Computer vision light path design method for large complex surface color detection

Abstract

The invention discloses a computer vision light path design method for large complex surface color detection, and relates to the technical field of computer vision and optical detection. Through comprehensive application of a self-adaptive light source system, a high-efficiency light splitting optical path design, a real-time image processing algorithm, an ambient light anti-interference technology and a multi-scale feature fusion technology, high-precision and high-efficiency detection of large complex surface colors is realized. The self-adaptive light source system dynamically adjusts the light source position and intensity to ensure the illumination uniformity; according to the efficient light splitting optical path design, a spectroscope and an optical filter are combined with a multispectral imaging technology, and the color detection precision is improved; the real-time image processing algorithm adopts an image segmentation and color recognition algorithm based on deep learning, and the real-time requirement is met; the ambient light anti-interference technology reduces the influence of ambient light change on a detection result through an ambient light sensor and an adaptive filtering algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and optical detection, and in particular to a computer vision optical path design method for large and complex surface color detection. Background Art

[0002] In the field of computer vision, color detection is one of the important tasks of image processing and analysis, and is widely used in industrial manufacturing, intelligent detection, and automated quality control. However, for large and complex surfaces (such as refrigerator shells), traditional computer vision methods face many challenges. First, due to the large curvature and size of the surface, it is difficult for a single light source to achieve uniform illumination, resulting in deviations in color detection results; second, the amount of high-resolution image data for large surfaces is huge, and the processing efficiency of traditional algorithms is low, making it difficult to meet real-time requirements; in addition, changes in ambient light and surface reflection interference will further affect the accuracy of detection, and traditional methods perform poorly when adapting to surfaces with different curvatures, textures, and materials. In existing technologies, color detection methods based on computer vision mostly rely on fixed light sources and simple image processing algorithms, which cannot effectively solve the above problems. Summary of the Invention

[0003] The purpose of the present invention is to address the above-mentioned shortcomings and propose a computer vision optical path design method that solves problems such as uneven lighting, large data volume, and environmental interference by optimizing the light source layout, optical path structure, and computer vision algorithm, thereby achieving high-precision and high-efficiency detection of the color of large and complex surfaces.

[0004] The present invention specifically adopts the following technical solutions:

[0005] A computer vision optical path design method for large and complex surface color detection, comprising:

[0006] Establish an adaptive light source system;

[0007] The efficient spectroscopic optical path design decomposes the incident light into light of different wavelengths. The split and filtered light enters the high-resolution image sensor for image acquisition and processing.

[0008] The real-time image processing design quickly analyzes and processes high-resolution image data, extracts color information, and provides real-time feedback. First, a deep learning-based image segmentation algorithm is used to segment the collected multispectral image data into different regions. The image segmentation algorithm is based on a convolutional neural network (CNN), whose basic structure includes convolutional layers, pooling layers, and fully connected layers. The calculation process of the convolutional layer is as follows:

[0009]

[0010] Among them, y i,j,krepresents the value of the kth channel of the output feature map at position i, j, w m,n,c,k represents the weight of the convolution kernel, x i+m-1，j+n-1，c Indicates the value of the cth channel of the input feature map at position i+m-1, j+n-1, b k Represents the bias term, which is operated through multiple layers of convolution and pooling;

[0011] On the basis of image segmentation, a color recognition algorithm based on support vector machine (SVM) is used to classify the color of each region. For the linearly separable case, the decision function of SVM is described by the following formula:

[0012]

[0013] Among them, x represents the input feature vector, y i represents the category label, α i represents the Lagrange multiplier, K(x i , x) represents the kernel function, b represents the bias term, and the radial basis function RBF is used as the kernel function, and its expression is:

[0014]

[0015] By training the SVM model, the color category of each area is accurately identified and a color distribution map is generated;

[0016] The ambient light anti-interference design uses a high-precision ambient light sensor to monitor the intensity and spectral distribution of ambient light in real time. The ambient light sensor is usually based on a photodiode or photomultiplier tube. Its output signal is proportional to the ambient light intensity. The voltage signal V output by the sensor can be described by the following formula:

[0017]

[0018] Where k is the sensitivity coefficient of the sensor, E(λ) is the spectral radiation intensity of the ambient light, S(λ) is the spectral response function of the sensor, λ1 and λ2 are the effective wavelength ranges of the sensor. By collecting the output signal of the ambient light sensor in real time, the intensity and spectral characteristics of the ambient light can be accurately obtained.

