Deep learning-based sample tube code reading identification method, code reading device and storage library

Abstract

The invention relates to the technical field of biomedical treatment, in particular to a sample tube code reading recognition method and device based on deep learning and a storage library, and aims to solve the problems that an existing code reading scheme cannot recognize abnormal conditions such as frosting and light reflection and a decoding recognition result is wrong. And correct prompt information cannot be given to the user. In order to achieve the purpose, the recognition method comprises the steps of obtaining a sample picture at the bottom of a sample box; according to the sample picture and a preset sample tube recognition model, judging whether a sample tube which is not recognized exists or not; if yes, inputting the pictures corresponding to the sample tubes which are not recognized into a preset sample tube recognition model for training to obtain an updated sample tube recognition model; and if not, obtaining corresponding information of the sample tube according to a sample tube identification model. According to the invention, normal management, abnormal management and non-management states can be directly distinguished, the two-dimensional code content can be quickly read, and the code identification efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and specifically provides a sample tube code reading and recognition method, a code reading device and a storage library based on deep learning. Background Art

[0002] -196 degrees deep cryogenic biological sample storage is an important basic equipment for current research in the medical and biological fields. Through the cryogenic storage of liquid nitrogen, biological tissues such as blood, stem cells and immune cells can remain active for a long time. In order to realize the information management of biological samples, each sample tube needs to be coded, and the code needs to be scanned to enter the information when the sample is stored, and the code also needs to be scanned for verification when it is out of the warehouse.

[0003] In a deep cryogenic environment, frost may form on the surface of the sample tube, and the surface of the sample tube may be reflective or blocked due to mechanical structure or other reasons. This will make it impossible to identify the bottom information of the sample tube, and then it will be impossible to know whether there is a tube at the current position, affecting the robot arm's grasping and the overall process. The current solution is to use the principle of scanning the code by taking a photo with a camera. The principle is to pre-process the image first, that is, grayscale and defrost it, and use the image segmentation algorithm on the processed image to obtain a pre-processed image containing a single QR code, and then decode and identify the QR code in the pre-processed image to obtain a decoding and recognition result. The disadvantage of this solution is that it cannot identify abnormal conditions such as frost, reflection or blockage, and there will be errors in the decoding and recognition results, and thus it is impossible to give the user correct prompt information.

[0004] Therefore, the present invention needs to provide a new technical solution to solve the above technical problems. Summary of the invention

[0005] The present invention aims to solve the above technical problem, that is, to solve the problem that the existing code reading solution cannot identify the situation of frost and reflection and may produce errors in the decoding and recognition results, and thus cannot provide the user with correct prompt information.

[0006] In a first aspect, the present invention provides a sample tube code reading and identification method based on deep learning, wherein the sample tube is stored in a sample box, and the sample tube code reading and identification method comprises: obtaining a sample image at the bottom of the sample box; judging whether there are unrecognizable sample tubes based on the sample image and a preset sample tube recognition model; if so, inputting the image corresponding to the unrecognizable sample tube into the preset sample tube recognition model for training to obtain an updated sample tube recognition model; if not, obtaining the corresponding information of the sample tube based on the sample tube recognition model.

[0007] In the specific implementation of the above-mentioned sample tube code reading and recognition method based on deep learning, the step of "determining whether there are unrecognizable sample tubes based on the sample image and the preset sample tube recognition model" specifically includes: using the image code reading algorithm to obtain the first feature information of the sample image; wherein the first feature information includes at least one of recognizable two-dimensional code information and two-dimensional code position information; inputting the sample image into the preset sample tube recognition model and comparing it to obtain the second feature information; wherein the second feature information includes at least one of the tube-with and tube-free states, tube-with position information and tube-free position information; based on the first feature information and the second feature information, determining whether there are unrecognizable sample tubes.

[0008] In the specific implementation of the above-mentioned sample tube code reading and recognition method based on deep learning, the step of "determining whether there are unrecognizable sample tubes based on the first feature information and the second feature information" specifically includes: performing coordinate conversion on the position information of the identifiable QR code, the tube position information and the tubeless position information to obtain the position information of the QR code, the tube position information and the tubeless position information in the same coordinate system; obtaining matching results based on the position information, the tube position information and the tubeless position information of the QR code in the same coordinate system; wherein the matching results include identifiable sample tubes and unrecognizable sample tubes; and determining whether there are unrecognizable sample tubes based on the matching results.

