Adaptive control method and device for sample identification in cryogenic environments, and biobank equipment
By acquiring sample identification status information in a deep cryogenic environment, determining the fault type, and executing corresponding control actions, the problem of abnormal sample identification was solved, the stability and success rate of identification were improved, and invalid operations and manual intervention were reduced.
Patent Information
- Application Number
- CN202611452586.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-17
- Publication Date
- 2026-11-17
AI Technical Summary
In deep cryogenic environments, sample identification equipment may experience abnormalities in identifying sample identification information due to factors such as frost, cold fog, reflection, changes in lighting, and freezing or jamming of mechanical actuators. Existing technologies are unable to effectively address different causes of these abnormalities, leading to frequent repetitive operations and manual intervention.
By acquiring the state information during the sample identification process, the target fault type causing the identification anomaly is determined, and corresponding control actions are executed according to the fault type to adjust the abnormal state. The identification is then performed again to verify the adjustment effect, thus forming a closed-loop control process.
It improves the adaptability of the sample identification process to dynamic low-temperature conditions, reduces the need for ineffective operations and manual intervention, and improves the stability and success rate of identification.
Smart Images

Figure CN123113373A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of biological sample identification technology, such as an adaptive control method and device for sample identification in deep cryogenic environments, and a biobank device. Background Technology
[0002] In the automated storage, transportation, and management of biological samples, it is often necessary to identify the labeling information on sample containers such as cryovials to confirm sample identity, match location, and facilitate subsequent automated operations. For example, an image acquisition device can be used to capture images of the two-dimensional barcode on the bottom of the cryovial or the label on the tube wall, and the corresponding sample identification information can be obtained through barcode decoding, optical character recognition, or other methods.
[0003] In cryogenic environments, samples and related operating equipment are subjected to prolonged periods of low temperature and high humidity. Sample surfaces, image acquisition devices, and mechanical actuators are susceptible to factors such as frost, cold fog, reflections, changes in lighting, and freezing, leading to blurred, overexposed, underexposed, obstructed, or sample pose deviations in the acquired images, thus affecting the normal identification of sample labeling information. Furthermore, these influencing factors typically change dynamically with environmental conditions and equipment operating status. Using fixed exposure parameters, fixed lighting methods, or fixed defrosting durations is insufficient to continuously adapt to varying actual operating conditions.
[0004] When existing automated equipment encounters sample recognition anomalies, it typically handles the situation by repeating the collection and recognition process or directly triggering an alarm and shutting down the machine. Although some vision-guided equipment can use visual information to complete sample positioning or mechanical operations, the visual information is mainly used to obtain position coordinates. It lacks the ability to further analyze the specific reasons for the current recognition anomalies, making it difficult to select appropriate adjustment methods for different anomaly causes. This can easily lead to problems such as repeatedly executing ineffective operations, frequent manual intervention, and untimely handling of local anomalies.
[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0006] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.
[0007] This disclosure provides an adaptive control method and apparatus for sample identification in cryogenic environments, as well as a biobank device, to improve the adaptability of the automatic sample identification process to dynamic cryogenic conditions.
[0008] In some embodiments, the adaptive control method for sample identification in cryogenic environments includes: identifying the sample to be identified to obtain a sample identification result; if the sample identification result does not meet preset identification conditions, acquiring state information related to the current sample identification and determining a target fault type that causes the current sample identification to be abnormal based on the state information; determining a target control action corresponding to the target fault type based on the target fault type; controlling the actuator corresponding to the target control action to execute the target control action to adjust the abnormal state corresponding to the target fault type; and re-identifying the sample to be identified after executing the target control action.
[0009] In some embodiments, the sample identification adaptive control device for cryogenic environments includes a processor and a memory storing program instructions, the processor being configured to execute the sample identification adaptive control method for cryogenic environments as described above when running the program instructions.
[0010] In some embodiments, the biobank device includes: a device body; and a sample identification adaptive control device for cryogenic environments, as described above, disposed on the device body.
[0011] The adaptive control method and apparatus for sample identification in cryogenic environments, and the biobank equipment provided in this disclosure can achieve the following technical effects: When a sample identification anomaly occurs, the target fault type causing the current identification anomaly is determined based on the status information related to the identification of the sample to be identified. The corresponding actuator is then controlled to adjust the abnormal state according to the target fault type. This ensures that the handling method for sample identification anomalies matches the actual cause of the anomaly, avoiding the use of fixed parameters or uniform retry methods to handle different types of identification anomalies. By re-identifying the sample after adjustment and verifying the adjustment effect, the adaptability of the sample identification process in cryogenic environments to dynamic conditions such as frost, changes in illumination, positional deviations, or equipment malfunctions is improved. This enhances the stability and success rate of sample identification, reducing unnecessary operations and the need for manual intervention.
[0012] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description
[0013] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1This is a flowchart of a sample identification adaptive control method for cryogenic environments provided in this disclosure. Figure 2 This is a schematic diagram of the identification process of the sample to be identified in an embodiment of this disclosure; Figure 3 This is a schematic diagram illustrating the process of determining the target fault type in an embodiment of this disclosure; Figure 4 This is a schematic diagram illustrating the process of determining the target control action corresponding to the target fault type in an embodiment of this disclosure; Figure 5 This is a schematic diagram illustrating the feedback verification process after executing the target control action in this embodiment of the disclosure; Figure 6 This is a schematic diagram of a sample identification adaptive control device for cryogenic environments provided in this embodiment of the present disclosure. Detailed Implementation
[0014] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.
[0015] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.
[0016] Unless otherwise stated, the term "multiple" means two or more.
[0017] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.
[0018] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.
[0019] The term "correspondence" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.
[0020] In the context of automated identification of cryogenic samples, existing equipment typically operates according to a fixed process of movement, photography, and identification, acquiring sample identification information through methods such as 2D barcode decoding and optical character recognition. When identification fails, it is usually handled by re-collecting, repeating the identification, or alarm shutdown. This type of solution can achieve basic automatic sample identification, but it often lacks the ability to differentiate and handle different abnormalities caused by dynamic changes in cryogenic environments, such as frost, cold fog, reflection, abnormal lighting, sample pose deviation, and freezing or jamming of mechanical actuators.
