Temperature data processing method, device and storage medium for peripheral circulation disorder monitoring

Abstract

The application provides a temperature data processing method, device and storage medium for peripheral circulation disorder monitoring, and relates to the technical field of biomedical engineering. The method comprises the following steps: acquiring original temperature data of a target monitoring object; obtaining a PI prediction value based on the original temperature data; constructing a CPTD index based on the original temperature data; and determining a peripheral circulation disorder evaluation result of the target monitoring object according to the PI prediction value and the CPTD index. The temperature data processing method can process the original temperature data of the target monitoring object to obtain the peripheral circulation disorder evaluation result of the target monitoring object, avoids the risk of invasive operation and contact infection, and is based on temperature data processing, which is not disturbed by environmental light. Therefore, the application provides a safe, stable and comprehensive method for obtaining the peripheral circulation disorder evaluation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioengineering, and particularly relates to a temperature data processing method, device and storage medium for monitoring peripheral circulation disorder. BACKGROUND

[0002] The peripheral circulation disorder evaluation methods in the prior art include a contact method and a non-contact method. The contact method has defects such as complex operation, endothelial damage or infection risk, and the non-contact method has defects such as the need for manual operation and low efficiency. Therefore, a method is needed that can overcome the invasiveness of the human body and efficiently process monitoring data to obtain the evaluation result of the peripheral circulation disorder.

[0003] Therefore, the prior art needs to be further improved. SUMMARY

[0004] In view of the defects in the prior art, the purpose of the present application is to provide a temperature data processing method, device and storage medium for monitoring peripheral circulation disorder, so as to overcome the defects of the prior art, such as complex operation, strong invasiveness and low efficiency, in the evaluation method of peripheral circulation disorder.

[0005] In a first aspect, the present application discloses a temperature data processing method for monitoring peripheral circulation disorder, which comprises the following steps:

[0006] Divide the original temperature data sequence of the target monitoring object into a plurality of sub-sequences;

[0007] For each sub-sequence, generate a sequence of grayscale image frames according to the original temperature data in the sub-sequence, calculate the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude, input the ratio into a preset regression model, and obtain a PI prediction value;

[0008] For each sub-sequence, extract the region temperature data of the target core region and the peripheral region in the original temperature data in the sub-sequence, calculate the core-peripheral temperature difference according to the extracted region temperature data, and construct a CPTD index;

[0009] Input a plurality of PI characteristic index data of a plurality of PI prediction values corresponding to the original temperature data sequence and a plurality of CPTD characteristic index data corresponding to a plurality of CPTD indexes into a preset classification model, and obtain the peripheral circulation disorder evaluation result output by the preset classification model.

[0010] Optionally, the original temperature data comprises first original temperature data corresponding to a peripheral region and second original temperature data corresponding to a core region; the peripheral region comprises a hand region and a chest region, and the core region comprises an abdomen region; the original temperature data is obtained by simultaneously shooting the peripheral region and the core region of the target monitoring object by using one infrared camera to obtain the combination of the first original temperature data and the second original temperature data.

[0011] In the step of calculating the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude, the hand region is selected as the ROI in the grayscale image frame; and in the step of extracting the region temperature data of the target core region and the peripheral region in the original temperature data in the subsequence, the chest region is selected as the peripheral region and the abdomen region is selected as the target core region.

[0012] Optionally, the step of calculating the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude comprises:

[0013] identifying the rest frames in the sequence of grayscale image frames and extracting the ROI in the rest frames;

[0014] extracting the pixel intensity time sequence of the ROI in all the rest frames respectively;

[0015] calculating the average value according to the pixel intensity time sequence of the ROI to obtain the average pixel intensity value corresponding to the ROI, and calculating the standard deviation according to the pixel intensity time sequence of the ROI to obtain the dynamic brightness fluctuation amplitude corresponding to the ROI;

[0016] determining the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude.

[0017] Optionally, the step of identifying the rest frames in the sequence of grayscale image frames and extracting the ROI in the rest frames comprises:

[0018] performing feature point extraction on the reference ROI demarcated from the first frame of the sequence of grayscale image frames to obtain at least one group of feature points;

[0019] distinguishing the rest frames and the dynamic frames according to the displacement of each feature point in each grayscale image frame;

[0020] extracting the ROI from each rest frame to obtain the ROI in each rest frame.

[0021] Optionally, the step of distinguishing the rest frames and the dynamic frames according to the displacement of each feature point in each grayscale image frame comprises:

[0022] sequentially tracking each of the feature points in a second grayscale image frame in the sequence of grayscale image frames except the first grayscale image frame, and calculating displacement of each of the tracked feature points between two adjacent grayscale image frames;

[0023] calculating displacement of each of the feature points in each of the second grayscale image frames according to the displacement of each of the feature points between two adjacent grayscale image frames;

[0024] accumulating the displacement of all the feature points in the same second grayscale image frame to calculate total motion intensity of all the feature points in each of the second grayscale image frames;

[0025] identifying a resting frame according to the total motion intensity of all the feature points in each of the second grayscale image frames.

[0026] Optionally, the method for constructing the preset regression model comprises:

[0027] obtaining a pixel intensity time sequence corresponding to each of a plurality of target sample data;

[0028] taking the pixel intensity time sequence of each of the target sample data as a sample subset, and calculating a sample ratio between an average pixel intensity value and a dynamic brightness fluctuation amplitude corresponding to each of the target sample subset;

[0029] performing linear regression fitting on the sample ratio and a reference PI value of each of the target sample subset to determine a regression coefficient;

[0030] obtaining the constructed regression model according to the regression coefficient.

[0031] Optionally, the step of inputting the ratio into a preset regression model to obtain a PI prediction value comprises:

[0032] determining a target regression model corresponding to the target monitoring object;

[0033] transmitting the ratio to the target regression model for prediction to obtain the PI prediction value output by the target regression model.

[0034] Optionally, the step of inputting a plurality of PI feature index data corresponding to a plurality of PI prediction values of the original temperature data sequence and a plurality of CPTD feature index data corresponding to a plurality of CPTD indexes into a preset classification model to obtain a peripheral circulation disorder evaluation result output by the preset classification model comprises:

[0035] extracting frequency domain features and time domain features corresponding to a plurality of PI prediction values to obtain a PI feature index set;

[0036] extracting frequency domain features and time domain features corresponding to the plurality of CPTD indexes to obtain a CPTD feature index set;

[0037] combining each PI feature index data in the PI feature index set with each CPTD feature index data in the CPTD feature index set to form a double-feature index space set;

[0038] selecting at least one PI feature index data and at least one CPTD feature index data from the double-feature index space set, and inputting the selected PI feature index data and CPTD feature index data into a preset classification model to obtain a peripheral circulation disorder evaluation classification result output by the preset classification model.

