Weak pressure measurement method based on photoelastic effect and image recognition

Abstract

The invention discloses a weak pressure measurement method based on a photoelastic effect and image recognition. The method comprises the following steps: S1, acquiring an initial image of a photoelastic sensing layer in a non-pressure state; s2, acquiring a measurement image of the photoelastic sensing layer in a pressure applying state of the to-be-measured object; s3, cutting the initial image and the measurement image, and reserving a stripe change area; s4, extracting grey-scale maps of RGB channels of the initial image and the measurement image, and performing subtraction to obtain a difference image; and S5, inputting the difference image into the pressure prediction model, and calculating to obtain a pressure estimation value or a pressure interval corresponding to the to-be-measured object. Compared with the prior art, the method has the advantages that a convolutional neural network (CNN) image recognition technology is creatively combined with a photoelastic mechanical effect, an end-to-end mapping model from optical interference image features to pressure information is established, and low-cost and rapid pressure (weight) interval measurement is realized.

Description

Technical Field

[0001] This invention relates to a method for measuring weak pressure based on photoelasticity and image recognition, belonging to the field of pressure testing technology for small objects. Background Technology

[0002] Pressure sensing technology is a core component of industrial automation and precision measurement, and its performance directly affects production efficiency and detection accuracy. In recent years, optical pressure sensors have gradually become a research hotspot due to their advantages such as resistance to electromagnetic interference and corrosion. Among them, sensing schemes based on photoelasticity have shown unique application potential because they directly convert mechanical stress into changes in optical signals.

[0003] However, existing technologies still have significant limitations in practical applications. Existing literature (CN110631746A) discloses a pressure detection component, detection method, and terminal. This scheme uses a polyurethane elastomer as the sensing element, converting pressure into an electrical signal through polarized light phase modulation combined with a photodiode. While this scheme has anti-interference advantages in mobile terminal touch scenarios, it relies on complex circuit signal processing and struggles to interpret the complex optical patterns generated by photoelastic materials, limiting its application in high-precision, small-range scenarios. Existing literature (CN108562385A) discloses a real-time force feedback device and method for micro-manipulation based on photoelastic effects, proposing to utilize the photoelastic effect of gelatin-glycerol composite materials to calculate micro-Newton-level forces by observing the order of birefringence fringes. Although this scheme can be applied to micro-nano scenarios such as cell manipulation, it requires manual interpretation of fringes and relies on empirical formulas, resulting in low efficiency and poor real-time performance. Furthermore, the insufficient environmental stability of gelatin materials makes it difficult to meet industrialization requirements.

[0004] Based on the above, it can be seen that existing pressure sensing technologies have the following common problems: First, high-sensitivity measurement often relies on complex photoelectric conversion and signal processing systems, resulting in complex system structures, high costs, and difficulty in achieving rapid and automated analysis while maintaining high accuracy; Second, existing technologies have poor adaptability to interference factors such as ambient light and changes in material tension, and lack stability; Third, most existing solutions still rely on special materials or complex preparation processes, which limits their promotion and application in actual industrial scenarios. Summary of the Invention

[0005] Based on the above, the present invention provides a method for measuring weak pressure based on photoelasticity and image recognition to overcome the shortcomings of the prior art.

[0006] The technical solution of this invention is: a weak pressure measurement method based on photoelastic effect and image recognition, comprising: S1 acquires an initial image of the photoelastic sensing layer under no-pressure conditions; S2 acquires a measurement image of the photoelastic sensing layer under pressure applied to the object under test; S3 performs cropping on the initial image and the measured image, preserving the areas of stripe variation; S4 extracts the grayscale images of the RGB three channels of the initial image and the measurement image and calculates the difference to obtain the difference image; S5 inputs the differential image into the pressure prediction model to calculate the estimated pressure value or pressure range corresponding to the object under test.

