Method and device for determining temperature of target area of infrared thermal image and storage medium

Abstract

The invention provides a method and device for determining the temperature of a target area of an infrared thermal image, equipment and a storage medium, and relates to the field of artificial intelligence. The method comprises the following steps: acquiring an infrared thermal image; performing contour extraction on the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image; inputting the contour map into a key point prediction model, and obtaining coordinate information of a target key point corresponding to the contour map according to the output of the key point prediction model; determining a target area in the infrared thermal image according to the coordinate information of the target key point; and determining the area temperature of the target area according to the temperature information of the pixel points in the target area. According to the method, the target area of the infrared thermal imaging image is divided by using the key point prediction model, automatic identification of the target area can be realized, and the identification accuracy and efficiency are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device, equipment and storage medium for determining the temperature of a target area of ​​an infrared thermal imaging image. Background Art

[0002] Infrared thermal imaging technology can convert the invisible infrared radiation (heat energy) emitted by the human body into visible thermal images. It is a non-destructive, non-invasive, real-time, intuitive and green functional imaging technology for observing the heat value of the whole body. With the help of thermal imaging technology, the shape and intensity of the heat source distribution of the human body can be observed to determine the area where the human lesion is located. At the same time, the nature of the disease and the degree of the lesion can be determined based on the thermal distribution, and then medical evaluation can be carried out.

[0003] At present, medical staff still manually identify the human organ areas in infrared thermal imaging images. This identification method is limited by the human body, is highly subjective, prone to errors, and time-consuming. It is difficult to efficiently and accurately locate the human organ areas. If the positioning is inaccurate, it is very easy to misdiagnose or miss the disease based on the thermal distribution of the inaccurate area. Therefore, how to efficiently and accurately determine the human organ areas in infrared thermal imaging images has become an urgent problem to be solved. Summary of the invention

[0004] In view of this, the present invention proposes a method, device, equipment and storage medium for determining the temperature of the target area of ​​an infrared thermal imaging image. Specifically, when analyzing the infrared thermal imaging image, the contour of the infrared thermal imaging image is first extracted to obtain a contour image corresponding to the infrared thermal imaging image. Then, the contour image corresponding to the infrared thermal imaging image is input into the key point prediction model, and the key point prediction model outputs the coordinate information of the target key points in the contour image. Then, based on the coordinate information of the target key points in the contour image, the original infrared thermal imaging image is matched to determine the key points in the infrared thermal imaging image, and based on the key points, the target area in the infrared thermal imaging image is determined, and then the target area temperature of the target area is determined according to the temperature information of the pixel points in the target area. The present invention uses the key point prediction model to divide the target area of ​​the infrared thermal imaging image, which can realize the automatic recognition of the target feature area and greatly improve the accuracy and efficiency of the recognition.

[0005] A first aspect of the present invention provides a method for determining the temperature of a target area of ​​an infrared thermal imaging image, the method comprising:

[0006] Acquire an infrared thermal image; the pixel points of the infrared thermal image include temperature information;

[0007] Perform contour extraction on the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image;

[0008] The contour map is input into the key point prediction model, and the coordinate information of the target key point corresponding to the contour map is obtained according to the output of the key point prediction model; the key point prediction model is used to determine the key points of the input image, and the key points are used to determine at least one key feature area in the input image; the key point prediction model is trained by the sample contour map;

[0009] The target area in the infrared thermal imaging image is determined according to the coordinate information of the target key points; and the target area temperature of the target area is determined according to the temperature information of the pixel points in the target area.

[0010] In an optional implementation, the training process of the key point prediction model is:

[0011] Acquire a sample contour map, the sample contour map includes a contour map image and key feature point marks; the key feature point marks are used to indicate coordinate information of key feature points in the contour map image; the key feature points in the contour map image are used to determine the key feature area of ​​the contour map image;

[0012] Inputting the contour image into the key point prediction model to obtain the coordinate information of the predicted key points output by the key point prediction model;

[0013] The coordinate information of the predicted key points and the key feature point marks are compared, and the model parameters of the key point prediction model are adjusted according to the comparison results.

[0014] In an optional embodiment, obtaining a sample profile map includes:

[0015] Get the standard contour map;

[0016] Determine the key feature areas of the standard contour map;

[0017] Determine the vertices of the key feature area of ​​the standard contour map as the key feature points of the standard contour map, and mark the key feature points of the standard contour map to obtain key feature point marks of the standard contour map;

[0018] The sample contour map is determined based on the standard contour map and the key feature point markings of the standard contour map.