[0039] In a second aspect, the present application provides a terminal device, comprising: a processor and a memory for storing a computer program, wherein the processor is configured to invoke and run the computer program stored in the memory, and execute the steps of the temperature data processing method for monitoring peripheral circulation disorder.

[0040] In a second aspect, the present application provides a computer readable storage medium for storing a computer program, wherein the computer program enables a computer to execute the steps of the temperature data processing method for monitoring peripheral circulation disorder.

[0041] Beneficial effects:

[0042] This invention provides a temperature data processing method, device, and storage medium for monitoring peripheral circulation obstruction. The method involves dividing the original temperature data sequence into several sub-sequences; generating a grayscale image frame sequence for each sub-sequence; and calculating the ratio between the average pixel intensity value of the ROI and the dynamic brightness fluctuation amplitude in the grayscale image frame sequence to determine the PI prediction value. For each sub-sequence, regional temperature data of the target core region and peripheral region are extracted, the core-periphery temperature difference is calculated, and a CPTD index is constructed. Multiple PI feature index data corresponding to the multiple PI prediction values ​​and multiple CPTD feature index data corresponding to the multiple CPTD indices are input into a preset classification model to obtain the peripheral circulation obstruction assessment result output by the preset classification model. The temperature data processing method provided in this application processes the original temperature data of the target monitoring object to obtain the assessment result of peripheral circulation obstruction, avoiding invasive operational risks and contact infection, and is not affected by ambient light interference due to processing based on temperature data. Furthermore, this method classifies data based on multiple PI predicted values ​​corresponding to multiple PI characteristic index data and multiple CPTD indexes corresponding to multiple CPTD characteristic index data, which improves the classification accuracy. The final temperature data classification results provide a safe, stable and comprehensive monitoring method for assessing peripheral circulation disorders. Attached Figure Description

[0043] Figure 1 This is a flowchart of the steps of a temperature data processing method for monitoring peripheral circulation obstruction provided by the present invention;

[0044] Figure 2 This is a schematic diagram illustrating the principle of a specific embodiment of the temperature data processing method provided by the present invention;

[0045] Figure 3 This is a schematic diagram illustrating the principle of data result classification in the temperature data processing method provided by this invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0047] Peripheral circulation is an important component of the circulatory system, referring to the blood pumped by the heart through the aorta to various tissues and organs, and then back to the heart through arterioles, capillaries and veins. Peripheral circulation disorder refers to poor blood circulation in peripheral tissues such as limbs (especially lower limbs) and skin of the human body. Therefore, if peripheral circulation disorder is monitored, it may indicate that peripheral blood vessels cannot normally transport blood, thereby affecting the oxygen supply and metabolism of tissues and organs. Therefore, as a physiological state, peripheral circulation disorder has clinical significance.

[0048] In the neonatal population, due to the immaturity of the cardiovascular system and limited compensatory capacity, the clinical significance of peripheral circulation monitoring is more prominent. In the neonatal intensive care unit (NICU), early abnormalities in peripheral circulation may indicate severe conditions such as perinatal asphyxia, infection or congenital heart defects, and timely intervention can significantly reduce the risk of organ damage. Therefore, real-time monitoring and accurate assessment of peripheral circulation in neonates is not only crucial for clinical decision-making, but also an important strategy to improve the prognosis of high-risk neonatal populations.

[0049] In the prior art, there are generally contact monitoring and non-contact monitoring, two different evaluation methods. The common way of contact monitoring is invasive arterial monitoring. Since invasive arterial monitoring involves directly inserting a catheter into the arterial blood vessel, it can continuously measure hemodynamic parameters such as arterial blood pressure and provide arterial blood samples for blood gas analysis. Although this technology can provide highly accurate information about the systemic circulation, it also has some limitations, including the complexity of the operation and the risk of endothelial damage, thrombosis and catheter-related infections. Therefore, this method is usually only suitable for critically ill neonates with severe hemodynamic instability.

[0050] Non-contact monitoring is a non-invasive peripheral perfusion assessment technique, including: capillary refill time (CRT), transcutaneous oxygen / carbon dioxide pressure monitoring (TcPO2 / TcPCO2), perfusion index (PI), and core-peripheral temperature difference (CPTD). Although capillary refill time is easy to operate, it is highly subjective and cannot be continuously monitored. Transcutaneous oxygen / carbon dioxide pressure monitoring has the risk of thermal injury to the skin and is more complex to operate. Perfusion index (PI) is a photoplethysmography (PPG)-based indicator that detects the intensity changes of transmitted light signals by a pulse oximeter probe emitting red and infrared light through peripheral tissues (such as fingers, feet, or earlobes). The PPG signal is converted into two components: the pulsatile alternating current (AC) component reflecting arterial blood flow during the cardiac cycle, and the non-pulsatile direct current (DC) component representing steady-state tissue absorption and venous blood volume. By calculating the ratio of these two components, the perfusion index can quantitatively assess perfusion status and is widely used in the monitoring of non-critical neonates due to its ease of use. However, as a contact method, perfusion index monitoring can cause skin irritation or discomfort, especially when used for a long time, and during infectious disease outbreaks, the use of such contact equipment may pose a risk of cross-infection. Core-peripheral temperature difference CPTD is similar to peripheral circulation, and in clinical practice, an infrared thermometer is usually used to assess the temperature difference between the core and peripheral areas. However, this method is less efficient and requires manual operation by nursing staff.

[0051] In recent years, camera-based photoplethysmography has shown significant potential. Remote photoplethysmography (rPPG) is a non-contact technique for measuring physiological indicators such as heart rate, respiration rate, and blood perfusion. RGB cameras capture skin reflectance and color changes caused by the periodic changes in blood volume driven by the cardiac cycle. Hemoglobin in blood has variable light absorption properties depending on blood volume, resulting in slight fluctuations in reflected light intensity. By extracting the ratio of AC to DC components in the rPPG signal, skin perfusion can be quantified. Compared to traditional contact PPG, rPPG has the advantage of being completely non-contact and can monitor multiple skin sites. However, rPPG also has significant limitations, particularly sensitivity to environmental light fluctuations, which can significantly affect signal quality.

