication unit 52. The acquired culture medium image data is image data obtained from the culture results based on the type of specimen, type of diluent, dilution rate, culture conditions, etc. Conditions such as type or dilution rate, culture conditions, etc. can be considered to be implicitly included. Therefore, when the learning model 61 learns culture medium image data as learning data, conditions such as the type of specimen, type of diluent, dilution rate, culture conditions, etc. are implicitly included in the learning data. It will be done.

However, by adjusting the detection processing parameters specified by the learning model 61 based on the supplementary information, it is possible to obtain more optimal detection processing parameters and further improve the colony detection accuracy.

The output processing unit 56 has a function as an output unit, and outputs a predetermined notification when the reliability of the type of culture medium identified by the learning model 61 is less than or equal to a predetermined threshold. The reliability is an index indicating how reliable the identification result of the learning model 61 is, and the threshold can be set to an appropriate value such as 98% or 95%, for example. The predetermined notification includes, for example, notification that the type of medium identified by the learning model 61 is incorrect or that colonies cannot be detected. As a result, if a colony cannot be detected, a notification to that effect is provided, so that erroneous colony detection can be prevented.

The detection unit 63 has a function as a setting unit, and can set necessary detection logic for detecting colonies when the output processing unit 56 outputs a predetermined notification. For example, even if the type of culture medium cannot be identified and the detection logic cannot be specified, the required detection logic can be set, so colonies can be detected.

If the detection unit 63 cannot detect colonies based on the acquired culture medium image data, the determination unit 55 determines the type of medium in the culture medium image data based on the medium type information (medium type) acquired by the communication unit 52. The type can be determined. Even when the learning model 61 cannot identify the type of medium, the type of medium can be determined based on the medium type information, so the learning effort of the learning model 61 can be reduced.

The control unit 51 can collect the culture medium image data input to the learning model 61 and the type of culture medium identified by the learning model 61 into the storage unit 53 as learning data. The learning processing unit 62 can update the learning model 61 using the collected learning data. The collected learning data can include data when the learning model 61 makes an incorrect identification. Thereby, the learning model 61 can be retrained, and the discrimination ability of the learning model 61 can be improved.

FIG. 9 is a schematic diagram showing a second example of the learning method of the learning model 61 by the learning processing unit 62. The difference from the first example illustrated in FIG. 6 is that detection logic is added as learning data. Specifically, detection logic information (detector information) for specifying each of a plurality of detection logics and detection processing parameter information for specifying each of various detection processing parameters can be used as learning data. Note that although the example in FIG. 9 shows a configuration in which the culture medium type is used as learning data, the culture medium type may not be included in the learning data.

The learning model 61 is generated using learning data including a culture medium image, detection logic information of a detector for detecting colonies, and detection processing parameter information. Thereby, when culture medium image data is input to the learning model 61, the learning model 61 can specify detection logic information and detection processing parameter information based on the input culture medium image data.

FIG. 10 is a schematic diagram showing a second example of the colony detection method by the processing unit 60. As shown in FIG. 9 described above, the learning model 61 is generated using learning data including a culture medium image, detection logic information of a detector for detecting colonies, and detection processing parameter information. information and detection processing parameter information can be specified. For example, if the detection logics are detection logics A, B, C, and D, the detection logic information is the optimality of each of the detection logics A, B, C, and D (e.g., how suitable is it to employ the detection logics? evaluation index) can be shown as a numerical value or a percentage. Further, the detection processing parameter information can indicate, for example, the degree of optimality when using each of various detection processing parameters as a numerical value or a ratio.

The detection unit 63 can generate a new detection logic 633 by combining the detection logics A, B, C, and D using the detection logic information specified by the learning model 61. As a result, there is no need to specify the culture medium type, and the detection logic and detection processing parameters can be specified together, making it possible to detect colonies while responding flexibly to new types in the future. Become.

FIG. 11 is a schematic diagram showing a third example of the colony detection method by the processing unit 60. For example, when medium image data of a medium in which an unknown specimen such as a specimen that has not been tested in the past has been cultured is input to the learning model 61, the medium type identified by the learning model 61 and the optimal Even if such detection logic is used, it is assumed that the detection unit 63 may not be able to detect a colony. In the example of FIG. 11, a symbol X schematically indicates that the learning model 61 is unable to provide, for example, optimal detection processing parameter information to the detection unit 63.