[0019] In an optional embodiment, determining the sample contour map according to the standard contour map and the key feature point marks of the standard contour map includes:

[0020] Performing deformation processing on the standard contour image to obtain a deformed contour image; the deformation processing includes at least one of rotation processing, translation processing, scaling processing, and waveform change processing;

[0021] Determine key feature points of the deformed contour image according to the deformation processing, and mark the key feature points of the deformed contour image to obtain key feature point marks of the deformed contour image;

[0022] The sample contour map is determined based on the deformation contour map and key feature point markings.

[0023] In an optional embodiment, rotation processing is used to rotate the standard contour image according to the rotation angle; translation processing is used to translate the standard contour image according to the translation amount and translation direction; scaling processing is used to scale the standard contour image according to the scaling factor; waveform change processing is used to deform the standard contour image according to amplitude parameters, frequency parameters and phase offset parameters.

[0024] In an optional embodiment, when the contour image is input into the key point prediction model, the method further includes:

[0025] Noise feature information is input into the key point prediction model; the noise feature information is used to add interference noise to the contour map image.

[0026] In an optional embodiment, when inputting noise feature information into the key point prediction model, the method further includes:

[0027] Determine the cropping ratio and cropping position;

[0028] The sample contour image is cropped according to the cropping ratio and the cropping position to obtain a cropped contour image;

[0029] The cropped contour image is input into the key information prediction model; the cropped contour image is used to add interference features to the contour image.

[0030] In an optional embodiment, the method further comprises:

[0031] When a preset condition is reached, the training process of the key point prediction model is stopped; wherein the preset condition is that the number of training times reaches a preset number of times or the residual of the coordinate information of the predicted key point and the key feature point mark is less than the residual threshold.

[0032] A second aspect of the present invention provides a device for determining the temperature of a target area of ​​an infrared thermal imaging image, the device comprising:

[0033] An acquisition unit, used for acquiring an infrared thermal imaging image; the pixel points of the infrared thermal imaging image include temperature information;

[0034] A processing unit, used for extracting the contour of the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image;

[0035] The processing unit is further used to input the contour map into the key point prediction model, and obtain the coordinate information of the target key point corresponding to the contour map according to the output of the key point prediction model; the key point prediction model is used to determine the key points of the input image, and the key points are used to determine at least one key feature area in the input image; the key point prediction model is obtained by training the sample contour map;

[0036] The determination unit is used to determine the target area in the infrared thermal imaging image according to the coordinate information of the target key point; and determine the target area temperature of the target area according to the temperature information of the pixel points in the target area.

[0037] In an optional embodiment, the device further comprises a training unit.

[0038] The training unit is specifically used to obtain a sample contour map, which includes a contour map image and key feature point marks; the key feature point marks are used to indicate the coordinate information of the key feature points in the contour map image; the key feature points in the contour map image are used to determine the key feature area of ​​the contour map image; the contour map image is input into the key point prediction model to obtain the coordinate information of the predicted key points output by the key point prediction model; the coordinate information of the predicted key points is compared with the key feature point marks, and the model parameters of the key point prediction model are adjusted according to the comparison results.

[0039] In an optional embodiment, the training unit is specifically used to obtain a standard contour map; determine the key feature area of ​​the standard contour map; determine the vertices of the key feature area of ​​the standard contour map as the key feature points of the standard contour map, and mark the key feature points of the standard contour map to obtain the key feature point markings of the standard contour map; determine the sample contour map based on the standard contour map and the key feature point markings of the standard contour map.

[0040] In an optional embodiment, the training unit is further used to perform deformation processing on the standard contour image to obtain a deformed contour image; the deformation processing includes at least one of rotation processing, translation processing, scaling processing, and waveform change processing. According to the deformation processing, key feature points of the deformed contour image are determined, and the key feature points of the deformed contour image are marked to obtain key feature point marks of the deformed contour image; according to the deformed contour image and the key feature point marks, a sample contour image is determined.

[0041] In an optional embodiment, rotation processing is used to rotate the standard contour image according to the rotation angle; translation processing is used to translate the standard contour image according to the translation amount and translation direction; scaling processing is used to scale the standard contour image according to the scaling factor; waveform change processing is used to deform the standard contour image according to amplitude parameters, frequency parameters and phase offset parameters.

[0042] In an optional implementation, the training unit is further used to input noise feature information into the key point prediction model; the noise feature information is used to add interference noise to the contour image.

[0043] In an optional embodiment, the training unit is also used to determine the cropping ratio and cropping position; crop the sample contour map according to the cropping ratio and cropping position to obtain a cropped contour map; input the cropped contour map into the key information prediction model; the cropped contour map is used to add interference features to the contour map image.

