Arch health assessment system and method based on non-contact thermal infrared imaging

Abstract

The invention belongs to the technical field of health monitoring and imaging analysis, and particularly relates to an arch health assessment system and method based on non-contact thermal infrared imaging. Comprising an infrared thermal information acquisition pad, an infrared sensor, a mobile device and an information processing module, the infrared thermal information acquisition pad is provided with a human foot positioning mark, the moving device is placed on one side of the infrared thermal information acquisition pad, the infrared sensor is slidably mounted on the moving device, and the infrared sensor is connected with an information processing table; and an HRVManba model is adopted in the information processing station to carry out data processing. A hardware system and an algorithm are deeply fused, a quantitative evaluation system of arch health is constructed, and data support is provided for personalized health management and early warning of foot diseases. The foot arch evaluation technology based on the non-contact infrared foot arch scanning system and the matched analysis algorithm has a wide application prospect, the accuracy and efficiency of foot arch analysis are improved, and the application range of the thermal infrared imaging technology in the health field is expanded.

Description

Technical Field

[0001] The present invention belongs to the technical field of health monitoring and imaging analysis, and specifically relates to a foot arch health assessment system and method based on non-contact thermal infrared imaging. Background Art

[0002] With the increasing focus on health in modern society, foot health has become a key area of concern. As a crucial support for human movement, the health of the foot directly impacts postural stability, athletic ability, and overall well-being. The arch of the foot is a core component of the foot's structure, and its form and function are crucial for its proper functioning. Abnormalities in the arch of the foot, such as flat feet and high arches, can not only cause foot pain but also trigger a series of consequential problems, such as dysfunction of the knee, hip, and spine. Therefore, accurate analysis and assessment of the arch of the foot has significant clinical and societal value.

[0003] Traditional methods of arch analysis usually rely on physical contact or imaging techniques, but they have many limitations. For example: (1) Footprint method: The footprint method infers the arch morphology by observing the pressure distribution on the sole of the foot. However, this method is greatly affected by the dynamic changes in pressure distribution and it is difficult to obtain stable and accurate results. In addition, the footprint method has high requirements for the operating environment and requires the subject to maintain a specific posture. It is easily interfered with by human factors during use.

[0004] (2) X-ray imaging: X-ray imaging determines the arch condition by capturing the bone structure of the foot. Although it is highly accurate, this method carries significant radiation risks, limiting its application in special groups such as children and pregnant women. Furthermore, the high cost of X-ray equipment makes it difficult to popularize and promote it in primary medical institutions and personal health management.

[0005] (3) 3D scanning technology: 3D scanning technology analyzes the arch of the foot by capturing the 3D morphology of the foot. Although it can provide rich morphological information, the equipment is expensive, the operation is complicated, and the scanning environment is demanding. This method usually requires the subject to stand or walk barefoot, which is inconvenient to operate and difficult to achieve daily use and portability.

[0006] (4) Thermal pressure-sensitive plate: The thermal pressure-sensitive plate infers the arch morphology by recording the contact pressure distribution of the sole of the foot. However, its resolution is limited and it is difficult to identify fine arch features. In addition, the thermal pressure-sensitive plate is highly dependent on the pressure range and distribution, and cannot effectively evaluate the arch characteristics in static and dynamic states.

[0007] In the field of arch analysis, there is an urgent need for a new technical means with high precision, non-contact, non-radiation, and convenient operation. In recent years, with the rapid development of thermal infrared imaging technology, its application in the field of medical health has gradually attracted attention. Thermal infrared imaging technology can obtain the surface temperature information of the human body non-contact by capturing the thermal radiation distribution on the surface of the object, providing a brand-new means for analyzing the physiological state of the human body.