[0007] In one example, the photoelastic sensing layer is mounted on a measuring bracket, a black placement box is provided on the photoelastic sensing layer, a light source is provided below the photoelastic sensing layer, and a camera is provided above the photoelastic sensing layer. The camera is connected to a computer via a data cable. The light source emits light onto the photoelastic sensing layer, and the camera captures a deformation image of the photoelastic sensing layer as the measurement image.

[0008] In one example, the photoelastic sensing layer is composed of two photoelastic material films orthogonally stacked and fixed on the measuring bracket. The light source is provided with a first polarizer, and the camera is provided with a second polarizer. The second polarizer and the first polarizer are at a preset angle.

[0009] In one example, the photoelastic material film is a PET film.

[0010] In one example, the second polarizer is positioned at a 45-degree angle to the first polarizer.

[0011] In one example, the stress prediction model is constructed as follows: S51 collects initial fringe images of the photoelastic sensing layer under no-pressure conditions; S52 collects measurement stripe images of the photoelastic sensing layer under pressure applied by objects of different weights, and the measurement stripe images are marked with weight. S53 Change the initial conditions and repeat step S52. The initial conditions include a slight change in the tension of the photoelastic sensing layer and ambient light. S54 inputs the collected initial stripe images and measured stripe images into the CNN model, and uses the corresponding pressure value or pressure range as the output to train the CNN model; S55 validates the CNN model, and once the accuracy reaches the predetermined requirements, the model is used as a stress prediction model.

[0012] The technical principle of this invention is as follows: In photoelasticity analysis, the observed interference fringes originate from the stress state within the material. When unpolarized light emitted from a light source passes through a polarizer, it forms plane-polarized light with a single vibration direction. When this plane-polarized light is incident on a transparent photoelastic material under stress, due to the birefringence effect (i.e., stress-induced photoelasticity) within the material, the light wave components vibrating along different principal stress directions will produce different propagation velocities. This velocity difference causes the two orthogonal vibration component waves to accumulate a defined optical path difference after penetrating the material. When the polarized light containing this optical path difference information subsequently passes through an analyzer whose polarization axis is at a 45-degree angle to the polarizer (i.e., in the analysis position), only the light vibration component along the polarization axis of the analyzer can pass through and satisfy the coherence condition, thus causing light interference. When the total phase difference at a specific point (proportional to the optical path difference) is an integer multiple of the wavelength of a specific monochromatic light, the interference of that monochromatic light exhibits constructive interference, manifesting as bright fringes; otherwise, it is destructive interference. Given that photoelastic materials are typically color-sensitive to this effect, and that white light sources encompass the entire visible spectrum, the interference pattern observed after the analyzer ultimately appears as a series of color bands characterizing a constant phase difference (or equivalently, a constant principal stress difference), known as isochromatic lines. The isochromatic line pattern visually depicts the distribution of the internal stress field of the material, forming the core physical basis for quantitative stress analysis (including the measurement of internal stress in materials such as glass) based on photoelastic methods.

[0013] The beneficial effects of this invention are as follows: This invention is the first to apply the stress-optical effect of photoelastic materials to the field of measuring weak pressure / weight. It abandons the traditional electrical signal conversion path and instead utilizes the visualized optical stripe patterns exhibited after material deformation under stress as the core sensing signal. Furthermore, it innovatively applies a convolutional neural network (CNN) model to automatically analyze the complex pressure information contained in the photoelastic image, learning and identifying subtle differences in stripe patterns under different pressure levels, thus achieving a mapping from optical image to pressure estimation. This is particularly suitable for measuring small, lightweight objects or weak pressure scenarios that are difficult to accurately measure with traditional sensors. The non-contact measurement also reduces potential interference or damage to the object being measured or the sensing element. In addition, because the photoelastic stripes themselves have the characteristic of visualizing stress, this invention can provide intuitive information for understanding pressure distribution.