[0044] In an optional embodiment, the training unit is also used to stop the training process of the key point prediction model when a preset condition is met; wherein the preset condition is that the number of training times reaches a preset number of times or the residual of the predicted key point coordinate information and the key feature point mark is less than a residual threshold.

[0045] The third aspect of the present invention provides a device for determining the temperature of a target area of ​​an infrared thermal imaging image, the device comprising a processor and a memory; the processor is coupled to the memory; the memory is used to store computer instructions, and the computer instructions are loaded and executed by the processor so that the reply device implements the method described in the first aspect and any one of the embodiments of the first aspect.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions; the computer instructions enable the computer to execute the method described in the first aspect and any one of the embodiments of the first aspect.

[0047] The method, device, equipment and storage medium for determining the temperature of a target area of ​​an infrared thermal imaging image provided by the present invention have the following beneficial effects compared with the prior art:

[0048] (1) When analyzing an infrared thermal imaging image, the present invention first extracts the contour of the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image. Then, the contour image corresponding to the infrared thermal imaging image is input into a key point prediction model, and the key point prediction model outputs the coordinate information of the target key points in the contour image. Then, based on the coordinate information of the target key points in the contour image, the original infrared thermal imaging image is matched to determine the key points in the infrared thermal imaging image, and the target area in the infrared thermal imaging image is determined based on the key points. In this way, the target area temperature of the target area can be determined based on the temperature information of the pixel points in the target area. The present invention uses a key point prediction model to divide the target area of ​​the infrared thermal imaging image, which can realize the automatic recognition of the target feature area and greatly improve the accuracy and efficiency of recognition.

[0049] (2) When training the key point prediction model, the present invention does not need to rely on a large number of manually labeled samples. Instead, a large number of sample contour maps are obtained by constructing a standard contour map and performing various deformations on the standard contour map. In this way, labor costs and annotation costs will be greatly saved. At the same time, when deforming the standard contour map, in addition to the traditional rotation and scaling deformation processing, the transformation (stretching and twisting) combination of different proportions of each part can also be used to cover more contour forms. In this way, the obtained sample contour map is richer and has more features, thereby improving the training effect of the key point prediction model.

[0050] (3) In the training process, the present invention also inputs noise feature information when inputting the sample contour map into the key point prediction model. In this way, the actual input can be better fitted, so that the output of the key point prediction model is more consistent with the actual result, thereby improving the prediction accuracy of the key point prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 A schematic flow chart of a method for determining the temperature of a target area of ​​an infrared thermal imaging image provided in an embodiment of the present application;

[0053] Figure 2 A schematic diagram of the structure of a key point prediction model provided in an embodiment of the present application;

[0054] Figure 3 A flowchart of a method for training a key point prediction model provided in an embodiment of the present application;

[0055] Figure 4 A schematic diagram of a flow chart of a method for obtaining a sample profile map provided in an embodiment of the present application;

[0056] Figure 5 A schematic diagram of a key feature area of ​​a standard contour map provided in an embodiment of the present application;

[0057] Figure 6 A schematic diagram of the structure of a device for determining the temperature of a target area of ​​an infrared thermal imaging image provided in an embodiment of the present application;

[0058] Figure 7 A schematic diagram of the structure of a device for determining the temperature of a target area of ​​an infrared thermal imaging image provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Infrared thermal imaging technology can convert the invisible infrared radiation (heat energy) emitted by the human body into visible thermal images. It is a non-destructive, non-invasive, real-time, intuitive and green functional imaging technology for observing the heat value of the whole body. With the help of thermal imaging technology, the shape and intensity of the heat source distribution of the human body can be observed to determine the area where the human lesion is located. At the same time, the nature of the disease and the degree of the lesion can be determined based on the thermal distribution, and then medical evaluation can be carried out.

[0061] At present, medical staff still manually identify the human organ areas in infrared thermal imaging images, and the manual identification method has the following disadvantages:

[0062] 1. Inefficiency: Manual recognition takes a lot of time and manpower. For scenarios that require rapid processing of a large number of infrared thermal images, such as medical diagnosis and real-time monitoring, manual recognition is difficult to meet the requirements of real-time and high efficiency.

[0063] 2. Strong subjectivity: Manual recognition is affected by many factors such as the observer’s experience, fatigue, and concentration, which may lead to inconsistency and subjectivity in the recognition results. There may be large recognition differences between different observers, affecting the accuracy and reliability of the results.

[0064] 3. Error-prone: Due to the limitations of the human eye and brain, manual recognition is prone to errors in complex or blurred images. In infrared thermal imaging images, it is difficult to accurately locate the human organ area, which is likely to lead to the risk of missed diagnosis or misdiagnosis.