[0021] During their research, the inventors discovered that different recognition anomalies correspond to different causes and require different handling methods. For example, overexposure in imaging may require adjusting the lighting conditions, frost obstruction may require local defrosting, pose deviation may require adjusting the sample orientation, and mechanical freezing and jamming require judgment based on the operating status of the actuator. Therefore, if a uniform retrieval is performed based solely on the recognition failure result, or a fixed remedial action is adopted, it is easy to encounter the problem that the adjustment measures do not match the actual cause of the anomaly.
[0022] Based on the above understanding, this application constructs an adaptive control mechanism for sample recognition in cryogenic environments. This mechanism first identifies the sample to be identified; when the sample identification result does not meet preset identification conditions, it acquires state information related to the current sample identification, and determines the target fault type causing the current sample identification anomaly based on the state information, thereby determining the corresponding target control action; subsequently, it controls the corresponding actuator to adjust the abnormal state corresponding to the target fault type, and re-identifies the sample after the action is executed.
[0023] By using the above methods, matching adjustments can be made for different causes of anomalies, reducing inefficiencies caused by invalid repeated identification or uniform processing. The adjustment effect can be verified by re-identification, thereby improving the stability and success rate of sample identification in deep cryogenic dynamic environments and reducing the need for manual intervention.
[0024] Figure 1 This is a flowchart of an adaptive control method for sample identification in cryogenic environments, provided by an embodiment of this disclosure. This method can be applied to automated storage, transport, or identification equipment for biological samples. The sample to be identified can be a cryovial or other sample container equipped with a two-dimensional barcode, text label, or other identifiable identifier. During sample identification, an image of the sample to be identified can be acquired using an image acquisition device, and the corresponding sample identification information can be obtained through barcode decoding, text recognition, or other methods. In one implementation, the identifiable identifier may include a two-dimensional barcode located at the bottom of the cryovial and / or a text label located on the wall of the cryovial.
[0025] In cryogenic environments, the samples to be identified and related equipment may be affected by factors such as frost, cold fog, reflection, abnormal lighting, changes in sample pose, or freezing and jamming of mechanical actuators. Different abnormal factors may cause sample images to become blurry, occluded, overexposed, underexposed, or misaligned, further resulting in sample identification results that do not meet the identification requirements. Therefore, in this embodiment, when sample identification is abnormal, it does not directly repeat the identification in a fixed manner, but determines the type of fault that caused the identification abnormality based on the status information related to the current identification process, and performs corresponding adjustments for that type of fault.
[0026] like Figure 1 As shown, the sample recognition adaptive control method includes: Step S101: Identify the sample to be identified and obtain the sample identification result.
[0027] In this step, a sample image of the sample to be identified can be acquired, and the two-dimensional barcode, text label, or other sample identifiers in the sample image can be identified to obtain the corresponding sample identification result. The sample identification result can be used to determine whether the current sample to be identified can be correctly identified.
[0028] Step S102: If the sample recognition result does not meet the preset recognition conditions, obtain the status information related to the current sample recognition, and determine the target fault type that caused the current sample recognition abnormality based on the status information.
[0029] Preset identification conditions are used to determine whether the currently obtained sample identification results can meet the requirements of subsequent sample confirmation or automated operations. When the sample identification results meet the preset identification conditions, the current sample identification results can be directly used for subsequent sample management or operations without entering the adaptive adjustment process for identification anomalies. When the sample identification results do not meet the preset identification conditions, it indicates that the current sample identification process may be affected by abnormal factors. At this time, status information related to the current sample identification is acquired. The status information can be used to characterize the status of the sample to be identified, related equipment, and / or the cryogenic environment during the current identification process, and the cause of the current sample identification anomaly can be analyzed based on the status information to determine the target fault type.
[0030] Step S103: Determine the target control action corresponding to the target fault type based on the target fault type.
[0031] Here, different fault types can correspond to different control actions, and the same fault type can correspond to one or more optional control actions. For example, for imaging anomalies caused by lighting conditions, the imaging conditions can be improved by adjusting the working state of the lighting device; for sample pose deviations, the relative pose between the sample to be identified and the image acquisition device can be adjusted by controlling the mechanical actuator; for frost obstruction on the sample surface, the corresponding area can be defrosted by a defrosting device.
[0032] This allows the determined target control action to be matched with the actual cause of the current fault, rather than using the same adjustment method for different types of identified anomalies.
[0033] Step S104: Control the actuator corresponding to the target control action to perform the target control action in order to adjust the abnormal state corresponding to the target fault type.
[0034] After determining the target control action, the target control action is converted into a control command that the corresponding actuator can execute, and the corresponding actuator is controlled to execute the control command. For example, the control command can be G-code that the mechanical actuator can parse, motor control signals, light source control parameters, or defrosting device control signals, etc.
[0035] For example, when the target control action is to adjust the lighting status, the lighting device can be controlled to change the brightness, lighting direction or other lighting parameters; when the target control action is to adjust the sample pose, the mechanical actuator can be controlled to move or rotate the sample to be identified; when the target control action is to defrost, the defrosting device can be controlled to process abnormal areas near the sample to be identified.
[0036] By executing target control actions, the relevant states causing the current identification anomaly can be altered, restoring the sample to a state conducive to identification. Adjusting the abnormal state corresponding to the target fault type is not limited to directly changing the state of the sample itself; it can also include adjusting lighting conditions, relative pose, local environmental conditions, or equipment operating conditions related to sample identification.
[0037] Step S105: After executing the target control action, the sample to be identified is re-identified.
[0038] After executing the target control action, the sample image of the sample to be identified can be re-acquired, and the barcode decoding, text recognition or other sample identification process can be executed again to obtain new sample identification results.
[0039] By comparing whether the sample recognition results after the action is executed meet the preset recognition conditions, it can be verified whether the target control action has effectively improved the current recognition status.
[0040] If the re-obtained sample recognition result meets the preset recognition conditions, the adjustment process for the current recognition anomaly can be ended; if it still does not meet the preset recognition conditions, fault judgment and control adjustment can continue based on the updated status information.
[0041] Through the above steps, this embodiment can determine the specific fault type based on the state information related to the current identification process when sample identification is abnormal. It then executes matching control actions based on the fault type, and re-identifies and verifies the adjustment results. This forms a closed-loop control process that uses state perception to attribute faults, determines action decisions for state adjustment, and finally performs feedback verification. Compared to directly repeating data collection or using fixed remedial actions after identification failure, this method improves the adaptability of the sample identification process to different abnormal conditions in cryogenic dynamic environments and reduces ineffective repetitive operations.