[0052] To overcome the above problems, the present application provides a temperature data processing method, device and storage medium for peripheral circulation disorder monitoring, by acquiring the original temperature data of the target monitoring object, based on the original temperature data, the PI prediction value is calculated and the CPTD index is calculated, and the dual-index fusion classification model is constructed combined with the PI prediction value and the CPTD index, and the final data processing result is obtained based on the classification model, that is, the evaluation result of peripheral circulation disorder is obtained. The method provided by the present application processes the temperature data of peripheral circulation disorder monitoring to obtain the evaluation result of peripheral circulation disorder. Since the measurement of temperature data is non-contact, it avoids the invasive damage to the human body and there is no risk of infection, so it is safe, and the temperature monitoring data is not disturbed by environmental light, and the evaluation result is stable. Only the temperature data is processed, the efficiency is high, so the method of the present application provides a safe, stable and more efficient data processing method.

[0053] The temperature data processing method, device and storage medium for peripheral circulation disorder monitoring provided by the present application will be described in further detail below in conjunction with the drawings of the specification.

[0054] In a first aspect, the present application provides a temperature data processing method for peripheral circulation disorder monitoring, as shown in Figure 1 The temperature data processing method comprises:

[0055] Step S1, dividing the original temperature data sequence of the target monitoring object into several subsequences.

[0056] Since the original temperature data of the target monitoring object obtained by monitoring is temperature data information for a period of time, the original temperature data sequence is preprocessed and divided into multiple subsequences according to a fixed window. For example, the original temperature data sequence: [T1, T2, T3, …, Tn] is divided into several subsequences: [T1, T2, T3], [T2, T3, T4], [T3, T4, T5], …, [Tn-2, Tn-1, Tn]1000 The original temperature data sequence is divided into a plurality of subsequences with a time period of 30 seconds, for example, subsequence 1: [T1, T2, T3, …, T30], subsequence 2: [T31, T32, T33, …, T60], …, subsequence 34: [T961, T962, T963, …, T990]. 30 31 32 33 60 991 992 993 1000

[0057] Specifically, the original temperature data includes first original temperature data corresponding to a peripheral region and second original temperature data corresponding to a core region; the peripheral region includes a hand region and a chest region, and the core region includes an abdominal region; the original temperature data is obtained by simultaneously photographing the peripheral region and the core region of a target monitoring object by using one infrared camera to obtain the first original temperature data and the second original temperature data.

[0058] In an implementation manner, the original temperature data is obtained by photographing the target monitoring object by using one infrared camera to obtain the original temperature data. Since the thermal images of the peripheral region and the core region are simultaneously photographed by using one infrared camera, and the processing is performed based on the same original temperature data, the asynchronous error of multiple devices can be eliminated in the temperature data processing, and the accuracy of the data processing is improved.

[0059] In step S2, for each subsequence, a grayscale image frame sequence is generated according to the original temperature data in the subsequence, a ratio between the average pixel intensity value of the ROI in the grayscale image frame sequence and the dynamic brightness fluctuation amplitude is calculated, the ratio is input into a preset regression model, and a PI prediction value is obtained.

[0060] In this step, for each subsequence divided in step S1, a grayscale image frame sequence is generated according to the temperature data sequence, a ratio between the average pixel intensity value of the ROI in the grayscale image frame sequence and the dynamic brightness fluctuation amplitude is calculated, and the PI prediction value is determined based on the ratio.

[0061] ​​​​​​​​​In one embodiment, since the sequence of grayscale image frames can contain motion frames, in order to reduce the influence of motion signals on physiological signals, the step first identifies the resting frames in the sequence of grayscale image frames, and then determines the PI prediction value by calculating the ratio between the average pixel intensity value of the ROI of the resting frames and the dynamic brightness waveguide amplitude. The method for identifying the resting frames is as follows: according to the reference ROI marked in the grayscale image frame of the first frame, based on the motion intensity characteristics of the selected feature points in the reference ROI (i.e. the displacement of the selected feature points in each grayscale image frame), the ROIs in the resting frames contained in the sequence of grayscale image frames are identified.

[0062] Specifically, the step first generates a sequence of grayscale image frames according to the original temperature data in the subsequence. Since the original temperature data of the target monitoring object collected by the infrared camera is in the form of a matrix composed of temperature values of each pixel, each frame of temperature data in the original temperature data can be converted into a visual grayscale image frame, and in this step, the temperature range needs to be mapped to the grayscale value. Generally, the temperature value is a continuous numerical value, for example: data between -20 degrees and 100 degrees, while each pixel in the grayscale image frame needs to be represented by an integer from 0 (black) to 255 (white), so in the specific implementation, the lowest temperature value is mapped to black, corresponding to 0, and the highest temperature value is mapped to white, corresponding to a grayscale value of 255.

[0063] In order to ensure that the temperature value after grayscale mapping is scaled to the range of 0 to 255 grayscale values, it is necessary to perform standardization processing on each frame of temperature data. Specifically, the standardization formula is generally to subtract the minimum temperature value from each temperature value, divide by the difference between the maximum temperature value and the minimum temperature value, and finally multiply by 255 to obtain the grayscale value of each pixel. In order to suppress background noise and enhance the contrast of high temperature regions in the image, a threshold segmentation strategy can also be used in the specific implementation. The threshold segmentation strategy sets a grayscale threshold, and the pixels below the threshold (which usually correspond to the background or low temperature region) are excluded, and the remaining high grayscale values represent the high temperature region, thereby effectively removing noise and improving the contrast of important regions.

[0064] If the thermal infrared camera only captures the temperature data of the target monitoring object at one time, then the single frame of temperature data, which is usually a two-dimensional number, corresponds to the temperature distribution in space, and after grayscale mapping processing, one grayscale image frame is generated. Combined with the above-mentioned method for identifying the resting frames, the sequence of grayscale image frames can be divided into two parts: the resting frames and the motion frames. Figure 2As shown, if the thermal infrared camera captures temperature data in a continuous time sequence, i.e. multiple frames of temperature data, each time corresponding to a two-dimensional temperature array, then each frame of temperature data will generate a corresponding grayscale image frame through grayscale mapping, and finally multiple grayscale image frames will be obtained. In this step, because the temperature change of the target monitoring object is to be analyzed, the thermal infrared camera captures temperature values at multiple times, so that these grayscale image frames are concatenated into a video sequence according to the time sequence, and a grayscale image frame sequence composed of continuous grayscale image frames is generated.

[0065] Further, the step of identifying the rest frame in the grayscale image frame sequence specifically comprises:

[0066] Step S21, feature point extraction is performed on the reference ROI demarcated from the first frame of the grayscale image frame sequence, and at least one group of feature points is obtained.

[0067] When the grayscale image frame sequence is generated according to the original temperature data in the subsequence, the first frame of the grayscale image frame sequence is subjected to ROI (Region of Interest) demarcation. The demarcation can be manual demarcation or automatic recognition demarcation using an algorithm. In the method disclosed in this embodiment, the ROI selected in this step is a hand region.

