A method for classifying material color based on LED perception

Abstract

The application discloses a material color classification method based on LED sensing, comprising the following steps: for a color classification system composed of one lighting LED, one single-color material and one multi-color sensing LED array, an LED sensing signal model is established, a color classification problem is equivalently converted into a spectrum reflectivity vector detection problem, and an optimization problem is established based on a generalized likelihood ratio test to realize robust detection; based on principal component analysis, detection on a high-dimensional spectrum reflectivity vector is converted into detection on a low-dimensional weight vector to reduce detection complexity and implementation cost; and by comparison with a feature vector set established in a training stage, material color classification is realized. The application utilizes LEDs in existing lighting infrastructure for signal sensing, is simple to deploy, and further reduces detection complexity and implementation cost of the color classification system by using principal component analysis and generalized likelihood ratio test, while guaranteeing robustness to link gain.

Description

Technical Field

[0001] This invention relates to the field of sensing technology, and in particular to a method for classifying material colors based on LED sensing. Background Technology

[0002] In today's world, the Internet of Things (IoT) has become a crucial concept in engineering and computing due to the vision of a globally interconnected infrastructure. IoT involves the combination of various emerging technologies, such as embedded sensors, positioning, near-field communication (NFC), and the Internet. IoT enables the integration of objects, devices, and systems equipped with sensors and microcontrollers to achieve interaction between them and their users. With the continuous development of information technology, the sensor network field involved in IoT is increasingly focused on creating environments that can seamlessly sense and respond to human needs—that is, the growing demand for sensing-based services. To meet specific needs, many sensing technologies have been proposed, including Wi-Fi, ultrasonic, and infrared technologies. However, traditional sensing technologies suffer from low sensing accuracy and high deployment costs. In contrast, Visible Light Sensing (VLS) can achieve higher sensing accuracy due to the strong directional transmission of visible light. At the same time, utilizing existing lighting infrastructure, VLS has relatively low deployment costs. These advantages make visible light sensing a promising sensing technology.

[0003] VLS can be implemented in two ways: active sensing and passive sensing. In recent years, to reduce the complexity and cost of VLS systems, most research has focused on passive methods. These passive systems can be categorized into three types: passive light sources, passive objects, and completely passive systems. If the light source is not modulated, the system is defined as passive. If the object does not require a light sensor, it is considered a passive object. If both the light source and the object are passive, the system is considered completely passive. Considering the need to reduce the introduction of additional equipment and lower the deployment cost and complexity of VLS, a fully passive sensing system would be an ideal solution.

[0004] VLS systems typically use photodetectors as sensing devices, which requires additional equipment deployment, resulting in higher system complexity and cost. LEDs are commonly used as transmitters in VLS systems, and can also function as signal receivers when operating in photovoltaic mode. Using LEDs as receivers offers advantages such as low cost, low power consumption, and high privacy protection, enabling low-complexity LED sensing systems. Summary of the Invention

[0005] In view of this, the main objective of the present invention is to provide a material color classification method based on LED sensing, in order to reduce the deployment cost and complexity of VLS systems and provide high privacy protection performance.

[0006] To achieve the above objectives, as one aspect of the present invention, a material color classification method based on LED sensing is provided, comprising the following steps: Step 1: For a color classification system consisting of a lighting LED, a monochromatic material, and a multicolor sensing LED array, establish an LED sensing signal model, transform the color classification problem into an equivalent spectral reflectance vector detection problem, and establish an optimization problem based on the generalized likelihood ratio test to achieve robust detection.

[0007] Step 2: Based on principal component analysis, the detection of high-dimensional spectral reflectance vectors is transformed into the detection of low-dimensional weight vectors to reduce detection complexity and implementation cost. By comparing with the feature vector set established in the training phase, material color classification is achieved.

[0008] Based on the above technical solution, it can be seen that the material color classification method based on LED sensing of the present invention will have the following benefits: (1) The present invention utilizes LEDs in the existing lighting infrastructure for signal sensing, without the need to add other detection equipment, which significantly reduces the deployment cost and complexity of VLS.

[0009] (2) This invention provides a material color classification method based on principal component analysis, which transforms the detection of high-dimensional spectral reflectance vectors into the detection of low-dimensional weight vectors, significantly reducing the detection complexity and implementation complexity of the color classification system.

[0010] (3) The present invention uses the generalized likelihood ratio test method to solve the optimization problem of color classification and improves the robustness of the color classification system to changes in link gain. Attached Figure Description

[0011] Figure 1 A block diagram of a color classification system provided as a preferred embodiment of the present invention.

[0012] Figure 2 These are the eight largest eigenvalues ​​of the covariance matrix of the spectral reflectance vector sample.

[0013] Figure 3 Based on Figure 1 The example shown illustrates the classification accuracy of the classification system under different training heights. Detailed Implementation

[0014] The main objective of this invention is to provide a material color classification method based on LED sensing. The LED-based color classification system significantly simplifies the deployment of VLS (Visual Synthetic Language). Furthermore, principal component analysis is used to optimize the classification problem, further reducing the detection complexity and implementation cost of the classification system. Robust color classification performance is obtained using the generalized likelihood ratio test method.