[0008] Thermal infrared imaging technology has the advantages of non-contact, non-radiation, and strong real-time performance, and is particularly suitable for the dynamic monitoring and evaluation of foot health. In the arch analysis task, thermal infrared images can reflect the thermal radiation distribution pattern on the sole of the foot, which is closely related to the biomechanical characteristics, blood circulation, and arch shape of the sole of the foot. For example, areas with higher plantar pressure are usually accompanied by higher thermal radiation, while areas with higher arches usually have lower thermal radiation due to less contact with the ground. Therefore, by analyzing the temperature distribution of the thermal infrared image of the sole of the foot, the morphological characteristics of the arch can be indirectly inferred. By obtaining the mask map of foot contact through a semantic segmentation model, analyzing the mask map can efficiently judge the health status of the arch. Summary of the Invention

[0009] The present invention provides an arch health assessment system and method based on non-contact thermal infrared imaging to solve the technical problems existing in the prior art, such as unstable results of traditional arch analysis methods, radiation risks of direct contact, expensive equipment and complex operation, and low analysis accuracy.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions: An arch health assessment system based on non-contact thermal infrared imaging includes an infrared thermal information acquisition pad, an infrared sensor, a mobile device, and an information processing module; human foot positioning marks are provided on the infrared thermal information acquisition pad, the mobile device is placed on one side of the infrared thermal information acquisition pad, the infrared sensor is slidably mounted on the mobile device, and the infrared sensor is connected to an information processing station; the HRVManba model is used for data processing in the information processing station.

[0011] A slide rail is provided on the mobile device, a stepping motor is provided in the slide rail, the infrared sensor is slidably arranged on the mobile device through the stepping motor, and the stepping motor is connected to the information processing station.

[0012] The lower end of the infrared sensor is an optical tripod with adjustable height and a spherical optical bracket with adjustable angle, and the spherical optical bracket is fixed at the upper end of the optical tripod.

[0013] A positioning status feedback device is further provided on the infrared sensor, and the positioning status feedback device is connected to the information processing station.

[0014] The data processing of the HRVMamba model is mainly divided into a sampling module and a core processing unit. The sampling module consists of two 3×3 convolutions with a stride of 2, and is used to perform downsampling on the image data monitored by the infrared sensor. The core processing unit is used to extract visual features from the data.

[0015] The visual feature extraction process is specifically divided into four stages. In the first stage, the core processing unit uses the same Bottleneck structure as HRNet to perform convolution operations on the larger feature maps in the image data monitored by the infrared sensor, and extracts preliminary visual features. In the subsequent three stages, the core processing unit uses the global-local joint visual state space module. The feature stream in each stage contains the resolution of the previous stage and an additional lower resolution, and finally generates features with four branches.

[0016] In the subsequent three stages, the input features are divided into branches with different resolutions by the core processing unit and processed separately. For the high-resolution branch, through the multi-scale local convolution and global scanning module in the GL-VSS module, using the local feature learning ability of convolution and the global receptive field learning ability of SS2D, the feature extraction is enhanced. For the low-resolution branch, it is also processed through the GL-VSS module. The traditional depthwise separable convolution is replaced by the MultiDW module, and multi-scale convolutional kernels are used to capture local features at different scales, enhancing the inductive bias of the model. Finally, features with four branches are generated.

[0017] The method of replacing the traditional depthwise separable convolution with the MultiDW module and using multi-scale convolutional kernels to capture local features at different scales and enhance the inductive bias of the model is shown in the following formula:

[0018]

[0019]

[0020] where is the result of dividing the input features along the channel dimension into G groups, and each group is the feature data of a part of the channels of the input features ; is the output feature after the depth convolution with a kernel size of on the g-th group of features ; is the final output feature.

[0021] A method for arch health assessment based on non-contact thermal infrared imaging. The subject stands at the human foot positioning mark on the infrared thermal information collection pad, and the infrared thermal information of the subject's sole is left on the pad. The information processing platform controls the infrared sensor on the mobile device to move from the subject's heel to the toe. The infrared sensor scans the infrared thermal information of the sole to form thermal infrared image data. The infrared thermal information collection pad collects the sole heat conduction information. The infrared thermal information collection pad and the infrared sensor transmit the thermal infrared image data and the sole heat conduction information to the information processing platform. The HRVManba model in the information processing platform performs data processing and feature extraction on the thermal infrared image data and the sole heat conduction information, and then generates arch feature analysis data.