[0014] Specifically, in this invention, the birefringence effect is enhanced by orthogonally stacked PET film layers. Combined with RGB three-channel differential image processing, the contrast of optical stripes can be significantly improved, effectively eliminating the influence of ambient light fluctuations and enabling automated measurement of weak pressure (weight). The sensitivity is significantly improved compared to existing electrical signal conversion schemes. The stripe image is automatically analyzed using a CNN pressure prediction model. Utilizing the powerful image feature extraction and pattern recognition capabilities of CNN, the complex information of photoelastic stripes can be processed to achieve automated estimation. Compared to traditional manual stripe interpretation and formula calculation, the processing speed reaches milliseconds, meeting real-time feedback requirements. Compared with existing technologies, this invention innovatively combines convolutional neural network (CNN) image recognition technology with photoelasticity effects to establish an end-to-end mapping model from optical interference image features to pressure information, achieving low-cost and rapid pressure (weight) range estimation. This invention eliminates complex optical systems, requiring no laser light source, monochromatic light source, precision lens group, waveplate, or complex optical path calibration device. This invention simplifies the electronic system, eliminating reliance on strain gauges, piezoelectric elements, and their associated signal conditioning circuits (such as signal amplifiers and ADC conversion modules). This invention reduces the requirements for sensing hardware, eliminating the need for high-precision spectrometers, fiber optic sensors, interferometers, or phase measurement equipment. This invention does not require the separate fabrication of special functional thin films (such as sensitive films containing noble metal nanoparticles or customized grating structure films) or the use of expensive materials. Attached Figure Description

[0015] Figure 1 This is a flowchart of the pressure measurement method; Figure 2 This is a schematic diagram of the pressure measuring device. Figure 3 This is a schematic diagram of the CNN model structure; Figure 4 This demonstrates the training performance of the model on a dataset. Explanation of reference numerals in the attached figures: 1. Object to be tested, 2. Black placement box, 3. Photoelastic sensing layer (double layer), 4. Measurement bracket, 5. Light source (with polarizer), 6. Camera (with polarizer), 7. Camera bracket, 8. Data cable, 9. Computer. Detailed Implementation

[0016] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0017] Please see Figure 1 This embodiment provides a method for measuring weak pressure based on photoelasticity and image recognition, comprising: S1 acquires the initial image of the black placement box 2 under no-pressure conditions.

[0018] In this embodiment, the black placement box 2 is composed of two PET films orthogonally stacked and fixedly mounted on the measuring bracket 4. The black placement box 2 is used to place the object to be measured 1. A light source 5 is located below the black placement box 2, and a first polarizer is mounted on the light source 5. A camera 6 is located above the black placement box 2, mounted on a camera bracket 7 with its lens pointing vertically downwards. A second polarizer is mounted on the camera 6, which is set at a 45-degree angle to the first polarizer. The camera 6 is connected to a computer 9 via a data cable 8, and can send the captured images to the computer 9 for analysis and processing. During operation, light is emitted from the light source 5 towards the black placement box 2, and the camera 6 captures images of the deformation of the black placement box 2 from above.

[0019] The following is a control experiment from the early stages of the study: 1. Comparative Experiments on Thin Film Structures Single-layer film structure: A single-layer PET film was used. It was found that limited stripe or color changes could only be observed under significant deformation conditions (large load). The specific performance was limited by the film thickness: thick films deformed very little, and the optical effect was weak; although thin films deformed more, the cumulative optical path difference was insufficient, and the phenomenon was also not significant.

[0020] Double-layer film structure (randomly stacked): Two layers of PET film are stacked at random angles in a tightly bonded manner. Clearer photoelastic stripes and color variations were observed compared to the single-layer film structure. The stripe morphology and color distribution are dependent on the relative angle of the optical axes of the two films.

[0021] Double-layer film structure (orthogonal optical axes): A double-layer PET film is used, with the optical axes of the two films precisely set to be perpendicular (90 degrees) to each other. The clearest interference fringes with the highest color contrast were observed. This configuration exhibits optimal ease of operation and structural stability under experimental conditions, and the phenomenon can be stably reproduced.