[0042] The following section further explains the identification process of the sample to be identified and the method for determining whether the sample identification result meets the preset identification conditions.
[0043] Figure 2 This is a schematic diagram of the identification process of the sample to be identified in an embodiment of this disclosure.
[0044] like Figure 2 As shown, the sample to be identified is identified, and the sample identification result is obtained, including: Step S201: Obtain the sample image of the sample to be identified.
[0045] The sample to be identified can be a cryopreservation tube or other sample container with an identifiable label. The image acquisition device can be set at the identification station corresponding to the sample to be identified, and will acquire the sample image after the sample enters the preset identification position.
[0046] The identifiable label can be located on the bottom and / or wall of the sample container. For example, the identifiable label may include a two-dimensional barcode on the bottom of the cryovial or a text label on the wall of the cryovial. Accordingly, the image acquisition device can acquire a sample image containing the area of the identifiable label based on its location.
[0047] Step S202: Perform identifier recognition on the sample image to obtain the identifier recognition content and the corresponding recognition confidence level.
[0048] In one implementation, when the identifiable identifier is a two-dimensional barcode, the barcode in the sample image can be decoded by a barcode decoding module to obtain the identifier content carried by the two-dimensional barcode. When the identifiable identifier is a text label, the text region in the sample image can be recognized by an optical character recognition module to obtain the corresponding text recognition content.
[0049] Specifically, the identification includes, but is not limited to, one or more of the following: decoding the two-dimensional barcode on the bottom of the cryopreservation tube, such as a DataMatrix code or a QR code; performing OCR character recognition on the label on the wall of the cryopreservation tube; or detecting and identifying other identification information in the sample image.
[0050] The identification content refers to the identification information itself obtained after successful identification, such as the string obtained after decoding a two-dimensional barcode or the character sequence obtained after OCR recognition.
[0051] Recognition confidence is used to characterize the credibility of the currently obtained identifier content. Different confidence determination methods can be adopted for different recognition methods. For example, for two-dimensional barcode recognition, the recognition confidence can be determined based on the error correction during the two-dimensional barcode decoding process; for text recognition or neural network-based image recognition, the recognition confidence can be determined based on the classification probability output by the recognition model.
[0052] In some implementations, the recognition confidence can be further corrected based on the image quality information of the sample images. For example, the initial recognition confidence can be weighted or penalized based on the blurriness, overexposure, and / or underexposure of the sample images to obtain the final recognition confidence. When the image quality is poor, the final recognition confidence can be reduced accordingly to avoid making overly optimistic judgments about the current recognition result based solely on the output of the identifier recognition module.
[0053] Step S203: Determine whether the sample recognition result meets the preset recognition conditions based on the identification content and recognition confidence level.
[0054] Preset recognition conditions may include obtaining the identification content and the corresponding recognition confidence level reaching a preset confidence threshold.
[0055] When the identification content is obtained and the identification confidence level reaches the preset confidence threshold, it can be determined that the sample identification result meets the preset identification conditions. At this time, the adaptive adjustment process for the current sample can be ended, and the identification content can be used for subsequent sample management or operation.
[0056] When no identification content is obtained, or when identification content is obtained but the identification confidence level does not reach the preset confidence threshold, it can be determined that the sample identification result does not meet the preset identification conditions. At this time, it indicates that the current sample identification may be affected by abnormal factors such as deep cryogenic environment, imaging status, and equipment operating status, triggering the subsequent process of obtaining status information and attributing faults for the current sample identification anomaly.
[0057] In one example, the preset reliability threshold can be set to 0.85; for samples with higher recognition reliability requirements, the preset reliability threshold can be increased, for example, to 0.95. The specific values mentioned above are only examples, and in practical applications, they can be set according to the sample type, identifier type, and recognition reliability requirements.
[0058] By employing the above method, before entering the adaptive control process, the conventional identification process can be used to determine whether the current sample to be identified can be reliably identified. For samples that have already obtained reliable identification results, there is no need to perform further fault diagnosis and hardware adjustments; only when the sample identification result does not meet the preset identification conditions is the subsequent state perception and fault attribution process triggered, thereby avoiding unnecessary state acquisition, inference, and control operations on normal samples.
[0059] Furthermore, in step S202, after obtaining the initial recognition confidence, the sample image can be further analyzed to obtain at least one image quality index. The quality weight is determined based on the image quality index and used as a penalty factor for the initial recognition confidence to correct the initial recognition confidence and obtain the final recognition confidence, which is then applied to the condition satisfaction determination in step S203.
[0060] Specifically, image quality metrics may include at least one of the following: Laplacian variance, global contrast, grayscale histogram features, percentage of overexposed pixels, and percentage of underexposed pixels. The Laplacian variance can be used to characterize the sharpness of the sample image; the percentage of overexposed pixels and the percentage of underexposed pixels can be determined based on the grayscale histogram of the sample image to characterize the degree of overexposure and underexposure, respectively.
[0061] In one specific implementation, the final identification confidence level can be determined in the following way: ; Where C represents the final recognition confidence level, C0 represents the initial recognition confidence level obtained during the identification process, and W represents the quality weight determined based on at least one image quality index.
[0062] Quality weights can be determined based on comparisons between each image quality metric and its corresponding preset quality threshold. For example, when the Laplacian variance is below the preset sharpness threshold, the sample image is considered to have a certain degree of blur, and the corresponding quality weight is set to 0.8; when the proportion of overexposed pixels in the sample image exceeds 15%, the corresponding quality weight can be set to 0.7. When multiple image quality metrics simultaneously fail to meet their respective preset quality conditions, the quality weights corresponding to each image quality metric can be multiplied together, or a smaller value can be selected from multiple quality weights as the final quality weight.
[0063] Therefore, when the sample image has obvious image quality abnormalities such as blurriness, overexposure, or underexposure, even if the label recognition module outputs a high initial recognition confidence, the initial recognition confidence can be corrected by quality weights, so that the final recognition confidence can reflect both the output result of the label recognition module itself and the actual imaging quality of the current sample image, reducing the possibility of high-confidence misidentification due to image quality abnormalities.
[0064] After obtaining the final recognition confidence score, when the identifier recognition content is obtained from the sample image and the recognition confidence score reaches the preset confidence threshold, it can be determined that the sample recognition result meets the preset recognition conditions. At this point, the currently obtained identifier recognition content can be used for subsequent sample management or operations without initiating an adaptive adjustment process for current sample recognition anomalies.
