Intelligent leather touch evaluation system and method based on mechanical spectrum feature fusion
By using an intelligent evaluation system based on the fusion of mechanical spectrum features, combined with optical imaging and deep learning, the problems of accuracy and efficiency in leather touch evaluation in traditional methods have been solved, and efficient and objective leather touch detection has been achieved.
Patent Information
- Application Number
- CN202511077885.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-01
- Publication Date
- 2025-11-14
AI Technical Summary
Traditional methods are insufficient to accurately and objectively evaluate the tactile properties of flexible materials such as leather, especially their dynamic mechanical response. Furthermore, existing instruments are cumbersome and inefficient, failing to meet the needs of large-scale industrial testing.
An intelligent evaluation system based on mechanical spectrum feature fusion is adopted, which combines dynamic mechanical loading, optical imaging and dual-task deep learning. The optical imaging module collects leather deformation data, the mechanical loading module collects mechanical data, and the dual-task neural network model is used for feature fusion and prediction.
It enables non-contact, multi-dimensional objective prediction of leather touch, improves the automation and objectivity of the evaluation, reduces subjective errors, and ensures the reliability and integrity of the test results.
Smart Images
Figure CN120953687A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and optical detection technology, and in particular to a smart evaluation system and method for leather tactile sensation based on the fusion of mechanical spectrum features. Background Technology
[0002] Flexible materials such as leather, rubber, and composite fibers are widely used in smart manufacturing, consumer goods, and automotive interiors. Their softness, elasticity, and tactile properties directly affect the quality evaluation and user experience. Tactile sensation, as an important indicator of material texture, has traditionally relied on human experience, involving subjective judgment through methods such as pressing, sliding, microscopic observation, and visual analysis. These methods are inefficient, unstable, and easily influenced by differences in the experience of testing personnel, making them unsuitable for large-scale, standardized industrial testing needs.
[0003] Due to differences in manufacturing processes and internal fiber structures, natural materials such as leather exhibit significant variations and non-uniformity in their dynamic mechanical responses. This makes it difficult for traditional methods to accurately capture their complex deformation characteristics, limiting the precision and objectivity of tactile evaluation. While some automated testing methods use image analysis technology to identify surface features, their ability to assess leather tactile feel at the mechanical level is limited, failing to accurately reflect the material's dynamic deformation characteristics under stress.
[0004] Existing instrument testing methods based on a single physical index (such as a softness tester or an elastic modulus tester) can obtain some tactile parameters, but they are difficult to comprehensively and quantitatively characterize multi-dimensional tactile indicators such as softness, elasticity, and tactile feel. Moreover, they often require the use of multiple instruments in combination, which is cumbersome and inefficient. Summary of the Invention
[0005] To overcome the problems of low efficiency, high subjectivity, reliance on multiple instruments, and cumbersome testing process in existing technologies for manual tactile evaluation, this invention provides an intelligent evaluation system and method for leather tactile sensation based on mechanical spectrum feature fusion. This method achieves non-contact, multi-dimensional objective prediction of leather tactile sensation through dynamic mechanical loading, optical imaging, and dual-task deep learning modeling, effectively improving the automation level and objectivity of tactile evaluation and reducing subjective errors.
[0006] To achieve the above objectives, this invention provides a leather tactile intelligence evaluation system based on mechanical spectrum feature fusion, comprising:
[0007] An optical imaging module, including a line laser emitter and a high-speed camera, is used to project light stripes onto the leather surface and acquire dynamic deformation data of the light stripes;
[0008] The mechanical loading and sensing module includes a linear reciprocating force application mechanism, a high-precision force sensor, and a contact rod. The linear reciprocating force application mechanism drives the contact rod to perform a loading-retraction force application operation on the leather, and the high-precision force sensor synchronously collects mechanical data during the force application process.
[0009] The mechanical fixing module includes a platform, an adjustable clamp, and an airbag adsorption device for fixing leather samples.
