Anesthesia Equipment Fault Diagnosis Method and System Based on Image Recognition

Abstract

The present invention discloses an anesthesia device fault diagnosis method and system based on image recognition, specifically relating to the technical field of medical device fault detection, and is used to solve the problem that in the existing non-contact image detection method for pipeline leakage detection of anesthesia devices, the recognition ability for weak leakage signals of transparent pipelines is poor; by combining multi-spectral dynamic imaging with polarized light scattering pattern analysis, microscopic deformation characteristics of the pipeline surface are obtained and a leakage path prediction result is generated; based on the spectral distribution characteristics of the reference image set, environmental reflection noise is suppressed, and combined with active pressure perturbation to trigger changes in the optical response of the candidate region, the inverse physical relationship between the penetration rate and the fluid conduction speed is verified; precise positioning of the target region is achieved through dynamic threshold screening and spatial continuity verification, and finally, a leakage risk assessment is generated by integrating spectral characteristics, path morphology, and real-time pressure parameters, realizing non-invasive and highly sensitive leakage recognition, and providing reliable guarantee for the safe operation of clinical anesthesia devices.

Description

Technical Field

[0001] The present invention relates to the technical field of medical device fault detection, and more specifically, to a method and system for fault diagnosis of anesthetic devices based on image recognition. Background Art

[0002] During clinical use, anesthetic devices need to ensure the tightness and safety of the gas circuit system. Among them, the transparent pipeline is prone to aging or micro-cracks due to long-term contact with gas and liquid, which may cause leakage risks. At present, the detection of pipeline leakage mainly relies on contact monitoring means such as pressure sensors and flow sensors. Such methods need to be directly connected to the device gas circuit, resulting in complex installation and high maintenance costs. In addition, their sensitivity to early trace leakage is limited. In addition, some non-contact detection technologies analyze the device status through image acquisition, but mostly focus on obvious fault characteristics such as deformation of mechanical components or abnormal display screen values, and lack targeted processing of visual signals of internal leakage in transparent pipelines.

[0003] When existing image recognition-based detection methods are used to handle transparent pipeline leakage, due to the weak optical signals of the liquid film or gas-liquid interface in the initial stage of leakage, and the optical interference (such as reflection and refraction) between the transparent material and the background environment is easy to cause confusion, it is difficult for traditional image processing algorithms to effectively distinguish real leakage features from noise, resulting in missed detection or misjudgment, making it impossible to timely identify early trace leakage, which may increase potential device failures and affect clinical safety. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method and system for fault diagnosis of anesthetic devices based on image recognition to solve the problems raised in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for fault diagnosis of anesthetic devices based on image recognition, comprising the following steps:

[0007] S1. Obtain a reference image set and a dynamic image set of the transparent pipeline area of the anesthetic device;

[0008] S2. Identify the microscopic deformation area based on the change in the polarization light scattering pattern on the surface of the transparent pipeline in the dynamic image set, and generate a prediction result of the potential leakage path in combination with the fluid conduction direction;

[0009] S3. According to the pixel distribution characteristics of the reference image set and the prediction result of the potential leakage path, suppress the environmental reflection noise in the non-leakage path area, and extract candidate areas within the leakage path range and having a spectral difference from the noise;

[0010] S4. Actively trigger the change in spectral reflectance of the candidate area by pressure perturbation, verify the inverse relationship between the penetration rate and the conduction speed in combination with the fluid conduction speed, and screen out the suspected fault areas;

[0011] S5. Determine the spectral intensity fluctuation threshold of the suspected fault area, and screen out the target areas where the spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and spatially overlaps with the microscopic deformation area;

[0012] S6. Match the change characteristics of the spectral reflectance of the target area and the spatial distribution pattern of the predicted results of the potential leakage path to generate the leakage risk assessment result.

[0013] In a preferred embodiment, obtain the reference image set and the dynamic image set of the transparent pipeline area of the anesthesia device, including:

[0014] Use a multispectral imaging device to collect a reference image set including visible light band and near-infrared band when the transparent pipeline is in a leak-free state, and the filter band covers the characteristic absorption peaks of the pipeline material in the normal sealed state;

[0015] Synchronously collect a dynamic image set including the polarization light scattering characteristics on the surface of the pipeline during the operation of the anesthesia device, and synchronously record the real-time pressure parameters inside the pipeline each time of collection;

[0016] Adjust the angle of the polarization filter to obtain a dynamic image set with the same filter band and polarization angle as the reference image set.

[0017] In a preferred embodiment, identify the microscopic deformation area based on the change in the polarization light scattering pattern on the surface of the transparent pipeline in the dynamic image set, and generate the predicted result of the potential leakage path in combination with the fluid conduction direction, including:

[0018] Extract the scattering images at different polarization angles in the dynamic image set, and calculate the gray value difference of the same area on the pipeline surface at each polarization angle;

[0019] Based on the area where the gray value difference exceeds the gray fluctuation range of the reference image set at the same polarization angle, mark it as the microscopic deformation area;

[0020] Combine the fluid conduction direction in the gas path of the anesthesia device with the spatial distribution of the microscopic deformation area, extend the boundary of the microscopic deformation area along the fluid conduction direction to generate a potential leakage path;

[0021] Eliminate the potential leakage paths whose matching degree with the normal deformation mode of the pipeline in the reference image set is higher than the preset threshold, and retain the predicted results of the potential leakage paths.

[0022] In a preferred embodiment, according to the pixel distribution characteristics of the reference image set and the potential leakage path prediction result, the environmental specular noise in the non-leakage path area is suppressed, and the candidate areas within the leakage path range and having a spectral difference from the noise are extracted, including:

[0023] Generating a noise mask based on the pixel gray distribution in the filter band of the reference image set, where the noise mask covers the environmental specular area in the dynamic image set and is outside the potential leakage path prediction result;

[0024] Comparing the spectral difference values between the dynamic image set and the reference image set in the same filter band, and screening the areas within the leakage path range where the spectral difference values exceed the upper limit of the gray distribution of the noise mask as the initial candidate areas;

[0025] Dynamically adjusting the noise suppression threshold according to the fluctuation direction of the real-time pressure parameter;

[0026] Verifying the spatial overlap ratio between the initial candidate area and the microscopic deformation area, excluding the candidate areas with a spatial overlap ratio lower than the preset value, and retaining the candidate areas that meet the spatial overlap ratio and the spectral difference values conform to the dynamic threshold.

[0027] In a preferred embodiment, if the pressure fluctuation amplitude increases, the noise suppression threshold is decreased; if the pressure fluctuation amplitude decreases, the noise suppression threshold is increased.

[0028] In a preferred embodiment, the active pressure perturbation triggers the change in the spectral reflectance of the candidate area, and combines the fluid conduction velocity to verify the inverse relationship between the penetration rate and the conduction velocity, and screen the suspected fault areas, including:

[0029] Applying a pressure perturbation with a preset duration to the candidate area, and synchronously recording the change amplitude of the spectral reflectance of the candidate area in the same filter band before and after the pressure perturbation;

[0030] Calculating the theoretical penetration rate threshold according to the fluid conduction velocity, where the fluid conduction velocity is derived based on the real-time pressure difference and the inner diameter of the pipeline through the Poiseuille's law;

[0031] Screening the candidate areas where the change amplitude of the spectral reflectance is inversely proportional to the theoretical penetration rate threshold as the suspected fault areas;

[0032] Excluding the interference areas in the candidate areas where the change direction of the spectral reflectance is inconsistent with the pressure perturbation direction.