[0019] After obtaining the ambient light data, an adaptive filtering algorithm is used to compensate for the ambient light interference. The adaptive filtering algorithm is based on the least mean square error (LMS) criterion. The weight update formula of the LMS algorithm is as follows:

[0020] w(n+1)=w(n)+μ·e(n)·x(n)

[0021] Where w(n) represents the weight vector of the filter, μ is the step size factor, e(n) is the error signal, and x(n) is the input signal;

[0022] Multi-scale feature fusion uses image pyramid to generate multi-scale image data. Image pyramid is a multi-resolution representation method that generates a series of images with different resolutions by downsampling the original image multiple times. The downsampling process is described by the following formula:

[0023]

[0024] Among them, I l (x, y) represents the pixel value of the I-th layer image at position (x, y), I l-1 (2x+m, 2y+n) represents the first l-1 The pixel value of the layer image at position (2x+m, 2y+n), w(m, n) is a Gaussian kernel function used to smooth the image to reduce information loss during downsampling;

[0025] After generating multi-scale image data, a convolutional neural network (CNN) is used to extract features of different scales. CNN can automatically learn high-level features of images through multi-layer convolution and pooling operations. For the I-th layer image, its feature extraction process is described by the following formula:

[0026]

[0027] Among them, F l (x, y) represents the eigenvalue of the layer I image at position (x, y), w(m, n) is the weight of the convolution kernel, b is the bias term, and f(·) is the activation function;

[0028] After extracting multi-scale features, the feature fusion algorithm is used to combine these features. The feature fusion algorithm is based on the weighted summation method, and its core formula is as follows:

[0029]

[0030] Among them, F fused (x, y) represents the fused feature value, α1 is the weight coefficient of the I-th layer feature, and L is the number of layers of the image pyramid.

[0031] Preferably, the adaptive light source system uses multiple adjustable light sources, which are arranged around the surface to be measured to form a multi-angle lighting layout. Each light source is equipped with an adjustable mechanical bracket and control system, which can dynamically adjust the position and angle of the light source according to the surface curvature and size.

[0032] Preferably, the efficient spectroscopic optical path design uses a spectroscope to decompose the incident light into multiple wavelength bands. The spectroscope is capable of separating the incident light into different spectral components according to the wavelength characteristics of the light. The design of the spectroscope is based on the principle of thin film interference, and its spectroscopic characteristics are described by the following formula:

[0033]

[0034] Where R(λ) represents the reflectivity, n1 and n2 are the refractive indices of the two media, δ is the phase difference, and λ is the wavelength of light;

[0035] The split light passes through the filter to further refine the spectral components. The design of the filter is based on the principle of multi-layer dielectric film interference. Its transmittance characteristics can be described by the following formula:

[0036]

[0037] Where T(λ) represents the transmittance, λ0 is the center wavelength, and Δλ is the bandwidth. By selecting different filters, the split light can be further refined into light of different wavelength bands to meet the needs of multispectral imaging.

[0038] Preferably, multispectral imaging technology and ambient light separation algorithm are combined to further reduce the interference of ambient light;

[0039] Multispectral imaging technology can simultaneously capture light information in different bands and generate multi-channel image data. The ambient light separation algorithm separates the ambient light component from the target surface reflected light component by analyzing the multispectral image data. The system uses multispectral imaging technology to obtain image data in the red, green, and blue bands, and uses the ambient light separation algorithm to remove the ambient light component, thereby obtaining pure target surface reflected light information. The core formula of the ambient light separation algorithm is as follows:

[0040] I target (λ)=I total (λ)-I ambient (λ)

[0041] Among them, I target Represents the spectral intensity of the light reflected from the target surface, I total Represents the total spectral intensity, I ambient (λ) represents the spectral intensity of the ambient light.

[0042] The present invention has the following beneficial effects:

[0043] The adaptive light source system uses multiple adjustable light sources, combined with computer vision algorithms to analyze surface curvature and reflective properties in real time, dynamically adjusting the position and intensity of the light source to ensure uniform lighting. This approach effectively avoids uneven lighting caused by surface curvature and size, providing high-quality optical input for subsequent color detection. Secondly, the efficient spectroscopic optical path design uses spectrometers and filters to decompose the incident light into different wavelengths, combined with multispectral imaging technology to improve the accuracy of color detection. This design not only captures richer color information but also reduces noise interference in the optical path, further improving the reliability of detection results.