[0010] The data processing and analysis module is used to perform data alignment, feature extraction, neural network inference, and result storage.
[0011] Preferably, the sampling frame rate of the high-speed camera in the optical imaging module is 250 FPS.
[0012] Preferably, the linear reciprocating force application mechanism is a stepper motor.
[0013] A method for intelligent evaluation of leather tactile sensation based on mechanical spectrum feature fusion, applied to the aforementioned intelligent evaluation system for leather tactile sensation based on mechanical spectrum feature fusion, comprises the following steps:
[0014] S1. The optical imaging module projects linear light stripes onto the leather surface and collects dynamic deformation data to form a deformation-time series. At the same time, the linear reciprocating force application mechanism of the mechanical loading and sensing module drives the touch rod to perform loading-retraction force application operations on the leather. The real-time force value of the touch rod is collected synchronously through a high-precision force sensor to form force-time series data.
[0015] S2. The deformation-time series data and force-time series data are time-aligned using the data processing and analysis module to form a deformation-mechanical response dataset.
[0016] S3. Extract the spectral features of time-varying mechanical parameters and frequency-domain system functions based on the deformation-mechanical response dataset, and fuse them to form mechanical spectrum features;
[0017] S4. Input the mechanical spectrum features into the dual-task neural network model, and output the tactile evaluation results of the leather through collaborative learning guided by the main task regression prediction and the auxiliary task classification.
[0018] Preferably, S1 further includes: adjusting the position and angle of the optical imaging module so that the line laser covers the target area of the leather; fixing the leather by the adjustable clamp of the mechanical fixing module and the airbag adsorption device to suppress vibration during the force application process and ensure that the force application point is at the edge of the line laser coverage area; and installing the sensing head of the high-precision force sensor at the end of the linear reciprocating force application mechanism.
[0019] Preferably, S2 specifically includes: extracting the center line of the light stripes in each frame image using an image processing algorithm, calculating the offset of the light stripes during the force application process, converting the pixel coordinates into real physical displacement coordinates, and obtaining deformation-time series data; and using an interpolation algorithm to perform time alignment between the deformation-time series and the force-time series.
[0020] Preferably, S3 specifically includes:
[0021] S3.1 Based on the aforementioned dynamic differential equation, a sliding window least squares method is used for time-domain fitting. By iteratively adjusting the parameters and minimizing the error objective function, three time-varying mechanical parameters a(t), b(t), and c(t) are obtained. The dynamic differential equation is expressed as:
[0022]
[0023] Where F(t) represents force-time series data, u(t) represents deformation-time series data, and a(t), b(t), and c(t) are time-varying mechanical parameters related to the dynamic characteristics, damping, and stiffness of the leather deformable body system in response to acceleration, respectively.
[0024] S3.2. Perform Fourier transform on the applied force signal F(t) and the deformation displacement signal u(t) to obtain the frequency domain representations F(ω) and U(ω). Calculate the frequency domain transfer function as follows:
[0025]
[0026] Then, the amplitude spectrum |H(ω)| and phase spectrum φ(ω) of the frequency domain transfer function are extracted;
[0027] S3.3 For each sampling point in the laser contour on the leather, extract the time-varying parameter vector [a(t), b(t), c(t)], the frequency-domain feature vector |H(ω)| and arg[H(ω)], and the force-induced deformation response vector [F(t), u(y)]. After preprocessing and normalizing the above vectors, they are directly concatenated to form the composite feature vector of each sampling point, expressed as:
[0028] [a(t), b(t), c(t), |H(ω)|, arg[H(ω)], F(t), u(t)]
[0029] S3.4. Centered on the point of force application, collect several sampling points evenly along the laser contour. The composite feature vectors of several sampling points together constitute the overall mechanical spectrum characteristics of the leather.