[0022] The data processing of the HRVMamba model is mainly divided into a sampling module and a core processing unit. The sampling module consists of two 3×3 convolutions with a stride of 2. The sampling module is used to reduce the feature resolution of the data to 1 / 4. The core processing unit is used to perform visual feature extraction processing on the data, and finally generates four branches, and the feature sizes are respectively: 。

[0023] Compared with the prior art, the present invention has the following beneficial effects: An arch health assessment system based on non-contact thermal infrared imaging proposed by the present invention uses an infrared thermal information collection pad and an infrared sensor to collect arch data, without direct contact with the device, avoiding the radiation risk that may be brought by the traditional direct contact detection method, and is safer and healthier for the detector. And the infrared thermal information collection pad and the infrared sensor have lower costs compared to traditional expensive and complex arch analysis devices. And the operation process is relatively simple. The operator only needs to set relevant parameters on the information processing platform and start the system to complete data collection, reducing the operation difficulty and the requirement for professional skills, and improving the detection efficiency.

[0024] Furthermore, when using the HRVMamba model for arch analysis, its innovative global-local joint visual state space module (GL-VSS) and multi-resolution parallel design can effectively integrate multi-scale information, enhance the ability to extract and model arch features. Compared with traditional methods, the obtained arch analysis results are more stable, reducing the result fluctuations caused by unreasonable data processing methods. The HRVMamba model can better capture the local and global features of the arch through operations such as multi-scale local convolution and multi-scale depth convolution, accurately judge the arch shape and health status, and greatly improve the analysis accuracy compared with traditional methods. Description of the Drawings

[0025] Figure 1 : Structural diagram of the arch data collection system based on non-contact thermal infrared imaging; Figure 2: Schematic diagram of infrared thermal information acquisition pad; Figure 3 : Schematic diagram of the structure of the mobile device; Figure 4 : Schematic diagram of the structure of the infrared sensor; Figure 5 : Schematic diagram of the data processing flow of the HRVManba model; Figure 6 : Schematic diagram of the internal system structure of the HRVManba model; Figure 7 : Schematic diagram of the original data of arch data acquisition; Figure 8 : Schematic diagram of the result of original data segmentation; Figure 9 : Standard comparison chart of the arch; Figure 10 : Schematic diagram of the arch data evaluation system for non-contact thermal infrared imaging.

[0026] Label description: 1. Infrared thermal information acquisition pad; 2. Infrared sensor; 3. Mobile device; 4. Spherical optical bracket; 5. Optical tripod; 6. Stepper motor; 7. Positioning status feedback device; 8. Information processing table. Detailed implementation manner

[0027] To further understand the content of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are only for explaining the present invention and not for limiting it.

[0028] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0029] As Figure 1 shown, it is the structure diagram of the arch data acquisition system for non-contact thermal infrared imaging of the present invention, which is mainly used to achieve high-precision acquisition of plantar thermal infrared data under the conditions of no radiation and non-contact, and provide reliable input data for arch analysis. The arch data acquisition mainly includes the following four parts: infrared thermal information acquisition pad 1, infrared sensor 2 and mobile device 3. The mobile device 3 is placed on one side of the infrared thermal information acquisition pad 1, and the infrared sensor 2 is slidably arranged on the mobile device 3, and the infrared sensor 2 slides along the mobile device 3. The functions and designs of each part will be described in detail below.

[0030] The infrared thermal information acquisition pad 1 is one of the core components of the entire arch data acquisition system. Its main function is to provide a standardized contact surface for the sole of the foot, and record the thermal radiation information of the sole of the foot after the sole of the foot contacts the pad. As Figure 2As shown in the figure, the infrared thermal information collection pad 1 is made of a material with high thermal conductivity and strong durability, with a length:width of 200 cm:70 cm. It can be made of thermoplastic elastomer to quickly and evenly conduct the heat from the sole of the foot, thereby improving the accuracy of thermal information collection. The surface of the infrared thermal information collection pad 1 is anti-slip treated to ensure that the subject will not slip when standing or walking, enhancing the safety of the testing process. The material of the infrared thermal information collection pad 1 can maintain stable performance at different ambient temperatures, ensuring the accurate collection of thermal radiation information. The surface of the infrared thermal information collection pad 1 has clear standardized human foot positioning marks to ensure that the subject can accurately place their feet. Thus, more consistent test data can be obtained.