[0065] When no identification content is obtained, or when the identification content is obtained but the final identification confidence level does not reach the preset confidence threshold, it can be determined that the sample identification result does not meet the preset identification conditions, and the subsequent process of obtaining status information and attributing faults for the current sample identification anomaly is triggered.
[0066] The following section further explains how to obtain status information related to the current sample recognition when the sample recognition result does not meet the preset recognition conditions, and how to determine the target fault type that caused the current sample recognition abnormality based on the status information.
[0067] Figure 3 This is a schematic diagram illustrating the process of determining the target fault type in an embodiment of this disclosure.
[0068] like Figure 3 As shown, when the sample recognition result does not meet the preset recognition conditions, the system acquires status information related to the current sample recognition and determines the target fault type causing the current sample recognition anomaly based on the status information, including: Step S301: Obtain at least two types of state information related to the current sample identification.
[0069] State information may include at least two of the following: visual state information, ontological state information, and environmental state information.
[0070] The visual state information reflects the imaging state of the current sample image and may include the sample image of the sample to be identified, as well as image quality information determined based on the sample image. The image quality information may include at least one of the following: Laplacian variance, global contrast, grayscale histogram features, overexposed pixel ratio, underexposed pixel ratio, image texture features, and sample localization features.
[0071] The physical status information reflects the hardware operating status related to the current sample recognition and may include at least one of the following: the operating status information of the mechanical actuator, the operating status information of the lighting device, and the operating status information of the defrosting device. For example, the operating status information of the mechanical actuator may include the real-time current feedback values of each joint of the robotic arm; the operating status information of the lighting device may include the current light source channel and its PWM duty cycle; and the operating status information of the defrosting device may include the current operating power.
[0072] Environmental status information is used to reflect the current operating conditions of the cryogenic environment and may include at least one of the temperature and humidity information of the space where the sample to be identified is located.
[0073] In one optional implementation, time alignment can be performed based on the timestamps corresponding to various types of state information, so that the visual state information, ontology state information, and environmental state information used correspond to the same recognition process of the current sample to be identified, and the time-aligned state information is encapsulated into unified state data.
[0074] Step S302: Perform joint analysis based on at least two types of state information to determine the target fault type that causes the current sample identification to be abnormal.
[0075] In this step, not only can imaging anomalies be determined based on the sample image itself, but also anomaly information from different sources can be comprehensively analyzed by combining the equipment operating status and environmental conditions.
[0076] In one implementation, the target fault type can be determined based on the state characteristics corresponding to the state information and preset fault determination rules.
[0077] Specifically, the operating status information of the mechanical actuator can be used to determine whether there is any abnormal operation of the mechanical actuator; based on the sample image and image quality information, it can be determined whether there are any imaging abnormalities, frost occlusion, cold fog blurring and / or sample pose deviation in the sample image.
[0078] For example, when the joint current value of the robotic arm fluctuates abnormally or exceeds the threshold corresponding to the normal holding current, it can be determined that there is a risk of freezing and jamming in the mechanical actuator; when the proportion of high grayscale pixels in the sample image is high, it can be determined that there is overexposure or abnormal reflection; when the proportion of low grayscale pixels is high, it can be determined that there is underexposure; when the gradient of the edge of the barcode area in the sample image decreases and irregular white crystal reflection features appear, it can be determined that there is frost layer occlusion; when the global Laplacian variance is lower than the sharpness threshold, it can be determined that there is cold fog blur; when the sample contour or positioning anchor point is offset relative to the preset position, it can be determined that there is sample pose deviation.
[0079] Based on the operational anomalies of the aforementioned mechanical actuators and / or anomalies present in the sample images, at least one candidate fault type can be identified. Candidate fault types may include at least one of actuator jamming, imaging anomalies, frost occlusion, cold fog blurring, and sample pose deviation, wherein imaging anomalies may further include overexposure and underexposure.
[0080] When only one candidate fault type exists, that candidate fault type can be identified as the target fault type. When multiple candidate fault types exist simultaneously, the fault priority corresponding to each candidate fault type can be obtained, and based on the fault priority, the candidate fault type to be processed first can be identified as the target fault type.
[0081] In this embodiment, the fault priorities, from highest to lowest, include actuator jamming, overexposure or underexposure, frost obstruction, and cold fog blurring or pose deviation. When multiple faults exist simultaneously, the faults with higher priority can be processed first.
[0082] In another implementation, an agent reasoning core based on a visual language model can be used to perform joint analysis based on visual state information and ontology state information to determine the target fault type that causes the current sample recognition anomaly.
[0083] Here, visual state information, ontological state information, and historical action retry sequences can be input into the agent's inference core. Visual state information can include sample images of the sample to be identified and image quality information obtained based on those images. Ontological state information can include at least one of the following: the operating state information of the mechanical actuator, the working state information of the lighting device, and the working state information of the defrosting device. The historical action retry sequence can include control actions already executed in response to the current sample identification anomaly and the corresponding execution results. Furthermore, environmental state information can also be used as auxiliary input.
[0084] Specifically, in this embodiment, the following three types of information are used as inputs to the agent's reasoning core: The sample image of the sample to be identified. The sample image can be input directly as raw pixel data, or it can be encoded, such as using Base64 encoding. In one implementation, image quality information calculated based on the sample image can also be input simultaneously, such as Laplacian variance, overexposed pixel percentage, and underexposed pixel percentage, as auxiliary input.
[0085] Device status parameters in the status information. The device status parameters in the status information can be input in a structured data format, such as JSON. Device status parameters may include at least one of the following: real-time current feedback values of each joint of the robotic arm, the PWM duty cycle of each channel of the current light source, and the current defrost power percentage. For example, the device status parameters can be represented as: {"joint_current": [0.5, 0.6, 1.2, 0.4, 0.3, 0.5], "light_pwm": {"R":80, "G": 80, "B": 80, "IR": 0}, "defrost_power": 0}. When environmental status information is used as auxiliary input, the real-time temperature and / or humidity of the cryogenic environment can also be further input.
[0086] Historical action retry sequence. The historical action retry sequence can be entered in a structured data format, for example: {"action_history": [{"action": "reduce coaxial light brightness", "result": "failure"}, {"action":"increase polarized light ratio", "result": "failure"}]}.
