Multi-range density meter intelligent detection system and method based on AI visual identification

Abstract

The invention discloses a multi-range density meter intelligent detection system and method based on AI visual identification, and belongs to the technical field of density meter intelligent detection. Behavior data of a user for interaction objects and function modules are collected, and key features such as user intention deviation degree and interface attention stability index are extracted; calculating and dynamically updating a user preference weight table; interface module sorting is further carried out in combination with a deep learning model, a fuzzy logic evaluation mechanism is introduced to carry out interpretable scoring on a sorting result, when a scoring result is lower than a threshold value, the system automatically corrects a sorting strategy based on user feedback, and finally optimized interface display content is output. Personalization, dynamics and controllability of interface display are achieved, and the man-machine interaction experience and the user satisfaction degree of the terminal equipment are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent detection of density meters, and in particular to a multi-range density meter intelligent detection system and method based on AI visual recognition. Background Art

[0002] Intelligent testing of multi-range density meters utilizes intelligent technology to automatically test and analyze density meters with multiple measurement ranges. This system intelligently switches between ranges based on the density characteristics of different substances. Through sensors and control algorithms, it achieves high-precision, real-time density measurement and data processing. It is widely used in industries such as petroleum, chemical, and food, improving testing efficiency and measurement accuracy.

[0003] The existing technology has the following shortcomings:

[0004] During the AI ​​visual recognition process of multi-range density meters, if range switch markers (such as dial codes, labels, or scale tape) cannot be accurately identified due to image blur, dust obstruction, or other reasons, the model may misjudge the current range interval, thereby incorrectly interpreting the same dial reading as the density value of a different range. Such errors are hidden, and the recognition results may appear reasonable on the surface, but in fact they deviate significantly from the actual physical quantity, easily misleading on-site operators and posing a significant risk to product quality control and process safety. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-range density meter intelligent detection system and method based on AI visual recognition to address the shortcomings of the background technology.

[0006] In order to achieve the above objectives, the present invention provides the following technical solutions: a multi-range density meter intelligent detection method based on AI visual recognition, comprising:

[0007] Capture images of the reading display area and range switching mark area of ​​the density meter, and pre-process the captured images;

[0008] The trained convolutional neural network model is used to recognize the preprocessed image, obtain the current reading value and the range interval of the density meter, and combine the reading value and the range interval to calculate the current density value;

[0009] By obtaining the density fluctuation range of the measured medium and setting the ratio value, a physical logical relationship model of density change is established to infer the current range.

[0010] The obtained range interval is compared for consistency. If they are consistent, the current density value is used as the recognition result; if they are inconsistent, the abnormal judgment logic is triggered;

[0011] When an abnormality judgment is triggered, if the confidence level of the AI ​​model output is lower than the preset threshold, the image re-acquisition process is automatically started, and the range switching area is partially magnified and re-identified. The built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make corrective judgments.

[0012] The corrected density value and range interval are output as the test result.

[0013] Preferably, the image acquisition includes photographing the reading display area and the range switching mark area of ​​the density meter using an industrial camera with a resolution of not less than 1280×1024.

[0014] Preferably, the convolutional neural network model is a multi-task recognition model built based on YOLOv5, ResNet or MobileNet architecture, which is used to simultaneously identify the density value in the reading display area and the range status mark in the range switching mark area.

[0015] Preferably, the calculation of the density value by combining the reading value with the range interval includes:

[0016] For a pointer-type density meter, the CNN model first identifies the pointer angle, and then uses a preset linear mapping function to map the dial reading R to the upper and lower limits of the identified range. The calculation expression is: Where ρ is the actual density value, R is the dial reading, and R max is the maximum scale value of the dial; ρ min , ρ max They are the lower limit and upper limit of the current range respectively.

[0017] Preferably, the physical logic relationship model includes a linear prediction model or an empirical fitting model constructed according to the medium ratio setting and the historical density fluctuation range, which is used to infer the theoretical density range of the current medium and determine the range range accordingly.