[0031] The infrared sensor 2 is responsible for capturing the thermal infrared image data of the sole of the foot and converting the thermal infrared image data into input data for analysis. The infrared sensor 2 uses a high-sensitivity uncooled infrared detector, which can capture the temperature distribution of the sole of the foot in real time under non-contact conditions. The imaging resolution of the infrared sensor 2 is as high as 0.1°C, which can capture the subtle changes in the thermal distribution of the sole of the foot, laying a foundation for the precise analysis of the arch shape. As Figure 4 As shown in the figure, the lower end of the infrared sensor 2 is a height-adjustable optical tripod 5 and an angle-adjustable spherical optical bracket 4. The spherical optical bracket 4 is fixed to the upper end of the optical tripod 5. The optical tripod 5 and the spherical optical bracket 4 enable the infrared sensor 2 to have a wide enough viewing angle to cover the entire sole area, avoiding image loss caused by field of view limitations. The imaging speed of the infrared sensor 2 reaches 30 frames per second, which can meet the needs of dynamic detection, such as recording the changes in the thermal distribution of the sole of the foot when the subject is walking.

[0032] In this embodiment, the mobile device 3 uses a slide rail, and the infrared sensor 2 is slidably arranged on the track of the slide rail. By sliding along its track, the infrared sensor 2 realizes the full-coverage scanning of the subject's sole area. The slide rail adopts high-precision electric control technology, which can move the sensor smoothly according to the preset path, thereby obtaining continuous and consistent image data. The high-precision electric control of the slide rail is specifically as follows: A high-precision stepper motor 6 and a positioning status feedback device 7 are equipped on the slide rail. The stepper motor 6 controls the moving speed of the infrared sensor 2 on the slide rail track, and the positioning status feedback device 7 monitors the position information of the infrared sensor 2 on the slide rail. Through the stepper motor 6 and the positioning status feedback device 7, it is ensured that the moving path and speed of the infrared sensor 2 are highly stable, avoiding data distortion caused by jitter or deviation of the infrared sensor 2 during the sliding process. The length and width of the slide rail are adjustable, which can adapt to the foot scanning needs of subjects with different body types, and at the same time support single-foot and double-foot modes. In this embodiment, the length of the slide rail is 210 cm and the width is 5 cm, and the width of the infrared sensor 2 slidably installed thereon is 10 cm. As Figure 3As shown. Through the above structural design, the modular design of the slide rail is realized, enabling the slide rail to be installed and debugged in a short time and facilitating its use in different places.

[0033] As Figure 10 As shown in the figure above, the stepper motor 6, the positioning status feedback device 7, and the infrared sensor 2 in the above-mentioned moving device 3 are all connected to the information processing platform 8. The staff can set the speed parameters through the information processing platform 8. The information processing platform 8 converts the set speed parameters into control signals and sends them to the stepper motor 6, controlling the stepper motor 6 to adjust its own rotation speed and number of steps according to the received control signals, so as to control the infrared sensor to move on the slide rail at a preset speed.

[0034] The subject stands according to the standardized human foot positioning marks on the infrared thermal information acquisition pad 1 to ensure that the sole of the foot completely fits the marked position. During the subject's standing, the subject remains stationary. The entire acquisition process lasts about 12 seconds. The heat emitted by the sole of the subject's foot and the heat emitted by different parts of the sole can be evenly conducted through the infrared thermal information acquisition pad 1, leaving corresponding sole infrared thermal information on the infrared thermal information acquisition pad 1. The staff controls the stepper motor 6 to start with the initial setting through the information processing platform 8. The stepper motor 6 drives the infrared sensor 2 to move from the heel to the toe on the slide rail of the moving device 3, sequentially capturing the sole infrared thermal information of the sole area. The infrared sensor 2 converts the acquired sole infrared thermal information into thermal infrared image data, and then generates a high-precision sole thermal infrared image. During the above acquisition process, the subject stands at the human foot positioning mark position on the infrared thermal information acquisition pad 1 to ensure the consistency of the foot position during each acquisition. This effectively avoids data errors caused by foot position deviation, makes the data collected at different times comparable, and provides a stable and reliable basis for subsequent analysis. The infrared sensor 2 scans the sole infrared thermal information left on the infrared thermal information acquisition pad 1 to further generate thermal infrared image data, avoiding abnormal heat accumulation or dissipation caused by uneven local pressure on the sole or individual foot shape differences, making the thermal signal of the entire sole presented in a more stable and uniform state, which helps the infrared sensor 2 capture comprehensive and accurate thermal infrared image data.