[0033] In a preferred embodiment, the inverse relationship verification satisfies that the change amplitude of the spectral reflectance is less than 1 / N times of the theoretical penetration rate threshold, where N is the linear coefficient of the fluid conduction velocity.

[0034] In a preferred embodiment, a spectral intensity fluctuation threshold for the suspected fault area is determined, and target areas with spectral intensity fluctuations exceeding the spectral intensity fluctuation threshold and spatially overlapping with the microscopic deformation area are screened, including:

[0035] Perform time-series analysis on the spectral reflectance changes in the suspected fault area, and calculate the standard deviation of the fluctuation amplitude within a preset time window as the spectral intensity fluctuation threshold;

[0036] Dynamically correct the spectral intensity fluctuation threshold based on the historical spectral fluctuation data of the normal deformation area in the reference image set, and the correction amplitude is positively correlated with the fluctuation amplitude of the real-time pressure parameter;

[0037] Screen candidate areas with spectral intensity fluctuation values exceeding the corrected spectral intensity fluctuation threshold and a spatial overlap rate with the microscopic deformation area higher than the preset overlap threshold as target areas;

[0038] Verify the consistency between the spectral fluctuation direction and the pressure perturbation direction of the target area, and eliminate candidate areas with reverse fluctuation directions.

[0039] In a preferred embodiment, match the spectral reflectance change characteristics of the target area and the spatial distribution pattern of the potential leakage path prediction result to generate a leakage risk assessment result, including:

[0040] Extract the spectral reflectance change characteristics of the target area, and match the spectral reflectance change characteristics with the spatial distribution pattern of the potential leakage path prediction result. The matching of the spatial distribution pattern includes path orientation, extension length, and overlap ratio with the microscopic deformation area;

[0041] Verify the negative correlation between the spectral reflectance change rate of the target area and the fluid conduction speed in the leakage path, and screen target areas that meet the negative correlation of rate - speed;

[0042] Analyze the spatial continuity of the target area on the leakage path, and eliminate areas with isolated distribution and inconsistent path conduction directions;

[0043] Dynamically assign weights to the spectral reflectance change amplitude and spatial continuity based on the real-time pressure parameter to generate a comprehensive leakage risk assessment result.

[0044] On the other hand, the present invention provides an anesthesia device fault diagnosis system based on image recognition, including:

[0045] Image acquisition module: Obtain a reference image set and a dynamic image set of the transparent pipeline area of the anesthesia device;

[0046] Polarization path module: Identify the microscopic deformation area based on the change in the polarization light scattering pattern on the surface of the transparent pipeline in the dynamic image set, and generate a potential leakage path prediction result in combination with the fluid conduction direction;

[0047] Noise reduction and extraction module: Based on the pixel distribution characteristics of the reference image set and the potential leakage path prediction results, the ambient reflection noise in the non-leakage path area is suppressed, and the candidate area within the leakage path range with spectral differences from the noise is extracted;

[0048] Disturbance verification module: Active pressure disturbance triggers the change of spectral reflectance in the candidate area, and verifies the inverse relationship between penetration rate and conduction velocity in combination with fluid conduction velocity, thus screening suspected fault areas;

[0049] Threshold screening module: determines the spectral intensity fluctuation threshold of the suspected fault area, and screens the target area whose spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and overlaps with the microscopic deformation area space;

[0050] Assessment generation module: Matches the spectral reflectance variation characteristics of the target area and the spatial distribution of the potential leakage path prediction results to generate leakage risk assessment results.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. Through the deep integration of multimodal optical features and fluid conduction characteristics, high-sensitivity and high-specificity detection of early leakage in transparent pipelines is achieved. Based on the collaborative analysis of polarized light scattering patterns and multi-spectral dynamic imaging, it can effectively capture microscopic deformation and liquid film diffusion characteristics that are difficult to identify with traditional visual methods. Through the closed-loop logic of leakage path prediction and noise suppression, the misjudgment caused by ambient light interference is significantly reduced. The active pressure perturbation design is combined with the inverse relationship verification of the permeation rate to dynamically bind the optical response with the fluid mechanics parameters, making the weak leakage signal explicit under the constraints of physical laws, solving the problem of missed detection caused by traditional image detection relying on static thresholds, and avoiding the interference of contact sensors on the operation status of the equipment, providing a non-invasive and highly reliable leakage monitoring method for clinical scenarios;

[0053] 2. Through the dual screening mechanism of spectral intensity fluctuation threshold and spatial overlap verification, the accurate positioning of real leakage characteristics is achieved. Its dynamic weight allocation and path continuity evaluation strategy can adapt to pressure fluctuations and fluid state changes under different working conditions to ensure the high consistency between risk assessment results and physical characteristics of leakage. Through cross-dimensional data fusion (optical, spatial, fluid conduction) and multi-stage logic verification, it breaks through the limitations of a single detection dimension in a complex interference environment, improves the detection rate of early trace leaks, and avoids false alarms caused by local noise or instantaneous disturbances, providing full-link closed-loop support from feature recognition to risk decision-making for the safety maintenance of anesthesia equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1Flow chart of the anesthesia equipment fault diagnosis method based on image recognition according to the present invention;

[0055] Figure 2 Structural schematic diagram of the anesthesia equipment fault diagnosis system based on image recognition according to the present invention. Detailed implementation manners

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] Embodiment 1: Figure 1 The anesthesia equipment fault diagnosis method based on image recognition according to the present invention is given, which includes the following steps:

[0058] S1. Obtain a reference image set and a dynamic image set of the transparent pipeline area of the anesthesia equipment;

[0059] S2. Identify the microscopic deformation area based on the change of the polarized light scattering pattern on the surface of the transparent pipeline in the dynamic image set, and generate a potential leakage path prediction result in combination with the fluid conduction direction;

[0060] S3. According to the pixel distribution characteristics of the reference image set and the potential leakage path prediction result, suppress the environmental reflection noise in the non-leakage path area, and extract the candidate area within the leakage path range and having a spectral difference from the noise;

[0061] S4. Actively trigger the change of the spectral reflectivity of the candidate area by pressure perturbation, and verify the inverse relationship between the penetration rate and the conduction speed in combination with the fluid conduction speed, and screen the suspected fault area;

[0062] S5. Determine the spectral intensity fluctuation threshold of the suspected fault area, and screen the target area whose spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and spatially overlaps with the microscopic deformation area;

[0063] S6. Match the spectral reflectivity change characteristics of the target area and the spatial distribution pattern of the potential leakage path prediction result to generate a leakage risk assessment result.