[0068] When the reference ROI is demarcated, feature point extraction is performed on the reference ROI, and the pixel points that are easily recognized and clear in the reference ROI are obtained, and the recognized pixel points are taken as feature points. Specifically, the method of recognizing multiple feature points can use Shi-Tomasi corner detector for extraction, or Harris corner detector or FAST corner detector, etc.

[0069] Step S22, each of the feature points is tracked in the second grayscale image frame of the grayscale image frame sequence other than the first grayscale image frame, and the displacement of each of the tracked feature points between two adjacent grayscale image frames is calculated.

[0070] When each of the feature points in the reference ROI is extracted, feature point tracking is performed on the subsequent frames of the first frame in sequence by using a feature point motion tracking algorithm, so that the positions of each of the feature points in each of the subsequent second grayscale image frames are obtained. In one implementation, the position information of each of the feature points in each of the second grayscale image frames can be tracked by using a pyramid Lucas-Kanade (LK) optical flow method.

[0071] After the positions of each of the feature points in different second grayscale image frames are obtained by tracking, the displacement of each of the feature points in two adjacent continuous second grayscale image frames can be calculated.

[0072] Specifically, since the Euclidean displacement is an important indicator for measuring the motion intensity, the Euclidean displacement of each feature point in the adjacent and continuous two second gray image frames can be calculated, and the Euclidean displacement is quantified as a measure of the motion intensity, and the motion degree of the feature point between the two continuous gray image frames is reflected by the Euclidean displacement of each feature point.

[0073] Step S23, according to the displacement of each feature point between the adjacent two gray image frames, the total displacement of each feature point in each second gray image frame is calculated.

[0074] Based on the displacement of each feature point between each adjacent gray image frame, the total displacement of each feature point in each second gray image frame can be calculated. For example, if there are 30 frames of gray image frames, when the displacement of each feature point between the first frame of gray image frame and the first second gray image frame, the displacement between the first second gray image frame and the adjacent second gray image frame, the displacement between the second second gray image frame and the adjacent third second image frame, …, the displacement between the twenty-eighth second gray image frame and the adjacent twenty-ninth second gray image frame is calculated, the displacement of each feature point between the current second gray image frame and each second gray image frame is superimposed, and the total displacement of each feature point in the corresponding second gray image frame is obtained.

[0075] Step S24, the total displacement of all the feature points in the same second gray image frame is accumulated, and the total motion intensity of all the feature points in each second gray image frame is calculated.

[0076] The motion intensity of each feature point in the same second gray image frame is accumulated, and the total motion intensity of each feature point in each second gray image frame is used to evaluate the motion change of each frame. That is, the total displacement of each feature point in the same gray image frame is accumulated, and the total motion intensity of each second gray image frame is obtained.

[0077] Step S25, according to the total motion intensity of each feature point in each second gray image frame, the rest frame is identified.

[0078] In the classification process, when the total motion intensity of a feature point in a certain gray image frame is lower than a preset threshold (the threshold is set to 200 pixels), it is considered that the gray image frame is a resting frame, that is, there is no significant motion. At this time, the gray image frame will be classified as a resting frame, which is suitable for subsequent physiological signal extraction. On the contrary, if the total motion intensity of a certain gray image frame exceeds the threshold, the gray image frame will be marked as a motion frame, indicating that the gray image frame contains significant motion information and cannot be used for signal extraction. Through this method, the resting frame and the motion frame in the video can be effectively distinguished, the interference of motion on signal extraction is reduced, and the accuracy and reliability of data processing are improved.

[0079] In addition, according to the calculation of the change amount of the total motion intensity corresponding to each second gray image frame with respect to time, the change amount of each total motion intensity (that is, the total displacement superposition) with respect to time is calculated, and whether the velocity of the feature point in each gray image frame exceeds a preset change threshold is analyzed. If it is lower than the preset change threshold, the second gray image frame is judged as a resting frame, otherwise it is a motion frame.

[0080] When the resting frames contained in the sequence of gray image frames are identified, the identified resting frames are sorted in time as a sequence of resting frames. After determining the resting frames in the sequence of resting frames, the ROIs in each resting frame are extracted. Since the reference ROI has been calibrated in the first frame, the ROI at the corresponding position can be extracted from each resting frame in turn according to the reference ROI calibrated in the first frame. Further, the way of extracting the ROI in each resting frame can also be selected from manual labeling, threshold-based segmentation labeling, edge detection and other ways.

[0081] In order to obtain the PI prediction value, the ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude of the ROIs in all the resting frames in the subsequence is calculated in this step, and the calculated ratio value is input into a preset regression model to obtain the PI prediction value output by the regression model. The ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude reflects the dynamic characteristics of the pixel points in the ROI with time.

[0082] Specifically, to calculate the ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude, it is necessary to first extract the pixel intensity time sequence corresponding to the ROIs in all the resting frames in the subsequence, and calculate the average pixel intensity value and the standard deviation of the pixel intensity time sequence corresponding to the ROIs respectively, to obtain the ratio of the average pixel intensity value and the standard deviation. The pixel intensity time sequence is the sequence data of the gray value of a single pixel point in the subsequence of all resting frames with time (frame sequence).

[0083] The average pixel intensity value (DC component, representing static brightness information) and the standard deviation (AC component, quantifying dynamic brightness fluctuation amplitude) are calculated. The AC / DC ratio reflects the dynamic characteristics of the relative skin temperature change in the ROI.

[0084] In detail, after the ROI in the rest frame is extracted, the step of calculating the ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude of the ROI in the sequence of grayscale image frames includes:

[0085] Step S26, the pixel intensity time sequence corresponding to the ROI in all rest frames is extracted.

[0086] The pixel intensity values of the ROI in all rest frames in the sub-sequence are extracted to obtain the pixel intensity time sequence of the ROI in all rest frames in the sub-sequence. In this step, in order to deal with the time discontinuity in the sequence of rest frames caused by removing the motion frames, the missing values are filled by cubic spline interpolation to ensure the integrity of the pixel intensity time sequence.

[0087] Step S27, the average value of the pixel intensity time sequence of the ROI is calculated respectively to obtain the average pixel intensity value corresponding to the ROI, and the standard deviation of the pixel intensity time sequence of the ROI is calculated to obtain the dynamic brightness fluctuation amplitude corresponding to each ROI.

[0088] The average value and the standard deviation of the pixel intensity are calculated respectively for the pixel intensity time sequence, wherein the average value of the pixel intensity is the DC component, which can represent the static brightness information, and the standard deviation is the AC component, which can quantify the dynamic brightness fluctuation amplitude.

[0089] Step S28, the ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude of the ROI in all rest frames is determined.