[0018] Preferably, the identification results of the range interval are obtained from two paths respectively:

[0019] The AI ​​visual recognition path uses a convolutional neural network model to identify the range identification information in the image, and the output range interval is recorded as L AI ;

[0020] The physical logic reasoning path, through medium property modeling and ratio calculation, derives a reasonable range interval, denoted as L PHY ;

[0021] Compare the range interval L of the two path identification AI With L PHY , make consistency judgment:

[0022] If L AI =L PHY , that is, if the recognition path is consistent with the judgment result of the logical reasoning path, the recognition is considered reliable, the density value ρ calculated in the AI ​​recognition result is used as the final effective density value of the current detection, and the recognition path is recorded as having passed the consistency check;

[0023] If L AI ≠L PHY , that is, the range judgments of the two are inconsistent, indicating that there is a risk of misjudgment in the AI ​​recognition results, and the abnormal judgment logic is automatically triggered.

[0024] Preferably, if the range recognition is inconsistent, first check the output confidence of the AI ​​model when recognizing the range, denoted as C AI , according to the pre-set confidence threshold C TH , to determine whether the recognition is reliable:

[0025] If C AI <C TH , that is, the recognition result confidence is insufficient, triggering the image re-collection mechanism;

[0026] If C AI ≥C TH , but it still conflicts with the logical judgment, then the current image is retained and the rule correction path is directly entered.

[0027] Preferably, a built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make a corrective judgment, specifically including:

[0028] The engine compares each optional range [L1, L2, ..., L i ,...,L n ], each L i Represents a specific range interval, n is the total number of range intervals; judge whether the reading falls within the normal range of a range interval, that is, whether it meets: is the lower limit of the i-th range interval, that is, L i The starting point of the interval; is the upper limit of the i-th range interval, that is, L i The end point of the interval;

[0029] If a range interval is found to match the reading but the current AI-recognized range does not match it, the AI ​​result is judged to be incorrect and corrected to the matching range; if multiple range intervals match, the one with the highest recommended probability in the logical reasoning path is selected;

[0030] Finally, the range interval and corresponding density value corrected after re-sampling or rule judgment are output as the optimized recognition result, and the source is marked as the recognition result after error correction.

[0031] The present invention also provides a multi-range density meter intelligent detection system based on AI visual recognition, including an image acquisition and preprocessing module, an AI recognition and density calculation module, a physical logic reasoning module, a consistency check and anomaly determination module, an image re-sampling and rule compensation module, and a detection result output module;

[0032] Image acquisition and preprocessing module: acquire images of the reading display area and range switching mark area of ​​the density meter, and preprocess the acquired images;

[0033] AI recognition and density calculation module: uses the trained convolutional neural network model to recognize the pre-processed image, obtains the current reading value and the range range of the density meter, and combines the reading value and the range range to calculate the current density value;

[0034] Physical logic reasoning module: By obtaining the density fluctuation range of the measured medium and setting the ratio value, a physical logic relationship model of density change is established to infer the current range;

[0035] Consistency check and abnormality judgment module: performs consistency comparison on the obtained range interval. If they are consistent, the current density value is used as the recognition result; if they are inconsistent, the abnormality judgment logic is triggered;

[0036] Image recapture and rule compensation module: When an abnormality judgment is triggered, if the confidence level of the AI ​​model output is lower than the preset threshold, the image recapture process is automatically initiated, and the range switching area is locally magnified and re-identified. The built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make corrective judgments;

[0037] Test result output module: outputs the corrected density value and range interval as the test result.

[0038] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0039] 1. This invention addresses the range misjudgment problem of traditional multi-range density meters during visual recognition by introducing a dual-path fusion mechanism combining AI visual recognition and physical logic reasoning. This improves the recognition system's ability to perceive range switching information and its accuracy, particularly in complex environments such as image blur, label obstruction, and light interference. Furthermore, by combining the CNN model's recognition results with the physical laws of medium density for consistency comparison, it effectively avoids hidden errors caused by mismatches between readings and ranges, ensuring the accuracy of density measurements and the safety of process operations.