[0064] S1. Obtain a reference image set and a dynamic image set of the transparent pipeline area of the anesthesia equipment, including:

[0065] Use a multispectral imaging device to collect a reference image set including visible light band and near-infrared band when the transparent pipeline is in a non-leaking state, and the filter band covers the characteristic absorption peaks of the pipeline material in a normal sealed state;

[0066] During the operation of the anesthesia device, a dynamic image set containing the polarized light scattering characteristics of the pipeline surface is synchronously collected, and the real-time pressure parameters inside the pipeline are synchronously recorded each time of collection.

[0067] Adjust the angle of the polarization filter to obtain a dynamic image set with the same filter band and polarization angle as the reference image set.

[0068] The filter bands of the multispectral imaging device cover the characteristic absorption peaks of the transparent pipeline material in the normal sealed state. For example, the central wavelengths of the filter bands include 515 nm and 550 nm in the visible light band, and 830 nm and 905 nm in the near-infrared band, and the bandwidth of each filter band is 10 nm to 20 nm.

[0069] The characteristic absorption peaks are determined by pre-performing Fourier transform infrared spectroscopy tests on the transparent pipeline material. Specifically, during the test, the transparent pipeline material is cut into thin slices with a thickness of 1 mm, placed in the sample chamber of the Fourier transform infrared spectrometer, the scanning wavelength range is 400 nm to 2500 nm, the positions of the characteristic absorption peaks of the material in the normal sealed state are identified, and the central wavelengths of the filter bands of the multispectral imaging device are set to be consistent with the positions of the characteristic absorption peaks.

[0070] For example, when there is a characteristic absorption peak at 830 nm in the near-infrared band for the material, the central wavelength of the filter band is set to 830 nm. The acquisition frequency of the reference image set is 5 frames per second, the exposure time of each frame of image is 20 milliseconds, and the ambient light intensity is controlled between 300 lux and 500 lux, for example, by adjusting the brightness of the operating room lights. During acquisition, the distance between the multispectral imaging device and the transparent pipeline is fixed at 30 cm, and the optical axis of the imaging device is perpendicular to the pipeline surface.

[0071] The dynamic image set uses the same multispectral imaging device as the reference image set, and the real-time pressure parameters inside the pipeline are recorded by a pressure sensor each time of collection. For example, the sampling frequency of the pressure sensor is consistent with the image acquisition frequency, and the pressure data is aligned with the time stamp of the dynamic image set. The polarization angles of the dynamic image set include 0 degrees, 45 degrees, and 90 degrees. Specifically, during implementation, the light rays in different polarization directions are captured by rotating the angle of the polarization filter. For example, the rotation of the polarization filter is driven by a stepper motor, the rotation accuracy is ±1 degree, and 100 milliseconds are waited after each switching of the polarization angle to ensure the stability of the filter, and then the image acquisition starts. The filter bands of the dynamic image set are kept consistent with the reference image set, for example, the filter band is set at 515 nm in the visible light band.

[0072] The exposure time of the dynamic image set is dynamically adjusted according to the ambient light intensity. For example, when the ambient light intensity is lower than 300 lux, the exposure time is increased to 30 milliseconds, and when it is higher than 500 lux, it is reduced to 10 milliseconds to ensure the consistency of the image brightness.

[0073] When adjusting the angle of the polarization filter, calibration is performed according to the polarization angle parameters of the reference image set. For example, the calibration process includes acquiring a set of calibration images in a leak-free state, and adjusting the angle of the polarization filter until the polarization light scattering pattern of the calibration image has an error less than 5% from the scattering pattern of the reference image set at the same polarization angle.

[0074] The root mean square error of pixel gray values is used for error calculation. For example, a 100×100 pixel block in the same area on the surface of the pipeline is selected to compare the gray value differences between the calibration image and the reference image. If the root mean square error exceeds 5%, the angle of the polarization filter is finely adjusted until it meets the standard. For example, after finely adjusting 0.5 degrees by a stepper motor, a new calibration image is acquired and the error is calculated. After calibration, the position of the filter is locked, and then the dynamic image set is acquired.

[0075] The pressure parameter is obtained by a piezoresistive pressure sensor installed at the gas path outlet of the anesthesia device. For example, the sensor range is from -10 kPa to 100 kPa, the accuracy is ±0.5% of the full scale, and the pressure data and the time stamp of the dynamic image are synchronized by the same clock source. For example, the clock source uses the internal crystal oscillator of the embedded system, with a frequency stability of ±50 ppm and a time stamp accuracy of 1 ms, ensuring the accurate correspondence between the pressure data and the image frames.

[0076] Through these steps, the optical conditions of the reference image set and the dynamic image set are consistent, including the filter band, polarization angle, ambient light intensity, and imaging distance, thus providing a comparable data basis for noise suppression and feature extraction in subsequent steps. For example, the consistent filter band makes the spectral difference analysis in subsequent steps comparable, the consistent polarization angle ensures that the change in the scattering pattern is only caused by the pipeline state, and the dynamic adjustment of the ambient light intensity avoids external light interference.

[0077] S2. Identify the microscopic deformation area based on the change in the polarization light scattering pattern on the surface of the transparent pipeline in the dynamic image set, and generate a potential leakage path prediction result in combination with the fluid conduction direction, including:

[0078] Extract the scattering images at different polarization angles in the dynamic image set, and calculate the gray value differences of the same area on the pipeline surface at each polarization angle;

[0079] Based on the area where the gray value difference exceeds the gray fluctuation range of the reference image set at the same polarization angle, it is marked as the microscopic deformation area;

[0080] Combine the fluid conduction direction in the anesthesia device gas path with the spatial distribution of the microscopic deformation area, and extend the boundary of the microscopic deformation area along the fluid conduction direction to generate a potential leakage path;

[0081] Eliminate the potential leakage paths that match the normal deformation mode of the pipelines in the reference image set with a matching degree higher than the preset threshold, and retain the prediction results of the potential leakage paths.

[0082] When extracting the scattering images of different polarization angles in the dynamic image set, use a multispectral imaging device to obtain the scattering images at polarization angles of 0°, 45°, and 90° respectively. The filter bands of the multispectral imaging device are consistent with those of the reference image set, such as the visible light band of 515 nm and the near-infrared band of 830 nm. The image acquisition interval for each polarization angle is 200 milliseconds to ensure that the changes in the surface state of the pipeline are completely recorded. When calculating the gray value difference of the same area on the pipeline surface at each polarization angle, select a 100×100 pixel area with fixed coordinates on the pipeline surface, and calculate the average gray value of this area in the images at polarization angles of 0°, 45°, and 90° respectively. Taking the gray value of the 0° polarization image as the reference, calculate the gray difference value of the corresponding area in the 45° and 90° images. For example, when the gray value of the 45° image is 15 to 20 units lower than that of the 0° image, it is determined that there is a change in the scattering mode.

[0083] When marking the microscopic deformation areas based on the areas where the gray value difference exceeds the gray value fluctuation range of the reference image set at the same polarization angle, the gray value fluctuation range of the reference image set is determined by statistically calculating the standard deviation of the gray values of the same pipeline area in multiple acquisitions in the leak-free state. For example, the standard deviation at the 0° polarization angle is ±5. When the gray difference value of a certain area in the dynamic image exceeds 3 times the reference standard deviation, it is marked as a microscopic deformation area. The boundary of the deformation area is determined by connected component analysis, and the areas where the gray difference between adjacent pixels exceeds the threshold are merged into the same deformation area.