[0090] The ratio between the standard deviation and the average value is calculated to obtain the ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude. The ratio between the standard deviation and the average value, i.e. the ratio between AC and DC, reflects the dynamic characteristics of the relative skin temperature change in the ROI.

[0091] In this step, in order to obtain a more accurate PI prediction value, the ratio is input into a preset regression model to obtain the PI prediction value by using the preset regression model. In specific implementation, the preset regression model is a linear regression model. Before use, the linear regression model needs to be trained to obtain a trained regression model, and then the ratio is input into the trained regression model to obtain the PI prediction value output by the regression model.

[0092] In detail, the linear regression model comprises an input layer, a regression layer and an output layer. The step of training the linear regression model to obtain the trained regression model comprises: transmitting the ratio data of the target sample and the reference PI data to the regression layer through the input layer; the regression layer adopts linear regression to fit the relationship between the sample ratio data and the reference PI data to obtain initial regression coefficients; the output layer outputs the PI prediction data according to the initial regression coefficients; the initial regression coefficients are adjusted according to the error between the PI prediction data and the actual PI data, and the steps of transmitting the sample ratio data and the reference PI data to the regression layer through the input layer, obtaining the initial regression coefficients by the regression layer, and adjusting the initial regression coefficients are repeated until the final regression coefficients are determined, and the trained regression model is obtained. The reference PI can be a perfusion index measured by a monitor, which is measured by photoplethysmography (PPG). The PI is the ratio of the alternating current component (AC, reflecting arterial pulsation) to the direct current component (DC, reflecting tissue static absorption) in the pulse wave signal.

[0093] Specifically, the pre-set regression model in this step is divided into two different types, one is a general regression model, and the other is a personalized regression model.

[0094] The construction of the general regression model strictly follows the principle of subject independence, and the training set and the test set are assigned to independent sample populations. The general regression model uses the least squares method to construct a linear regression model to evaluate the general prediction ability of the model among individuals. Its structure usually includes three main components: input layer, regression layer and output layer. During training, the input layer is used to receive the target sample ratio data and the reference PI data. The target sample ratio data and the sample reference PI data are used to train the general regression model, thereby helping to establish the mapping relationship between the target sample ratio data and the sample reference PI data. In the regression layer, linear regression method is used to process the input data, and least squares method is used to fit the model to calculate the relationship between the target sample ratio data and the sample reference PI data. The goal of this layer is to learn a globally applicable regression coefficient, which can remain consistent among multiple individuals, thereby providing unified prediction ability.

[0095] In specific application, the output layer obtains the trained general regression model according to the regression coefficients calculated by the regression layer. The step of inputting the ratio into the pre-set regression model to obtain the PI prediction value comprises:

[0096] The ratio is transmitted to the regression model for prediction to obtain the PI prediction value output by the regression model. Since the regression model comprises an input layer, a regression layer and an output layer, the ratio is transmitted to the regression layer through the input layer, the regression layer predicts the PI according to the determined regression coefficients, and the PI prediction value is output from the output layer.

[0097] The regression model described above can be trained based on general sample data, and thus the trained regression model is a general regression model, which is applicable to population-level prediction.

[0098] To obtain more accurate PI prediction values for different monitoring targets, a personalized regression model can also be trained. The personalized regression model is trained based on time series data of target individuals to obtain personalized PI prediction values. The personalized regression model constructs time series features of target individual data, divides the data set into multiple target sample subsets corresponding to multiple target individuals, for example, each target sample subset contains 10 consecutive data samples and is segmented at a fixed step. Linear regression fitting is performed on each target sample subset to construct a personalized regression model. The personalized regression model is optimized for target individual sample data, aiming to consider the physiological differences of each target sample individual to provide more accurate predictions. The structure of the personalized regression model also includes an input layer, a regression layer, and an output layer.

[0099] The training method of the personalized regression model is similar to that of the general regression model, and still includes target sample ratios and reference PI values corresponding to target individual sample subsets from the TIR camera. The difference is that the personalized regression model focuses on the data of each target individual, and thus each input is a pixel intensity time series of a target individual sample subset. In the regression layer, local linear regression is used to enable the personalized regression model to learn individualized regression coefficients based on the local features of each target sample subset. The output layer outputs the PI prediction value for the target individual based on the calculation results of the regression layer. Since the personalized regression model is trained based on target sample subsets corresponding to target individuals, it can provide personalized PI prediction values.

[0100] Specifically, the construction method of the personalized regression model is as follows: obtaining pixel intensity time series corresponding to multiple target sample data; taking the pixel intensity time series of each target sample data as a target sample subset, and calculating the sample ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude corresponding to each target sample subset; performing linear regression fitting on the sample ratio and the reference PI value of each target sample subset to determine the regression coefficient; and obtaining the constructed regression model according to the regression coefficient.

[0101] In specific implementation, different regression models for different types of target objects can also be trained respectively, and the construction method is as follows: a plurality of target data sets are obtained, each target data set corresponding to an object category; a regression model for an object category is constructed based on each target data set to obtain a regression model set. The object category can be different categories of people, or a single target individual identified based on an ID or a unique label.

[0102] In the step of constructing a regression model for an object category based on each target data set to obtain a regression model set, first, the pixel intensity time series of each target sample data is taken as a target sample subset, and the sample ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude corresponding to each target sample subset is calculated. The sample ratio and the reference PI value of each target sample subset are linearly regressed and fitted to determine the personalized regression coefficient; and the personalized regression model is obtained according to the personalized regression coefficient. Since each target data set corresponds to a regression model, a plurality of target data sets correspond to a regression model set.

[0103] Further, when the regression model is constructed, the step of inputting the ratio into the preset regression model to obtain the PI prediction value includes:

[0104] Step S29, obtaining a target regression model of the target monitoring object.

[0105] When the personalized regression model is used to predict the PI value of a single target individual, since the regression model for different object categories or the regression model for different targets is trained, in use, the regression model that is most suitable for the single target monitoring object needs to be matched from the regression model set obtained by training. When the most suitable regression model is matched, the object category used in the training of the regression model can be matched, and if the object categories are the same, the matching is successful. For example, the types of target monitoring objects include the elderly population and the newborn population, and therefore, in order to obtain a regression model that is most suitable for the target monitoring object based on object category matching, the target object category of the target monitoring object needs to be obtained first.

[0106] The target object category of the target monitoring object can be determined from the basic information of the target monitoring object, or the target object category information input by the user can be received to obtain the target object category corresponding to the target monitoring object.

[0107] In the training of the personalized regression model, a sample subset can also be made using the monitoring data of the target monitoring object, and a personalized regression model is trained for the target regression model training. When the regression model is matched, the regression model corresponding to the target monitoring object can be matched according to the unique identifier corresponding to the training sample data, such as the ID or other unique identifier information.