[0065] 4. Poor scalability: With the expansion of application scenarios and the increase in data volume, the scalability of manual recognition is limited. When the amount of infrared thermal images that need to be processed exceeds the limit of manual processing capabilities, the recognition efficiency and accuracy will drop sharply.

[0066] 5. High cost: Long-term reliance on manual recognition requires hiring and training a large number of professionals, resulting in high labor costs.

[0067] Therefore, how to efficiently and accurately determine the human organ areas in infrared thermal imaging images has become an urgent problem to be solved.

[0068] In response to the above problems, the present invention proposes a method, device and storage medium for determining the temperature of the target area of ​​an infrared thermal imaging image. Specifically, when analyzing an infrared thermal imaging image, the contour of the infrared thermal imaging image is first extracted to obtain a contour image corresponding to the infrared thermal imaging image. The contour image corresponding to the infrared thermal imaging image is then input into a key point prediction model, and the key point prediction model outputs the coordinate information of the target key points in the contour image. Then, based on the coordinate information of the target key points in the contour image, the original infrared thermal imaging image is matched to determine the key points in the infrared thermal imaging image, and the target area in the infrared thermal imaging image is determined based on the key points. In this way, the target area temperature of the target area can be determined based on the temperature information of the pixel points in the target area. The present invention uses a key point prediction model to divide the target area of ​​the infrared thermal imaging image, which can realize automatic identification of the target area and greatly improve the accuracy and efficiency of identification. Below, the embodiments of the present application are described in detail:

[0069] Figure 1 The present invention provides a flow chart of a method for determining the temperature of a target area of ​​an infrared thermal imaging image. Figure 1 As shown, the method includes:

[0070] 101. Obtain infrared thermal imaging images.

[0071] First, the infrared thermal image to be predicted is collected to obtain an infrared thermal image with temperature information. For example, the infrared thermal image may be an infrared thermal image of a human body. It is understandable that the pixel points of the infrared thermal image include temperature information.

[0072] 102. Perform contour extraction on the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image.

[0073] Next, contour extraction is performed on the infrared thermal imaging image to obtain a contour map corresponding to the infrared thermal imaging image. Exemplarily, if the infrared thermal imaging image is a human infrared thermal imaging image, the contour detection and contour extraction methods in OpenCV can be used to extract the human contour from the human infrared thermal imaging image. Specifically, contour information is obtained and a single-channel human contour map is generated based on the contour information.

[0074] 103. Input the contour map into the key point prediction model, and obtain the coordinate information of the target key point corresponding to the contour map according to the output of the key point prediction model.

[0075] Among them, the key point prediction model is used to determine the key points of the input image, and the key points are used to determine the key areas in the input image. Therefore, after the contour map corresponding to the infrared thermal imaging image is input into the key point prediction model, the key point prediction model will predict the coordinates of each key point of the contour map and output the coordinate information of the target key point corresponding to the contour map.

[0076] The structure of the key point prediction model is introduced below. Figure 2 A schematic diagram of the structure of a key point prediction model provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the key point prediction model includes an input layer, a convolutional layer 1, a convolutional layer 2, a pooling layer, a discard layer, a flattening layer, a fully connected layer 1, a fully connected layer 2, and an output layer in sequence.

[0077] The input layer is used to receive the contour map, and the two subsequent convolutional layers are used for feature extraction. For example, the convolution kernel size of convolutional layer 1 and convolutional layer 2 can be set to 3×3, and the activation function thereof adopts the ReLU function.

[0078] Exemplarily, the key point prediction model is based on the residual network model ResNet-50. The first convolution layer of the ResNet-50 model can be modified to an input layer that accepts a single channel input to ensure that the key point prediction model can process the input contour map.

[0079] The two convolution layers can include Depthwise convolution and Pointwise convolution. Depthwise convolution is used to perform convolution on each input channel independently, which can significantly reduce model parameters and calculation amount. Assuming that the input is Cin channels and the convolution kernel size is k×k, the number of parameters of Depthwise convolution is:

[0080] Parameters depthwise =C in ×k×k (1)

[0081] Pointwise convolution can use 1×1 convolution to combine the output of Depthwise convolution into a new feature map, effectively extract features and increase the network's expressive power. Its parameter quantity is:

[0082] Parameters pointwise =C in ×C out (2)

[0083] For example, the last fully connected layer of ResNet-50 can be replaced with a layer that outputs 50 coordinate values ​​to meet the positioning requirements of key feature points of the human body. A linear activation function is used to directly output the key point coordinates predicted by the key point prediction model.

