Mobile phone appearance full-inspection system and method based on visual twinning

Abstract

The invention discloses a mobile phone appearance full-inspection system and method based on visual twinning, and relates to the technical field of mobile phone appearance detection.The method comprises the steps that a bionic optical sensor array is deployed, and mobile phone appearance data are collected; feature parameters are extracted, the optical features are coded into DNA sequences, and abnormal sites of the DNA sequences are automatically marked at material junctions; constructing a mobile phone appearance exclusive color gamut model; detection parameters are dynamically adjusted in the color gamut model, and the capture sensitivity to tiny chromatic aberration is enhanced; comparing chromaticity relevance of a standard prototype and a to-be-tested machine through quantum measurement, and automatically marking a problem area; and in combination with a mutation site in the DNA sequence, associating to a specific production process link through a pre-constructed causal association model. According to the invention, a four-dimensional color gamut model containing space, wavelength, time and observation angle is constructed, accurate detection of a gradient area is realized, and non-classical chromatic aberration correlation which cannot be captured by a traditional method is identified.

Description

Technical Field

[0001] This invention relates to the field of mobile phone appearance inspection technology, specifically a mobile phone appearance full inspection system and method based on visual twins. Background Technology

[0002] Traditional mobile phone appearance inspection relies on manual visual inspection or a single optical sensor (such as an RGB camera or spectrometer), which has problems such as strong subjectivity, low efficiency, and insensitivity to subtle color differences and material transitions.

[0003] Existing automated inspection systems struggle to simultaneously meet the demands of detecting composite defects in multiple materials (such as metals, glass, and coatings), and they are unable to correlate defects with production process parameters. When inspecting the smoothness of color transitions at the junctions of gradient back covers and multi-material joints, they lack dynamic modeling capabilities, easily missing subtle color differences. They can only identify surface defects and cannot use inspection data to deduce abnormalities in production parameters, making it difficult to achieve closed-loop quality control. Summary of the Invention

[0004] The purpose of this invention is to provide a mobile phone appearance full inspection system and method based on visual twins to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for full inspection of mobile phone appearance based on visual twins, the method comprising the following steps: S1. Deploy a biomimetic optical sensor array to collect mobile phone appearance data; the array includes a quantum dot fluorescence channel, a terahertz thermal imaging channel, a polarization holographic channel, and a plasma resonance channel; based on the output signals of each detection channel, competitive fusion is performed through a brain-like spiking neural network to output the mobile phone appearance material classification result; S2. Based on the thermal image obtained by the bionic optical sensor array, extract feature parameters, encode the optical features of each mobile phone exterior material into a DNA sequence, and automatically mark abnormal sites of the DNA sequence at the material interface to record the color difference diffusion trend. S3. Combining spatial coordinates, wavelength, time, and observation angle, a color gamut model exclusive to the mobile phone appearance is constructed; the material transition zone is accurately located through the plasma resonance channel, and the detection parameters are dynamically adjusted in the color gamut model to enhance the sensitivity to capture subtle color differences, so that the color transition smoothness of the gradient area meets the standard. S4. In the digital twin, the color states of the standard prototype and the test machine are quantum entangled. The color correlation between the standard prototype and the test machine is compared by quantum measurement. When the result violates the limit of Bell's inequality, it is determined that there is an unacceptable color difference and the problem area is automatically marked. S5. By combining mutation sites in the DNA sequence and using a pre-constructed causal relationship model, the abnormal production parameters are inferred in reverse based on the DNA sequence mutation type.

[0006] Based on the above scheme, the ultraviolet fluorescence channel identifies the molecular structure characteristics of organic coatings by capturing fluorescence signals of specific wavelengths; the terahertz thermal imaging channel identifies different metal materials by utilizing the characteristic spectrum generated by metal lattice vibrations; the polarization holographic channel analyzes the changes in the polarization state of light waves and reconstructs the stress distribution on the glass surface; and the plasma resonance channel is used to detect color abrupt change signals at the nanoscale.