[0108] Step S210, transmitting the ratio to the target regression model for prediction to obtain the PI prediction value output by the regression model.

[0109] When the regression model corresponding to the target monitoring object is obtained, the ratio is input into the regression model to obtain the PI prediction value output by the regression model.

[0110] Further, in order to improve the accuracy of the PI prediction value, this step also includes using the Pearson correlation coefficient (R) to analyze the linear correlation between the ratio and the reference PI measured by the monitor. The MAE (Mean Absolute Error) is used as a quantitative indicator to quantify the prediction accuracy of the PI by calculating the average absolute deviation between the PI prediction value and the reference PI value.

[0111] Step S3, for each sub-sequence, extracting the region temperature data of the target core region and the peripheral region in the original temperature data in the sub-sequence, calculating the core-peripheral temperature difference according to the extracted region temperature data, and constructing the CPTD index.

[0112] This step extracts the temperature data of the core region and the temperature data of the peripheral region for each sub-sequence, so as to calculate the temperature difference between the core region and the peripheral region, that is, the core-peripheral temperature difference. Since each sub-sequence contains a plurality of original temperature data, a plurality of region temperature data corresponding to the target core region and the peripheral region can be extracted from the original temperature data. In specific implementation, the average temperature value of the target core region and the average temperature value of the peripheral region in the sub-sequence can be selected to calculate the temperature difference between the core region and the peripheral region. In specific implementation, the temperature of the target core region and the temperature of the peripheral region corresponding to the same time point can also be selected to calculate the core-peripheral temperature difference.

[0113] In combination Figure 2 As shown, by extracting the temperature data of the core region and the peripheral region, the temperature difference between the temperature data of the core region and the peripheral region is calculated to construct the CPTD index. The core region is the abdominal region, and the peripheral region is the chest region.

[0114] Specifically, in the step of calculating the ratio between the average pixel intensity value of the ROI in all resting frames and the dynamic brightness fluctuation amplitude, and determining the PI prediction value according to the ratio, the hand region is selected as the ROI; in the step of extracting the temperature data of the target core region and the peripheral region, calculating the core-peripheral temperature difference, and constructing the CPTD index, the chest region is selected as the peripheral region and the abdominal region is selected as the core region.

[0115] In order to reduce the influence of abnormal temperature on CPTD calculation, a three-step data quality control mechanism is established: first, remove the temperature points outside the human body bearing range, second, remove the sudden data whose temperature difference between adjacent measurement points exceeds the preset threshold, and third, identify the abnormal values deviating from the standard value.

[0116] In detail, first, the data points whose temperature values exceed the human body bearing range are excluded according to the physiological threshold, for example, the data points in the range of 32-42℃. Second, the data points whose temperature difference between adjacent measurement points exceeds a certain preset threshold are filtered out, for example, the sudden data whose temperature difference exceeds 2℃. Third, the time domain sliding window method (for example, the window width is 30 frames) is used, and the observation values deviating from the window mean value by more than three times the standard deviation are classified as abnormal. For the identified abnormal data, three times spline interpolation is used for data reconstruction to ensure the time continuity of the temperature sequence.

[0117] Step S4, inputting the multiple PI feature index data corresponding to the multiple PI prediction values of the original temperature data sequence and the multiple CPTD feature index data corresponding to the multiple CPTD indexes into a preset classification model to obtain a peripheral circulation disorder evaluation result output by the preset classification model.

[0118] Since each subsequence obtains a PI prediction value and a CPTD index, the original temperature data sequence can obtain multiple PI prediction values and multiple CPTD indexes. In this step, the multiple PI feature index data corresponding to the multiple PI prediction values and the multiple CPTD feature index data corresponding to the multiple CPTD indexes are combined to determine the processing result of the temperature monitoring data, that is, the peripheral circulation disorder classification result corresponding to the target monitoring object.

[0119] Further, in combination with Figure 3 As shown in FIG. 8, the step of inputting the multiple PI feature index data corresponding to the multiple PI prediction values of the original temperature data sequence and the multiple CPTD feature index data corresponding to the multiple CPTD indexes into a preset classification model to obtain a peripheral circulation disorder evaluation result output by the preset classification model includes:

[0120] Step S41, extracting the frequency domain features and time domain features corresponding to the multiple PI prediction values to obtain a PI feature index set.

[0121] Step S42, extract the frequency domain features and time domain features corresponding to the plurality of CPTD indexes to obtain a CPTD index feature index set.

[0122] In the feature selection, 14 statistical features are extracted from the PI feature indexes and the CPTD feature indexes, respectively. These features represent the distribution characteristics and fluctuation structure of the indexes, including time domain features and frequency domain features, thereby forming a multi-dimensional original feature space. The feature indexes are listed in Table 1.

[0123] Table 1

[0124]

[0125] In order to obtain the independent discrimination ability of different feature indexes and the accuracy of the double-index method analysis, step S41 extracts 14-dimensional features of the PI index to construct a PI feature index set, and step S42 extracts 14-dimensional features of the CPTD index to construct a CPTD feature index set of the CPTD index. Therefore, two 14-dimensional feature index sets are constructed.

[0126] Step S43, combine each PI feature index in the PI feature index set with each CPTD feature index in the CPTD feature index set to form a double-feature index space set.

[0127] After the PI feature index set and the CPTD feature index set are constructed, this step constructs a double-feature index space set combining PI and CPTD based on the PI feature index set and the CPTD feature index set, that is, all feature indexes of CPTD and PI indexes are connected to form a double-feature index space (28 dimensions). Specifically, the PI feature index data and the CPTD feature index data are directly merged into the same set space, thereby obtaining a 28-dimensional double-feature index space set.

[0128] Step S44, select at least one PI feature index data and at least one CPTD feature index data from the double-feature index space set, and input the selected PI feature indexes and CPTD feature indexes into a preset classification model to obtain a peripheral circulation disorder evaluation classification result output by the preset classification model.

[0129] On this basis, the importance of the double feature indicators in the double feature indicator space set is evaluated, and finally a predetermined number of double feature indicators are selected, for example, 6 most discriminative feature indicators are selected. In this step, at least one PI feature indicator and at least one CPTD feature indicator are contained in each of the selected most discriminative feature indicators, so that the feature indicators input into the preset classification model contain data of both PI and CPTD aspects.

[0130] In specific implementation, the way of selecting the most discriminative feature indicator data is automatically defined by machine learning, the importance of each feature indicator data to the classification evaluation result is obtained, and then the most important feature indicator is obtained. The machine learning model can be a decision tree model or a linear regression model.