[0044] In terms of image processing, this solution uses a real-time image processing algorithm based on deep learning to rapidly analyze and process high-resolution images. Image segmentation and color recognition algorithms significantly reduce data volume and improve detection efficiency, meeting the real-time requirements of large-scale surface detection. Furthermore, the introduction of ambient light interference reduction technology further enhances the robustness of the system. The ambient light sensor monitors ambient light intensity in real time, and the detection data is compensated using an adaptive filtering algorithm to reduce the impact of ambient light changes on the detection results, ensuring the stability and accuracy of the detection results. Finally, the application of multi-scale feature fusion technology enables the system to adapt to surfaces of varying curvature, texture, and material. By combining multi-scale image feature extraction technology, surface details can be more comprehensively captured, improving the adaptability and robustness of detection.

[0045] The optical path design method described in this application features high precision, high efficiency, strong anti-interference capabilities, and wide adaptability. Through an adaptive light source system and efficient spectroscopic optical path design, the accuracy of color detection can be significantly improved. Real-time image processing algorithms and ambient light anti-interference technology ensure high efficiency and stability of detection. Multi-scale feature fusion technology further enhances the system's adaptability to different surfaces. This invention can be widely applied in industrial manufacturing, intelligent detection, automated quality control, and other fields, providing an efficient and reliable solution for color detection on large and complex surfaces, with significant practical value and broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A framework diagram of the computer vision optical path design method for large and complex surface color detection;

[0047] Figure 2 This is the working principle diagram of the adaptive light source system;

[0048] Figure 3 is a flowchart of the real-time image processing algorithm;

[0049] Figure 4 This is the implementation diagram of ambient light anti-interference technology. DETAILED DESCRIPTION

[0050] The specific implementation of the present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0051] Combine Figure 1-Figure 4, establish an adaptive light source system; the adaptive light source system is one of the core components of this solution, and its main function is to ensure the uniformity of lighting on large and complex surfaces by dynamically adjusting the position and intensity of the light source, thereby providing high-quality optical input for subsequent color detection. First, the system uses multiple adjustable light sources, which are arranged around the surface to be measured to form a multi-angle lighting layout. The type of light source can be LED, halogen lamp or other high-brightness light source, and the specific choice depends on the needs of the detection scenario. Each light source is equipped with an adjustable mechanical bracket and control system, which can dynamically adjust the position and angle of the light source according to the curvature and size of the surface. For example, when inspecting a car body, the light source can automatically adjust the illumination angle according to the curved structure of the car body to ensure that the light evenly covers the entire surface.

[0052] Secondly, the adaptive lighting system is closely integrated with computer vision algorithms to achieve intelligent light source control. The system uses a high-resolution image sensor to capture real-time images of the surface being measured and utilizes computer vision algorithms to analyze the surface's curvature, reflectivity, and illumination distribution. Based on this analysis, the control system dynamically adjusts the brightness and position of each light source to optimize illumination uniformity. For example, if insufficient illumination is detected on a specific surface area, the system automatically increases the brightness of the light source at that location or adjusts the angle of the light source to increase illumination intensity. Furthermore, the color temperature and spectral characteristics of the light source can be adjusted based on the surface material to further enhance color detection accuracy.

[0053] In practical applications, the adaptive lighting system also features scene adaptation capabilities. For example, in scenes with significant ambient light fluctuations, the ambient light sensor can monitor the ambient light intensity in real time and, combined with an adaptive filtering algorithm, dynamically adjust the light source's output power to offset the impact of ambient light on detection results. Furthermore, it supports multi-mode switching, enabling the selection of different lighting modes based on the requirements of different detection tasks. For example, when inspecting highly reflective surfaces, a low-angle lighting mode can be used to reduce interference from reflected light; when inspecting low-reflective surfaces, a high-angle lighting mode can be used to enhance light intensity.

[0054] The adaptive light source system effectively addresses uneven illumination on large, complex surfaces, providing high-quality optical input for color detection. Its intelligent, dynamic light source control not only improves detection accuracy but also enhances the system's adaptability and practicality, laying a solid foundation for subsequent image processing and analysis.