[0087] The above three types of information can be organized according to a preset multimodal Prompt template to form the input of the agent's reasoning core. The Prompt template may include, for example, task description, sample images, device status parameters, historical action retry sequences, and output format requirements.
[0088] After receiving the input, the intelligent agent reasoning core uses the cross-modal understanding capability of the visual language model to identify anomalies in the current sample and attribute the faults to the faults. It then obtains fault labels to characterize the fault attribution results and determines the target fault type based on the fault labels.
[0089] For example, when there are obvious overexposed areas in the sample image and the lighting device is currently in a high output state, the intelligent agent inference core can combine the imaging features in the sample image and the working state of the lighting device to attribute the current identification anomaly to the fault and determine the fault label corresponding to the imaging anomaly. When the sample image is blurry or has abnormal texture, and the mechanical actuator is in an abnormal operating state, the cause of the current identification anomaly can be further distinguished by combining the visual state and the body state.
[0090] Furthermore, the agent inference core can also combine historical action retry sequences to generate action suggestions and action parameters corresponding to the current target fault type. By introducing historical action retry sequences, the agent inference core can learn about the control actions that have already been executed for the current identified anomaly and their execution results, thereby avoiding the repeated selection of control actions that have been executed but failed to effectively improve the identification state during subsequent policy generation.
[0091] For example, if the fault is attributed to excessive light or reflection, and the historical actions have attempted to reduce the brightness of the coaxial light but failed, then action suggestions such as "switch side light source" or "increase polarization ratio" will be generated. If the fault is attributed to surface frost and defrosting has not been attempted in the historical actions, then action suggestions and action parameters will be generated for "start local hot air nozzle spraying for 3 seconds" or "control the robotic arm to rotate the sample 5 degrees to avoid the frost layer". If the fault is attributed to actuator jamming, an action suggestion of "pause the current action and start local heating to break the ice" is generated.
[0092] When generating a policy, the core of the agent reasoning system also refers to the historical action retry sequence to avoid repeatedly generating actions that have failed, and generates alternative policies that are different from previous actions when necessary.
[0093] In one implementation, the agent's inference core can output inference results using a structured data format. These results may include at least one of fault labels, action suggestions, and action parameters. Fault labels are used to determine the target fault type, while action suggestions and action parameters can serve as inputs for subsequently determining the target control action.
[0094] In another implementation, the reasoning result of the agent's reasoning core also includes a reasoning confidence level, with a value ranging from 0 to 1. If the output reasoning confidence level is lower than a preset reasoning confidence level threshold, the controller can refuse to accept the reasoning result and instead use preset rules for fault attribution, or request manual intervention. In this embodiment, the reasoning confidence level threshold is set to 0.7.
[0095] Therefore, step S302 can determine the target fault type based on state characteristics and preset fault judgment rules, or it can use the intelligent agent reasoning core based on visual language model to jointly analyze visual state information and ontological state information to complete fault attribution. In the latter implementation, it can also combine historical action retry sequences to generate subsequent control strategies, thereby providing a basis for determining the target control action.
[0096] Step S303: Enter the corresponding control strategy generation process according to the target fault type.
[0097] After determining the target fault type, the target fault type can be used as the input for the subsequent target control action determination process, and the corresponding candidate control actions, action parameters, and execution methods can be determined for the current target fault type.
[0098] For example, when the target fault type is imaging anomaly, candidate control actions related to illumination adjustment can be further determined; when the target fault type is frost obstruction, candidate control actions related to local defrosting or local heating can be further determined; when the target fault type is sample pose deviation, pose adjustment actions corresponding to the mechanical actuator can be further determined.
[0099] In this way, when the sample recognition result does not meet the preset recognition conditions, joint analysis can be performed based on at least two types of information from visual state, ontology state and environmental state, and the target fault type that causes the current sample recognition abnormality can be determined by rule judgment or intelligent agent reasoning, thereby providing a basis for generating corresponding control strategies for different fault types.
[0100] The following section further explains the specific implementation method of determining the target control action corresponding to the target fault type based on the target fault type.
[0101] Figure 4 This is a schematic diagram illustrating the process of determining the target control action corresponding to the target fault type in an embodiment of this disclosure.
[0102] like Figure 4 As shown, the target control actions corresponding to the target fault type are determined based on the target fault type, including: Step S401: Determine the set of candidate control actions corresponding to the target fault type based on the target fault type.
[0103] Different fault types can be pre-associated with different sets of candidate control actions so that subsequent control actions can be matched with the specific cause of the currently identified anomaly.
[0104] For example, when the target fault type is overexposure or reflection, candidate control actions may include reducing the brightness of coaxial light, increasing the proportion of polarized light, and fine-tuning the relative angle of the sample or image acquisition device; when the target fault type is frost obstruction, candidate control actions may include activating the targeted defrosting air knife or activating local micro-heating.
[0105] In other implementations, for situations such as cold fog, thin ice, or unclear 3D etching features, the imaging state of the sample can be improved by switching the illumination mode. For example, near-infrared light or red light can be switched for frost or thin ice, polarized light can be used for water fog reflection, and low-angle lateral dark field illumination can be used for 3D deformation or insufficient etching depth.
[0106] Therefore, the purpose of this step is to limit the control actions that can be used to eliminate or mitigate the impact of the current target fault to the corresponding candidate range based on the current target fault type, so as to avoid indiscriminate selection from all executable actions.
[0107] Step S402: Determine the action parameters corresponding to each candidate control action based on the current state information.
[0108] After identifying candidate control actions, corresponding action parameters can be generated by combining the current state information.
[0109] For example, to address overexposure in imaging, the adjustment range for illumination brightness can be determined based on the percentage of overexposed pixels or the area of the overexposed region in the current sample image. In one example, the light source brightness can be gradually reduced in preset increments, such as decreasing by 10% each time.
[0110] For cases of frost cover, the duration of the defrosting action can be determined based on the detected frost coverage area. For example, the duration of localized heating or defrosting can be extended as the frost coverage area increases.
[0111] Action parameters may also include the pose adjustment of the mechanical actuator, the light source channel, the light source duty cycle, the polarization state, the heating power, or the defrosting duration. Specific action parameters can be determined based on the current fault severity, the current equipment status, and the execution range of the corresponding candidate control actions.
[0112] In one implementation, when the target control action involves fine-tuning of the mechanical actuator, the corresponding pose adjustment amount can be determined based on the coordinate system of the center point of the robotic arm tool.