[0007] Based on the above scheme, step S1 includes: S1-1. The raw signals acquired by each channel are converted into pulse timing patterns. The quantum dot fluorescence channel maps fluorescence lifetime to pulse interval, modulates pulse amplitude with spectral peaks, and outputs a pulse sequence characterized as the decay of excited states of coating molecules. The terahertz thermal imaging channel maps characteristic frequencies to pulse firing frequencies and thermal diffusivity to pulse widths, outputting pulse clusters reflecting lattice vibrational energy level transitions. The polarization holographic channel maps phase difference to pulse phase shift and polarization degree to pulse duty cycle, outputting pulse wave packets carrying stress birefringence phase information. The plasma resonance channel maps wavelength shift to pulse time shift and resonance intensity to pulse number. Output a pulse group modulated with the corresponding surface plasmon resonance frequency; S1-2. Input the pulse timing signal into the brain-like spiking neural network. The brain-like spiking neural network adopts a hierarchical spatiotemporal coding architecture, which includes an input layer, a hidden layer and an output layer. The input layer corresponds to quantum dot fluorescence, terahertz thermal imaging, polarization holography, and plasma resonance channels; each channel contains a neuron for receiving the pulse timing signals converted from each channel. The neuron membrane potential satisfies the following formula: ; in, V is expressed as the membrane potential decay constant; V(t) represents the membrane potential at time t; V rest V(t) represents the resting potential; I(t) represents the input pulse current; when V(t) ≥ V th Pulse is emitted at time, V th This is represented as the threshold potential, and then reset to V. rest ; The hidden layer includes adaptive threshold LIF neurons, which achieve channel signal competition through lateral inhibition connections; The output layer contains C classification neurons, where C represents the number of mobile phone appearance material categories; the output is the material classification probability distribution, which is obtained by normalizing the pulse firing frequency. In the described neuromorphic spiking neural network, each neuron dynamically adjusts its connection weights based on the precise timing and intensity of the input pulses, strengthening the transmission of channel signals closely related to specific material features and suppressing irrelevant or interfering signals, as shown in the following formula: ; Among them, w ij Let represent the connection weight from neuron i to neuron j, and represent the connection strength between the two neurons; η represents the learning rate, which controls the step size for weight updates; t i t represents the pulse firing time of neuron i; j Let K be the pulse firing time of neuron j; K() represents the pulse time-dependent plasticity kernel function. S1-3. During the competitive fusion process, the contribution of each channel is quantified by the pulse emission synchronization index, as shown in the following formula: ; Among them, Γ i Let represent the pulse synchronization index of the i-th channel, indicating the degree of synchronization between the pulse output of this channel and other channels; N represents the total number of channels participating in the competition; C ij (τ) represents the pulse cross-correlation function between channel i and channel j; C ii (0) and C jj (0) represents the autocorrelation function of the channel. The larger the autocorrelation value, the more frequently the pulses are emitted by the channel and the higher the signal strength. τ represents the time delay variable. The pulse cross-correlation function between channel i and channel j is given by the following formula: ; Among them, C ij (τ) represents the pulse cross-correlation function between channel i and channel j; T represents the upper limit of the time interval; t represents the time index; τ represents the time delay variable; S i (t) represents the pulse signal of channel i at time t; S j (t+τ) represents the pulse signal of channel j at time t+τ; During the competitive fusion process, the contribution of each channel is quantified by the pulse emission synchronization index Γ. The higher the value of Γ, the greater the weight of the channel in the final decision. S1-4. Output the material classification results based on the pulse synchronization index of each channel.

[0008] Based on the above scheme, step S2 includes: S2-1. Based on the thermal images of quantum dot fluorescence channel, terahertz thermal imaging channel, polarization holographic channel and plasma resonance channel, extract the optical characteristic parameters of the mobile phone appearance material; the characteristic parameters include fluorescence spectrum peak, fluorescence lifetime, lattice vibration characteristic frequency, thermal diffusion coefficient, stress birefringence phase difference, degree of polarization, surface plasmon resonance wavelength shift and resonance intensity. S2-2. Construct a mapping function between feature parameters and DNA bases to convert continuous feature parameters into discrete DNA sequences; S2-3. Based on the gradient changes in the heatmap, determine the material boundary line; extract the DNA sequences D from both sides of the boundary line. left and D right The boundary sequence D pre-existing in the standard twin std Perform dynamic time-normalized alignment and calculate the Hamming distance using the following formula: ; Where H represents the Hamming distance; M represents the DNA sequence length; δ() represents the indicator function used to determine whether bases are identical; D det (k) represents the k-th base of the detected sequence; k represents the index of the base; When the Hamming distance is greater than a preset threshold, the site k in the labeled sequence that differs from the standard value is defined as a mutation site, and the mutation type is recorded. S2-4. For consecutive samples of the same batch of mobile phones, collect DNA sequences in chronological order of production time, count the frequency of occurrence of each abnormal site in the time series, and construct a mutation hotspot map. A color difference diffusion model is constructed using the Fokker-Planck equation, as shown in the following formula: ; Where f(k,t) represents the mutation frequency at position k in the time sequence at time t; D diff V(k) is represented by the color difference diffusion coefficient, used to describe the propagation speed; V(k) is represented by the drift velocity, used to describe the migration direction of the mutation site. S2-5. Based on the color difference diffusion model, predict the diffusion range and severity of color difference in future production batches and generate early warning thresholds.