In such a case, the detection logic selection unit 634 can select (generate) a new detection logic based on the supplementary information related to the culture medium type and the culture medium image data. As a result, even when testing an unknown specimen, the detection logic is automatically adjusted, making it possible to detect colonies.

FIG. 12 is a schematic diagram showing an example of a method for learning a learning model by the learning server 200. The learning server 200 has the same configuration as the colony detection device 50 illustrated in FIG. 1 . The learning server 200 acquires learning information from Company H's hygiene management system 100h, acquires learning information from Company J's hygiene management system 100j, and acquires learning information from Company K's hygiene management system 100k. . In addition, although the example of FIG. 12 illustrates the hygiene management systems of three companies, the sources from which learning information is obtained are not limited to three companies.

Information such as detection logic and detection parameters is recorded in the culture medium information DB 210. Moreover, acquired learning information (culture medium image data, etc.) can be recorded in the culture medium information DB 210.

The learning server 200 learns and updates (generates) a learning model using the learning information acquired from the hygiene management system 100h of Company H, and uses the learned and updated learning model as a learning model for Company H. can be introduced into the colony detection device 50h. Similarly, the learning server 200 learns and updates (generates) a learning model using the learning information acquired from the hygiene management system 100j of Company J, and uses the learned and updated learning model as a learning model for Company J. It can be introduced into the colony detection device 50j for Company J. The learning server 200 also learns and updates (generates) a learning model using the learning information acquired from the hygiene management system 100k of company K, and uses the learned and updated learning model as a learning model for company K. It can be introduced into a corporate colony detection device 50k.

As described above, the acquired culture medium image data is classified by acquisition source. The acquisition source can be, for example, a company or a company. Different companies handle different specimens and culture conditions. A learning model is generated for each acquisition source using learning data including culture medium image data classified for each acquisition source. Thereby, it is possible to generate a learning model optimized for each company or company, and it is possible to improve the discrimination ability of the learning model.

FIG. 13 is a flowchart showing an example of a learning process procedure for the learning model 61 by the learning processing unit 62. In the following, for convenience, the main body of processing will be described as the learning processing section 62. The learning processing unit 62 acquires culture medium image data (S11) and acquires the type of culture medium (medium type) (S12). The learning processing unit 62 acquires colony position information and colony detection reliability from the detection unit 63 (S13), and acquires detection processing parameters used by the detection unit 63 (S14).

The learning processing unit 62 generates teacher data based on the acquired culture medium image data, type of culture medium, detection processing parameters, colony position information, and colony detection reliability (S15), and performs learning and updating of the learning model 61. (S16).

The learning processing unit 62 determines whether or not to update the detection processing parameters (S17), and when updating (YES in S17), updates the detection processing parameters (S18) and continues the processing from step S13 onwards. Thereby, the learning model 61 can be trained by changing the detection processing parameters little by little.

If the detection processing parameters are not updated (NO in S17), the learning processing unit 62 determines whether there is other culture medium image data for learning (S19), and if there is other culture medium image data (YES in S19), The processing from step S11 onwards is continued. If there is no other culture medium image data (NO in S19), the learning processing unit 62 ends the process.

FIG. 14 is a flowchart showing an example of a colony detection processing procedure by the colony detection device 50. Hereinafter, for convenience, the main body of processing will be described as the control unit 51. The control unit 51 acquires culture medium image data (S31), and determines the type of culture medium (medium type) (S32). The control unit 51 specifies detection processing parameters (S33) and specifies detection logic (S34). Note that the detection processing parameters are specified by the learning model 61, and the detection logic can be specified according to the type of culture medium identified by the learning model 61.

The control unit 51 performs image processing on the acquired culture medium image data using the specified detection logic and detection processing parameters to detect colonies (S35). Detection of colonies includes, for example, detecting location information of colonies and determining reliability of colony detection.

The control unit 51 determines whether the reliability of colony detection is less than or equal to a predetermined threshold (S36), and if the reliability is not less than or equal to the predetermined threshold (NO in S36), colony detection result information (for example, FIG. 8) is generated and output (S37), and the process ends. If the reliability is less than or equal to the predetermined threshold (YES in S36), the control unit 51 outputs a notification that colonies cannot be detected (S38), and ends the process.

In addition, when the notification that colony detection is not possible is output in step S38, the control unit 51 (detection unit 63) may set the necessary detection logic for detecting colonies. This allows colony detection

The control unit 51 and the processing unit 60 of this embodiment can also be realized using a general-purpose computer equipped with a CPU (processor), GPU, RAM (memory), and the like. In other words, by loading a computer program that defines the procedure of each process as shown in FIGS. The control section 51 and the processing section 60 can be realized by the following. The computer program may be recorded on a recording medium and distributed. The learning model 61 learned by the learning server 200 and the computer program based on the learning model 61 may be distributed to and installed on the colony detection device 50 via the network 1.