[0113] In one implementation, when the candidate control action includes fine-tuning the pose of a mechanical actuator, the corresponding pose adjustment amount can be determined based on the coordinate system of the robot arm tool center point. The pose adjustment amount may include at least one of translational adjustment amounts along the X, Y, and Z axes, and rotational adjustment amounts around the X, Y, and Z axes.
[0114] Specifically, the robotic arm pose when the axis of the sample to be identified is in a preset relative position to the optical axis of the image acquisition device can be used as the reference pose, and small-range adjustments can be made around the reference pose. For example, the translation adjustment range along the X, Y, and Z axes can be limited to ±2 mm, the rotation adjustment range around the X and Y axes can be limited to ±5°, and the rotation adjustment range around the Z axis can be determined within the range of 0° to 360°. The pose corresponding to when the axis of the sample to be identified is parallel to the optical axis of the image acquisition device can be used as the zero-angle reference pose.
[0115] By constraining the pose adjustment range of the mechanical actuator, the relative position or angle between the sample to be identified and the image acquisition device can be changed, and the imaging state can be improved, while reducing the risk of excessive displacement or collision of the mechanical actuator in a deep, low-temperature, and confined working space.
[0116] Step S403: Based on the action priority and / or action mutual exclusion relationship between each candidate control action, determine the target control action and the corresponding target action parameters from the candidate control action set.
[0117] When there are multiple candidate control actions, the target control action to be executed can be determined based on the preset priority of different actions and the mutual exclusion relationship between actions.
[0118] In one implementation, the action priority can be set in the order of safety-level actions, optical-level actions, mechanical-level actions, and thermal-level actions. Among them, safety-level actions can be executed before other types of actions to reduce the risk of mechanical freezing or equipment malfunction.
[0119] Furthermore, mutually exclusive conditions can be set based on the physical relationships between actions. For example, thermal heating and the introduction of cold air can be set to be mutually exclusive to avoid executing control actions with opposite effects simultaneously; optical adjustment and mechanical fine-tuning can be combined as needed.
[0120] For example, if the current target fault type is overexposure of the image, and the candidate control actions include reducing the illumination brightness and fine-tuning the sample yaw angle, the action parameters corresponding to reducing the illumination brightness can be determined first based on the current degree of overexposure, and mechanical fine-tuning actions can be further combined and executed when necessary.
[0121] For example, if the target fault type is frost obstruction, and the candidate actions include both hot air defrosting and local micro-heating, the appropriate action for the current operating conditions can be selected as the target control action based on the current equipment status, the degree of frost coverage, and safety constraints.
[0122] After determining the target control action and its parameters, they can be converted into control commands executable by the corresponding actuators. For example, they can be converted into motion control commands for mechanical actuators, light source control parameters for lighting devices, or start / stop and power control parameters for defrosting devices, and then executed by the corresponding actuators.
[0123] In a further implementation, when the target fault type is frost obstruction and the target control action is determined to be local defrosting according to steps S401 to S403, the specific operation location corresponding to the local defrosting action can be further determined.
[0124] Specifically, the two-dimensional location of the frost-covered area in the sample image can be determined first. For example, the two-dimensional bounding region corresponding to the frost-covered area can be obtained based on the visual recognition results, and the corresponding target pixel position can be determined based on the two-dimensional bounding region.
[0125] Subsequently, based on the imaging parameters of the image acquisition device and the current pose of the mechanical actuator, the target pixel position can be mapped to the three-dimensional spatial position in the corresponding coordinate system of the mechanical actuator. The imaging parameters can include camera intrinsic parameters and distortion parameters, and the current pose of the mechanical actuator can be used to determine the spatial relationship between the image acquisition device and the mechanical actuator.
[0126] After obtaining the three-dimensional spatial position, the corresponding working position of the defrosting device can be determined based on the structural parameters of the defrosting device and the preset working distance. The defrosting device can then be controlled to move or act on the working position to perform local defrosting on the area covered by frost.
[0127] Using the above method, the frost-covered areas identified in the sample image can be converted into defrosting target locations in actual physical space, enabling local defrosting actions to be performed on the currently frost-covered areas without requiring large-scale heating of the entire sample or surrounding area. In one implementation, fault area location can be achieved based on a mapping from two-dimensional pixel coordinates to three-dimensional physical coordinates, and the local hot air nozzle or defrosting conduit can be controlled to move to the corresponding working position according to the mapping result.
[0128] It should be noted that when using a visual language model-based agent inference core for fault attribution, the agent inference core can output action suggestions and corresponding action parameters. In this case, the action suggestions and action parameters can serve as inputs for determining candidate control actions and target action parameters in this embodiment. The control system can further combine preset action priorities, action mutual exclusion relationships, and hardware execution constraints to verify and filter the action suggestions before determining the final target control action.
[0129] In this way, after determining the target fault type, the corresponding candidate control actions and action parameters can be further determined based on the current state information. The final target control action can be determined by combining action priority, action mutual exclusion relationship and equipment execution constraints. This allows the fault attribution result to be transformed into a control strategy that can be actually executed by mechanical, optical or thermal actuators, thereby achieving targeted adjustment for the current sample identification anomaly.
[0130] The following section provides further explanation of the feedback verification and readjustment process after the target control action is executed.
[0131] Figure 5 This is a schematic diagram illustrating the feedback verification process after executing the target control action in this embodiment of the disclosure.
[0132] like Figure 5 As shown, after executing the target control action, the sample to be identified is re-identified, including: Step S501: After executing the target control action, the sample to be identified is re-identified, and a new sample identification result is obtained.
[0133] After executing the target control action, the sample image of the sample to be identified can be re-acquired, and the identifier recognition can be performed according to the method described in the aforementioned embodiment to obtain new identifier recognition content and corresponding recognition confidence. Subsequently, it can be determined whether the new sample recognition result meets the preset recognition conditions based on whether identifier recognition content has been obtained and whether the recognition confidence reaches the preset confidence threshold.
[0134] When the new sample recognition result meets the preset recognition conditions, it can be determined that the target control action has effectively improved the current recognition anomaly, and the adaptive adjustment process for the current sample to be recognized ends.
[0135] Step S502: If the new sample recognition result still does not meet the preset recognition conditions, reacquire the status information related to the current sample recognition.
[0136] If the preset recognition conditions are still not met after re-recognition, it indicates that the previous target control action has not completely eliminated the current recognition anomaly, or that the current anomaly state has changed after the action was executed. In this case, at least two of the following can be reacquired: visual state information, ontological state information, and environmental state information, to form updated state information corresponding to the current recognition cycle.