[0040] 2. This invention further introduces a confidence-based image recapture mechanism and rule-based judgment engine. When an anomaly is detected, the recognition process is automatically restarted and errors are corrected through logical judgment, creating a complete intelligent error-correction closed loop. This method possesses adaptive, self-diagnostic, and self-correcting capabilities, and can output reliable test results without human intervention. This significantly improves the robustness, automation level, and engineering application value of multi-range density detection systems, making them suitable for industrial scenarios requiring high density accuracy, such as those in the petrochemical, pharmaceutical, and food industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0042] Figure 1 This is a mind map of the method of the present invention.

[0043] Figure 2 This is a mind map of the system modules of the present invention. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] Example 1, please refer to Figure 1 As shown, the multi-range density meter intelligent detection method based on AI visual recognition described in this embodiment includes:

[0046] Capture images of the reading display area and range switching mark area of ​​the density meter, and pre-process the captured images;

[0047] The trained convolutional neural network model is used to recognize the preprocessed image, obtain the current reading value and the range interval of the density meter, and combine the reading value and the range interval to calculate the current density value;

[0048] By obtaining the density fluctuation range of the measured medium and setting the ratio value, a physical logical relationship model of density change is established to infer the current range.

[0049] The obtained range interval is compared for consistency. If they are consistent, the current density value is used as the recognition result; if they are inconsistent, the abnormal judgment logic is triggered;

[0050] When an abnormality judgment is triggered, if the confidence level of the AI ​​model output is lower than the preset threshold, the image re-acquisition process is automatically started, and the range switching area is partially magnified and re-identified. The built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make corrective judgments.

[0051] The corrected density value and range interval are output as the test result.

[0052] In an embodiment of the present invention, the following image acquisition and preprocessing steps are specifically included:

[0053] In this embodiment, industrial-grade imaging equipment (such as industrial cameras, edge computing terminal integrated camera modules, etc.) is used to capture images of the density table. The image acquisition target area mainly includes:

[0054] Reading display area: the area where the density meter displays the value, which may be a pointer dial or a digital display;

[0055] Range switching identification area: This is the display area that indicates the current range of the density meter, usually in the form of a label, scale ring, knob status or LCD symbol.

[0056] The camera equipment should have the following features to adapt to industrial scenarios:

[0057] The resolution should be no less than 1280×1024 to ensure that small font scales or labels can be recognized;

[0058] It has adaptive exposure and strong light suppression capabilities to handle reflective, oily or low-light environments;

[0059] The installation position should be fixed within the range of appropriate viewing angle and focal length from the density meter (e.g., 30cm–1m) to ensure that the acquired image is unobstructed and free of distortion.

[0060] After image acquisition is completed, the following preprocessing operations are performed on the original image to improve the accuracy and stability of AI model recognition:

[0061] Convert the original color image into a grayscale image to reduce computational complexity, remove background color interference, and make the main recognition targets (such as pointers, numbers, and label text) more prominent.

[0062] Edge enhancement algorithms (such as Laplace operator and Canny edge detection) are used to highlight the pointer outline, digital character boundaries and scale lines, which is particularly suitable for identifying the relationship between the pointer angle and the scale.

[0063] Based on the spatial distribution of the target locations in the image, the image is segmented using traditional image segmentation algorithms (such as thresholding and contour analysis) or deep learning object detection models (such as YOLO and Mask R-CNN). The "reading area" and "range marking area" are cropped into two separate image blocks for subsequent analysis by the recognition module.

[0064] In addition, gamma correction, histogram equalization or denoising filtering operations can be applied according to the actual acquisition environment (such as uneven lighting and mirror reflection) to improve image quality.

[0065] After completing image acquisition and preprocessing, the present invention further analyzes the preprocessed image to obtain the density meter reading and the current range interval, and calculates the actual density value based on this. The specific process is as follows:

[0066] This paper uses a trained Convolutional Neural Network (CNN) model to intelligently recognize density meter images. The CNN model can use ResNet, YOLOv5, EfficientNet, or MobileNet architectures, and is configured differently depending on the dial type (analog or digital):

[0067] For pointer-type density meters, the model is used to detect the pointer position and scale distribution;

[0068] For digital density meters, the model is used to identify digital characters or LED segment codes;

[0069] For the range identification area, the model detects the label, characters or dial status to determine the range.

[0070] The CNN model training data comes from density table images collected under various working conditions, including image samples under different angles, lighting, pollution levels, etc., to ensure that the model has good robustness and generalization capabilities.