[0131] After obtaining the target feature indicator space set for each subsequence, the target feature indicators in the target feature indicator space set are input into the preset classification model, and the classification model classifies according to the input target feature indicators to obtain the corresponding peripheral circulation disorder classification result. That is, the double feature space set composed of CPTD+PI is input into the classification model, and then the classification result output by the classification model is obtained.

[0132] In specific implementation, the preset classification model can be any one of K-nearest neighbor (KNN), support vector machine (SVM), random forest (RF) and logistic regression (LR). The four mainstream supervised classification models cover different discriminant mechanisms. All classification models are embedded with a standardizer StandardScaler. When the classification model is evaluated, 5-fold stratified cross-validation is adopted to ensure consistent distribution of each class in each fold, thereby enhancing the robustness and fairness of performance evaluation. In each fold of cross-validation, feature selection is only performed on the training target sample subset, and the optimal feature set of CPTD, PI and CPTD+PI is constructed.

[0133] Before using the classification model, the classification model can be trained using the training subset, and the training result is evaluated on the corresponding verification subset, so as to obtain the trained classification model. In order to comprehensively evaluate the actual performance of the classification model in the multi-class task, the present application selects four widely used evaluation indicators: accuracy, recall, precision and F1 score. Accuracy is used to measure the overall correctness of the model classification as a basic performance indicator. Recall reflects the ability of the model to identify positive samples, indicating the coverage of the model for key categories. Precision is used to evaluate the reliability of positive prediction, indicating the proportion of true positive samples in the samples predicted as positive. F1 score, as the harmonic mean of precision and recall, is a key indicator for balancing model performance, especially suitable for scenarios with uneven class distribution or low tolerance for misclassification.

[0134] The processing method provided by the application overcomes the limitations of traditional contact PPG by quantifying PI through the normalized time-varying characteristics (AC / DC ratio) of the DC component in the resting frame converted from the original temperature data, and also estimates CPTD by using the spatial distribution characteristics of the temperature data, thereby realizing the synchronous acquisition of the spatial and temporal domain indicators by a single device, and finally establishing a peripheral circulation disorder evaluation model based on temperature processing by combining the two.

[0135] The verification results show that there is a significant positive correlation between the AC / DC ratio and the reference PI (R = 0.82). The personalized AC / DC ratio calibration achieves a mean absolute error (MAE) of 0.23%, which is better than the general regression model (MAE = 0.33%). Correlation analysis shows that there is a significant negative correlation between the AC / DC ratio and the CPTD, which enhances its clinical interpretability. In 35 newborns (21 cases of normal circulation and 14 cases of abnormal circulation), the accuracy of the two-index classification model reaches 88.57%, which is significantly better than the best single-index model (the accuracy of CPTD and PI alone is 80.00% and 77.14%, respectively), so the two-index model (PI+CPTD) significantly improves the classification performance compared with the single-index (PI and CPTD) model.

[0136] The temperature data processing method for monitoring peripheral circulation disorders provided by the application will be further described below. Figure 2 And Figure 3 The temperature data processing method for monitoring peripheral circulation disorders provided by the application will be further described below.

[0137] The temperature data for monitoring peripheral circulation disorders of the human body provided by the application is the temperature data obtained by using a thermal infrared camera to capture the peripheral and core regions of the human body. The method processes the temperature data of the peripheral and core regions of the human body to determine whether the peripheral circulation disorder of the human body is abnormal.

[0138] In specific implementation, the following steps are included:

[0139] 1. Thermal infrared video input. In order to obtain the temperature data for monitoring peripheral circulation disorders of the target monitoring object, a thermal infrared camera (TIR camera) is used to capture the thermal infrared video of the target monitoring object. It generates dynamic images with temperature distribution as the core information by continuously capturing the dynamic changes of the thermal radiation of the target monitoring object surface.

[0140] 2. Skin region of interest selection and tracking. Convert raw temperature data contained in thermal infrared videos into grayscale image frames, and segment the region of interest from the grayscale image frames. Calibrate the ROI in the first frame image, and identify feature points according to the calibrated ROI in the first frame, and track the positions of the feature points in other image frames.

[0141] 3. According to the positions of the feature points in different feature frames, calculate the total displacement of each feature point, thereby classifying the motion frames and the resting frames.

[0142] 4. Extract the pixel intensity time series from the ROI of the resting frame.

[0143] 5. Calculate the AC / DC ratio according to the extracted pixel intensity time series.

[0144] 6. Input the AC / DC ratio and the reference PI into the pre-established regression model to obtain the PI prediction value output by the regression model.

[0145] While calculating the PI prediction value from the raw temperature data, the CPTD index is also obtained by processing the raw temperature data. The CPTD index is the temperature difference between the core and peripheral regions calculated based on the temperature of the core region and the temperature of the peripheral region, respectively.

[0146] In combination Figure 3 When the PI prediction value and the CPTD index are calculated respectively, the PI feature index set, the CPTD feature index set and the double feature index space set are established respectively, and the index features in the double feature index space set are input into the pre-trained classification model to obtain the classification result output by the classification model, which is the abnormal cycle and the normal cycle.

[0147] The present application proposes a non-contact monitoring scheme based on a thermal infrared (TIR) camera to solve the problems of invasive operation risk (such as arterial catheterization), contact infection risk (such as long-term wearing of pulse oximeter) and environmental light interference (such as dependence on visible light by RGB camera) in existing neonatal peripheral circulation monitoring technology. By innovatively quantifying the direct current normalized time-varying features (AC / DC ratio) and CPTD of thermal infrared signals, a double-index fusion classification model is constructed to realize non-invasive assessment of human peripheral circulation disorders. Compared with traditional technology, the present application uses a single TIR camera to simultaneously obtain PI prediction value and CPTD index, eliminates multi-device data asynchronous error, verifies the high accuracy of PI calibration based on the PI personalized model obtained by the TIR camera, and improves the prediction accuracy through double-index joint classification, providing a safer, more stable and comprehensive circulation monitoring method.

[0148] In a second aspect, the present application provides a terminal device, comprising: a processor and a memory, the memory being configured to store a computer program, and the processor being configured to invoke and run the computer program stored in the memory to perform the steps of the temperature data processing method for peripheral circulation disorder monitoring.

[0149] In a third aspect, the present application provides a computer readable storage medium, configured to store a computer program, and the computer program is configured to make a computer perform the steps of the temperature data processing method for peripheral circulation disorder monitoring.