[0009] In the specific implementation of the above-mentioned sample tube code reading and recognition method based on deep learning, the step of "inputting the images corresponding to the unrecognizable sample tubes into the preset sample tube recognition model for training" specifically includes: first marking the images corresponding to the unrecognizable sample tubes as abnormal images; and inputting the marked abnormal images into the preset sample tube recognition model for training.

[0010] In the specific implementation of the above-mentioned sample tube code reading and recognition method based on deep learning, the step of "obtaining the corresponding information of the sample tube according to the sample tube recognition model" specifically includes: inputting the sample image into the updated sample tube recognition model and comparing it to obtain the third feature information; wherein the third feature information includes at least one of the states of with tube, without tube and abnormal tube, with tube position information, without tube position information and abnormal tube position information; according to the first feature information and the third feature information, the corresponding information of the sample tube is respectively obtained as the tube with code and the corresponding QR code information, the tube without code and the tube without code.

[0011] In the specific implementation of the above-mentioned sample tube code reading and identification method based on deep learning, the sample tube code reading and identification method also includes: before obtaining the tube, code and corresponding QR code information, first obtaining the pre-entered actual information; comparing the position information of the sample tube and the corresponding QR code information with the actual information; if the comparison results are consistent, directly outputting the tube, code and corresponding QR code information as the corresponding information of the sample tube.

[0012] In a specific implementation of the above-mentioned sample tube code reading and recognition method based on deep learning, the sample tube code reading and recognition method also includes: if the comparison result is inconsistent, the image corresponding to the sample tube with inconsistent comparison result is marked as an abnormal image, and the marked abnormal image is input into the preset sample tube recognition model for training to obtain an updated sample tube recognition model.

[0013] In the specific implementation of the above-mentioned sample tube code reading and recognition method based on deep learning, the sample tube code reading and recognition method also includes: the sample tube code reading and recognition method also includes: before marking the abnormal picture, pushing the abnormal picture to the user and sending a marking request; when the marking request is "yes", marking the abnormal picture.

[0014] In a second aspect, the present invention further provides a sample tube code reading device, comprising a controller, wherein the controller is configured to be able to execute the sample tube code reading and identification method as described above.

[0015] In a third aspect, the present invention also provides a biological automated storage library, comprising the sample tube code reading device as described above.

[0016] In the case of adopting the above technical solution, the present invention can obtain the sample image at the bottom of the sample box; according to the sample image and the preset sample tube recognition model, it is determined whether there are sample tubes that cannot be recognized; if there are, the image corresponding to the sample tube that cannot be recognized is input into the preset sample tube recognition model for training to obtain an updated sample tube recognition model; if not, the corresponding information of the sample tube is obtained according to the sample tube recognition model. The present invention uses a deep learning algorithm to input the sample image into the sample tube recognition model to treat abnormal conditions such as frost, reflection or occlusion as normal conditions, and can directly distinguish between normal tubes, abnormal tubes and no tube states, and quickly read the content of the QR code. Compared with the prior art of graying and defrosting the image before recognizing the QR code, the code recognition efficiency and accuracy are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings, in which:

[0018] Figure 1It is a flow chart of the main steps of the sample tube code reading and recognition method based on deep learning of the present invention;

[0019] Figure 2 This is a possible complete step flow chart of the sample tube code reading and recognition method based on deep learning of the present invention. DETAILED DESCRIPTION

[0020] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not used to limit the scope of protection of the present invention. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0021] It should be noted that in the description of the present invention, the terms such as "bottom" indicating direction or position relationship are based on the direction or position relationship shown in the drawings, which is only for the convenience of description, and does not indicate or imply that the relevant device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the ordinal numbers "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0022] In addition, it should be noted that in the description of the present invention, although the various steps of the control method of the present invention are described in this application in a specific order, these orders are not restrictive. Without departing from the basic principles of the present invention, those skilled in the art can perform the steps in a different order.