[0015] Specifically, this invention discloses a material color classification method based on LED sensing, comprising the following steps: Step 1: For a color classification system consisting of a lighting LED, a monochromatic material, and a multicolor sensing LED array, establish an LED sensing signal model, transform the color classification problem into an equivalent spectral reflectance vector detection problem, and establish an optimization problem based on the generalized likelihood ratio test to achieve robust detection.

[0016] Step 2: Based on principal component analysis, the detection of high-dimensional spectral reflectance vectors is transformed into the detection of low-dimensional weight vectors to reduce detection complexity and implementation cost. By comparing with the feature vector set established in the training phase, material color classification is achieved.

[0017] Specifically, step 1 includes: 1.1 A color classification system based on LED sensing is constructed using one lighting LED, one monochromatic material, and one multicolor sensing LED array. The monochromatic material includes... The sensing LED array contains a variety of possible colors. LEDs of different colors.

[0018] 1.2, Order For optical link gain, For the first A sensing LED at wavelength Response rate at the location The spectral reflectance of a monochromatic material For the spectral power distribution of lighting LEDs, If it is additive noise, then the first The response output of an LED can be expressed as:

[0019] Sensing wavelength range Divided into Each width is The center wavelength is subintervals, ; Assumption The distance between the sensing LEDs is extremely small, taking ,in Indicates the first The and the first The optical link gain of the sensing LED. Let For observing scalars, The spectral power distribution vector of the lighting LED. The spectral reflectance vector, For the response matrix, If the noise vector is given, then the output vector of the sensing LED array can be approximated as:

[0020] make Represents the spectral reflectance vector set of a monochromatic material, where For the first The spectral reflectance vector of a monochromatic material is used to transform the color classification problem into an equivalent spectral reflectance vector detection problem. A generalized likelihood ratio test is employed to jointly estimate the scalar. and detection spectral reflectance vector The following optimization problem is established:

[0021] in Represents vector The detected value, Represents scalar The estimated value, Represents a probability function. The variable represents the value that maximizes the expression described later. This represents the value of the variable that minimizes the expression described later. Represents vector Norm.

[0022] Furthermore, step 2 specifically includes: 2.1 Covariance matrix of spectral reflectance vector samples Perform eigenvalue decomposition, let Indicates that the diagonal element is The diagonal matrix of eigenvalues, Representing an orthogonal matrix whose column vectors are eigenvectors, we obtain:

[0023] superscript To represent the transpose of a matrix, let Indicates by A matrix composed of principal direction basis vectors Let the weight vector be represented, then the spectral reflectance vector can be approximately represented as:

[0024] Definition corresponds to weight vector set ,in Indicates the first The weight vector corresponding to each monochromatic material. Using the updated output vector expression of the sensing LED array, the weight vector is... To perform the test, the problem (3) in step 1 is equivalently transformed into:

[0025] in Represents the weight vector The detected value, Represent the transformation matrix and define the vector. With corresponding set Problem (6) can be transformed into:

[0026] in Represents vector The detected value. Using vector Given time observation scalar The estimated expression transforms the optimization problem (7) into:

[0027] According to the optimization problem (8), the material color classification problem is equivalent to the problem of classifying materials in the set. Find the output vector in exist The vector with the largest projection in the direction Define the eigenvector. ,use As a key feature for detecting the color of materials, it achieves robustness to changes in link gain.

[0028] Problem (8) can be solved in two stages: an offline training stage and a real-time testing stage. In the offline training stage, a set of feature vectors corresponding to each monochromatic material to be tested is established. , For the first The feature vector corresponding to each monochromatic material. During the real-time testing phase, the output vector of the sensing LED array is calculated. With sets The inner product of all feature vectors is used to select the color corresponding to the feature vector that maximizes the inner product, thereby classifying the material colors.

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0030] Figure 1 The block diagram of the color classification system shown is a schematic diagram of a preferred embodiment. The color classification system is constructed using an illumination LED, a monochromatic material, and a multicolor sensing LED array. The transmitting end consists of an illumination LED, which receives a DC signal. The light emitted by the illumination LED is incident on the surface of the monochromatic material and reflected to the receiving end sensing LED array. Filtering and signal processing are then performed, and finally, the color classification result of the monochromatic material is obtained based on the response of the sensing LED array.

[0031] This invention uses common monochrome materials as the analysis objects of the color classification method, including monochrome paper and monochrome fabric. Figure 2 The eight largest eigenvalues ​​of the sample covariance matrix of the spectral reflectance vector are shown. It can be seen that the sample covariance matrix exhibits three principal eigenvalues ​​for the three sample combinations: paper, fabric, and mixed samples. Therefore, the color classification problem can be transformed from detecting a high-dimensional spectral reflectance vector to detecting a three-dimensional weight vector. Based on this fact, the color classification method of this invention uses a three-color LED array to construct a sensing LED array. It is worth noting that, ideally, the three-color LEDs used should each have their own independent response wavelength range.