[0022] Based on the above, it can be seen that a single-layer film structure is difficult to meet the requirements of high-sensitivity optical response for weak pressure measurement. A double-layer film structure (with mutually orthogonal optical axes) has the advantages of high-contrast stripes and ease of operation. Therefore, a double-layer orthogonal optical axis PET film structure was selected as the standard sensing layer configuration of this invention.

[0023] 2. Polarizer Angle Configuration Experiment Polarizer angle test relative to the thin film optical axis: When the polarization axis of the polarizer / analyzer forms a 45-degree angle with the thin film optical axis, interference fringes and their dynamic response to pressure changes can be clearly observed. At other angle configurations, although clear fringes and changes can still be observed, the initial fringe morphology and color distribution differ. Therefore, a 45-degree angle between the polarizer's polarization axis and the thin film optical axis is selected as the standard optical configuration of this invention.

[0024] S2 acquires a measurement image of the black placement box 2 under pressure applied to the object 1; Place the object to be tested 1 inside the black placement box 2, and take images as described above.

[0025] S3 performs cropping on the initial image and the measured image, preserving the areas of stripe variation.

[0026] In this embodiment, classic image processing techniques are used to crop the image of irrelevant areas, retaining only the image of the areas with stripe variations.

[0027] S4 extracts the grayscale images of the RGB three channels of the initial image and the measurement image and calculates the difference to obtain the difference image.

[0028] S5 inputs the differential image into the pressure prediction model to calculate the estimated pressure value or pressure range corresponding to the object to be tested 1.

[0029] In this embodiment, the pressure prediction model is constructed as follows: S51 collects the initial stripe image of the black placement box 2 under no-pressure conditions; S52 collects measurement stripe images of the black placement box 2 under pressure applied by objects of different weights, the measurement stripe images being marked with weights; S53 Change the initial conditions and repeat step S52. The initial conditions include a slight change in the tension of the black placement box 2 and ambient light. S54 inputs the collected initial stripe images and measured stripe images into the CNN model, and uses the corresponding pressure value or pressure range as the output to train the CNN model; S55 validates the CNN model, and once the accuracy reaches the predetermined requirements, the model is used as a stress prediction model.

[0030] Through model training, a mapping relationship can be established from the visual features of striped images (color distribution, stripe density, shape, texture, etc.) to the corresponding pressure level.

[0031] The structure and parameters of each layer of the Convolutional Neural Network (CNN) model are as follows: Input layer: Input data format: RGB three-channel image with dimensions of 128 pixels * 128 pixels.

[0032] Data preprocessing: The input image is an RGB image after difference processing (difference image), where each channel carries the grayscale information of R / G / B colors respectively, so the input tensor dimension is (128, 128, 3).

[0033] Feature extraction module (convolutional and pooling layers): First convolutional-pooling layer group: Convolutional layer: Uses 8 convolutional kernels of size 13*13 with a stride of 1, and uses the Rectified Linear Unit (ReLU) activation function. The output feature map size is (116, 116, 8).

[0034] Pooling layer: Max pooling is used, with a pooling kernel size of 2*2, a stride of 2, and an output feature map size of (58, 58, 8).

[0035] Second convolutional-pooling layer group: Convolutional layer: 16 convolutional kernels of size 8*8 with a stride of 1 and ReLU activation are used. The output feature map size is (51, 51, 16).

[0036] Pooling layer: Max pooling, pooling kernel size 2*2, stride 2, output feature map size (25, 25, 16).

[0037] Third convolutional-pooling layer group: Convolutional layer: 32 convolutional kernels of size 3*3 are used with a stride of 1 and ReLU activation. The output feature map size is (23, 23, 32).

[0038] Pooling layer: Max pooling, pooling kernel size 2*2, stride 2, output feature map size (11, 11, 32).

[0039] Fourth convolutional-pooling layer group: Convolutional layer: 32 convolutional kernels of size 3*3 are used, with a stride of 1 and ReLU activation. The output feature map size is (9, 9, 32).