[0084] 104. Determine a target area in the infrared thermal imaging image according to the coordinate information of the target key point; and determine a target area temperature of the target area according to the temperature information of the pixel points in the target area.

[0085] Then, the coordinate information of the target key points is matched with the original infrared thermal image to obtain the temperature information corresponding to the target key points. At the same time, the coordinate information of the target key points can be used to determine the target area according to the given number, and then obtain the temperature information of the target area.

[0086] When analyzing an infrared thermal imaging image, the present invention first extracts the contour of the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image. Then, the contour image corresponding to the infrared thermal imaging image is input into a key point prediction model, and the key point prediction model outputs the coordinate information of the target key points in the contour image. Then, based on the coordinate information of the target key points in the contour image, the original infrared thermal imaging image is matched to determine the key points in the infrared thermal imaging image, and the target area in the infrared thermal imaging image is determined based on the key points. In this way, the target area temperature of the target area can be determined based on the temperature information of the pixel points in the target area. The present invention uses a key point prediction model to divide the target area of ​​the infrared thermal imaging image, which can realize automatic recognition of the target area and greatly improve the accuracy and efficiency of recognition.

[0087] The following is an introduction to the training process of the key point prediction model. Figure 3 A flowchart of a key point prediction model training method provided in an embodiment of the present application. Figure 3 As shown, the training method includes:

[0088] 301. Obtain a sample contour map.

[0089] The sample contour map includes a contour map image and key feature point marks. The key feature point marks are used to indicate the coordinate information of the key feature points in the contour map image, and the key feature points in the contour map image are used to determine the key feature area of ​​the contour map image. In this embodiment, the sample contour map is not manually annotated. Instead, it is obtained by constructing a standard contour map and performing multiple deformations on the standard contour map. In this way, the labor cost and annotation cost will be greatly saved. Please refer to the content shown below for the specific construction method, which will not be introduced here.

[0090] 302. Input the contour image into the key point prediction model to obtain the coordinate information of the predicted key points output by the key point prediction model.

[0091] Exemplarily, when the contour image is input into the key point prediction model, noise feature information may also be input into the key point prediction model so as to better fit the actual input situation.

[0092] Exemplarily, the real infrared thermal imaging image can be evaluated to increase the interference term of the input contour image. This is because of the particularity of infrared thermal imaging images, and many problems will arise when actually extracting contours, such as unclear contour extraction, blurred boundaries of human body contour lines, and large interference from noisy background lines. Therefore, inputting noise feature information into the key point prediction model can further improve the training effect and improve the prediction performance of the key point prediction model. Based on the above-mentioned interference factors unique to infrared thermal imaging images, a specific noise type that fits the contour map of the infrared thermal imaging image can be added to the input of the key point prediction model to optimize the training of the key point prediction model.

[0093] In the process of training the key point model using infrared images collected from real scenes, contour images with low model prediction accuracy can be collected based on the test results of the key point model; then these contour images are compared with the contour images with accurate predictions to analyze the interference factors that affect the prediction accuracy. Next, noise fitting the interference factor can be added to the image with accurate predictions to verify whether the interference factor is the reason that affects the model prediction accuracy. If so, the interference factor is added to the input contour image when the key point model is trained later.

[0094] In one case, the difference in real human body morphology, that is, the angle between human limbs, has an impact on the model prediction results of the key point prediction model. For example, random cropping can be used to reduce the interference of the position of human limbs. That is, the input contour image is randomly cropped to obtain a cropped contour image, and the cropped contour image and the original contour image are input into the key point prediction model together. In this way, the generalization ability of the key point prediction model can be improved.

[0095] Specifically, the cropping ratio and the cropping position may be determined, and the sample contour image may be cropped according to the cropping ratio and the cropping position to obtain a cropped contour image. The cropping process is described below with a specific example:

[0096] Assuming the input image is I, whose size is H×W, the random cropping process can be defined as:

[0097] The probability is randomly selected, where the randomly generated probability is between 0 and 1. If the probability is greater than the probability threshold, the original image I is directly returned, that is, I′=I. If it is less than the probability threshold, the cropping ratio is randomly selected, and the cropping ratio is illustratively between 0.3 and 0.4. Then, the cropping position is randomly selected, such as selecting the left cropping or the right cropping, to obtain the cropping contour image I′.

[0098] 303. Compare the coordinate information of the predicted key points and the key feature point marks, and adjust the model parameters of the key point prediction model according to the comparison result.