[0071] For the reading display area in the image, the CNN model outputs the following key parameters after identification:

[0072] If it is a pointer type, output the pointer angle value (θ), and calculate the corresponding density value reading in combination with the calibrated scale mapping model;

[0073] If it is digital, the string recognition result (such as "1.86") is output and directly used as the density reading.

[0074] The recognition result may be accompanied by a confidence indicator for subsequent credibility judgment.

[0075] The system identifies the current range of the density meter (for example, 0–1g / cm³, 1–2g / cm³, 2–3g / cm³, etc.) by performing image recognition on the "range switching mark area." Identification methods include, but are not limited to, label text recognition (OCR), coder or switch mark status recognition, and color / graphic coding recognition.

[0076] The system combines the recognized reading value with the range interval to generate the final density value. The calculation method is as follows:

[0077] If it is a relative reading (such as a dial reading range of 0-100), a linear mapping is performed in combination with the current upper and lower limits of the range. The calculation expression is: Where ρ is the actual density value, R is the dial reading, and R max is the maximum scale value of the dial; ρ mim , ρ max They are the lower limit and upper limit of the current range respectively.

[0078] If it is a digital density meter, directly regard the identified value as the density value.

[0079] After identifying the density chart image and obtaining a preliminary reading, to enhance the accuracy of the system's range recognition results, the present invention further introduces a physical logic reasoning path. By analyzing the property information of the measured medium, a density-related change model is established to infer the current range interval. The specific steps are as follows:

[0080] The system automatically obtains the process configuration information of the current measured medium by connecting to the host computer system, control unit or production database, including but not limited to:

[0081] Theoretical density range of the liquid or gas medium being measured (e.g., 1.10–1.25 g / cm3);

[0082] Set raw material ratio or production formula (such as water-to-alcohol ratio, acid-base concentration setting);

[0083] Actual density fluctuation range in historical operating data (e.g., density values ​​within the past 5 minutes were between 1.15 ± 0.02 g / cm3);

[0084] Current production process stage (e.g., feeding, mixing, constant temperature stage);

[0085] The above information can be called in real time by PLC, DCS system or production database, or can be manually entered by on-site operators.

[0086] Based on the acquired medium data and ratio settings, the system automatically builds a physical logic relationship model to predict the reasonable range of the current density. The model forms include:

[0087] Linear model: suitable for scenarios where density and ratio have a linear relationship;

[0088] Empirical fitting model: a regression curve or distribution model trained based on historical data;

[0089] Stage logic model: Set different density expectation ranges for different process stages.

[0090] For example, for a saline solution, its density value ρ y The relationship between the salt mass fraction w can be expressed approximately linearly: ρ y =ρ0+k·w; where ρ0 is the density of pure water and k is a coefficient determined by fitting experimental data.

[0091] The system compares the expected density value range output by the above model with all possible ranges of the density table to infer the most likely range at the moment. For example:

[0092] If the model estimates the density to be 1.83–1.87 g / cm3, it should be in the 1–2 g / cm3 range;

[0093] If the model's estimated value significantly exceeds the upper and lower limits of the AI ​​recognition result (such as 0–1g / cm3), it can be determined that there is a risk of AI misjudgment.

[0094] The range interval obtained by reasoning is used as the recognition result of the physical logic path, and is compared with the results of the AI ​​visual recognition path for consistency, providing a reference basis for subsequent decision-making.

[0095] To improve the accuracy of density table recognition results and the system's self-checking capabilities, the present invention designs a consistency comparison mechanism and anomaly determination logic after the visual recognition path and the physical logic reasoning path are completed in parallel. This mechanism is used to automatically determine whether there are potential errors in the recognition process and trigger optimization processing accordingly. The details are as follows:

[0096] The system obtains the range interval recognition results from the following two paths:

[0097] AI visual recognition path: Use the convolutional neural network (CNN) model to identify the range identification information in the image, and the output range interval is recorded as L AI ;

[0098] Physical logic reasoning path: Through medium property modeling and ratio calculation, a reasonable range interval is derived, denoted as L PHY .