[0023] See first Figure 1 , which is a flow chart of the main steps of the sample tube code recognition method based on deep learning of the present invention. Figure 1 As shown, the sample tube code reading and identification method of the present invention includes:

[0024] S1, obtain the sample image at the bottom of the sample box;

[0025] S2, judging whether there are unrecognized sample tubes according to the sample image and the preset sample tube recognition model;

[0026] S3, if there is, input the image corresponding to the sample tube that cannot be identified into a preset sample tube recognition model for training to obtain an updated sample tube recognition model;

[0027] S4: If it does not exist, obtain the corresponding information of the sample tube according to the sample tube identification model.

[0028] In step S1, in some embodiments, a sample image at the bottom of the sample box is obtained by a camera. Using a camera to take pictures helps reduce costs.

[0029] In step S2, the step of “determining whether there are unrecognized sample tubes based on the sample image and the preset sample tube recognition model” specifically includes:

[0030] S21, using an image code reading algorithm to obtain first feature information from the sample image; wherein the first feature information includes at least one of recognizable two-dimensional code information and two-dimensional code position information;

[0031] S22, inputting the sample image into a preset sample tube recognition model and performing comparison to obtain second characteristic information; wherein the second characteristic information includes at least one of a tube-with and tube-without state, tube-with position information, and tube-without position information;

[0032] S23, judging whether there are any sample tubes that cannot be identified based on the first characteristic information and the second characteristic information.

[0033] In step S21, in some embodiments, the image code reading algorithm first obtains the position information of each two-dimensional code in the sample image, then decodes each two-dimensional code, and finally outputs all recognizable two-dimensional code information and corresponding position information. The first feature information obtained by the image code reading algorithm may include one or both of the recognizable two-dimensional code information and the two-dimensional code position information. Exemplarily, the first feature information includes the recognizable two-dimensional code information and the two-dimensional code position information.

[0034] In step S22, in some embodiments, the preset sample tube recognition model is a sample tube recognition model obtained by training through a deep learning algorithm based on sufficient sample data images collected in advance. Specifically, the sufficient sample data images collected in the above-mentioned manner are sample images of sample tubes stored in different positions on each type of sample box, which are collected in advance in a preset number of sheets. There are many types of sample boxes for storing sample tubes in the prior art, mainly 10*10 (10 rows and 10 columns), 12*8 (12 rows and 8 columns), 8*6 (8 rows and 6 columns), and 6*4 (6 rows and 4 columns), so one of each of these types of sample boxes is found. Then, for each type of sample box storing sample tubes in different positions, take pictures to obtain sample pictures. Here, different positions refer to different positions of the sample tubes on the sample box in each sample picture. The number of sample tubes in each sample picture is not limited in this application and can be set arbitrarily. The placement position of the sample tubes can also be set arbitrarily. For example, sample tubes are placed in the four corners of the sample box for taking pictures. However, these sample images must include images of empty boxes and full boxes. Among them, regarding the preset number of sample images recorded above, each type of sample box needs to take a preset number of pictures, which can be 100, 200, 500, etc. The specific value of the preset number of pictures is flexibly set according to the accuracy and difficulty of recognition of the trained sample recognition model.

[0035] Further, the preset sample tube identification model is a data model including the state of tube and no tube, the position information of tube and no tube. The second characteristic information obtained according to the sample tube identification model may include one or two or three or four of the state of tube and no tube, the position information of tube and no tube. Exemplarily, the second characteristic information includes the state of tube and no tube, the position information of tube and no tube.

[0036] It should be noted that the above-mentioned QR code position information, piped position information and unpiloted position information can be the coordinate information (X, Y) in the sample image, or can be positioned based on the rows and columns in the sample image, that is, the row and column. Technical personnel in this field can set it as needed, and no limitation is made in the present invention.

[0037] In step S23, the step of “determining whether there is an unrecognized sample tube according to the first characteristic information and the second characteristic information” specifically includes:

[0038] S231, coordinate conversion is performed on the position information of the identifiable two-dimensional code, the position information with control, and the position information without control to obtain the position information of the two-dimensional code, the position information with control, and the position information without control in the same coordinate system;

[0039] S232, obtaining a matching result according to the position information of the two-dimensional code, the position information with the tube, and the position information without the tube in the same coordinate system; wherein the matching result includes the sample tube that can be identified and the sample tube that cannot be identified;

[0040] S233, determine whether there are any sample tubes that cannot be identified based on the matching results.