[0055] The efficient spectroscopic optical path design uses a spectroscope to decompose the incident light into multiple wavelength bands. The spectroscope is able to separate the incident light into different spectral components according to the wavelength characteristics of the light. The design of the spectroscope is based on the principle of thin film interference. Its spectroscopic characteristics are described by the following formula:

[0056]

[0057] Where R(λ) represents the reflectivity, n1 and n2 are the refractive indices of the two media, δ is the phase difference, and λ is the wavelength of light;

[0058] The split light passes through the filter to further refine the spectral components. The design of the filter is based on the principle of multi-layer dielectric film interference. Its transmittance characteristics can be described by the following formula:

[0059]

[0060] Where T(λ) represents the transmittance, λ0 is the center wavelength, and Δλ is the bandwidth. By selecting different filters, the split light can be further refined into light of different wavelength bands to meet the needs of multispectral imaging.

[0061] The efficient spectroscopic optical path design decomposes the incident light into light of different wavelengths. The split and filtered light enters the high-resolution image sensor for image acquisition and processing.

[0062] The real-time image processing design quickly analyzes and processes high-resolution image data, extracts color information, and provides real-time feedback. First, a deep learning-based image segmentation algorithm is used to segment the collected multispectral image data into different regions. The image segmentation algorithm is based on a convolutional neural network (CNN), whose basic structure includes convolutional layers, pooling layers, and fully connected layers. The calculation process of the convolutional layer is as follows:

[0063]

[0064] Among them, y i，j，k represents the value of the kth channel of the output feature map at position i, j, w m,n,c,k represents the weight of the convolution kernel, x i+m-1,j+n-1,c Indicates the value of the cth channel of the input feature map at position i+m-1,j+n-1, b k Represents the bias term, which is operated through multiple layers of convolution and pooling;

[0065] On the basis of image segmentation, a color recognition algorithm based on support vector machine (SVM) is used to classify the color of each region. For the linearly separable case, the decision function of SVM is described by the following formula:

[0066]

[0067] Among them, x represents the input feature vector, y i represents the category label, α i represents the Lagrange multiplier, K(x i, x) represents the kernel function, b represents the bias term, and the radial basis function RBF is used as the kernel function, and its expression is:

[0068]

[0069] By training the SVM model, the color category of each area is accurately identified and a color distribution map is generated.

[0070] To improve the algorithm's real-time performance, image pyramid and multi-scale feature fusion techniques are also employed. Image pyramid is a multi-resolution representation method that generates a series of images at different resolutions by downsampling the original image multiple times. Multi-scale feature fusion combines feature maps of different resolutions to improve the algorithm's robustness and accuracy. For example, when detecting large, complex surfaces, rapid segmentation and recognition can be performed first on a low-resolution image, followed by refined processing of key areas on a high-resolution image. This improves processing efficiency while maintaining accuracy.

[0071] In practical applications, the real-time image processing algorithm also possesses adaptive learning capabilities. For example, when inspecting surfaces of varying materials, the system can dynamically update the SVM model based on historical data to improve color recognition accuracy. Furthermore, the system supports parallel computing and hardware acceleration, fully utilizing hardware resources such as GPUs and FPGAs to further enhance algorithm efficiency.

[0072] The real-time image processing algorithm enables rapid analysis and processing of high-resolution image data, extracting color information and providing real-time feedback. Its deep learning-based image segmentation algorithm and SVM-based color recognition algorithm ensure high-precision and efficient color detection. The application of image pyramids and multi-scale feature fusion technology further enhances the algorithm's robustness and adaptability, providing an efficient and reliable solution for color detection on large and complex surfaces.

[0073] The ambient light anti-interference design uses a high-precision ambient light sensor to monitor the intensity and spectral distribution of ambient light in real time. The ambient light sensor is usually based on a photodiode or photomultiplier tube. Its output signal is proportional to the ambient light intensity. The voltage signal V output by the sensor can be described by the following formula:

[0074]

[0075] Where k is the sensitivity coefficient of the sensor, E(λ) is the spectral radiation intensity of the ambient light, S(λ) is the spectral response function of the sensor, λ1 and λ2 are the effective wavelength ranges of the sensor. By collecting the output signal of the ambient light sensor in real time, the intensity and spectral characteristics of the ambient light can be accurately obtained.