[0035] During the arch data acquisition process, the positioning status feedback device 7 continuously monitors the position information of the infrared sensor 2 on the slide rail, and the positioning status feedback device 7 feeds back the monitored position information to the information processing platform 8 in real time. After receiving this information, the information processing platform 8 compares it with the preset scanning path to determine whether the infrared sensor 2 moves along the predetermined trajectory. If the information processing platform 8 finds that there is a deviation between the actual position of the infrared sensor 2 and the preset path, it will immediately generate an adjustment instruction according to the magnitude and direction of the deviation and send it to the stepper motor 6. The stepper motor 6 controls the movement of the infrared sensor 2 according to the adjustment instruction. Through continuous monitoring and adjustment, the information processing platform 8 can ensure that the infrared sensor 2 moves smoothly along the preset path on the slide rail, thereby achieving full coverage scanning of the subject's plantar area and obtaining continuous and consistent image data. At the same time, the information processing platform 8 can also accurately control the infrared sensor 2 to collect data at specific positions according to the position information provided by the positioning status feedback device 7, improving the accuracy and reliability of the scanning.

[0036] After the acquisition is completed, the infrared sensor 2 transmits the thermal infrared image data to the information processing platform 8 in real time. As Figure 7 shown, the information processing platform 8 takes the thermal infrared image data collected by the infrared sensor 2 as the original data. After preprocessing the received thermal infrared image data, the information processing platform 8 forms an input sequence for data processing: , where represents the set of real numbers. It means that each element in is a real number. represents the number of pixels of the image in the vertical direction. represents the number of pixels of the image in the horizontal direction. 3 is the number of image data channels. The HRVManba (High-Resolution Visual State-Space Model) model is used in the information processing platform 8 for arch data operation. The HRVMamba model is a brand-new deep learning framework designed for dense prediction tasks. Its core innovation lies in the introduction of the Global-Local Visual State Space (GL-VSS) module and its optimized application in the high-resolution multi-scale framework. The following content will introduce the specific operations and implementations of the HRVMamba model in detail.

[0037] Existing Vision Mamba models usually generate single-scale, low-resolution features, resulting in serious information loss and making it difficult to capture the details and multi-scale variations required in dense prediction tasks. To address this issue, this method draws on the multi-resolution parallel design of HRNet and proposes a High-Resolution Vision State Space Model (HRVMamba) to better adapt to dense prediction tasks. As Figure 6 shown, the internal structure of the HRVMamba model includes: a linear projection layer (Linear), a depthwise separable convolution (DWConv), a SiLU activation function, an SS2D operation, a Conv convolution operation, and a Concat operation. By combining convolution and SS2D operations, local and global feature learning is enhanced. As Figure 5 shown, the data processing system of the HRVMamba model includes a sampling module (Stem) and a core processing unit (HR mamba block). The sampling module (Stem) receives the input sequence: , and the sampling module consists of two convolutions with a stride of 2 . The sampling module reduces the feature resolution of the input sequence: to 1 / 4. Subsequently, the data processing of the HRVMamba model performs feature extraction in four stages on the feature data processed by the sampling module. Each stage's feature stream includes the resolution of the previous stage and an additional lower resolution. Finally, the model generates four branches with feature sizes of: .

[0038] Previous studies have shown that performing convolution operations on larger feature maps in the early stages of the HRVMamba model is more conducive to visual feature extraction. Therefore, in the first stage, the same Bottleneck structure as HRNet is used to perform convolution operations on larger feature maps to extract preliminary visual features. In the subsequent three stages, the core processing unit (HR mamba block) uses a Global-Local Joint Vision State Space Module (GL-VSS). In each subsequent stage, the feature stream includes the resolution of the previous stage and an additional lower resolution. The GL-VSS module is built based on the VSS module of VMamba and combines multi-scale local convolution and global scanning modules, multi-scale depth convolution (MultiDW) blocks, and a feed-forward network (FFN) as feature extraction units, enhancing the network's local feature extraction ability.