[0099] Next, the coordinate information of the predicted key points and the key feature point marks are compared, and the model parameters of the key point prediction model are adjusted according to the comparison results so that the coordinate information of the predicted key points output by the key point prediction model and the actual key feature point coordinate information (key feature point marks) are as close as possible.

[0100] 304. When the preset condition is reached, the training process of the key point prediction model is stopped.

[0101] Exemplarily, when the number of training times reaches a preset number or the residual of the predicted key point coordinate information and the key feature point mark is less than the residual threshold, the training of the key point prediction model is stopped.

[0102] For example, the loss function of training can be Mean Squared Error (MSE). The optimizer can be Adam optimizer, because it shows good convergence and stability during training. The number of training epochs, that is, the preset number of training epochs, can be set to 20.

[0103] Below Figure 3 The method for obtaining the sample contour map for training the key point prediction model in the illustrated embodiment is received in detail. Figure 4 A schematic diagram of a method for obtaining a sample profile provided in an embodiment of the present application. Figure 4 As shown, the method includes:

[0104] 401. Obtain a standard contour map.

[0105] Taking the infrared thermal imaging image of the human body as an example, it is necessary to construct a standard human body contour image of the infrared thermal imaging image of the human body.

[0106] 402. Determine the key feature areas of the standard contour map.

[0107] For example, according to the theory of TCM syndrome differentiation, the human body trunk can be divided into nine key characteristic areas: chest, heart, stomach, abdomen, left flank, right flank, lower abdomen, left lower abdomen, and right lower abdomen. Figure 5 shown.

[0108] 403. Determine the vertices of the key feature area of ​​the standard contour map as the key feature points of the standard contour map, and mark the key feature points of the standard contour map to obtain key feature point marks of the standard contour map.

[0109] The human body can be divided into multiple regions based on the position of organs. Then, according to the distribution range of each region in the human body, multiple polygons of key feature regions are constructed, and then the vertices of each polygon are marked to form a key feature point group. The key feature points are marked to obtain the key feature point markings of the standard contour map.

[0110] In this way, the standard contour image and the key feature point marks of the standard contour image constitute the final standard sample, and the standard sample can be encoded using three-channel color, where the standard contour image is placed in the blue channel B255, the polygon group of the key feature area is placed in the green channel G255, and the key feature point is placed in the red channel R255. Exemplarily, the size of the standard sample can be set to 640×480 pixels, and after extracting the vertex coordinates of the key feature point group, a separate number is generated for each vertex.

[0111] 404. Determine the sample contour map according to the standard contour map and the key feature point marks of the standard contour map.

[0112] Next, a large number of sample contours are generated based on the standard samples. Exemplarily, the standard contour can be deformed to obtain a deformed contour; then the key feature points of the deformed contour are determined based on the deformation, and the key feature points of the deformed contour are marked to obtain the key feature point marks of the deformed contour; then the sample contour is determined based on the deformed contour and the key feature point marks. The deformation includes at least one of rotation, translation, scaling, and waveform change.

[0113] For example, the data of the standard sample can be constructed into a triple format: a standard contour map, a key feature area of ​​the standard contour map, and a key feature point mark of the standard contour map. Then the triple of the standard sample is deformed, and random transformation parameters are set. The deformation process involves rotation, translation, scaling, and waveform transformation. At the same time, it is necessary to ensure the consistency of the deformation parameters of each vertex during each deformation. The following is a brief introduction to the deformation process:

[0114] 1. Rotation processing:

[0115] For example, a normally distributed random angle can be generated as a rotation parameter to achieve random rotation of the standard contour map. The rotation process can be expressed as:

[0116]

[0117] Among them, (x', y') is the rotated coordinate, (x, y) is the original coordinate, and θ is the randomly generated rotation angle.

[0118] 2. Translation processing:

[0119] The standard contour map can be randomly translated horizontally and vertically, expressed as:

[0120]

[0121] Among them, (x', y') is the coordinate after translation, (x, y) is the original coordinate, and (Δx, Δy) is the randomly generated translation amount, which controls the range of translation.

[0122] 3. Scaling processing:

[0123] By generating a random scaling factor, the image is scaled. It is expressed as:

[0124]

[0125] Among them, (x', y') is the scaled coordinate, (x, y) is the original coordinate, and s is the randomly generated scaling factor.

[0126] 4. Waveform change processing:

[0127] Apply a waveform transformation based on a sine function to produce a deformation with periodic changes. This transformation can be achieved by the following formula:

[0128] M x =x+A·sin(a·x+b)

[0129] M y =y+A·sin(a·y+b) (6)

[0130] Among them, Mx and My are the transformed coordinates, x and y are the original coordinates, A is the amplitude, a is the frequency, and b is the phase offset.