[0076] After obtaining the ambient light data, an adaptive filtering algorithm is used to compensate for the ambient light interference. The adaptive filtering algorithm is based on the least mean square error (LMS) criterion. The weight update formula of the LMS algorithm is as follows:

[0077] w(n+1)=w(n)+μ·e(n)·x(n)

[0078] Where w(n) represents the weight vector of the filter, μ is the step size factor, e(n) is the error signal, and x(n) is the input signal;

[0079] It also combines multispectral imaging technology and ambient light separation algorithm to further reduce the interference of ambient light;

[0080] Multispectral imaging technology can simultaneously capture light information in different bands and generate multi-channel image data. The ambient light separation algorithm separates the ambient light component from the target surface reflected light component by analyzing the multispectral image data. The system uses multispectral imaging technology to obtain image data in the red, green, and blue bands, and uses the ambient light separation algorithm to remove the ambient light component, thereby obtaining pure target surface reflected light information. The core formula of the ambient light separation algorithm is as follows:

[0081] I target (λ)=I total (λ)-I ambient (λ)

[0082] Among them, I target Represents the spectral intensity of the light reflected from the target surface, I total Represents the total spectral intensity, I ambient (λ) represents the spectral intensity of the ambient light.

[0083] Multi-scale feature fusion uses image pyramid to generate multi-scale image data. Image pyramid is a multi-resolution representation method that generates a series of images with different resolutions by downsampling the original image multiple times. The downsampling process is described by the following formula:

[0084]

[0085] Among them, I l (x, y) represents the pixel value of the I-th layer image at position (x, y), I l-1 (2x+m, 2y+n) represents the first l-1 The pixel value of the layer image at position (2x+m, 2y+n), w(m, n) is a Gaussian kernel function used to smooth the image to reduce information loss during downsampling;

[0086] After generating multi-scale image data, a convolutional neural network (CNN) is used to extract features of different scales. CNN can automatically learn high-level features of images through multi-layer convolution and pooling operations. For the I-th layer image, its feature extraction process is described by the following formula:

[0087]

[0088] Among them, F l (x, y) represents the eigenvalue of the layer I image at position (x, y), w(m, n) is the weight of the convolution kernel, b is the bias term, and f(·) is the activation function;

[0089] After extracting multi-scale features, the feature fusion algorithm is used to combine these features. The feature fusion algorithm is based on the weighted summation method, and its core formula is as follows:

[0090]

[0091] Among them, F fused (x, y) represents the fused eigenvalue, α l is the weight coefficient of the I-th layer feature, and L is the number of layers of the image pyramid.

[0092] Weight coefficient α l It can automatically learn from training data to ensure that the contribution of features at different scales in the fusion process is optimal. Through feature fusion, the system can combine global information and local details to improve the adaptability and robustness of the color detection algorithm.

[0093] In practical applications, multi-scale feature fusion technology also offers dynamic adjustment capabilities. For example, when detecting surfaces with varying curvatures and textures, the system can dynamically adjust the number of image pyramid layers and feature fusion weights based on surface characteristics to optimize detection results. Furthermore, it supports parallel computing and hardware acceleration, fully leveraging hardware resources like GPUs and FPGAs to further improve the efficiency of feature extraction and fusion.

[0094] Multi-scale feature fusion technology can significantly improve the adaptability and robustness of color detection algorithms for complex surfaces. Its multi-scale data generation based on image pyramids and feature extraction based on CNNs ensure the richness and diversity of feature information. The feature fusion algorithm based on weighted summation further enhances the algorithm's overall performance, providing an efficient and reliable solution for color detection on large and complex surfaces.