[0041] In some embodiments, in a deep cryogenic environment, it is inaccurate to identify whether there is a tube in the sample box according to a preset sample tube recognition model, that is, when outputting a tubeless state, it may also include a tubeless state, but due to reflection or frost, the recognition result is a tubeless state, which makes it impossible to determine whether the failure to recognize the QR code is because the code is not recognized because there is a tube or because the code is not recognized because there is no tube. The specific matching process is to obtain the tube-with-code and tube-without-code and the corresponding confidence based on the position information of the QR code, the tube position information and the tubeless position information in the same coordinate system, and determine whether there is an unrecognized sample tube according to the size of the confidence. When there is an image with a confidence less than the confidence threshold, it is determined that there is an unrecognized sample tube. When there is no image with a confidence less than the confidence threshold, it is determined that there is no unrecognized sample tube.

[0042] In step S3, the step of “inputting the image corresponding to the unrecognized sample tube into the preset sample tube recognition model for training” specifically includes:

[0043] S31, first marking the image corresponding to the sample tube that cannot be identified as an abnormal image;

[0044] S32, inputting the marked abnormal images into a preset sample tube recognition model for training.

[0045] In some embodiments, the sample tube that cannot be identified specifically refers to the sample tube whose QR code cannot be identified, that is, the existence of the sample tube can be identified at a certain location where the sample tube is stored in the sample image, but the QR code information on the sample tube cannot be read. In actual operation, the sample box is transported between a deep cryogenic environment, a cryogenic environment, and a normal temperature environment. Therefore, during the transportation process, frost may appear on the surface of the sample tube, and due to the mechanical structure or other conditions, the surface of the sample tube may also reflect or be blocked. The frost, reflection, or blockage may affect the recognition of the QR code on the surface of the sample tube, resulting in a situation where the sample tube cannot be identified. In the preset sample tube recognition model, only the state of having a tube and the state of not having a tube can be identified, and the abnormal situation of frosting, reflection, or blockage cannot be identified. Therefore, when it is determined that there is an unrecognized sample tube, the corresponding picture is marked as an abnormal picture as a new training set, and the training set is input into the preset sample tube recognition model for training to obtain an updated sample tube recognition model. In addition to recognizing the state of having a tube and the state of not having a tube, the model can also recognize the abnormal situation of frosting or reflection. Specifically, the abnormal images mentioned above are sample images of frosting, reflection or occlusion under different circumstances, which are collected in a preset number of images. Similarly, regarding the preset number of sample images recorded above, the preset number of images can be 100, 200, 500, etc. The specific value of the preset number of images is flexibly set according to the accuracy and difficulty of recognition of the trained sample recognition model.

[0046] It should be noted that although the abnormal pictures mentioned above are described in conjunction with frosting, reflection or obstruction, this is not restrictive and may also be abnormal pictures in other situations, such as the situation where the QR code does not appear on the sample tube, that is, the situation where the QR code is omitted on the sample tube.

[0047] In step S4, the step of “obtaining corresponding information of the sample tube according to the sample tube identification model” specifically includes:

[0048] S41, inputting the sample image into the updated sample tube recognition model and performing comparison to obtain third feature information; wherein the third feature information includes at least one of the states of tube presence, tube absence and abnormal tube presence, tube presence position information, tube absence position information and abnormal tube position information;

[0049] S42, according to the first characteristic information and the third characteristic information, respectively obtain the corresponding information of the sample tube, namely, the tube with code and the corresponding QR code information, the tube without code and the tube without code.

[0050] In step S41, in some embodiments, the updated sample tube identification model is a data model including the states of tubes, tubes without and abnormal tubes, tube position information, tubeless position information, and abnormal tube position information. The third feature information obtained according to the updated sample tube identification model may include one or two or three or four or five or six of the states of tubes, tubes without and abnormal tubes, tube position information, tubeless position information, and abnormal tube position information. Exemplarily, the third feature information includes the states of tubes, tubes without and abnormal tubes, tube position information, tubeless position information, and abnormal tube position information. It should be noted that the abnormal tube position information described above may be the coordinate information (X, Y) in the sample image, or it may be positioned according to the rows and columns in the sample image, that is, the row and column. Those skilled in the art may set it as needed, and it is not limited in the present invention.