[0030] Preferably, in S4, the auxiliary task of the dual-task neural network model is the classification task of the leather region location. By classifying the region location labels of the leather samples, the objective correlation between the region location and the material mechanical properties is used to guide the learning of the main task. The main task is the regression prediction of the three-dimensional tactile index of the leather, which includes softness, elasticity, and feel.
[0031] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0032] 1. Achieve non-destructive testing and high-precision evaluation: Optical imaging is used to collect dynamic deformation data of leather, avoiding damage to the sample caused by contact testing. At the same time, multi-dimensional data collection and precise processing ensure the reliability of the evaluation results.
[0033] 2. Achieve complementary and synergistic effects of multimodal features: By integrating time-varying mechanical parameters in the time domain and system function spectrum features in the frequency domain to construct mechanical spectrum features, the dynamic mechanical response of leather during the stress process can be fully captured, overcoming the limitations of single features in representing complex tactile sensations and improving the completeness of feature description.
[0034] 3. Optimize evaluation performance through dual-task learning: Introduce leather region location labels as an auxiliary task, which are trained in conjunction with the main task of tactile index regression prediction. This guides the model to learn the physical consistency of leather, effectively reducing the impact of subjective label noise and improving the model's generalization ability and the objectivity of the evaluation results.
[0035] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is an architecture diagram of a leather tactile intelligent evaluation system based on mechanical spectrum feature fusion according to an embodiment of the present invention;
[0038] Figure 2 This is a schematic diagram of a leather tactile intelligent evaluation method based on mechanical spectrum feature fusion according to an embodiment of the present invention;
[0039] Figure 3 This is a schematic diagram of the optical imaging module, mechanical loading and sensing module, and mechanical fixing module according to an embodiment of the present invention.
[0040] Figure 4 This is a schematic diagram illustrating the design of the main task and auxiliary task in the dual-task neural network of this invention.
[0041] Figure Labels
[0042] 1. Line laser emitter; 2. High-speed camera; 3. Linear reciprocating force application mechanism; 4. Force sensor; 5. Leather to be tested. Detailed Implementation
[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0045] Example
[0046] like Figure 1 and Figure 3 As shown, the method of this invention relies on a leather tactile intelligent evaluation system composed of multiple modules, including an optical imaging module, a mechanical loading and sensing module, a mechanical fixing module, and a data processing and analysis module. Wherein:
[0047] The optical imaging module consists of a line laser emitter 1 and a high-speed camera 2. The line laser emitter 1 projects line laser stripes onto the leather surface, and the high-speed camera 2 captures the dynamic deformation of the light stripes during the application of force in real time, thereby forming a continuous sequence of image frames.
[0048] The mechanical loading and sensing module includes a linear reciprocating force application mechanism 3, a high-precision force sensor 4, and a contact rod. The linear reciprocating force application mechanism 3 uses a stepper motor, which drives the contact rod to perform loading-retraction motion along a fixed path. At the same time, the force sensor 4 records the pressure between the contact and the leather surface in real time during this process, generating corresponding force-time series data.
[0049] The mechanical fixing module consists of a platform, an adjustable clamp, and an airbag adsorption device. The clamping method is adjusted according to the size of the leather, and the airbag adsorption ensures that the leather remains stable during loading, avoiding mechanical vibration from interfering with optical imaging and force data acquisition.
[0050] The data processing and analysis module runs on the PC and completes deformation and force data processing, mechanical feature extraction, and deep learning inference tasks.
[0051] Based on the above system, authenticity can be determined by the touch of the leather, such as... Figure 2 and Figure 4 As shown, the intelligent evaluation method for leather touch proposed in this invention includes the following steps:
[0052] S1. First, set up the equipment and initialize the system. Fix the positions of the line laser emitter 1 and high-speed camera 2 in the optical imaging module and determine their position parameters. Adjust the angle and position of the line laser so that its light stripe covers the target detection area of the leather 5 to be tested. Start the high-speed camera 2 and adjust the lens focal length and exposure time to ensure that the deformation of the light stripe on the leather surface as force is applied can be clearly captured.