[0040] Pooling layer: Max pooling, pooling kernel size 2*2, stride 2, output feature map size (4, 4, 32).

[0041] Feature transformation and classification module (fully connected layer): Flattening layer: The three-dimensional feature map tensor (4, 4, 32) output by the fourth pooling layer is flattened into a one-dimensional vector with dimension (512).

[0042] The first fully connected layer contains 128 neuron nodes and uses the ReLU activation function.

[0043] The second fully connected layer contains 64 neuron nodes and uses the ReLU activation function.

[0044] The third fully connected layer contains 32 neuron nodes and uses the ReLU activation function.

[0045] Output layer: Dimensions: 10 neuron nodes.

[0046] Activation function: Softmax.

[0047] Functional mapping: Each node corresponds to a preset pressure range category. In this embodiment, the mapping relationship is as follows: the weak pressure range of 0 grams to 100 grams is divided into 10 consecutive intervals with 10-gram intervals (i.e., 0-10g, 10-20g, ..., 90-100g). The model output is the probability distribution of the pressure range to which the input image belongs.

[0048] Training and optimization: Loss function: Categorical Crossentropy.

[0049] Optimizer: Adaptive Moments Estimation Optimizer (Adam) The model was trained using a small amount of data, and the results are as follows: Figure 4 As shown, the final accuracy rate of the training set was 99.06%, and the final accuracy rate of the test set was 94.95%. It should be noted that the devices in this embodiment were all manually built, and the dataset was relatively small. The accuracy of the measurement and the application scenarios can be further improved by using more sophisticated devices and larger datasets.

[0050] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for measuring weak pressure based on the photoelastic effect and image recognition, characterized in that, The method comprises the following steps: S1: obtaining an initial image of the photoelastic sensing layer under a pressure-free state; S2: obtaining a measurement image of the photoelastic sensing layer under a pressure state applied by a to-be-measured object; S3: performing a clipping process on the initial image and the measurement image to retain a fringe change region; S4: extracting a gray-scale image of three channels of the initial image and the measurement image and performing a difference operation to obtain a difference image; S5: inputting the difference image into a pressure prediction model to calculate an estimated pressure value or a pressure interval corresponding to the to-be-measured object.

2. The weak pressure measurement method according to claim 1, characterized in that: the photoelastic sensing layer is mounted on a measurement support, a black placement box is arranged on the photoelastic sensing layer, a light source is arranged below the photoelastic sensing layer, a camera is arranged above the photoelastic sensing layer, and the camera is connected to a computer through a data line; the light source emits light to the photoelastic sensing layer, and the camera captures a deformation picture of the photoelastic sensing layer as the measurement image.

3. The weak pressure measurement method according to claim 2, characterized in that: the photoelastic sensing layer is composed of two pieces of orthogonally stacked photoelastic material films and is fixed on the measurement support, a first polarizer is arranged on the light source, a second polarizer is arranged on the camera, and the second polarizer is arranged at a preset angle with the first polarizer.

4. The method of weak pressure measurement according to claim 3, wherein, The photoelastic material film is a PET film.

5. The method of weak pressure measurement according to claim 3, wherein, The second polarizer is arranged at an angle of 45 degrees with the first polarizer.

6. The method of weak pressure measurement according to claim 1, wherein, The pressure prediction model is constructed in the following manner: S51: collecting an initial fringe picture of the photoelastic sensing layer under a pressure-free state; S52: collecting measurement fringe pictures of the photoelastic sensing layer under a pressure state applied by different weights, the measurement fringe pictures being provided with weight marks; S53: changing initial conditions, repeating step S52, and the initial conditions including a slight change in the tension of the photoelastic sensing layer and ambient light; S54: inputting the collected initial fringe picture and the measurement fringe picture into a CNN model, taking a corresponding pressure value or a pressure interval as an output, and training the CNN model; S55: verifying the CNN model, and taking the model as the pressure prediction model when the accuracy reaches a predetermined requirement.

Images