[0051] The sample tube code reading and identification method of the present invention also includes:

[0052] S5, before obtaining the tube and code and the corresponding QR code information, first obtain the actual information entered in advance.

[0053] In some embodiments, the actual information pre-entered refers to the mapping relationship between the position of the sample tube in the sample box, the QR code on the sample tube and the corresponding QR code information stored in the database in advance. By way of example, the QR code information includes but is not limited to the type of biological sample in the sample tube, the number of the sample tube, etc.

[0054] S6, comparing the location information of the sample tube and the corresponding QR code information with the actual information.

[0055] S7, if the comparison results are consistent, directly output the tube and code and the corresponding QR code information as the corresponding information of the sample tube.

[0056] S8, if the comparison results are inconsistent, the pictures corresponding to the sample tubes with inconsistent comparison results are marked as abnormal pictures, and the marked abnormal pictures are input into the preset sample tube recognition model for training to obtain an updated sample tube recognition model.

[0057] In some embodiments, for sample tubes with slight frost, reflection, or obstruction, such sample tubes will be considered to have tubes and codes, but when identifying the QR code, the surface is unclear, resulting in deviations in the read QR code. Therefore, the position information of the sample tube and the corresponding QR code information are compared with the actual information to ensure the accuracy of information recognition. For sample tubes with inconsistent comparison results, the corresponding images are marked, new training sets are added, and input into the preset sample tube recognition model for training to obtain an updated sample tube recognition model, so that the model can also recognize abnormal conditions such as slight frost, reflection, or obstruction.

[0058] The sample tube code reading and identification method of the present invention also includes:

[0059] S9, before marking the abnormal picture, push the abnormal picture to the user and send a marking request.

[0060] In some embodiments, when an abnormal picture appears, a marking request is sent to the user, such as "confirm whether to mark the picture".

[0061] S10: When the marking request is “yes”, the abnormal picture is marked.

[0062] In some embodiments, the present invention implements interaction with users, fits the actual needs of users, meets user demands, and at the same time improves the recognition accuracy and flexibility of the trained sample recognition model.

[0063] Refer to the following Figure 2 , which is a possible complete step flow chart of the sample tube code reading and recognition method based on deep learning of the present invention. Figure 2 As shown, after the washing machine finishes washing, the sample tube code reading and identification method of the present invention includes:

[0064] S101, obtaining a sample image at the bottom of the sample box;

[0065] S102, using an image code reading algorithm to obtain first feature information from the sample image;

[0066] S103, inputting the sample image into the sample tube recognition model and performing comparison to obtain the second feature information / the third feature information;

[0067] S104, judging whether there is an unrecognized sample tube according to the first characteristic information and the second characteristic information / the third characteristic information;

[0068] If yes, execute step S105;

[0069] S105, push the abnormal picture to the user and send a marking request;

[0070] If the marking request is "yes", execute step S106;

[0071] S106, inputting the image corresponding to the sample tube that cannot be identified into a preset sample tube identification model for training to obtain an updated sample tube identification model;

[0072] If not, execute step S107;

[0073] S107, obtaining actual information entered in advance;

[0074] S108, comparing the position information of the sample tube and the corresponding QR code information with the actual information;

[0075] If not, execute step S105;

[0076] If they are consistent, execute step S109;

[0077] S109, respectively outputting the information of the sample tube with the code and the corresponding QR code information, the sample tube without the code, and the sample tube without the code.

[0078] In summary, in the preferred technical solution of the present invention, a sample image at the bottom of the sample box is obtained; based on the sample image and the preset sample tube recognition model, it is determined whether there are sample tubes that cannot be recognized; if there are, the image corresponding to the sample tube that cannot be recognized is input into the preset sample tube recognition model for training to obtain an updated sample tube recognition model; if not, the corresponding information of the sample tube is obtained according to the sample tube recognition model. The present invention uses a deep learning algorithm to input the sample image into the sample tube recognition model to treat abnormal conditions such as frost, reflection or occlusion as normal conditions, and can directly distinguish between normal tubes, abnormal tubes and no tube states, and quickly read the content of the QR code. Compared with the prior art of graying and defrosting the image before recognizing the QR code, the code recognition efficiency and accuracy are improved.