[0039] The multi-scale local convolution and global scanning module includes a linear projection layer, a Local-Conv module, an SS2D module, a Hadamard product operation, etc. Among them, the Local-Conv module is a module we proposed to enhance the local feature extraction ability of SS2D. It mainly consists of two paths, including a LayerNorm layer, a standard convolution with a convolution kernel of 3, an SS2D operation, and a Concat operation. After passing through the linear projection layer, convolution operations are performed on half of the channels of the feature, and SS2D operations are performed on the other half of the channels. Finally, the output is obtained through the Concat operation. Combining the excellent local feature learning ability of convolution and the excellent learning ability of SS2D in the global receptive field can enhance the comprehensive feature learning ability of the network.

[0040]

[0041]

[0042]

[0043] Among them, LN represents the LayerNorm layer, and Cat is the Concat operation.

[0044] In addition, the Vision Mamba model usually processes images into token sequences through two-way or four-way scanning to establish a global receptive field. However, this method destroys the 2D spatial relationship and lacks an inductive bias for local features. To solve this problem, this method introduces the MultiDW module to replace the traditional depthwise separable convolution (DWconv), and uses multi-scale convolution kernels to capture local features of different scales, thereby enhancing the inductive bias of the model. Specifically, as shown in the following formula, the input feature X is first divided into G groups along the channel dimension. The g-th group of features is processed through a depth convolution with a convolution kernel size of , and then the resulting features are concatenated and shuffled to enhance the feature interaction between groups:

[0045]

[0046]

[0047] Among them, is the result of dividing the input feature into G groups along the channel dimension. Each group is the feature data of a part of the channels of the input feature ; is the output feature of the g-th group of features after passing through the depth convolution with a convolution kernel size of ; is the final output feature. First, are concatenated along the channel dimension (Concat), and then the concatenated features are shuffled to enhance the feature interaction between groups. Finally, the final output is obtained through the GELU activation function .

[0048] In the subsequent stage, the input features are divided into branches with different resolutions by the core processing unit and processed separately. Different branches represent features with different resolutions. For the high-resolution branch, through the multi-scale local convolution and global scanning module in the GL-VSS module, the local feature learning ability of convolution and the global receptive field learning ability of SS2D are utilized to enhance feature extraction. For the low-resolution branch, it is also processed through the GL-VSS module. The MultiDW module is mainly used to replace the traditional depthwise separable convolution, and multi-scale convolutional kernels are used to capture local features at different scales, enhancing the inductive bias of the model. The same operations are performed in the third and fourth stages, and finally the model generates the above four branches. The multi-scale feature fusion method continues the design of HRNet. Through a series of upsampling and downsampling modules, the features from different parallel branches are fused, so as to achieve the efficient utilization of multi-resolution and improve the visual feature modeling ability of the model in complex scenarios.

[0049] After being processed by the HRVMamba model, the arch feature analysis data is output. The arch feature analysis data includes the exact shape of the arch (such as whether it is flatfoot, high arch and its degree), the correlation analysis of the temperature distribution and pressure distribution in different areas of the sole, the evaluation of the arch health status, etc. Compare the arch feature analysis data output by the HRVMamba model with Figure 9 the different arch types in the described arch standard comparison chart for control analysis. The model determines the corresponding type of the arch condition in the standard comparison chart according to the correlation analysis of the sole heat distribution and pressure distribution. Combining the analysis output of the HRVMamba model and the standard comparison chart, and integrating the expert knowledge of professionals in sports medicine research institutions, the final diagnosis result is generated, providing a basis for the foot health assessment and sports training advice of the subject according to the diagnosis result.

[0050] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only to illustrate the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the claims of the present invention.

Claims

1. An arch health assessment system based on non-contact thermal infrared imaging, characterized in that It includes an infrared thermal information collection pad (1), an infrared sensor (2), a mobile device (3), and an information processing module (8); there are human foot positioning marks on the infrared thermal information collection pad (1), the mobile device (3) is placed on one side of the infrared thermal information collection pad (1), the infrared sensor (2) is slidably mounted on the mobile device (3), and the infrared sensor (2) is connected to an information processing station (8); the HRVManba model is used for data processing in the information processing station (8).