[0053] Next, the leather to be tested 5 is placed on the platform of the mechanical fixing module. The leather to be tested 5 consists of several genuine leather samples and similar-looking imitation samples. The adjustable clamp specifications are adjusted according to the size of the leather, and the airbag adsorption device is used to keep the leather surface flat and stable, avoiding mechanical vibration caused by force from interfering with optical imaging. A suitable flat-head adapter component is selected, and the sensing contact of the force sensor 4 is installed on the flat end of the contact rod.
[0054] Next, leather sample data acquisition begins. The force application program is initiated, with a stepper motor driving a contact rod along a loading-retraction path to apply dynamic mechanical input to the leather. During the force application process, the leather surface deforms, typically within 2 cm. A high-precision force sensor 4 with a fixed sampling rate synchronously records the real-time force value applied by the contact, generating force-time series data F(t), which is then transmitted to the PC. A high-speed camera 2 continuously records the changes in the light stripes on the leather surface at a frame rate of 250 FPS.
[0055] S2. Data Processing and Alignment. Image processing algorithms are used to extract the center line of the light stripes in each frame of the image. The offset of the light stripes during the applied force process is calculated, and the pixel coordinates are converted into actual physical displacements to obtain deformation-time series data u(t), representing the deformation signal of the leather and describing its deformation over time. Interpolation algorithms are used to align the force-time series and deformation-time series data, ensuring that the mechanical input is aligned with the deformation response time of the leather surface.
[0056] S3. Extract time-domain time-varying mechanical parameters and frequency-domain system function spectral features from the deformation-mechanical response dataset, and fuse them to form mechanical spectrum features; specifically including:
[0057] S3.1. Regarding time-domain feature extraction, based on the dynamic differential equation, a sliding window least squares method is used for time-domain fitting. By iteratively adjusting the parameters and minimizing the error objective function, three time-varying mechanical parameters a(t), b(t), and c(t) are obtained. The dynamic differential equation is expressed as:
[0058]
[0059] Where F(t) represents force-time series data, u(t) represents deformation-time series data, and the term containing u(t) in the equation is the displacement term. The term in question is the velocity term. The term in question is the acceleration term. a(t), b(t), and c(t) are time-varying mechanical parameters reflecting the dynamic characteristics, damping, and stiffness-related properties of the leather deformation system in response to acceleration, respectively, and the time-domain mechanical characteristics reflecting the properties of leather.
[0060] S3.2. Regarding frequency domain feature extraction, the applied force signal F(t) and the deformation displacement signal u(t) are subjected to Fourier transform processing to obtain the frequency domain representations F(ω) and U(ω). The frequency domain transfer function is calculated as follows:
[0061]
[0062] H(ω) describes the system's response characteristics to the input force signal at various frequencies.
[0063] F(ω) is the input, U(ω) is the output, and H(ω) is the dynamic characteristic of the leather material. The relationship between these three in the frequency domain reveals how the leather responds to forces of different frequencies, reflecting its tactile essence such as elasticity, damping, and inertia. The magnitude and phase of the frequency domain transfer function are calculated, yielding the amplitude spectrum |H(ω)| and the phase spectrum arg[H(ω)]. These two are then concatenated to form the frequency domain mechanical characteristic vector of the leather. The amplitude spectrum represents the gain of each frequency component, reflecting the amplification or attenuation effect of the system on inputs of different frequencies. The phase spectrum represents the deformation displacement at different points on the laser profile of the leather line, indicating the lag or lead characteristics of the response to applied force in time.
[0064] S3.3. Fusion of temporal and frequency domain features: Specifically, for each sampling point in the laser contour on the leather, extract the time-varying parameter vector [a(t), b(t), c(t)], the frequency domain feature vectors |H(ω)| and arg[H(ω)], and the force-induced deformation response vector [F(t), u(t)]. After preprocessing and normalizing the above vectors, they are directly concatenated to form the composite feature vector of each sampling point, expressed as:
[0065] [a(t), b(t), c(t), |H(ω)|, arg[H(ω)], F(t), u(t)]
[0066] S3.4. Centered on the point of force application, 64 sampling points are uniformly collected along the laser contour. The composite feature vectors of the 64 sampling points together constitute the overall mechanical spectrum characteristics of the leather. This multi-level fusion feature can more comprehensively characterize the mechanical behavior of the leather.