[0084] When generating potential leakage paths by combining the fluid conduction direction in the gas circuit of the anesthesia device and the spatial distribution of the microscopic deformation areas, the fluid conduction direction is calculated based on the pressure difference between the inlet and outlet of the gas circuit of the anesthesia device. For example, when the inlet pressure is 40 kPa and the outlet pressure is 35 kPa, the fluid conduction direction is from the inlet to the outlet. Along this direction, extend the boundary of the microscopic deformation area downstream of the fluid, and the extension step size is 0.5 times the pipeline diameter. For example, when the pipeline diameter is 10 mm, the extension step size is 5 mm. The number of extensions is dynamically adjusted according to the fluid velocity, and the fluid velocity is calculated by the pressure difference and the inner diameter of the pipeline. For example, when the pressure difference is 5 kPa and the inner diameter is 10 mm, the fluid velocity is 0.2 m per second, and the boundary is extended once per second to generate a continuously extended potential leakage path.

[0085] When eliminating potential leakage paths whose matching degree with the normal deformation mode of the pipeline in the reference image set is higher than the preset threshold, the normal deformation mode is determined by statistically analyzing the shape parameters of all deformation regions in the reference image set. The shape parameters include the aspect ratio, area, and perimeter of the region. For example, the average aspect ratio of the normal deformation region is 1.2, and the average area is 50 square pixels. Calculate the similarity between the deformation region corresponding to the potential leakage path and the normal deformation mode. The similarity calculation uses the weighted sum of the aspect ratio and area, and the weights are 0.6 and 0.4 respectively. When the weighted similarity exceeds the preset threshold of 90%, it is determined as normal deformation and eliminated, and the abnormal leakage paths with a weighted similarity lower than 90% are retained as the prediction result.

[0086] The weight distribution of the aspect ratio and area is determined based on the contribution degrees of the two types of parameters to leakage determination in historical leakage data. For example, through statistical analysis of historical leakage cases, it is found that the contribution ratio of the aspect ratio to the discrimination of abnormal leakage is about 60%, and the area accounts for about 40%. Therefore, the weights are set to 0.6 and 0.4. The weight distribution process includes: performing logistic regression analysis on the aspect ratio and area of all deformation regions in the historical leakage data, calculating the ratio of the regression coefficients of the two types of parameters, and normalizing the ratio to obtain the weights.

[0087] The preset threshold is determined through experimental verification, specifically including: collecting multiple groups of normal deformation data in a non-leakage state, calculating the similarity distribution between the normal deformation region and the reference data, and taking the upper limit value of the similarity distribution as the threshold. For example, if the highest similarity of the normal deformation region is statistically 89%, the threshold is set to 90% to cover the normal fluctuation range.

[0088] When obtaining the similarity distribution between the normal deformation region and the reference data, the specific method is as follows: in a non-leakage state, collect multiple groups of normal deformation image data, extract the contours and characteristic parameters of each deformation region on the pipeline surface through image segmentation; match the geometric structure, texture distribution, and edge features of each deformation region with the standardized data at the corresponding position in the reference database region by region, and use the feature weighted fusion algorithm to calculate the overall similarity between each region and the reference data; perform frequency statistics on the similarity calculation results of all normal samples, draw the similarity probability distribution curve, determine its concentration interval and discrete boundary, and finally obtain the similarity distribution result covering the normal deformation fluctuation range, such as determining the upper limit threshold of the distribution through normal distribution fitting or non-parametric statistical methods.

[0089] The predicted results of the retained potential leakage paths are stored in the form of a sequence of coordinate points. Each coordinate point corresponds to the position on the pipeline surface, and the sequence direction is consistent with the fluid conduction direction. The path length is calculated based on the fluid velocity and the number of extensions. For example, when the fluid velocity is 0.2 meters per second and the number of extensions is 10 times, the path length is 2 meters. Finally, a leakage prediction report including path coordinates, length, and risk level is generated.

[0090] S3. According to the pixel distribution characteristics of the reference image set and the potential leakage path prediction results, suppress the environmental reflection noise in the non-leakage path area, and extract candidate areas within the leakage path range and having a spectral difference from the noise, including:

[0091] Generate a noise mask based on the pixel gray-scale distribution in the filter band of the reference image set, where the noise mask covers the environmental reflection area in the dynamic image set and is outside the potential leakage path prediction result area;

[0092] Compare the spectral difference values between the dynamic image set and the reference image set in the same filter band, and screen out the areas within the leakage path range where the spectral difference value exceeds the upper limit of the gray-scale distribution of the noise mask as the initial candidate areas;

[0093] Dynamically adjust the noise suppression threshold according to the fluctuation direction of the real-time pressure parameter. If the pressure fluctuation amplitude increases, the noise suppression threshold is decreased; if the pressure fluctuation amplitude decreases, the noise suppression threshold is increased;

[0094] Verify the spatial overlap ratio between the initial candidate area and the microscopic deformation area, eliminate the candidate areas with a spatial overlap ratio lower than the preset value, and retain the candidate areas that meet the spatial overlap ratio and the spectral difference value conforms to the dynamic threshold.

[0095] When generating the noise mask, the filter band of the reference image set is consistent with that of the dynamic image set. For example, the visible light band of 515 nm and the near-infrared band of 830 nm. The reference image set collects multiple groups of images in the non-leakage state, and statistically analyzes the pixel gray-scale distribution of each area on the pipeline surface in the same filter band. The gray-scale distribution is determined by calculating the range of gray-scale values in multiple acquisitions of each area. For example, in the 515 nm filter band, the gray-scale value of a certain area fluctuates between 120 and 150, and the range is 30. The noise mask covers the area in the dynamic image set where the gray-scale value is within this range and is outside the potential leakage path prediction result area. For example, if the gray-scale value of a certain area in the dynamic image is 140 and is outside the leakage path range, it is marked as an environmental reflection noise area. When screening the areas within the leakage path range where the spectral difference value exceeds the upper limit of the gray-scale distribution of the noise mask, the pixel-level gray-scale value comparison is performed between the dynamic image set and the reference image set in the same filter band. For example, in the 830 nm filter band, the gray-scale value of a certain pixel in the dynamic image is 25 lower than that of the reference image. The upper limit of the gray-scale distribution of the noise mask is determined by statistically analyzing the 95% quantile of the gray-scale values in the reference image set in the same filter band. For example, the upper limit is 160. When the dynamic pixel gray-scale difference value exceeds 25 and is within the leakage path range, it is marked as an initial candidate area.

[0096] When dynamically adjusting the noise suppression threshold according to the fluctuation direction of real-time pressure parameters, the pressure fluctuation amplitude is determined by calculating the pressure difference between adjacent time points. For example, if the current pressure is 40 kPa and the previous time point is 38 kPa, the fluctuation amplitude is an increase of 2 kPa. If the pressure fluctuation amplitude increases, it indicates that the fluid state in the pipeline is unstable and may be accompanied by a leakage risk. At this time, the noise suppression threshold is reduced to capture weaker spectral difference signals. For example, the threshold is adjusted from 25 to 20. If the pressure fluctuation amplitude decreases, it indicates that the fluid state tends to be stable. At this time, the noise suppression threshold is increased to reduce false detections. For example, the threshold is adjusted from 25 to 30. The adjusted dynamic threshold is applied to screen the spectral difference values of the initial candidate regions. For example, when the pressure fluctuation amplitude increases, only the candidate regions with a gray-scale difference value exceeding 20 are retained.