2. The arch health assessment system based on non-contact thermal infrared imaging according to claim 1, wherein A slide rail is provided on the mobile device (3), a stepper motor (6) is provided in the slide rail, the infrared sensor (2) is slidably arranged on the mobile device (3) through the stepper motor (6), and the stepper motor (6) is connected to the information processing station (8).

3. The arch health assessment system based on non-contact thermal infrared imaging according to claim 2, characterized in that, The lower end of the infrared sensor (2) is an optical tripod (5) with adjustable height and a spherical optical bracket (4) with adjustable angle, and the spherical optical bracket (4) is fixed to the upper end of the optical tripod (5).

4. The arch health assessment system based on non-contact thermal infrared imaging according to claim 3, characterized in that A positioning status feedback device (7) is also provided on the infrared sensor (2), and the positioning status feedback device (7) is connected to the information processing station (8).

5. The arch health assessment system based on non-contact thermal infrared imaging according to claim 1, characterized in that, The data processing of the HRVMamba model is mainly divided into a sampling module and a core processing unit. The sampling module consists of two 3×3 convolutions with a stride of 2. The sampling module is used to perform downsampling of the feature resolution on the image data monitored by the infrared sensor (2), and the core processing unit is used to perform visual feature extraction processing on the data.

6. The arch health assessment system based on non-contact thermal infrared imaging according to claim 5, wherein The visual feature extraction processing is specifically divided into four stages. In the first stage, the core processing unit uses the same BottleNeck structure as HRNet to perform convolution operations on the larger feature maps in the image data monitored by the infrared sensor (2) to extract preliminary visual features; in the subsequent three stages, the core processing unit uses a global-local joint visual state space module. The feature stream in each stage contains the resolution of the previous stage and an additional lower resolution, and finally generates features with four branches.

7. The arch health assessment system based on non-contact thermal infrared imaging according to claim 6, characterized in that, In the subsequent three stages, the input features are divided into branches with different resolutions by the core processing unit and processed separately. For the high-resolution branch, through the multi-scale local convolution and global scanning module in the GL-VSS module, using the local feature learning ability of convolution and the global receptive field learning ability of SS2D, the feature extraction is enhanced; for the low-resolution branch, it is also processed through the GL-VSS module. The traditional depthwise separable convolution is replaced by the MultiDW module, and multi-scale convolutional kernels are used to capture local features at different scales, enhancing the inductive bias of the model, and finally generating features with four branches.

8. The arch health assessment system based on non-contact thermal infrared imaging according to claim 7, characterized in that, The method of replacing the traditional depthwise separable convolution with the MultiDW module and using multi-scale convolutional kernels to capture local features at different scales and enhance the inductive bias of the model is specifically shown in the following formula Among them, is the result after dividing the input features into G groups along the channel dimension, and each group is the feature data of a part of the channels of the input features ; is the feature of the g-th group after depth convolution with a convolution kernel size of ; is the final output feature.

9. A method for evaluating the health of the foot arch based on non-contact thermal infrared imaging, based on a system for evaluating the health of the foot arch based on non-contact thermal infrared imaging according to any one of claims 1 to 8, characterized in that, The subject stands at the human foot positioning mark position of the infrared thermal information collection pad (1), and the infrared thermal information of the subject's sole is left on the infrared thermal information collection pad (1). The information processing station (8) controls the infrared sensor (2) on the mobile device to move from the subject's heel to the toe. The infrared sensor (2) scans the infrared thermal information of the sole to form thermal infrared image data. The infrared sensor (2) transmits the thermal infrared image data to the information processing station (8). After the HRVManba model in the information processing station (8) processes the data and extracts features from the thermal infrared image data, arch feature analysis data is generated.

10. The method for arch health assessment based on non-contact thermal infrared imaging according to claim 9, characterized in that, The data processing of the HRVMamba model is mainly divided into a sampling module and a core processing unit. The sampling module consists of two 3×3 convolutions with a stride of 2, and the sampling module is used to reduce the feature resolution of the data to 1 / 4; the core processing unit is used to perform visual feature extraction processing on the data, and finally generates four branches, and the feature sizes are respectively: .

Images