[0137] Specifically, visual state information can be determined based on re-acquired sample images; body state information can reflect the current working status of mechanical actuators, lighting devices, and / or defrosting devices after the previous target control action is executed; and environmental state information can reflect the current temperature and humidity status in the cryogenic environment after the action is executed. Therefore, subsequent fault judgment is based on the current state after the action is executed, rather than continuing to use the old state information before the action is executed.
[0138] Step S503: Based on the reacquired status information, determine the target fault type that caused the current sample identification abnormality again, and determine the next target control action based on the re-determined target fault type.
[0139] In this step, the at least two types of reacquired status information can be jointly analyzed as described in the aforementioned embodiments to redetermine the target fault type corresponding to the currently identified anomaly. It should be noted that the re-determined target fault type can be the same as or different from the previous target fault type.
[0140] For example, if the previous round of anomaly identification was affected by both reflection and frost obstruction, the system would first identify overexposure as the target fault type and execute a target control action to reduce the illumination brightness. After the action was executed, the overexposure in the sample image was improved, but frost obstruction still prevented the marker from being properly identified. At this point, the reacquired visual state information could indicate that the overexposure anomaly had been reduced, while the frost texture still existed. Therefore, frost obstruction could be re-identified as the current target fault type, and the next target control action corresponding to local defrosting could be further determined.
[0141] For example, if the recognition results cannot be improved after repeatedly performing the same lighting adjustment action, the historical action retry sequence can be combined to avoid continuing to repeat the same control action that has become invalid, and other control strategies can be selected from the corresponding candidate control action set.
[0142] Step S504: Control the actuator corresponding to the next target control action to execute the next target control action.
[0143] The process involves repeatedly performing steps such as sample identification, reacquisition of status information, re-determination of target fault type, and execution of target control actions until the sample identification result meets the preset identification conditions or the preset stop conditions.
[0144] After each target control action is executed, sample identification can be performed again, and a decision can be made based on the current identification result to proceed to the next round of adaptive adjustment. When the sample identification result meets the preset identification conditions, the current adaptive adjustment process can be terminated.
[0145] If the sample identification result still does not meet the preset identification conditions, the status information can be reacquired, the target fault type can be re-determined, and the next target control action can be generated in accordance with steps S502 and S503.
[0146] In one implementation, the preset stop condition may include: the number of times the control action corresponding to the same fault type is executed reaches a preset limit, or the hardware protection condition is triggered.
[0147] For example, a maximum number of retries can be set for the same fault type. If the sample identification result still does not meet the preset identification conditions after a preset number of control actions have been executed for the same fault, the automatic adjustment corresponding to that fault will be stopped, and an alarm message or manual handling request will be output.
[0148] For example, to avoid prolonged exposure of samples in non-frozen environments, a safe exposure time for samples can be set. When the cumulative exposure time reaches a preset safe threshold, even if the current sample has not been successfully identified, subsequent heating, defrosting, or other control actions that may further affect the sample temperature can be terminated, and the process can be switched to manual processing.
[0149] Furthermore, if the sample identification results meet the preset identification conditions, the corresponding status information, target fault type, executed target control actions and their action parameters during this adaptive adjustment process can be associated and stored to form a historical processing record, which can be called in the subsequent process of determining control strategies for similar or identical identified anomalies.
[0150] By employing the above method, after each target control action is executed, the current fault state can be re-evaluated using the new identification results and updated state information, and subsequent control strategies can be adjusted accordingly. This allows the entire adaptive control process to dynamically update as the abnormal state changes, rather than simply repeating the same remedial action. Furthermore, by setting the number of retries and hardware protection conditions, the risk of excessive intervention affecting the sample or equipment can be reduced while improving automatic recovery capabilities.
[0151] Combination Figure 6 As shown, this disclosure provides an adaptive control device for sample identification in cryogenic environments, including a processor 600 and a memory 601. Optionally, the device may further include a communication interface 602 and a bus 603. The processor 600, communication interface 602, and memory 601 can communicate with each other via the bus 603. The communication interface 602 can be used for information transmission. The processor 600 can call logical instructions in the memory 601 to execute the adaptive control method for sample identification in cryogenic environments described in the above embodiment.
[0152] Furthermore, the logic instructions in the aforementioned memory 601 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.
[0153] The memory 601, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 600 executes functional applications and data processing by running the program instructions / modules stored in the memory 601, thereby realizing the sample identification adaptive control method for cryogenic environments described in the above embodiments.
[0154] The memory 601 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 601 may include high-speed random access memory and may also include non-volatile memory.
[0155] This embodiment also provides a biobank device, including a device body and the aforementioned sample identification adaptive control device for cryogenic environments. The sample identification adaptive control device for cryogenic environments can be installed on the device body or communicate with the device body.
[0156] For example, biobank equipment can be automated cryogenic biobank equipment, automated liquid nitrogen biobank equipment, automated cryopreservation tube storage and retrieval equipment, or cryogenic sample identification and transport workstations. This equipment can automatically store, retrieve, transport, and identify biological samples such as cryopreservation tubes in a deep cryogenic environment. The equipment can include a sample identification station where a mechanical actuator moves the sample to be identified. An image acquisition device then captures the sample identification image, and an adaptive control device adjusts the status and provides feedback in case of identification anomalies.
[0157] The device itself may include a cryogenic storage area for storing biological samples, a mechanical actuator for manipulating the samples to be identified, an image acquisition device for acquiring images of the samples to be identified, and an illumination device and / or a defrosting device for improving the sample identification process. The samples to be identified may be cryovials or other sample containers equipped with two-dimensional barcodes, text labels, or other identifiable markings.
[0158] The sample recognition adaptive control device for cryogenic environments can communicate with an image acquisition device, a mechanical actuator, a lighting device, and a defrosting device to obtain sample recognition results and status information related to the current sample recognition. If the sample recognition result does not meet the preset recognition conditions, it can determine the target fault type that caused the current sample recognition abnormality and the corresponding target control action, and then control the corresponding actuator to perform the target control action.
[0159] For example, when the target fault type is imaging abnormality, the lighting device can be controlled to adjust the brightness, type, or direction of the light source; when the target fault type is sample pose deviation, the mechanical actuator can be controlled to adjust the relative pose between the sample to be identified and the image acquisition device; when the target fault type is frost occlusion, the defrosting device can be controlled to perform local defrosting on the corresponding frost area.