[0079] Based on the above-mentioned embodiments, the present invention further provides a sample tube code reading device, including a controller, and the controller is configured to be able to execute the sample tube code reading and identification method described above.

[0080] Based on the above-mentioned embodiments, the present invention also provides a biological automated storage library, including the sample tube code reading device as described above.

[0081] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A sample tube code recognition method based on deep learning, characterized in that: The sample tube is stored in a sample box, and the sample tube code reading and identification method includes: Obtaining a sample image at the bottom of the sample box; According to the sample image and a preset sample tube recognition model, determining whether there are unrecognized sample tubes; If so, the image corresponding to the sample tube that cannot be identified is input into the preset sample tube identification model for training to obtain an updated sample tube identification model; If it does not exist, the corresponding information of the sample tube is obtained according to the sample tube identification model.

2. The sample tube code recognition method based on deep learning according to claim 1 is characterized in that: The step of "determining whether there are unrecognized sample tubes according to the sample image and the preset sample tube recognition model" specifically includes: Using an image code reading algorithm to obtain first feature information from the sample image; wherein the first feature information includes at least one of recognizable two-dimensional code information and two-dimensional code position information; Input the sample image into a preset sample tube recognition model and compare it to obtain second feature information; wherein the second feature information includes at least one of a tube-with and tube-without state, tube-with position information, and tube-without position information; Based on the first characteristic information and the second characteristic information, determine whether there are any sample tubes that cannot be identified.

3. The sample tube code recognition method based on deep learning according to claim 2 is characterized in that: The step of "determining whether there is an unrecognized sample tube based on the first characteristic information and the second characteristic information" specifically includes: The coordinates of the position information of the identifiable QR code, the position information with control and the position information without control are converted to obtain the position information of the QR code, the position information with control and the position information without control in the same coordinate system; According to the position information of the two-dimensional code, the position information with the tube and the position information without the tube in the same coordinate system, a matching result is obtained; wherein the matching result includes identifiable sample tubes and unidentifiable sample tubes; Determine whether there are any unidentified sample tubes based on the matching results.

4. The sample tube code recognition method based on deep learning according to claim 1, characterized in that: The step of "inputting the images corresponding to the sample tubes that cannot be identified into the preset sample tube identification model for training" specifically includes: First, the images corresponding to the sample tubes that cannot be identified are marked as abnormal images; The marked abnormal images are input into the preset sample tube recognition model for training.

5. The sample tube code reading method based on deep learning according to any one of claims 2 to 4, characterized in that: The step of "obtaining corresponding information of the sample tube according to the sample tube identification model" specifically includes: The sample image is input into the updated sample tube recognition model and compared to obtain third feature information; wherein the third feature information includes at least one of the states of tube presence, tube absence and abnormal tube presence, tube absence and abnormal tube position information; According to the first characteristic information and the third characteristic information, the corresponding information of the sample tube is obtained as follows: a tube with a code and the corresponding two-dimensional code information, a tube without a code, and no tube without a code.

6. The sample tube code reading and recognition method based on deep learning according to claim 5 is characterized in that: The sample tube code reading and identification method further includes: Before obtaining the tube and code and the corresponding QR code information, first obtain the actual information entered in advance; Comparing the position information of the sample tube and the corresponding QR code information with the actual information; If the comparison results are consistent, the tube, code and corresponding QR code information are directly output as the corresponding information of the sample tube.

7. The sample tube code reading and recognition method based on deep learning according to claim 6 is characterized in that: The sample tube code reading and identification method further includes: If the comparison result is inconsistent, the image corresponding to the sample tube with the inconsistent comparison result is marked as an abnormal image, and the marked abnormal image is input into a preset sample tube recognition model for training to obtain an updated sample tube recognition model.

8. The sample tube code reading and recognition method based on deep learning according to claim 4 or 7, characterized in that: The sample tube code reading and identification method further includes: Before marking the abnormal picture, push the abnormal picture to the user and send a marking request; If the marking request is "yes", the abnormal picture is marked.

9. A sample tube code reading device, characterized in that: It includes a controller, which is configured to be able to execute the sample tube code reading and identification method as described in any one of claims 1 to 8.

10. A biological automated storage library, characterized in that: Includes a sample tube code reading device as described in claim 9.

Images