In the above-described embodiment, the culture medium image data is acquired by reading the film-type culture medium 20 with a scanner, but the invention is not limited to this. Culture medium image data may be acquired by imaging with.

In the embodiment described above, if there are no colonies, the following processing can be performed, for example. As a first process, if the culture medium type is not included in the supplementary information, it is possible to manually input, for example, AC as 0 and EC as 0. As a second process, the culture medium type can be recorded only when there are colonies, such as 100 SA and 0 otherwise.

The colony detection device of this embodiment includes a culture medium image acquisition unit that acquires culture medium image data of a culture medium in which microorganisms are cultured, and a culture medium image data acquired by the culture medium image acquisition unit, and learning data including a culture medium image and a culture medium type. a specifying unit that specifies a detector for detecting colonies based on the type of culture medium identified by the discriminator; and a detection unit that detects colonies based on the culture medium image data acquired by the image acquisition unit.

The computer program of this implementation includes processing for acquiring culture medium image data of a culture medium in which microorganisms are cultured, and learning data generated using the acquired culture medium image data and learning data including culture medium image data and culture medium type. A process of identifying a detection algorithm for detecting colonies based on the culture medium type input to a model and identified by the learning model, and a process of detecting colonies based on culture medium image data acquired using the specified detection algorithm. Execute.

The colony detection method of this implementation acquires medium image data of a medium in which microorganisms are cultured, and inputs the acquired medium image data to a discriminator generated using learning data including the medium image data and the medium type. Then, a detector for detecting colonies is specified based on the medium type identified by the discriminator, and colonies are detected based on medium image data acquired using the specified detector.

The discriminator generation method of this embodiment acquires medium image data of a medium in which microorganisms are cultured, acquires the medium type of the medium image data, and uses learning data including the acquired medium image data and medium type. Generate a classifier.

The discriminator of this embodiment is generated using learning data including medium image data of a medium in which microorganisms are cultured and the medium type of the medium image data.

The medium image acquisition unit acquires medium image data of a medium in which microorganisms are cultured. Specifically, for example, an appropriate sterilized diluent is added to a sample (specimen) of a film-type medium, the sample liquid is diluted, and the film-type medium is cultured under the required culture conditions using an optical reader (e.g., a scanner, etc.). ) can acquire the culture medium image data obtained by reading.

The classifier is generated using learning data including culture medium image data and culture medium type. The discriminator can use a multilayer neural network (deep learning), for example, a convolutional neural network, but may also use other machine learning. The medium type indicates the type of medium, for example, a medium for measuring the number of viable bacteria (also referred to as AC), a medium for measuring the number of coliform bacteria (also referred to as CC), a medium for measuring the number of E. coli/coliform bacteria (EC). (also referred to as SA), medium for measuring the number of Staphylococcus aureus (also referred to as SA), etc.

The discriminator is generated (learning) using learning data including culture medium image data and culture medium type, so for example, if culture medium image data of unknown culture medium type is input to the discriminator, the discriminator will The medium type can be identified based on the medium image data.

The identification unit inputs the culture medium image data acquired by the culture medium image acquisition unit to the discriminator, and specifies a detector for detecting colonies based on the culture medium type identified by the discriminator. A detector is a detection logic or detection algorithm. Thereby, the optimal detector can be specified according to the type of medium (medium type).

The detection unit detects colonies based on the culture medium image data acquired by the culture medium image acquisition unit using the detector specified by the identification unit. A colony is a collection of bacteria that is created by culturing bacteria so small that they are invisible to the naked eye and growing to the point where they can be seen. The number of bacteria is determined by counting the colonies.

With the above configuration, by simply inputting culture medium image data into the discriminator, the culture medium type can be identified, and the optimal detector for the identified culture medium type can be identified to detect colonies, making it possible to select the detection logic. This eliminates the need for such complicated operations, and improves the accuracy of colony detection.

In the colony detection device of this embodiment, the detection unit further generates the culture medium image data acquired by the culture medium image acquisition unit using learning data including detection processing parameters for detecting colonies and colony detection result information. colony is detected based on the detection processing parameters identified by the discriminator and the detector.