[0067] S4. To reduce the subjectivity of manually created labels used for regression tasks, the mechanical spectrum features are input into a dual-task neural network model. Through collaborative learning guided by the main task regression prediction and the auxiliary task classification, the tactile evaluation results of the leather are output.
[0068] The primary task is regression prediction of three-dimensional tactile indices of leather (softness, elasticity, and feel), while the auxiliary task is classification prediction of leather region location labels. The auxiliary task labels, representing different regions of multiple leather samples (e.g., center, left, right), are randomly encoded, such as 0, 5, and 10, indicating location labels from different regions of the same piece of leather. Although the region labels are randomly distributed, they essentially reflect the physical consistency of the same leather in different regions. Therefore, leather type information is not directly used as a training objective. Instead, the classification training of region labels guides the model to automatically learn the mechanical properties of different regions of the same leather, driving the model to consider the physical consistency of leather in different regions. This suppresses the subjectivity of human sensory rating labels and improves the model's objectivity and generalization ability.
[0069] Training was conducted using 80% data for training and 20% data for testing. To verify whether the model truly learned the objective mechanical properties of leather, rather than simply fitting subjective scores, a "leather consistency" verification method was proposed: the feature vectors output by the fully connected layer for all samples in the test set were calculated, and the average similarity between label categories at each location was analyzed using cosine similarity. The verification results showed that labels from different regions (within the same leather) belonging to the same piece of leather had high average feature similarity; labels from different leathers had lower average similarity, indicating that the model's learning focused on the objective factors of the mechanical properties of leather belonging to the same piece, rather than relying solely on subjectively biased label training.
[0070] The trained neural network can directly take the collected mechanical spectrum features as input and output the corresponding three-dimensional tactile index prediction value, accurately and objectively reflecting the softness, elasticity and feel of the leather surface.
[0071] Users view the detection results on the display terminal. The detection results are uploaded to the database to update the leather material tactile feature library and model parameters, and to continuously optimize the dual-task neural network model.
[0072] The remaining technical features in the above embodiments can be flexibly selected by those skilled in the art to meet different specific practical needs according to actual circumstances. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims. In the above description, numerous specific details have been set forth to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to implement the present invention. In other instances, to avoid obscuring the present invention, well-known techniques, such as specific construction details, operating conditions, and other technical conditions, have not been specifically described.
[0073] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A smart evaluation system for leather tactile sensation based on the fusion of mechanical spectrum features, characterized in that, include: An optical imaging module, including a line laser emitter and a high-speed camera, is used to project light stripes onto the leather surface and acquire dynamic deformation data of the light stripes; The mechanical loading and sensing module includes a linear reciprocating force application mechanism, a high-precision force sensor, and a contact rod. The linear reciprocating force application mechanism drives the contact rod to perform a loading-retraction force application operation on the leather, and the high-precision force sensor synchronously collects mechanical data during the force application process. The mechanical fixing module includes a platform, an adjustable clamp, and an airbag adsorption device for fixing leather samples. The data processing and analysis module is used to perform data alignment, feature extraction, and storage of neural network inference results.
2. The intelligent evaluation system for leather tactile sensation based on mechanical spectrum feature fusion according to claim 1, characterized in that, The sampling frame rate of the high-speed camera in the optical imaging module is 250 FPS.
3. The intelligent evaluation system for leather tactile sensation based on mechanical spectrum feature fusion according to claim 1, characterized in that, The linear reciprocating force application mechanism uses a stepper motor.