[0032] In a specific embodiment, the lighting LED is a white LED, and the sensing LED array consists of three LED chips with peak wavelengths of 465nm, 550nm, and 660nm respectively. The monochromatic material is a monochromatic fabric of eight colors. The transceiver is located at a given height above the monochromatic material; in this embodiment, three sensing height scenarios are set: 0.5m, 1.0m, and 1.5m. In the specific implementation process, for each monochromatic material, the training phase uses... The feature vector is obtained by averaging the sampling points. During the testing phase, The average of each sampling point is used to obtain the output vector for testing. .

[0033] Figure 3 The classification accuracy is shown under different training heights. The settings include... , It can be clearly seen that the color classification method proposed in this invention possesses stable classification performance. Specifically, using training results at different heights, the classification accuracy is greater than 95%, and reaches 100% in most test scenarios. This result fully verifies the robustness of this invention to changes in link gain. The above results demonstrate that this color classification method achieves a near 100% classification accuracy, exhibits stable classification performance, and is robust to changes in link gain. Therefore, the color classification method of this invention has high practical application value.

[0034] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A material color classification method based on LED sensing, characterized in that, Includes the following steps: Step 1: For a color classification system consisting of a lighting LED, a monochromatic material, and a multicolor sensing LED array, establish an LED sensing signal model, transform the color classification problem into an equivalent spectral reflectance vector detection problem, and establish an optimization problem based on the generalized likelihood ratio test to achieve robust detection. Step 1 includes: 1.1 A color classification system based on LED sensing is constructed using one lighting LED, one monochromatic material, and one multicolor sensing LED array. The monochromatic material includes... The sensing LED array contains a variety of possible colors. LEDs of different colors; 1.2, Order For optical link gain, For the first A sensing LED at wavelength Response rate at the location The spectral reflectance of a monochromatic material For the spectral power distribution of lighting LEDs, If it is additive noise, then the first The response output of each LED is represented as follows: (1) Sensing wavelength range Divided into Each width is The center wavelength is subintervals, ; Assumption The distance between the sensing LEDs is extremely small, taking ,in They represent the first The and the first The optical link gain of the sensing LED; let For observing scalars, The spectral power distribution vector of the lighting LED. The spectral reflectance vector, For the response matrix, If the noise vector is given, then the output vector of the sensing LED array is approximately: (2) make Represents the spectral reflectance vector set of a monochromatic material, where For the first The spectral reflectance vector of a monochromatic material is used to transform the color classification problem into a spectral reflectance vector detection problem. A generalized likelihood ratio test is then employed to jointly estimate the scalar. and detection spectral reflectance vector The following optimization problem (3) is established: (3) in Represents vector The detected value, Represents scalar The estimated value, Represents a probability function. The variable represents the value that maximizes the expression described later. This represents the value of the variable that minimizes the expression described later. Represents vector Norm; Step 2: Based on principal component analysis, the detection of high-dimensional spectral reflectance vectors is transformed into the detection of low-dimensional weight vectors to reduce detection complexity and implementation cost. By comparing with the feature vector set established in the training phase, material color classification is achieved. Step 2 includes: 2.1 Covariance matrix of spectral reflectance vector samples Perform eigenvalue decomposition, let Indicates that the diagonal element is The diagonal matrix of eigenvalues, Representing an orthogonal matrix whose column vectors are eigenvectors, we obtain: (4) superscript To represent the transpose of a matrix, let Indicates by A matrix composed of principal direction basis vectors Let the weight vector be represented, then the spectral reflectance vector is approximately expressed as: (5) Definition corresponds to weight vector set ,in This represents the weight vector corresponding to the i-th monochromatic material; using the updated output vector expression of the sensing LED array, the weight vector is... The test is performed, and problem (3) is equivalently transformed into problem (6): (6) in Represents the weight vector The detected value, Represent the transformation matrix; define a vector. With corresponding set Problem (6) can be transformed into: (7) in Represents vector The detected value; using vector Given time observation scalar The estimated expression transforms the optimization problem (7) into: (8)。 2. The material color classification method based on LED sensing according to claim 1, characterized in that: According to the optimization problem (8), the material color classification problem is equivalent to the problem of classifying materials in the set. Find the output vector in exist The vector with the largest projection in the direction Define the eigenvector ,use As a key feature for detecting the color of materials, robustness to changes in link gain is achieved; Problem (8) can be solved in two stages: an offline training stage and a real-time testing stage. In the offline training stage, a set of feature vectors corresponding to each monochromatic material to be tested is established. , For the first The feature vector corresponding to each monochromatic material; during the real-time testing phase, the output vector of the sensing LED array is calculated. With sets The inner product of all feature vectors is used to select the color corresponding to the feature vector that maximizes the inner product, thereby classifying the material colors.

Images