The discriminator generation method of this embodiment acquires detection result information of colonies corresponding to the acquired culture medium image data, and generates detection processing parameters for detecting colonies, the detection processing parameters corresponding to the detection result information. is acquired, and the discriminator is generated using learning data including the acquired detection result information and detection processing parameters.

The classifier is generated (learned) using learning data including detection processing parameters for detecting colonies and colony detection result information. The detection result information can include colony position information and the reliability of colony detection (for example, the probability of how accurate the colony detection result is). For example, each time a discriminator is given culture medium image data as learning data and the detection processing parameters are updated, the updated detection processing parameters and colony detection result information obtained as a result of the update are given to the discriminator as learning data. can be learned. Thereby, the discriminator can identify the detection processing parameters necessary to make the reliability of colony detection equal to or higher than a predetermined threshold.

The detection unit inputs the culture medium image data acquired by the culture medium image acquisition unit into the discriminator, and detects colonies based on the detection processing parameters and the detector identified by the discriminator.

With the above configuration, just by inputting culture medium image data into the discriminator, the optimal detection processing parameters for the detector can be identified and colonies can be detected, eliminating the need for complicated work such as specifying detection processing parameters. This eliminates the need for additional work and improves the accuracy of colony detection.

The colony detection device of this embodiment includes an additional information acquisition unit that acquires additional information related to the culture medium image data acquired by the medium image acquisition unit, and a detection processing parameter specified by the discriminator, which is acquired by the additional information acquisition unit. and an adjustment unit that performs adjustment based on the incidental information obtained.

The supplementary information acquisition unit obtains supplementary information related to the acquired culture medium image data. For example, by providing an area in which characters or symbols can be written in the margins of a film-type medium excluding the culture layer, and writing additional information in this area, the film-type medium can be read by an optical reader. , additional information can be obtained. The supplementary information can include, for example, the type of specimen, the type or dilution rate of diluent, culture conditions, and the like.

The adjustment unit adjusts the detection processing parameters specified by the classifier based on the additional information acquired by the additional information acquisition unit. The medium image data acquired by the medium image acquisition unit is image data obtained from the culture results based on the type of specimen, type of diluent, dilution rate, culture conditions, etc. It can be considered that conditions such as the type of diluent, the type or dilution rate of the diluent, and culture conditions are implicitly included. Therefore, when a discriminator learns culture medium image data as learning data, conditions such as the type of specimen, type of diluent or dilution rate, and culture conditions are considered to be implicitly included in the learning data. .

However, by adjusting the detection processing parameters specified by the discriminator based on the supplementary information, it is possible to obtain even more optimal detection processing parameters, and it is possible to further improve the accuracy of colony detection.

The colony detection device of this embodiment includes an output unit that outputs a predetermined notification when the reliability of the culture medium type identified by the discriminator is less than or equal to a predetermined threshold.

The output unit outputs a predetermined notification when the reliability of the culture medium type identified by the discriminator is less than or equal to a predetermined threshold. The reliability is an index indicating how reliable the classification result of the classifier is, and the threshold can be set to an appropriate value such as 98% or 95%, for example. The predetermined notification includes, for example, notification that the culture medium type identified by the discriminator is incorrect or that colonies cannot be detected. As a result, if a colony cannot be detected, a notification to that effect is provided, so that erroneous colony detection can be prevented.

The colony detection device of this embodiment includes a setting section that sets a required detector for detecting colonies when the output section outputs the notification.

The setting unit sets a required detector for detecting colonies when the output unit outputs the notification. For example, even if the medium type cannot be identified and the detector cannot be specified, colonies can be detected because the required detector can be set.

The colony detection device of this embodiment includes a type information acquisition unit that acquires medium type information of the culture medium image data acquired by the medium image acquisition unit, and a case where no colony is detected based on the medium image data acquired by the medium image acquisition unit. , a determination unit that determines the culture medium type of the culture medium image data based on the culture medium type information acquired by the type information acquisition unit.

The type information acquisition unit acquires culture medium type information of the culture medium image data acquired by the culture medium image acquisition unit. For example, medium type information can be written in the margin of a film-type medium excluding the culture layer. The medium type information includes the medium type, and can be written in, for example, letters or symbols.

The determination unit determines the culture medium type of the culture medium image data based on the culture medium type information acquired by the type information acquisition unit when no colony is detected based on the culture medium image data acquired by the culture medium image acquisition unit. Even when the discriminator cannot identify the culture medium type, the type of culture medium can be determined based on the culture medium type information, so the learning effort of the discriminator can be reduced.