[0097] When verifying the spatial overlap ratio between the initial candidate region and the microscopic deformation region, the spatial overlap ratio is determined by calculating the proportion of overlapping pixels between the candidate region and the microscopic deformation region. For example, if the candidate region contains 100 pixels and 60 of them overlap with the microscopic deformation region, the overlap ratio is 60%. The preset value is set according to the statistical overlap ratio of the normal deformation regions in the reference image set. For example, the average overlap ratio of the normal deformation regions in the non-leakage state is statistically 55%, and the preset value is set to 50%. When the overlap ratio of the candidate region is lower than 50%, it is determined as noise interference and excluded. The candidate regions with a qualified overlap ratio and spectral difference values meeting the dynamic threshold are retained. For example, if a candidate region has an overlap ratio of 55%, a spectral difference value of 22, and a dynamic threshold of 20, it is retained. The coordinates of the retained candidate regions are matched with the spatial distribution pattern of the leakage path prediction results. For example, if the candidate region is located in the extension direction of the leakage path and is continuously adjacent to the path coordinate sequence, it is determined as an effective leakage feature region.

[0098] The noise suppression and candidate region extraction process combines the filter band consistency, leakage path spatial constraint, pressure parameter dynamic threshold, and deformation region overlap verification to ensure high-sensitivity capture of weak leakage signals in transparent pipelines while excluding the interference of environmental reflection and normal deformation. For example, in the 515-nm filter band, after the initial candidate regions are screened by the dynamic threshold and overlap verification, the spectral difference values of the finally retained regions are between 20 and 30, and the overlap ratio with the microscopic deformation region exceeds 50%, thus providing reliable input for subsequent leakage risk assessment.

[0099] S4. Actively trigger the change in the spectral reflectance of the candidate region by pressure perturbation, and verify the inverse relationship between the penetration rate and the conduction speed in combination with the fluid conduction speed, and screen the suspected fault regions, including:

[0100] Apply a pressure perturbation with a preset duration to the candidate region, and synchronously record the change amplitude of the spectral reflectance of the candidate region in the same filter band before and after the pressure perturbation;

[0101] Calculate the theoretical penetration rate threshold according to the fluid conduction velocity, where the fluid conduction velocity is derived from the real-time pressure difference and the inner diameter of the pipeline through Poiseuille's law;

[0102] Select the candidate areas where the change amplitude of spectral reflectance is inversely proportional to the theoretical penetration rate threshold as the suspected fault areas; the inverse proportional relationship is verified to satisfy that the change amplitude of spectral reflectance is less than 1 / N times of the theoretical penetration rate threshold, where N is the linear coefficient of the fluid conduction velocity;

[0103] Exclude the interference areas in the candidate areas where the change direction of spectral reflectance is inconsistent with the pressure disturbance direction.

[0104] When applying a pressure disturbance for a preset duration in the candidate areas, short-term adjustment of the internal pressure of the pipeline is performed through the gas path control system of the anesthesia device. For example, the pressure is increased from 40 kPa to 45 kPa within 1 second and maintained for 2 seconds and then restored to the original pressure, and the spectral reflectance data of the candidate areas before and after the pressure disturbance are synchronously collected in the same filter band. The change amplitude of spectral reflectance is determined by calculating the difference in the average gray value of the same candidate area before and after the pressure disturbance. For example, the average gray value of the candidate area in the 515 nm filter band before the disturbance is 120, and it drops to 100 after the disturbance, then the change amplitude is 20 units. The filter band is consistent with the reference image set, such as the visible light band of 515 nm and the near-infrared band of 830 nm, to ensure the comparability of spectral data.

[0105] When calculating the theoretical penetration rate threshold according to the fluid conduction velocity, the fluid conduction velocity is derived from the real-time pressure difference and the inner diameter of the pipeline through Poiseuille's law, specifically including: calculating the flow velocity according to the real-time pressure difference between the gas path inlet and outlet of the anesthesia device. For example, when the inlet pressure is 45 kPa and the outlet pressure is 40 kPa, the pressure difference is 5 kPa. Combining with the inner diameter of the pipeline of 10 mm, the fluid conduction velocity is calculated to be 0.25 m per second through Poiseuille's law. The theoretical penetration rate threshold is inversely proportional to the fluid conduction velocity. For example, when the conduction velocity is 0.25 m per second, the theoretical penetration rate threshold is set to 0.05 ml per second, and when the conduction velocity is increased to 0.5 m per second, the threshold is reduced to 0.025 ml per second. The basis for threshold calculation is the physical law that the penetration rate of the leakage substance in the pipeline decreases with the increase of the flow velocity. For example, the faster the flow velocity, the shorter the residence time of the leakage substance on the pipeline surface and the less the penetration amount.

[0106] When screening candidate regions where the amplitude of spectral reflectance change is inversely proportional to the theoretical penetration rate threshold, the inverse relationship verification needs to satisfy that the amplitude of spectral reflectance change is less than 1 / N times the theoretical penetration rate threshold, where N is the linear coefficient of the fluid conduction velocity. For example, when the fluid conduction velocity is 0.25 meters per second, N is set to 5, and the theoretical penetration rate threshold is 0.05 milliliters per second, then the amplitude of spectral reflectance change needs to be less than 0.01 milliliter equivalent (0.05 / 5). If the change amplitude of the candidate region is 0.008 milliliter equivalent, it is determined to conform to the inverse relationship. The linear coefficient N is determined through experimental calibration. For example, at a known leakage rate, the reflectance change amplitudes corresponding to different flow rates are measured, and the linear relationship between the N value and the flow rate is fitted as N = flow rate × pipeline length. When the pipeline length is 2 meters, a flow rate of 0.25 meters per second corresponds to N = 0.5.

[0107] When excluding interference regions in the candidate regions where the direction of spectral reflectance change is inconsistent with the direction of pressure perturbation, the correlation between the direction of pressure perturbation and the direction of spectral reflectance change is verified through physical laws. For example, when a positive pressure perturbation (pressure increase) is applied, the liquid film thickness in the leakage region becomes thinner due to the increased pressure, resulting in a decrease in spectral reflectance. If the reflectance of the candidate region increases after pressurization, it is determined to be interference; conversely, when a negative pressure perturbation (pressure reduction) is applied, the reflectance should increase, and if it decreases, it is excluded. During verification, the corresponding relationship between the direction of pressure perturbation and the direction of reflectance change is recorded. For example, if the decrease amplitude of reflectance exceeds 5 units after pressurization, it is an effective signal, and candidate regions with a change amplitude lower than 5 or in the opposite direction are excluded. The remaining suspected fault regions need to satisfy both the inverse relationship verification and the direction consistency. For example, if the reflectance of a certain candidate region decreases by 18 units after pressurization, corresponding to a penetration rate equivalent of 0.009 milliliters, which is less than the theoretical threshold of 0.01 milliliters, and the change direction meets the expectation, it is marked as a suspected fault region.