[0150] The present application provides a temperature data processing method, device and storage medium for peripheral circulation disorder monitoring. The method comprises the following steps: performing gray scale mapping processing on original temperature data of a target monitoring object to generate a gray scale image frame, and extracting an ROI in a resting frame based on the gray scale image frame; calculating a ratio of AC and DC in the ROI in the resting frame, and inputting the ratio and a reference PI value into a preset regression model to obtain a PI prediction value; extracting time sequence temperature data of a target core region and a peripheral region in the original temperature data, calculating a core-peripheral temperature difference, and constructing a CPTD index; and inputting a plurality of characteristic index data corresponding to the PI prediction value and the CPTD index into a preset classification model to obtain a peripheral circulation disorder evaluation result output by the classification model. The temperature data processing method provided by the present application processes original temperature data of a target monitoring object to obtain a peripheral circulation disorder evaluation result, avoids the risk of invasive operation and contact infection, and processes the temperature data without being affected by environmental light. The present application provides a safe, stable and comprehensive method for obtaining a peripheral circulation disorder evaluation result.

[0151] Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. The present application is intended to cover any variations, uses or adaptive changes of the present application following the general principles thereof and including those expressly stated or implied herein.

[0152] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are contained in at least one embodiment or example of the present application. In the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or N embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0153] It can be understood that the above embodiments are exemplary and cannot be understood as a limitation of the present application, and those of ordinary skill in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A temperature data processing method for peripheral circulatory disorder monitoring, characterized by, The method comprises the following steps: dividing a raw temperature data sequence of a target monitoring object into a plurality of subsequences; for each subsequence, generating a sequence of grayscale image frames from the raw temperature data in the subsequence, calculating the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude, inputting the ratio into a preset regression model to obtain a PI prediction value; for each subsequence, extracting the regional temperature data of the target core region and the peripheral region in the raw temperature data in the subsequence, calculating the core-peripheral temperature difference according to the extracted regional temperature data to construct a CPTD index; inputting a plurality of PI characteristic index data corresponding to a plurality of PI prediction values of the raw temperature data sequence and a plurality of CPTD characteristic index data corresponding to a plurality of CPTD indexes into a preset classification model to obtain a peripheral circulation disorder evaluation result output by the preset classification model; the step of calculating the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude comprises: identifying the resting frames in the sequence of grayscale image frames and extracting the ROI in the resting frames; extracting the pixel intensity time sequence of the ROI in all resting frames respectively; calculating the average value according to the pixel intensity time sequence of the ROI to obtain the average pixel intensity value corresponding to the ROI, and calculating the standard deviation according to the pixel intensity time sequence of the ROI to obtain the dynamic brightness fluctuation amplitude corresponding to the ROI; determining the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude; the step of inputting a plurality of PI characteristic index data corresponding to a plurality of PI prediction values of the raw temperature data sequence and a plurality of CPTD characteristic index data corresponding to a plurality of CPTD indexes into a preset classification model to obtain a peripheral circulation disorder evaluation result output by the preset classification model comprises: extracting the frequency domain features and time domain features corresponding to a plurality of PI prediction values to obtain a PI characteristic index set; extracting the frequency domain features and time domain features corresponding to a plurality of CPTD indexes to obtain a CPTD characteristic index set; combining each PI characteristic index data in the PI characteristic index set with each CPTD characteristic index data in the CPTD characteristic index set to form a double-feature index space set; selecting at least one PI characteristic index data and at least one CPTD characteristic index data from the double-feature index space set, and inputting the selected PI characteristic index data and CPTD characteristic index data into a preset classification model to obtain a peripheral circulation disorder evaluation classification result output by the preset classification model.

2. The temperature data processing method for peripheral circulatory disorder monitoring according to claim 1, characterized in that, The raw temperature data includes first raw temperature data corresponding to a peripheral region and second raw temperature data corresponding to a core region; the peripheral region includes a hand region and a chest region, and the core region includes an abdominal region; the raw temperature data is obtained by simultaneously shooting the peripheral region and the core region of the target monitoring object with one infrared camera, and the first raw temperature data and the second raw temperature data are combined to form; In the step of calculating the ratio between the average pixel intensity value of the ROI in the sequence of grayscale image frames and the dynamic brightness fluctuation amplitude, a hand region is selected as the ROI in the grayscale image frame; in the step of extracting the region temperature data of the target core region and the peripheral region in the original temperature data in the sub-sequence, the peripheral region is selected as a chest region and the target core region is selected as an abdominal region.

3. The temperature data processing method for peripheral circulatory disorder monitoring according to claim 2, characterized in that, The step of identifying the rest frames in the sequence of grayscale image frames and extracting the ROI in the rest frames comprises: feature point extraction is performed on the reference ROI demarcated from the first frame of the sequence of grayscale image frames, to obtain at least one group of feature points; rest frames and dynamic frames are distinguished according to the displacement of each feature point in each grayscale image frame; the ROI in each rest frame is extracted to obtain the ROI in each rest frame.

4. The temperature data processing method for peripheral circulatory disorder monitoring according to claim 3, characterized in that, The step of distinguishing rest frames and dynamic frames according to the displacement of each feature point in each grayscale image frame comprises: each feature point in the sequence of grayscale image frames except the first grayscale image frame is tracked in sequence, and the displacement of each feature point between adjacent two grayscale image frames is calculated; the total displacement of each feature point in each second grayscale image frame is calculated according to the displacement of each feature point between adjacent two grayscale image frames; the total motion intensity of all the feature points in each second grayscale image frame is calculated by accumulating the total displacement of all the feature points in the same second grayscale image frame; rest frames are identified according to the total motion intensity of all the feature points in each second grayscale image frame.

5. The temperature data processing method for peripheral circulatory disorder monitoring according to claim 4, characterized in that, The method for constructing the preset regression model comprises: obtaining a pixel intensity time sequence corresponding to each target sample data; taking the pixel intensity time sequence of each target sample data as a target sample subset, and calculating the sample ratio between the average pixel intensity value and the dynamic brightness fluctuation amplitude corresponding to each target sample subset; performing linear regression fitting on the sample ratio and the reference PI value of each target sample subset to determine a regression coefficient; obtaining the constructed regression model according to the regression coefficient.

6. The temperature data processing method for peripheral circulatory disorder monitoring according to claim 5, characterized in that, The step of inputting the ratio into the preset regression model to obtain the PI prediction value comprises: determining a target regression model corresponding to the target monitoring object; transmitting the ratio to the target regression model for prediction to obtain the PI prediction value output by the target regression model.

7. A terminal device, characterized by, comprise: a processor and a memory for storing a computer program, the processor being configured to invoke and run the computer program stored in the memory to execute the steps of the temperature data processing method for peripheral circulation disorder monitoring according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, a computer program for causing a computer to execute the steps of the temperature data processing method for peripheral circulation disorder monitoring according to any one of claims 1 to 6.

Images