4. A method for intelligent evaluation of leather tactile sensation based on mechanical spectrum feature fusion, applied to the intelligent evaluation system for leather tactile sensation based on mechanical spectrum feature fusion as described in claims 1-3, characterized in that, The steps are as follows: S1. The optical imaging module projects linear light stripes onto the leather surface and collects dynamic deformation data to form a deformation-time series. At the same time, the linear reciprocating force application mechanism of the mechanical loading and sensing module drives the touch rod to perform loading-retraction force application operations on the leather. The real-time force value of the touch rod is collected synchronously through a high-precision force sensor to form force-time series data. S2. The deformation-time series data and force-time series data are time-aligned through the data processing and analysis module to form a deformation-mechanical response dataset. S3. Extract time-domain time-varying mechanical parameters and frequency-domain system function spectral features based on the deformation-mechanical response dataset, and fuse them to form mechanical spectrum features; S4. Input the mechanical spectrum features into the dual-task neural network model, and output the tactile evaluation results of the leather through collaborative learning guided by the main task regression prediction and the auxiliary task classification.
5. The intelligent evaluation method for leather tactile sensation based on mechanical spectrum feature fusion according to claim 4, characterized in that, S1 also includes: adjusting the position and angle of the optical imaging module so that the line laser covers the target area of the leather; fixing the leather with the adjustable clamp of the mechanical fixing module and the airbag adsorption device to suppress vibration during the force application process and ensure that the force application point is at the edge of the line laser coverage area; and installing the sensing head of the high-precision force sensor at the end of the linear reciprocating force application mechanism.
6. The intelligent evaluation method for leather tactile sensation based on mechanical spectrum feature fusion according to claim 4, characterized in that, S2 specifically includes: extracting the center line of the light stripes in each frame of the image through an image processing algorithm, calculating the offset of the light stripes during the force application process, converting the pixel coordinates into real physical displacement coordinates, and obtaining deformation-time series data; and using an interpolation algorithm to perform time alignment between the deformation-time series and the force-time series.
7. The intelligent evaluation method for leather tactile sensation based on mechanical spectrum feature fusion according to claim 6, characterized in that, S3 specifically includes: S3.1 Based on the aforementioned dynamic differential equation, a sliding window least squares method is used for time-domain fitting. By iteratively adjusting the parameters and minimizing the error objective function, three time-varying mechanical parameters a(t), b(t), and c(t) are obtained. The dynamic differential equation is expressed as: Where F(t) represents force-time series data, u(t) represents deformation-time series data, and a(t), b(t), and c(t) are time-varying mechanical parameters related to the dynamic characteristics, damping, and stiffness of the leather deformable body system in response to acceleration, respectively. S3.
2. Perform Fourier transform on the applied force signal F(t) and the deformation displacement signal u(t) to obtain the frequency domain representations F(ω) and U(ω). Calculate the frequency domain transfer function as follows: Then, the amplitude spectrum |H(ω)| and phase spectrum φ(ω) of the frequency domain transfer function are extracted; S3.3 For each sampling point in the laser contour on the leather, extract the time-varying parameter vector [a(t), b(t), c(t)], the frequency-domain feature vector |H(ω)| and arg[H(ω)], and the force-induced deformation response vector [F(t), u(t)]. After preprocessing and normalizing the above vectors, they are directly concatenated to form the composite feature vector of each sampling point, expressed as: [a(t), b(t), c(t), |H(ω)|, arg[H(ω)], F(t), u(t)] S3.
4. Centered on the point of force application, collect several sampling points evenly along the laser contour. The composite feature vectors of several sampling points together constitute the overall mechanical spectrum characteristics of the leather.
8. The intelligent evaluation method for leather tactile sensation based on mechanical spectrum feature fusion according to claim 4, characterized in that: In S4, the auxiliary task of the dual-task neural network model is the classification task of the leather region location; the main task is the regression prediction of the three-dimensional tactile indicators of the leather, which include softness, elasticity, and feel.