[0099] Here, both range intervals can be expressed as an upper and lower limit interval (such as 1.0–2.0 g / cm3), or represented by an interval number (such as the range number "Section II").

[0100] The system compares the range interval L of the two paths AI With L PHY , make consistency judgment:

[0101] If L AI =L PHY , that is, if the recognition path is consistent with the judgment result of the logical reasoning path, the recognition is considered reliable; the system uses the density value ρ calculated in the AI ​​recognition result as the final effective density value of the current detection; and records the recognition path as having passed the consistency check.

[0102] If L AI ≠L PHY , that is, the range judgments of the two are inconsistent, indicating that there is a risk of misjudgment in the AI ​​recognition results, and the system automatically triggers the abnormal judgment logic.

[0103] When the results of the recognition path and the inference path are inconsistent, the system will place the recognition process into an abnormal pending confirmation state and enter the next remedial mechanism (see the subsequent image re-acquisition and rule correction section for details).

[0104] In abnormal state:

[0105] The original density value is not directly used as the output result; the system records the current status as a range recognition anomaly; and simultaneously records the results of the two paths, confidence parameters, and related images for subsequent analysis.

[0106] To prevent frequent triggering of anomalies due to minor model deviations or critical value errors, this system can set a tolerable offset threshold: if the boundaries of two range intervals overlap or are adjacent, and the recognition confidence is high, it can be set to partially consistent; strict or loose judgment rules can be set according to actual process requirements.

[0107] When the consistency comparison step determines that there is a range conflict between the visual recognition path and the logical reasoning path, it indicates that there is a risk of misjudgment in the recognition result. To further improve the recognition accuracy and system stability, this invention sets up a remedial process based on AI confidence judgment, image re-sampling enhancement and rule engine error correction mechanism. The details are as follows:

[0108] If there is inconsistency in range recognition, the system first checks the output confidence of the AI ​​model when recognizing the range, which is recorded as C AI The system pre-sets the confidence threshold C TH (e.g. 0.85), used to determine whether the recognition is reliable enough:

[0109] If C AI <C TH , that is, the recognition result confidence is insufficient, and the system triggers the image re-collection mechanism;

[0110] If CAI ≥C TH , but it still conflicts with the logical judgment, the system retains the current image and directly enters the rule correction path.

[0111] If C AI <C TH The system controls the camera to recapture images of the range switch area. Optimizing the image acquisition process includes but is not limited to: improving resolution; adjusting exposure or light source angle; capturing only key areas (local ROIs) to reduce interference; and optionally applying image enhancement algorithms such as super-resolution reconstruction (Super-Resolution), contrast enhancement, and edge sharpening. After acquisition is complete, the recaptured images are fed back into the CNN model for range recognition, and a new range recognition result and confidence level are output.

[0112] If the range recognition error still exists after image re-sampling, or the system is in a non-re-sampling condition (such as image device limitations), the built-in rule-based range judgment engine will be automatically called to correct the results. The specific processing is as follows:

[0113] The engine compares each optional range [L1, L2, ..., L i ,...,L n ], each L i Represents a specific range interval, n is the total number of range intervals; determine whether the reading falls within the normal range of a range interval, that is, whether it meets the following conditions: is the lower limit of the i-th range interval, that is, L i The starting point of the interval; is the upper limit of the i-th range interval, that is, L i The end point of the interval;

[0114] If a certain range interval is found to match the reading but the current AI recognition range does not match it, the AI ​​result is judged to be incorrect and corrected to the matching range;

[0115] If multiple range intervals match, the one with the highest recommendation probability in the logical reasoning path is selected.

[0116] For example, if the density value is set to ρ = 1.83, the system's range set is: L1 = [0.0, 1.0]; L2 = [1.0, 2.0]; L3 = [2.0, 3.0].

[0117] The judgment logic is: check whether It is found that ρ = 1.83∈[1.0,2.0], which matches L2; ​​therefore, the system can assume that the current density value most likely belongs to the second range (1–2 g / cm3). If the AI ​​model identifies it as another range (such as 0–1), it means that there is a misidentification and should be corrected.