[0095] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. A computer vision optical path design method for large and complex surface color detection, characterized in that: include: Establish an adaptive light source system; The efficient spectroscopic optical path design decomposes the incident light into light of different wavelengths. The split and filtered light enters the high-resolution image sensor for image acquisition and processing. The real-time image processing design quickly analyzes and processes high-resolution image data, extracts color information, and provides real-time feedback. First, a deep learning-based image segmentation algorithm is used to segment the collected multispectral image data into different regions. The image segmentation algorithm is based on a convolutional neural network (CNN), whose basic structure includes convolutional layers, pooling layers, and fully connected layers. The calculation process of the convolutional layer is as follows: Among them, y i，j，k Represents the value of the kth channel of the output feature map at position i, j, w m，n，c，k represents the weight of the convolution kernel, x i+m-1，j+n-1，c Indicates the value of the cth channel of the input feature map at position i+m-1, j+n-1, b k Represents the bias term, which is operated through multiple layers of convolution and pooling; On the basis of image segmentation, a color recognition algorithm based on support vector machine (SVM) is used to classify the color of each region. For the linearly separable case, the decision function of SVM is described by the following formula: Among them, x represents the input feature vector, y i represents the category label, α i represents the Lagrange multiplier, K(x i , x) represents the kernel function, b represents the bias term, and the radial basis function RBF is used as the kernel function, and its expression is: By training the SVM model, the color category of each area is accurately identified and a color distribution map is generated; The ambient light anti-interference design uses a high-precision ambient light sensor to monitor the intensity and spectral distribution of ambient light in real time. The ambient light sensor is usually based on a photodiode or photomultiplier tube. Its output signal is proportional to the ambient light intensity. The voltage signal V output by the sensor can be described by the following formula: Where k is the sensitivity coefficient of the sensor, E(λ) is the spectral radiation intensity of the ambient light, S(λ) is the spectral response function of the sensor, λ1 and λ2 are the effective wavelength ranges of the sensor. By collecting the output signal of the ambient light sensor in real time, the intensity and spectral characteristics of the ambient light can be accurately obtained. After obtaining the ambient light data, an adaptive filtering algorithm is used to compensate for the ambient light interference. The adaptive filtering algorithm is based on the least mean square error (LMS) criterion. The weight update formula of the LMS algorithm is as follows: w(n+1)=w(n)+μ·e(n)·x(n) Where w(n) represents the weight vector of the filter, μ is the step size factor, e(n) is the error signal, and x(n) is the input signal; Multi-scale feature fusion uses image pyramid to generate multi-scale image data. Image pyramid is a multi-resolution representation method that generates a series of images with different resolutions by downsampling the original image multiple times. The downsampling process is described by the following formula: Among them, I l (x, y) represents the pixel value of the I-th layer image at position (x, y), I l-1 (2x+m, 2y+n) represents the first l-1 The pixel value of the layer image at position (2x+m, 2y+n), w(m, n) is a Gaussian kernel function used to smooth the image to reduce information loss during downsampling; After generating multi-scale image data, a convolutional neural network (CNN) is used to extract features of different scales. CNN can automatically learn high-level features of images through multi-layer convolution and pooling operations. For the I-th layer image, its feature extraction process is described by the following formula: Among them, F l (x, y) represents the eigenvalue of the layer I image at position (x, y), w(m, n) is the weight of the convolution kernel, b is the bias term, and f(·) is the activation function; After extracting multi-scale features, the feature fusion algorithm is used to combine these features. The feature fusion algorithm is based on the weighted summation method, and its core formula is as follows: Among them, F fused (x, y) represents the fused eigenvalue, α l is the weight coefficient of the I-th layer feature, and L is the number of layers of the image pyramid.

2. A computer vision optical path design method for large and complex surface color detection according to claim 1, characterized in that: The adaptive light source system uses multiple adjustable light sources, which are arranged around the surface to be measured to form a multi-angle lighting layout. Each light source is equipped with an adjustable mechanical bracket and control system, which can dynamically adjust the position and angle of the light source according to the surface curvature and size.

3. The computer vision optical path design method for large and complex surface color detection according to claim 1, characterized in that: The efficient spectroscopic optical path design uses a spectroscope to decompose the incident light into multiple wavelength bands. The spectroscope is able to separate the incident light into different spectral components according to the wavelength characteristics of the light. The design of the spectroscope is based on the principle of thin film interference. Its spectroscopic characteristics are described by the following formula: Where R(λ) represents the reflectivity, n1 and n2 are the refractive indices of the two media, δ is the phase difference, and λ is the wavelength of light; The split light passes through the filter to further refine the spectral components. The design of the filter is based on the principle of multi-layer dielectric film interference. Its transmittance characteristics can be described by the following formula: Where T(λ) represents the transmittance, λ0 is the center wavelength, and Δλ is the bandwidth. By selecting different filters, the split light can be further refined into light of different wavelength bands to meet the needs of multispectral imaging.

4. The computer vision optical path design method for large and complex surface color detection according to claim 1, characterized in that: It also combines multispectral imaging technology and ambient light separation algorithm to further reduce the interference of ambient light; Multispectral imaging technology can simultaneously capture light information in different bands and generate multi-channel image data. The ambient light separation algorithm separates the ambient light component from the target surface reflected light component by analyzing the multispectral image data. The system uses multispectral imaging technology to obtain image data in the red, green, and blue bands, and uses the ambient light separation algorithm to remove the ambient light component, thereby obtaining pure target surface reflected light information. The core formula of the ambient light separation algorithm is as follows: I target (λ)=I total (-I ambient (λ) Among them, I target Represents the spectral intensity of the light reflected from the target surface, I total Represents the total spectral intensity, I ambient (λ) represents the spectral intensity of the ambient light.

Images