[0108] The change characteristics of the spectral reflectance of the candidate region are bound to the physical laws of the fluid conduction velocity. Combining with the verification of the pressure perturbation direction can effectively distinguish real leakage signals from random noise. For example, in the 830-nanometer filter band, the amplitude of spectral reflectance change of a certain candidate region is 15 units, corresponding to a penetration rate equivalent of 0.007 milliliters, which is less than the theoretical threshold of 0.01 milliliters, and the reflectance decreases in the same direction after pressurization, then it is determined to be a suspected fault region and enters the subsequent processing steps.

[0109] S5. Determine the spectral intensity fluctuation threshold of the suspected fault region, and screen the target regions where the spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and spatially overlaps with the micro-deformation region, including:

[0110] Conduct time series analysis on the spectral reflectance change of the suspected fault region, and calculate the standard deviation of the fluctuation amplitude within the preset time window as the spectral intensity fluctuation threshold;

[0111] Dynamically correct the spectral intensity fluctuation threshold based on the historical spectral fluctuation data of the normal deformation region in the reference image set, and the correction amplitude is positively correlated with the fluctuation amplitude of the real-time pressure parameter;

[0112] Select the candidate regions with the spectral intensity fluctuation value exceeding the corrected spectral intensity fluctuation threshold and the spatial overlap rate with the microscopic deformation region higher than the preset overlap threshold as the target regions;

[0113] Verify the consistency between the spectral fluctuation direction and the pressure perturbation direction of the target region, and eliminate the candidate regions with the reverse fluctuation direction.

[0114] When performing time series analysis on the spectral reflectance change of the suspected fault region, the preset time window is set according to the typical leakage response time of the anesthesia device. For example, the time window is 5 seconds, and the spectral reflectance data of the candidate region within this time period is continuously collected. The standard deviation of the spectral reflectance change amplitude is determined by calculating the fluctuation of the average gray value of each frame image within the time window. For example, the average gray value of a certain candidate region fluctuates from 120 to 110 within 5 seconds, and the standard deviation is 8.5. The spectral intensity fluctuation threshold is set to 1.5 times the standard deviation, that is, 12.75. The fluctuation exceeding this threshold is regarded as abnormal. The historical spectral fluctuation data of the normal deformation region in the reference image set is determined by statistically calculating the standard deviation of the spectral reflectance in the same filter band under the leak-free state. For example, the average standard deviation of the normal deformation region in the 515-nm filter band is 5, and the reference value of the fluctuation threshold is 7.5.

[0115] When dynamically correcting the spectral intensity fluctuation threshold based on the fluctuation amplitude of the real-time pressure parameter, the pressure fluctuation amplitude is determined by calculating the absolute value of the pressure difference between adjacent time points. For example, the current pressure is 42 kPa, and the previous time point is 40 kPa, and the fluctuation amplitude is 2 kPa. The rule for the correction amplitude to be positively correlated with the pressure fluctuation amplitude is: for every 1 kPa increase in pressure fluctuation, the fluctuation threshold is increased by 5%. For example, the reference threshold is 7.5. When the pressure fluctuation amplitude is 2 kPa, the corrected threshold is 7.5×(1 + 0.05×2) = 8.25. If the pressure fluctuation amplitude is 5 kPa, the threshold is adjusted to 7.5×(1 + 0.05×5) = 9.375. The threshold correction process is realized by the look-up table method. The correction coefficients corresponding to different pressure fluctuation amplitudes are pre-stored in the table. For example, the coefficient corresponding to a pressure fluctuation of 1 kPa is 1.05, 2 kPa corresponds to 1.10, and so on.

[0116] When screening candidate regions where the spectral intensity fluctuation value exceeds the corrected threshold and the spatial overlap rate with the microscopic deformation region is higher than the preset overlap threshold, the preset overlap threshold is set according to the average overlap ratio of the normal deformation regions in the reference image set. For example, statistics show that the average overlap ratio of the normal deformation regions and the microscopic deformation regions is 55%, and the preset overlap threshold is set to 50%. For example, if the spectral intensity fluctuation value of a candidate region is 9.0, exceeding the corrected threshold of 8.25, and the proportion of overlapping pixels with the microscopic deformation region is 60%, higher than the preset value of 50%, it is marked as the target region. If the fluctuation value of the candidate region is 8.0 and the overlap ratio is 45%, it is excluded. The coordinate range of the target region and the boundary error of the microscopic deformation region are controlled within 10% of the pipeline diameter. For example, when the pipeline diameter is 10 mm, a boundary error of ±1 mm is allowed.

[0117] When verifying the consistency between the spectral fluctuation direction of the target region and the pressure perturbation direction, the pressure perturbation direction is divided into two modes: pressure increase and pressure decrease. For example, during the pressure increase stage (the pressure rises from 40 kPa to 45 kPa), the spectral reflectivity of the normal leakage region should show a downward trend. If the reflectivity of the target region rises after the pressure increase, it is determined that the directions are inconsistent and it is excluded. The direction consistency is verified by comparing the spectral reflectivity changes of two frames of images before and after the pressure perturbation. For example, if the reflectivity drops by more than 5 units after the pressure increase, it is considered that the directions are consistent; if the rise or fall is less than 5 units, it is excluded. The excluded target regions need to meet the triple conditions of the fluctuation threshold, spatial overlap rate, and direction consistency. For example, if the fluctuation value of a region is 9.5 (threshold 8.25), the overlap rate is 60%, and the reflectivity drops by 6 units after the pressure increase, it is retained as a valid target region.

[0118] The screening of the target region combines dynamic threshold correction, spatial overlap constraint, and physical law verification. For example, in the 830 nm filter band, if the fluctuation value of a candidate region is 9.8, the overlap rate is 65%, and the reflectivity drops by 7 units after the pressure increase, it enters the final leakage risk assessment after meeting all the conditions. During the screening process, interference signals with fluctuation values lower than the threshold, insufficient overlap rates, or abnormal directions are excluded. For example, a region with a fluctuation value of 7.0 (threshold 8.25) is excluded, and another region with a fluctuation value of 9.0 but a reflectivity increase of 3 units is also excluded. The list of finally retained target regions is transmitted to the subsequent steps to generate the leakage risk assessment result.

[0119] S6. Match the spectral reflectivity change characteristics of the target region and the spatial distribution pattern of the potential leakage path prediction result to generate the leakage risk assessment result, including:

[0120] Extract the spectral reflectivity change characteristics of the target region, match the spectral reflectivity change characteristics with the spatial distribution pattern of the potential leakage path prediction result. The matching of the spatial distribution pattern includes the path direction, extension length, and the overlap ratio with the microscopic deformation region;

[0121] Verify the negative correlation between the spectral reflectance change rate of the target area and the fluid conduction velocity in the leakage path, and screen the target areas that meet the negative correlation of rate-velocity;

[0122] Analyze the spatial continuity of the target area on the leakage path, and eliminate the areas with isolated distribution and inconsistent path conduction direction;

[0123] Based on the real-time pressure parameters, dynamically assign weights to the spectral reflectance change amplitude and spatial continuity to generate a comprehensive leakage risk assessment result.