[0118] Ultimately, the system outputs the corrected range and density values ​​after re-sampling or rule-based judgment as the optimized recognition result, marking the source as the corrected recognition result. The system also automatically saves the original recognition record and the correction path to facilitate model training optimization and fault tracing.

[0119] The present invention outputs the final confirmed density reading value and range interval for use by the upper system or reference by the operator. The output content includes the following two core elements:

[0120] Based on the final, confirmed identification path, the system outputs the current medium's density value, ρfinal. This density value may be derived from: direct identification by the AI ​​identification path and passing consistency verification; updated results from image re-sampling; or corrections based on a rule-based engine. This value is system-verified and valid measurement data that can be used for process control, alarm detection, and quality tracking.

[0121] The system also outputs the range Lfinal corresponding to the current density value, which is the actual working range of the density meter. For example:

[0122] If ρfinal=1.83, Lfinal=[1.0,2.0];

[0123] If it is a numbering method, the output is the range number "Section II" or "High Range" etc.

[0124] The system supports outputting final inspection results as structured data, such as JSON or CSV formats, and transmitting them to external systems through the following methods: data docking with industrial control systems such as PLC and DCS; uploading to the cloud platform through edge computing devices; and displaying them on a local visual interface for operators to view.

[0125] Example 2, please refer to Figure 2 As shown, the multi-range density meter intelligent detection system based on AI visual recognition described in this embodiment includes an image acquisition and preprocessing module, an AI recognition and density calculation module, a physical logic reasoning module, a consistency check and anomaly determination module, an image re-sampling and rule compensation module, and a detection result output module;

[0126] Image acquisition and preprocessing module: acquire images of the reading display area and range switching mark area of ​​the density meter, and preprocess the acquired images;

[0127] AI recognition and density calculation module: uses the trained convolutional neural network model to recognize the pre-processed image, obtains the current reading value and the range range of the density meter, and combines the reading value and the range range to calculate the current density value;

[0128] Physical logic reasoning module: By obtaining the density fluctuation range of the measured medium and setting the ratio value, a physical logic relationship model of density change is established to infer the current range;

[0129] Consistency check and abnormality judgment module: performs consistency comparison on the obtained range interval. If they are consistent, the current density value is used as the recognition result; if they are inconsistent, the abnormality judgment logic is triggered;

[0130] Image recapture and rule compensation module: When an abnormality judgment is triggered, if the confidence level of the AI ​​model output is lower than the preset threshold, the image recapture process is automatically initiated, and the range switching area is locally magnified and re-identified. The built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make corrective judgments;

[0131] Test result output module: outputs the corrected density value and range interval as the test result.

[0132] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0133] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0134] Those skilled in the art will appreciate that the units and algorithm steps of each example 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 performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0135] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. The intelligent detection method for multi-range density meters based on AI visual recognition is characterized by: include: Capture images of the reading display area and range switching mark area of ​​the density meter, and pre-process the captured images; The trained convolutional neural network model is used to recognize the preprocessed image, obtain the current reading value and the range interval of the density meter, and combine the reading value and the range interval to calculate the current density value; By obtaining the density fluctuation range of the measured medium and setting the ratio value, a physical logical relationship model of density change is established to infer the current range. The obtained range interval is compared for consistency. If they are consistent, the current density value is used as the recognition result; if they are inconsistent, the abnormal judgment logic is triggered; When an abnormality judgment is triggered, if the confidence level of the AI ​​model output is lower than the preset threshold, the image re-acquisition process is automatically started, and the range switching area is partially magnified and re-identified. The built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make corrective judgments. The corrected density value and range interval are output as the test result.

2. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 1 is characterized in that: The image acquisition includes using an industrial camera with a resolution of not less than 1280×1024 to shoot the reading display area and the range switching mark area of ​​the density meter.

3. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 1 is characterized in that: The convolutional neural network model is a multi-task recognition model built based on the YOLOv5, ResNet or MobileNet architecture, which is used to simultaneously identify the density value in the reading display area and the range status mark in the range switching mark area.

4. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 1 is characterized in that: The calculation of the density value by combining the reading value with the range interval includes: For a pointer-type density meter, the CNN model first identifies the pointer angle, and then uses a preset linear mapping function to map the dial reading R to the upper and lower limits of the identified range. The calculation expression is: Where ρ is the actual density value, R is the dial reading, and R max is the maximum scale value of the dial; ρ min , ρ max They are the lower limit and upper limit of the current range respectively.

5. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 4 is characterized in that: The physical logic relationship model includes a linear prediction model or an empirical fitting model constructed according to the medium ratio setting and the historical density fluctuation range, which is used to infer the theoretical density range of the current medium and determine the range range accordingly.

6. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 5 is characterized in that: Get the range interval recognition results from two paths respectively: The AI ​​visual recognition path uses a convolutional neural network model to identify the range identification information in the image, and the output range interval is recorded as L AI ; The physical logic reasoning path, through medium property modeling and ratio calculation, derives a reasonable range interval, denoted as L PHY ; Compare the range interval L of the two path identification AI With L PHY , make consistency judgment: If L AI =L PHY , that is, if the recognition path is consistent with the judgment result of the logical reasoning path, the recognition is considered reliable, the density value ρ calculated in the AI ​​recognition result is used as the final effective density value of the current detection, and the recognition path is recorded as having passed the consistency check; If L AI ≠L PHY , that is, the range judgments of the two are inconsistent, indicating that there is a risk of misjudgment in the AI ​​recognition results, and the abnormal judgment logic is automatically triggered.

7. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 6 is characterized in that: If the range recognition is inconsistent, first check the output confidence of the AI ​​model when recognizing the range, denoted as C AI , according to the pre-set confidence threshold C TH , to determine whether the recognition is reliable: If C AI <C TH , that is, the recognition result confidence is insufficient, triggering the image re-collection mechanism; If C AI ≥C TH , but it still conflicts with the logical judgment, then the current image is retained and the rule correction path is directly entered.

8. The multi-range density meter intelligent detection method based on AI visual recognition according to claim 7 is characterized in that: The built-in rule-based range judgment engine compares the reading value with the upper and lower limits of different ranges to make corrective judgments, including: The engine compares each optional range [L1, L2, ..., L i ,...,L n ], each L i Represents a specific range interval, n is the total number of range intervals; judge whether the reading falls within the normal range of a range interval, that is, whether it meets: is the lower limit of the i-th range interval, that is, L i The starting point of the interval; is the upper limit of the i-th range interval, that is, L i The end point of the interval; If a range interval is found to match the reading but the current AI-recognized range does not match it, the AI ​​result is judged to be incorrect and corrected to the matching range; if multiple range intervals match, the one with the highest recommended probability in the logical reasoning path is selected; Finally, the range interval and corresponding density value corrected after re-sampling or rule judgment are output as the optimized recognition result, and the source is marked as the recognition result after error correction.

9. An intelligent detection system for a multi-range density meter based on AI visual recognition, for implementing the intelligent detection method for a multi-range density meter based on AI visual recognition according to any one of claims 1 to 8, characterized in that: It includes image acquisition and preprocessing module, AI recognition and density calculation module, physical logic reasoning module, consistency verification and anomaly judgment module, image re-acquisition and rule compensation module, and detection result output module; Image acquisition and preprocessing module: acquire images of the reading display area and range switching mark area of ​​the density meter, and preprocess the acquired images; AI recognition and density calculation module: uses the trained convolutional neural network model to recognize the pre-processed image, obtains the current reading value and the range range of the density meter, and combines the reading value and the range range to calculate the current density value; Physical logic reasoning module: By obtaining the density fluctuation range of the measured medium and setting the ratio value, a physical logic relationship model of density change is established to infer the current range; Consistency check and abnormality judgment module: performs consistency comparison on the obtained range interval. If they are consistent, the current density value is used as the recognition result; if they are inconsistent, the abnormality judgment logic is triggered; Image recapture and rule compensation module: When an abnormality judgment is triggered, if the confidence level of the AI ​​model output is lower than the preset threshold, the image recapture process is automatically initiated, and the range switching area is locally magnified and re-identified. The built-in rule-based range judgment engine is used to compare the reading value with the upper and lower limits of different ranges to make corrective judgments; Test result output module: outputs the corrected density value and range interval as the test result.

Images