[0009] Based on the above scheme, step S3 includes: S3-1. Construct a four-dimensional color gamut model that includes spatial coordinates, wavelength, time, and observation angle, as shown in the following formula: ; Wherein, Ω(x,y,λ,t,θ) represents the four-dimensional color gamut model; B(λ) represents the spectral power distribution of the light source, used to simulate different ambient light; R(x,y,λ,t,θ) represents the reflectivity of the mobile phone appearance under spatial position, wavelength, time, and observation angle, obtained by fusing multi-channel data collected by a bionic optical sensor array; Y(λ) represents the CIE standard photometric function of the human eye, used to describe the sensitivity of the human eye to light of different wavelengths; λ1 and λ2 represent the lower and upper wavelength limits of the visible spectrum, respectively. S3-2. Obtain the spatial gradient of the plasma resonance channel signal. The gradient abrupt change is defined as the boundary of the material transition zone. The transition zone is represented as a parameterized curve L(α), where α∈[0,1] is the curve parameter, corresponding to the color transition progress. S3-3. Apply a multi-angle adjustable light source to the gradient region, ensuring the light source direction and observation angle satisfy the Brewster angle condition to suppress specular reflection interference. The formula is: θ i =arctan(n2 / n1); where θ i The direction of the light source is indicated by n; n1 and n2 represent the refractive indices of the materials on both sides of the transition zone, respectively. In the material transition zone region, the spectral sampling interval is increased, and the sampling is focused on wavelength regions that are sensitive to the human eye; S3-4. Obtain the color gradient along the transition zone curve L(α), using the following formula: ; Where C represents the color value in the CIE Lab color space; Represented as a color gradient; L * Represented as luminance coordinates in the CIELab color space; a * Represented as red-green axis coordinates in the CIE Lab color space; b * Represented as the yellow and blue axis coordinates in the CIE Lab color space; If the maximum value of the color gradient is less than the preset threshold, the color transition smoothness is determined to meet the standard; otherwise, it is marked as an abnormal area.

[0010] Based on the above scheme, step S4 includes: S4-1. Encode the color states of the standard prototype and the device under test into color quantum bit states respectively. Prepare the color entangled states of the standard prototype and the device under test through controlled gate operations, as shown in the following formula: ; It is represented as the color entanglement state between the standard prototype and the device under test, used to represent the quantum correlation of the color states between the standard prototype and the device under test; and These represent the quantum state representations of whether the standard prototype is qualified or unqualified, respectively. and These represent the quantum states of the tested machine as qualified and unqualified, respectively. S4-2. Measure the correlation function of entangled states under different operators, as shown in the following formula: ; Where E(c,d) represents the correlation function; c and d represent the measurement direction vectors; and σ represents the Pauli matrix vector. Represented as a tensor product operation, it is used to represent the joint measurement of a standard prototype and a machine under test; S4-3. Verify Bell's inequality, the formula is as follows: ; Where S represents the Bell parameter; c1 and c2 represent two different measurement direction vectors for the standard prototype; d1 and d2 represent two different measurement direction vectors for the device under test. S4-4. When the measurement result |S|>2, it is determined that there is a quantum correlation between the color state of the standard sample and the test sample, that is, the color difference exceeds the detection range, which is an unacceptable anomaly. The density matrix of the entangled state is reconstructed by quantum tomography to obtain the quantum variance of the color feature, as shown in the following formula: ; Where A represents the color feature operator; The quantum variance of color features is used to represent the quantum uncertainty of color difference; ρ represents the density matrix of entangled states reconstructed by quantum tomography. By combining a four-dimensional color gamut model, quantum variance is mapped to spatial coordinates of the phone's appearance, marking out areas of color difference abnormalities.