[0124] When matching the spatial distribution pattern of the spectral reflectance change characteristics of the target area and the predicted result of the potential leakage path, the spectral reflectance change characteristics include the change amplitude, change rate and change direction. For example, after applying a pressure perturbation, the spectral reflectance of a certain target area shows a continuous downward trend, while the reflectance of another area rebounds after the pressure recovers. Such dynamic change characteristics are used to correlate with the spatial pattern of the leakage path. The spatial distribution pattern of the predicted result of the potential leakage path is generated according to the fluid conduction direction. For example, the leakage path extends from the gas path inlet to the outlet along the fluid flow direction, the path direction is consistent with the internal pressure gradient direction of the equipment, and the extension length covers the range that the fluid may diffuse during the monitoring period. When matching, if the coincidence degree between the spectral reflectance change characteristics of the target area and the spatial trend of the leakage path is relatively high, and its extension length covers at least 70% of the path prediction range, it is determined to be a spatial pattern match.

[0125] When verifying the negative correlation between the spectral reflectance change rate and the fluid conduction velocity, the negative correlation is manifested as: when the fluid conduction velocity increases due to the increase of the pressure difference, the residence time of the leakage substance on the pipeline surface is shortened, resulting in a slowdown of the spectral reflectance change rate. For example, at a high pressure difference, the fluid velocity is fast, and the spectral reflectance change rate of the target area should be lower than that at a low pressure difference. If a certain area still shows a high rate of change at a high pressure difference, it is determined as an abnormal interference signal and eliminated. The verification process is realized by comparing the spectral reflectance change rates of the same target area under different pressure conditions. For example, if the rate is significantly lower at a high pressure difference than at a low pressure difference, it conforms to the negative correlation.

[0126] When analyzing the spatial continuity of the target area on the leakage path, the spatial continuity requires that the target area be distributed along the leakage path direction, and the interval between adjacent areas does not exceed the preset distance. For example, the distance between the center points of adjacent target areas on the leakage path needs to be less than twice the pipeline diameter. If a certain area is isolated outside the path or the interval exceeds the preset distance, it is regarded as discontinuous and excluded. The verification of the path conduction direction consistency needs to ensure that the included angle between the distribution trend of the target area and the fluid conduction direction is less than the allowable deviation. For example, when the included angle between the connection direction of the center points of the target area and the fluid conduction direction exceeds 30 degrees, it is determined that the direction is inconsistent.

[0127] When dynamically allocating the weights of the spectral reflectance change amplitude and spatial continuity based on real-time pressure parameters, the weight allocation rule is dynamically adjusted according to the pressure state. For example, in the high-pressure state, the contribution weight of the spectral reflectance change amplitude to the leakage risk is increased, while the weight of spatial continuity is relatively decreased; in the low-pressure state, it is the opposite. In specific implementation, the system automatically switches the weight allocation mode according to the range of the real-time pressure value. For example, the "spectral weight priority" mode is adopted in the high-pressure range, and the "spatial continuity priority" mode is adopted in the low-pressure range. When generating the comprehensive leakage risk assessment result, the weighted spectral reflectance change score and spatial continuity score are combined to divide the risk level. For example, a certain target area has a high spectral score and a medium spatial continuity score under the high-pressure state, and is classified as a high risk after weighting; another area has a high spatial continuity score but a low spectral score under the low-pressure state, and is classified as a medium risk after weighting. The final assessment result marks the risk level and location information of the target area for the operation and maintenance personnel to handle targeted.

[0128] The optical characteristics, spatial distribution and fluid mechanics laws of the target area are deeply integrated. For example, a certain area has a continuous decrease in spectral reflectance, is continuously distributed along the leakage path and has a consistent direction, and is given a high weight under high pressure and determined as a high-risk leakage point; another area has significant spectral changes but scattered distribution and deviated direction, and the risk level is reduced after weight adjustment. The assessment result output includes the risk level, location description and disposal suggestions, such as "the third bend at the inlet section, high risk, it is recommended to detect leaks first", realizing the closed-loop from data to decision-making.

[0129] Embodiment 2: Figure 2 The structural schematic diagram of the anesthesia equipment fault diagnosis system based on image recognition according to the present invention is given, which includes:

[0130] Image acquisition module: Obtain the reference image set and dynamic image set of the transparent pipeline area of the anesthesia equipment;

[0131] Polarization path module: Based on the change of the polarization light scattering pattern on the surface of the transparent pipeline in the dynamic image set, identify the microscopic deformation area, and generate the prediction result of the potential leakage path in combination with the fluid conduction direction;

[0132] Noise reduction and extraction module: According to the pixel distribution characteristics of the reference image set and the prediction results of potential leakage paths, suppress the environmental reflection noise in the non-leakage path area, and extract candidate areas within the leakage path range and having spectral differences from the noise.

[0133] Perturbation verification module: Actively trigger the change in spectral reflectance of the candidate area by pressure perturbation, and combine the fluid conduction speed to verify the inverse relationship between the penetration rate and the conduction speed, and screen out suspected fault areas.

[0134] Threshold screening module: Determine the spectral intensity fluctuation threshold of the suspected fault area, and screen out the target area where the spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and spatially overlaps with the microscopic deformation area.

[0135] Evaluation and generation module: Match the spectral reflectance change characteristics of the target area and the spatial distribution pattern of the potential leakage path prediction results, and generate a leakage risk assessment result.

[0136] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0137] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0138] In addition, in each embodiment of the present application, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0139] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0140] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An anesthesia equipment fault diagnosis method based on image recognition, characterized in that: The steps include: S1, obtaining a reference image set and a dynamic image set of a transparent pipeline area of ​​an anesthesia device; S2, based on the change of polarized light scattering pattern on the transparent pipe surface in the dynamic image set, the micro deformation area is identified, and the potential leakage path prediction result is generated in combination with the fluid conduction direction; S3, based on the pixel distribution characteristics of the reference image set and the potential leakage path prediction results, suppress the ambient reflection noise in the non-leakage path area, and extract the candidate area within the leakage path range and with a spectral difference from the noise; S4. Active pressure disturbance triggers the change of spectral reflectance in the candidate area, and verifies the inverse relationship between the permeation rate and the conduction velocity in combination with the fluid conduction velocity, thus screening the suspected fault area; S5. Determine the spectral intensity fluctuation threshold of the suspected fault area, and select the target area where the spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and overlaps with the microscopic deformation area space; S6. Match the spectral reflectance variation characteristics of the target area and the spatial distribution pattern of the potential leakage path prediction results to generate leakage risk assessment results.