[0160] After executing the target control action, the image acquisition device can re-acquire the sample image of the sample to be identified, and the sample identification adaptive control device for deep cryogenic environment can re-identify the sample to verify the execution effect of the target control action. If the re-acquired sample identification result still does not meet the preset identification conditions, the status information acquisition, fault attribution and control action determination can be carried out again, thereby forming a closed-loop adaptive control for the deep cryogenic sample identification process.
[0161] This disclosure provides a computer-readable storage medium storing computer-executable instructions configured to execute the aforementioned adaptive control method for sample identification in cryogenic environments. The aforementioned computer-readable storage medium can be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.
[0162] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of this disclosure. The aforementioned storage medium can be a non-transitory storage medium, including: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code; it can also be a transient storage medium.
[0163] The foregoing description and accompanying drawings fully illustrate embodiments of this disclosure to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in this application is for describing embodiments only and is not intended to limit the claims. As used in the description of embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or” as used in this application means including one or more of the associated listed items and all possible combinations thereof. Additionally, when used in this application, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., disclosed in the embodiments, if they correspond to the method section disclosed in the embodiments, the relevant parts can be referred to the description of the method section.
[0164] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0165] The methods and products disclosed in the embodiments herein (including but not limited to devices and equipment) can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units may be merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to implement this embodiment according to actual needs. In addition, the functional units in the embodiments of this disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
[0166] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
Claims
1. An adaptive control method for sample identification in cryogenic environments, characterized in that, include: The sample to be identified is identified to obtain the sample identification result; If the sample identification result does not meet the preset identification conditions, obtain the status information related to the current sample identification, and determine the target fault type that caused the current sample identification abnormality based on the status information; Based on the target fault type, determine the target control action corresponding to the target fault type; Control the actuator corresponding to the target control action to execute the target control action, so as to adjust the abnormal state corresponding to the target fault type; After the target control action is executed, the sample to be identified is re-identified.
2. The sample recognition adaptive control method according to claim 1, characterized in that, The process of identifying the sample to be identified and obtaining the sample identification result includes: The sample image of the sample to be identified is subjected to label recognition to obtain the label recognition content and the corresponding recognition confidence level; Based on the identification content and identification confidence level, determine whether the sample identification result meets the preset identification conditions.
3. The sample recognition adaptive control method according to claim 1, characterized in that, The step of acquiring state information related to the current sample identification and determining the target fault type causing the current sample identification abnormality based on the state information includes: Acquire at least two of the following: visual state information, ontological state information, and environmental state information; The target fault type is determined by jointly analyzing at least two of the visual state information, the ontological state information, and the environmental state information.
4. The sample recognition adaptive control method according to claim 3, characterized in that, The visual state information includes a sample image of the sample to be identified and image quality information determined based on the sample image; The body status information includes at least one of the following: the operating status information of the mechanical actuator, the working status information of the lighting device, and the working status information of the defrosting device; The environmental status information includes at least one of the temperature information and humidity information of the deep cryogenic environment.
5. The sample recognition adaptive control method according to claim 4, characterized in that, The step of determining the target fault type that causes the current sample recognition anomaly based on the status information includes: Determine whether the mechanical actuator has any operational abnormalities based on the aforementioned operational status information; Based on the sample image and the image quality information, determine whether the sample image has imaging anomalies, frost occlusion, cold fog blurring and / or sample pose deviation; Based on the operational abnormalities of the mechanical actuator and / or the abnormalities present in the sample images, at least one candidate fault type is determined; In the presence of multiple candidate fault types, the candidate fault type to be processed first is determined as the target fault type based on the fault priority corresponding to each candidate fault type.
6. The sample recognition adaptive control method according to claim 3, characterized in that, The step of determining the target fault type that causes the current sample recognition anomaly based on the status information includes: The visual state information, the ontology state information, and the historical action retry sequence are input into the agent reasoning core based on the visual language model. The intelligent agent reasoning core performs joint analysis on the visual state information and the ontology state information to attribute the fault to the current sample identification anomaly and obtain the fault label. The target fault type is determined based on the fault label; The historical action retry sequence includes control actions that have been executed for identifying anomalies in the current sample and the corresponding execution results.
7. The sample recognition adaptive control method according to claim 1, characterized in that, The step of determining the target control action corresponding to the target fault type based on the target fault type includes: Based on the target fault type, determine a set of candidate control actions corresponding to the target fault type; Determine the action parameters corresponding to each candidate control action based on the status information; Based on the action priority and / or action mutual exclusion relationship between each candidate control action, the target control action and the corresponding target action parameters are determined from the set of candidate control actions.
8. The sample recognition adaptive control method according to claim 7, characterized in that, When the target fault type is frost obstruction and the target control action includes local defrosting, the actuator corresponding to the control performs the target control action, including: Determine the two-dimensional location of the frost-covered area in the sample image; Based on the imaging parameters of the image acquisition device and the current pose of the mechanical actuator, the two-dimensional position is converted into a three-dimensional spatial position corresponding to the frost-covered area. The operating position of the defrosting device is determined based on the three-dimensional spatial position. The defrosting device is controlled to perform local defrosting on the area covered by frost at the working position.
9. The sample recognition adaptive control method according to any one of claims 1 to 8, characterized in that, The step of re-identifying the sample to be identified after executing the target control action includes: The sample to be identified is re-identified, and a new sample identification result is obtained; If the new sample identification result still does not meet the preset identification conditions, the status information related to the current sample identification will be reacquired. Based on the reacquired status information, the target fault type causing the current sample identification anomaly is determined again, and the next target control action is determined based on the re-determined target fault type. Control the actuator corresponding to the next target control action to execute the next target control action; The steps of sample identification, status information reacquisition, target fault type re-determination, and target control action execution are repeated until the sample identification result meets the preset identification condition or the preset stop condition. The preset stop conditions include the number of times the control action corresponding to the same fault type is executed reaches a preset limit or the hardware protection condition is triggered.
10. An adaptive control device for sample identification in cryogenic environments, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute, when running the program instructions, the sample identification adaptive control method for cryogenic environments as described in any one of claims 1 to 9.
11. A biobank device, characterized in that, include: Equipment body; The sample identification adaptive control device for cryogenic environments as described in claim 10 is disposed on the device body.