[0131] For example, some real human body contour images can be selected to match the standard contour images in key feature areas to obtain deformation parameters, including rotation angle, scaling ratio and offset, and the deformation parameter data can be stored. These deformation parameters will be used for further sample generation and model training to ensure the diversity and representativeness of the samples.

[0132] The present invention does not need to rely on a large number of manually labeled samples when training the key point prediction model. Instead, a large number of sample contour maps are obtained by constructing a standard contour map and performing multiple deformations on the standard contour map. In this way, labor costs and annotation costs will be greatly saved. At the same time, when deforming the standard contour map, in addition to the traditional rotation and telescopic deformation processing, the transformation (stretching and twisting) combination of different proportions of each part can also be used to cover more contour forms. In this way, the obtained sample contour map is richer and has more features, thereby improving the training effect of the key point prediction model.

[0133] Based on the above description, Figure 6 This is a schematic diagram of the structure of a device for determining the temperature of a target area of ​​an infrared thermal imaging image provided in an embodiment of the present application. Figure 6 As shown, the device comprises:

[0134] The acquisition unit 601 is used to acquire an infrared thermal image; the pixel points of the infrared thermal image include temperature information;

[0135] The processing unit 602 is used to extract the contour of the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image;

[0136] The processing unit 602 is further used to input the contour image into the key point prediction model, and obtain the coordinate information of the target key point corresponding to the contour image according to the output of the key point prediction model; the key point prediction model is used to determine the key points of the input image, and the key points are used to determine at least one key area in the input image; the key point prediction model is obtained by training the sample contour image;

[0137] The determination unit 603 is used to determine the target area in the infrared thermal imaging image according to the coordinate information of the target key point; and determine the target area temperature of the target area according to the temperature information of the pixel points in the target area.

[0138] In an optional implementation, the device further includes a training unit 604 .

[0139] The training unit 604 is specifically used to obtain a sample contour map, which includes a contour map image and key feature point marks; the key feature point marks are used to indicate the coordinate information of the key feature points in the contour map image; the key feature points in the contour map image are used to determine the key feature area of ​​the contour map image; the contour map image is input into the key point prediction model to obtain the coordinate information of the predicted key points output by the key point prediction model; the coordinate information of the predicted key points is compared with the key feature point marks, and the model parameters of the key point prediction model are adjusted according to the comparison results.

[0140] In an optional embodiment, the training unit 604 is also used to obtain a standard contour map; determine the key feature area of ​​the standard contour map; determine the vertices of the key feature area of ​​the standard contour map as the key feature points of the standard contour map, and mark the key feature points of the standard contour map to obtain the key feature point markings of the standard contour map; determine the sample contour map based on the standard contour map and the key feature point markings of the standard contour map.

[0141] In an optional embodiment, the training unit 604 is further used to perform deformation processing on the standard contour image to obtain a deformed contour image; the deformation processing includes at least one of rotation processing, translation processing, scaling processing, and waveform change processing. According to the deformation processing, key feature points of the deformed contour image are determined, and the key feature points of the deformed contour image are marked to obtain key feature point marks of the deformed contour image; according to the deformed contour image and the key feature point marks, a sample contour image is determined.

[0142] In an optional embodiment, rotation processing is used to rotate the standard contour image according to the rotation angle; translation processing is used to translate the standard contour image according to the translation amount and translation direction; scaling processing is used to scale the standard contour image according to the scaling factor; waveform change processing is used to deform the standard contour image according to amplitude parameters, frequency parameters and phase offset parameters.

[0143] In an optional implementation, the training unit 604 is further used to input noise feature information into the key point prediction model; the noise feature information is used to add interference noise to the contour image.

[0144] In an optional embodiment, the training unit 604 is also used to determine the cropping ratio and cropping position; crop the sample contour map according to the cropping ratio and cropping position to obtain a cropped contour map; input the cropped contour map into the key information prediction model; and the cropped contour map is used to add interference features to the contour map image.

[0145] In an optional embodiment, the training unit 604 is also used to stop the training process of the key point prediction model when a preset condition is met; wherein the preset condition is that the number of training times reaches a preset number of times or the residual of the predicted key point coordinate information and the key feature point mark is less than a residual threshold.

[0146] See also Figure 5 The embodiment of the present application also shows a device 1000 for determining the temperature of a target area of ​​an infrared thermal imaging image. The device 1000 may include one or more central processors 1001 and a communication interface 1003. Optionally, a memory 1002 may also be included.