2. The anesthesia equipment fault diagnosis method based on image recognition according to claim 1, characterized in that: Obtain baseline image sets and dynamic image sets of the transparent pipeline area of ​​the anesthesia equipment, including: Multispectral imaging equipment is used to collect a reference image set containing visible light bands and near-infrared bands when the transparent pipeline is leak-free. The filter band covers the characteristic absorption peak of the pipeline material under normal sealing conditions. During the operation of the anesthesia equipment, a dynamic image set containing the polarized light scattering characteristics of the pipeline surface is synchronously acquired, and the real-time pressure parameters inside the pipeline are synchronously recorded during each acquisition; The polarization filter angle is adjusted to obtain a dynamic image set with the same filter band and polarization angle as the reference image set.

3. The anesthesia equipment fault diagnosis method based on image recognition according to claim 1, characterized in that: Based on the changes in polarized light scattering patterns on the transparent pipe surface in the dynamic image set, the microscopic deformation area is identified, and the potential leakage path prediction results are generated in combination with the fluid conduction direction, including: Extract the scattered images at different polarization angles in the dynamic image set, and calculate the gray value difference of the same area on the pipeline surface at each polarization angle; Areas based on grayscale value differences exceeding the grayscale fluctuation range of the reference image set at the same polarization angle are marked as micro-deformation areas; Combining the fluid conduction direction in the gas circuit of the anesthesia equipment with the spatial distribution of the microscopic deformation area, the boundary of the microscopic deformation area is extended along the fluid conduction direction to generate a potential leakage path; The potential leakage paths whose matching degree of normal deformation pattern of pipeline in the reference image set is higher than the preset threshold are eliminated, and the prediction results of potential leakage paths are retained.

4. The anesthesia equipment fault diagnosis method based on image recognition according to claim 1, characterized in that: According to the pixel distribution characteristics of the reference image set and the potential leakage path prediction results, the ambient reflection noise in the non-leakage path area is suppressed, and the candidate areas within the leakage path range and with spectral differences from the noise are extracted, including: A noise mask is generated based on the pixel grayscale distribution in the filter band of the reference image set, where the noise mask covers the environmental reflection area in the dynamic image set and the area outside the potential leakage path prediction result; Compare the spectral difference values ​​of the dynamic image set and the reference image set in the same filter band, and select the area within the leakage path where the spectral difference value exceeds the upper limit of the grayscale distribution of the noise mask as the initial candidate area; Dynamically adjust the noise suppression threshold according to the real-time pressure parameter fluctuation direction; Verify the spatial overlap ratio between the initial candidate region and the microscopic deformation region, eliminate candidate regions whose spatial overlap ratio is lower than the preset value, and retain candidate regions that meet the spatial overlap ratio and whose spectral difference values ​​meet the dynamic threshold.

5. The method for anaesthesia equipment fault diagnosis based on image recognition according to claim 4, characterized in that: If the pressure fluctuation amplitude increases, the noise suppression threshold is lowered, and if the pressure fluctuation amplitude decreases, the noise suppression threshold is increased.

6. The method for anaesthesia equipment fault diagnosis based on image recognition according to claim 1, characterized in that: Active pressure disturbance triggers the change of spectral reflectance in the candidate area, and combines the fluid conduction velocity to verify the inverse relationship between the permeation rate and the conduction velocity, and screen the suspected fault area, including: Apply a pressure disturbance of a preset duration to the candidate area, and simultaneously record the change amplitude of the spectral reflectance of the candidate area in the same filter band before and after the pressure disturbance; The theoretical permeation rate threshold is calculated based on the fluid conductivity velocity, which is derived from Poiseuille's law based on the real-time pressure difference and the inner diameter of the pipeline; The candidate areas whose spectral reflectance variation amplitude is inversely proportional to the theoretical penetration rate threshold are selected as suspected fault areas; The interference areas in the candidate areas where the direction of spectral reflectance change is inconsistent with the direction of pressure disturbance are excluded.

7. The method for anaesthesia equipment fault diagnosis based on image recognition according to claim 6, characterized in that: The inverse relationship verifies that the change in spectral reflectance is less than 1 / N times the theoretical permeability threshold, where N is the linear coefficient of the fluid conduction velocity.

8. The method for anaesthesia equipment fault diagnosis based on image recognition according to claim 1, characterized in that: Determine the spectral intensity fluctuation threshold of the suspected fault area, and screen the target area whose spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and overlaps with the microscopic deformation area space, including: Perform time series analysis on the spectral reflectance changes in the suspected fault area, and calculate the standard deviation of the fluctuation amplitude within the preset time window as the spectral intensity fluctuation threshold; Dynamically correct the spectral intensity fluctuation threshold based on the historical spectral fluctuation data of the normal deformation area in the reference image set, and the correction amplitude is positively correlated with the real-time pressure parameter fluctuation amplitude; The candidate regions whose spectral intensity fluctuation values ​​exceed the corrected spectral intensity fluctuation threshold and whose spatial overlap rate with the microscopic deformation region is higher than the preset overlap threshold are selected as target regions; Verify the consistency between the spectral fluctuation direction and the pressure disturbance direction of the target area, and eliminate candidate areas with reverse fluctuation directions.

9. The anesthesia equipment fault diagnosis method based on image recognition according to claim 1, characterized in that: Match the spectral reflectance variation characteristics of the target area and the spatial distribution of the potential leakage path prediction results to generate leakage risk assessment results, including: Extract the spectral reflectance change characteristics of the target area, and match the spectral reflectance change characteristics with the spatial distribution form of the potential leakage path prediction results. The matching of the spatial distribution form includes the path direction, extension length and the overlap ratio with the micro deformation area; Verify the negative correlation between the rate of change of the spectral reflectance of the target area and the fluid conduction velocity in the leakage path, and select the target area that meets the rate-velocity negative correlation; Analyze the spatial continuity of the target area on the leakage path and eliminate the isolated areas that do not meet the consistency of the path conduction direction; Based on the real-time pressure parameters, dynamic weight allocation is performed on the spectral reflectance variation amplitude and spatial continuity to generate a comprehensive leakage risk assessment result.

10. An anesthesia equipment fault diagnosis system based on image recognition, used to implement the anesthesia equipment fault diagnosis method based on image recognition according to any one of claims 1 to 9, characterized in that: include: Image acquisition module: obtains the reference image set and dynamic image set of the transparent pipeline area of ​​the anesthesia equipment; Polarization path module: identifies microscopic deformation areas based on the changes in polarized light scattering patterns on the surface of transparent pipes in dynamic image sets, and generates potential leakage path prediction results based on the fluid conduction direction; Noise reduction and extraction module: Based on the pixel distribution characteristics of the reference image set and the potential leakage path prediction results, the ambient reflection noise in the non-leakage path area is suppressed, and the candidate area within the leakage path range with spectral differences from the noise is extracted; Disturbance verification module: Active pressure disturbance triggers the change of spectral reflectance in the candidate area, and verifies the inverse relationship between penetration rate and conduction velocity in combination with fluid conduction velocity, thus screening suspected fault areas; Threshold screening module: determines the spectral intensity fluctuation threshold of the suspected fault area, and screens the target area whose spectral intensity fluctuation exceeds the spectral intensity fluctuation threshold and overlaps with the microscopic deformation area space; Assessment generation module: Matches the spectral reflectance variation characteristics of the target area and the spatial distribution of the potential leakage path prediction results to generate leakage risk assessment results.

Images