The colony detecting device of this embodiment detects a colony based on the culture medium type identified by the discriminator and supplementary information related to the culture medium image data acquired by the medium image acquisition unit when the colony cannot be detected by the detection unit. The apparatus includes a generation unit that generates a detector for the generation unit, and uses the detector generated by the generation unit to detect colonies based on the culture medium image data acquired by the culture medium image acquisition unit.

For example, when medium image data of a medium in which an unknown specimen, such as a specimen that has not been tested in the past, has been cultured is input to a discriminator, the optimal detection for the medium type identified by the discriminator and the medium type. Even if a container is used, there may be cases where the detection unit cannot detect colonies. Therefore, if a colony cannot be detected by the detection unit, a new detector is generated based on the culture medium type and the accompanying information related to the culture medium image data, so that the detector can be used even when testing an unknown specimen. Automatic adjustment makes it possible to detect colonies.

The colony detection device of this embodiment includes a medium image acquisition section that acquires medium image data of a medium in which microorganisms are cultured, and a medium image data acquisition section that acquires medium image data of a medium in which microorganisms are cultured. input into a discriminator generated using learning data including detector data and detection processing parameter data of the detector, and using a detector generated based on the detector information and detection processing parameter information specified by the discriminator. and a detection section that detects colonies based on the culture medium image data acquired by the culture medium image acquisition section.

The medium image acquisition unit acquires medium image data of a medium in which microorganisms are cultured. Specifically, for example, an appropriate sterilized diluent is added to a sample (specimen) of a film-type medium, the sample liquid is diluted, and the film-type medium is cultured under the required culture conditions using an optical reader (e.g., a scanner, etc.). ) can acquire the culture medium image data obtained by reading.

The classifier is generated using learning data including a culture medium image, detector information of a detector for detecting colonies, and detection processing parameter information. Thereby, when culture medium image data is input to the discriminator, the discriminator can specify detector information and detection processing parameter information based on the input culture medium image data.

The detection unit can detect colonies based on the culture medium image data acquired by the culture medium image acquisition unit using the detector information and detection processing parameter information specified by the discriminator. This eliminates the need to specify the type of culture medium and allows the detector and detection processing parameters to be specified at the same time, making it possible to detect colonies while responding flexibly to new types in the future. Become.

The colony detection method of this implementation involves reading a medium in which microorganisms are cultured in a culture layer formed on a film-like base sheet with an optical reader, and acquiring medium image data obtained by reading with the optical reader. do.

Culture medium image data can be obtained by reading a medium in which microorganisms are cultured in a culture layer formed on a film-like base sheet with an optical reader.

The discriminator generation method of the present embodiment collects culture medium image data input to the discriminator and the culture medium type identified by the discriminator as learning data, and updates the discriminator using the collected learning data.

The culture medium image data input to the classifier and the culture medium type identified by the classifier are collected as learning data, and the classifier is updated using the collected learning data. The learning data to be collected can include data when the classifier makes an incorrect classification. Thereby, the classifier can be retrained and the discrimination ability of the classifier can be improved.

In the discriminator generation method of the present embodiment, acquired culture medium image data is classified for each acquisition source, and the discriminator is generated for each acquisition source using learning data including the classified culture medium image data for each acquisition source.

The acquired culture medium image data is classified by acquisition source. The acquisition source can be, for example, a company or a company. Different companies handle different specimens and culture conditions. A classifier is generated for each acquisition source using learning data including the classified culture medium image data for each acquisition source. Thereby, it is possible to generate a discriminator optimized for each company or company, and it is possible to improve the discrimination ability of the discriminator.

20 Film-type medium 21 Base sheet 22 Frame 23 Culture layer 24 Cover sheet 50 Colony detection device 51 Control section 52 Communication section 53 Storage section 54 Adjustment section 55 Judgment section 56 Output processing section 60 Processing section 61 Learning model 62 Learning processing section 63 Detection unit 631 Detection logic unit 632 Detection processing parameter unit

Claims (1)

a medium image acquisition unit that acquires medium image data of a medium in which microorganisms are cultured;
The medium image data acquired by the medium image acquisition unit is input to the discriminator generated using the learning data including the medium image, detector information of a detector for detecting colonies, and detection processing parameter information, and A colony detection device comprising: a detection section that detects colonies based on culture medium image data acquired by the culture medium image acquisition section using a detector generated based on detector information and detection processing parameter information specified by a discriminator.

Images