[0147] The memory 1002 may be a temporary storage or a permanent storage. Furthermore, the central processor 1001 may be configured to communicate with the memory 1002 and execute a series of instruction operations in the memory 1002 on the footprint recognition device.

[0148] In this embodiment, the CPU 1001 can execute the aforementioned Figures 1 to 5 The steps performed in the illustrated embodiment will not be described in detail here.

[0149] In this embodiment, the specific functional module division in the central processing unit 1001 can be the same as the above Figure 6 The division method of each module described in is similar and will not be repeated here.

[0150] It should be noted that the information execution process of the central processor 1001 of the above-mentioned device 1000 and other contents can be specifically referred to in the aforementioned Figures 1 to 5 The description in the method embodiment shown will not be repeated here.

[0151] The present invention also provides a computer-readable storage medium, which stores computer instructions; the computer instructions enable the computer to execute the above Figures 1 to 5 The steps performed in any one of the embodiments shown.

[0152] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for determining the temperature of a target area of ​​an infrared thermal imaging image, characterized in that: The method comprises: Acquire an infrared thermal imaging image; the pixel points of the infrared thermal imaging image include temperature information; Performing contour extraction on the infrared thermal imaging image to obtain a contour image corresponding to the infrared thermal imaging image; The contour map is input into a key point prediction model, and the coordinate information of the target key point corresponding to the contour map is obtained according to the output of the key point prediction model; the key point prediction model is used to determine the key points of the input image, and the key points are used to determine at least one key feature area in the input image; the key point prediction model is trained by the sample contour map; The target area in the infrared thermal imaging image is determined according to the coordinate information of the target key point; and the target area temperature of the target area is determined according to the temperature information of the pixel points in the target area.

2. The method according to claim 1, characterized in that The training process of the key point prediction model is: Acquire a sample contour map, wherein the sample contour map includes a contour map image and key feature point marks; the key feature point marks are used to indicate coordinate information of key feature points in the contour map image; the key feature points in the contour map image are used to determine a key feature area of ​​the contour map image; Inputting the contour image into the key point prediction model to obtain coordinate information of the predicted key points output by the key point prediction model; The coordinate information of the predicted key point is compared with the key feature point mark, and the model parameters of the key point prediction model are adjusted according to the comparison result.

3. The method according to claim 2, characterized in that The step of obtaining a sample profile diagram comprises: Get the standard contour map; Determining key feature areas of the standard contour map; Determine the vertices of the key feature area of ​​the standard contour map as the key feature points of the standard contour map, and mark the key feature points of the standard contour map to obtain key feature point marks of the standard contour map; The sample contour map is determined according to the standard contour map and key feature point marks of the standard contour map.

4. The method according to claim 3, characterized in that: The step of determining the sample contour map according to the standard contour map and key feature point marks of the standard contour map comprises: Performing deformation processing on the standard contour image to obtain a deformed contour image; the deformation processing includes at least one of rotation processing, translation processing, scaling processing, and waveform change processing; Determining key feature points of the deformed contour image according to the deformation processing, and marking the key feature points of the deformed contour image to obtain key feature point marks of the deformed contour image; The sample contour map is determined according to the deformation contour map and the key feature point marks.

5. The method according to claim 4, characterized in that The rotation processing is used to rotate the standard contour image according to a rotation angle; The translation process is used to translate the standard contour image according to the translation amount and translation direction; The scaling process is used to scale the standard contour map according to a scaling factor; The waveform change process is used to deform the standard contour image according to an amplitude parameter, a frequency parameter and a phase offset parameter.

6. The method according to any one of claims 2 to 5, characterized in that: When the contour map image is input into the key point prediction model, the method further comprises: Noise feature information is input into the key point prediction model; the noise feature information is used to add interference noise to the contour image.

7. The method according to claim 6, characterized in that When inputting noise feature information into the key point prediction model, the method further includes: Determine the cropping ratio and cropping position; Clipping the sample contour image according to the clipping ratio and the clipping position to obtain the clipping contour image; The cropped contour image is input into the key information prediction model; the cropped contour image is used to add interference features to the contour image.

8. The method according to claim 7, characterized in that The method further comprises: When a preset condition is reached, the training process of the key point prediction model is stopped; wherein the preset condition is that the number of training times reaches a preset number of times or the residual of the coordinate information of the predicted key point and the key feature point mark is less than a residual threshold.

9. A device for dividing an infrared thermal imaging image into regions, characterized in that: The device comprises a processor and a memory; the processor is coupled to the memory; the memory is used to store computer instructions, and the computer instructions are loaded and executed by the processor, so that the device implements the method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions; the computer instructions enable the computer to execute the method according to any one of claims 1 to 8.

Images