[0011] Based on the above scheme, step S5 includes: S5-1. Classify mutation sites in DNA sequences, including base substitutions, base insertions or deletions, and continuous mutation hotspots; quantify the impact of different process parameters on DNA mutations using historical production data to form a causal relationship map; use the causal relationship map to calculate the probability of association between each process step and mutation, and screen out process steps with high probability of association. S5-2. Evaluate the impact of each process parameter on DNA mutation, compare the actual production parameters with the standard values, and identify abnormal parameters. S5-3. Calculate the optimal abnormal process parameters using Bayesian inference, as shown in the following formula: ; in, This represents the optimal abnormal process parameters; Q represents the observed mutation type. This is represented as process parameter v under mutation type Q conditions. z The posterior probability; z represents the index of the process parameter; S5-4. Based on the optimal abnormal process parameters, output the set of process parameters that need to be adjusted, and give the direction and recommended range of parameter adjustment.

[0012] Based on the above scheme, the causal relationship graph includes: The causal relationship diagram is used to represent the causal relationship between process parameters and defects, and the formula is as follows: ; Where G represents the causal relationship graph, which is a directed acyclic graph; W represents the set of nodes used to represent production process parameters; and F represents the edit sum used to represent process parameter v. g For defect v h The causal relationship; The strength of a causal effect is expressed by the following formula: ; in, Represented as process parameter v g For defect v h The strength of the causal effect; This is represented as the process parameter v g Defect v under conditions h The probability of occurrence.

[0013] A mobile phone appearance full inspection system based on visual twins, the system includes: a data acquisition module, a material classification module, a visual twin modeling module, a color difference detection module, and a process optimization module; The data acquisition module includes a biomimetic optical sensor module and a signal preprocessing module; the biomimetic optical sensor module deploys four detection channels: quantum dot fluorescence, terahertz thermal imaging, polarization holography, and plasma resonance, to acquire mobile phone appearance data; the signal preprocessing module converts the raw signals from each channel into pulse timing patterns. The material classification module includes a spiking neural network and a channel competition module. The spiking neural network uses the pulse time-dependent plasticity mechanism to dynamically adjust the connection weights of neurons and strengthen the channel signals related to material features. The channel competition module quantifies the contribution of each channel through the pulse synchronization index and outputs the material classification result. The visual twin modeling module includes a DNA encoding module and a four-dimensional color gamut module. The DNA encoding module maps optical feature parameters to DNA sequences and marks mutation sites at material boundaries. The four-dimensional color gamut module integrates spatial coordinates, wavelength, time, and observation angle to construct a four-dimensional color gamut model, simulates color performance under different environments, and enhances the sensitivity of color difference detection in gradient areas. The color difference detection module includes a quantum state preparation module and a quantum decision module. The quantum state preparation module encodes the color parameters of the standard sample and the test machine into quantum bit states, prepares Bell entangled states, and establishes quantum correlations of color states. The quantum decision module verifies and determines color differences through Bell inequality and marks abnormal areas. The process optimization module includes a causal relationship mapping module and an anomaly reasoning module. The causal relationship mapping module uses historical production data to establish a causal relationship mapping and quantifies the causal effect strength between process parameters and DNA mutations. The anomaly reasoning module obtains the optimal anomaly parameters based on the causal relationship mapping and outputs an adjustment plan.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention collects multi-dimensional data through a biomimetic optical sensor array and combines it with competitive fusion of a pulse neural network to achieve accurate classification of mobile phone exterior materials; 2. This invention constructs a four-dimensional color gamut model that includes space, wavelength, time, and observation angle, enabling accurate detection of gradient regions and identification of non-classical color difference correlations that traditional methods cannot capture; 3. Based on causal relationship mapping, this invention reverse-engineers DNA mutation sites to specific production process parameters, thereby achieving precise location of defects. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of a mobile phone appearance full inspection method based on visual twins according to the present invention. Figure 2 This is a schematic diagram of the structure of a mobile phone appearance full inspection system based on visual twins according to the present invention. Detailed Implementation

[0016] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example: Figures 1-2 As shown, the present invention provides a technical solution, a method for full inspection of mobile phone appearance based on visual twins, the method comprising the following steps: S1. Deploy a biomimetic optical sensor array to collect mobile phone appearance data; the array includes a quantum dot fluorescence channel, a terahertz thermal imaging channel, a polarization holographic channel, and a plasma resonance channel; based on the output signals of each detection channel, competitive fusion is performed through a brain-like spiking neural network to output the mobile phone appearance material classification results; For example, deploying a biomimetic optical sensor array on a mobile phone production line to detect the back cover of a mobile phone made of magnesium alloy and glass with a blue-purple gradient color. Specifically, the ultraviolet fluorescence channel identifies the molecular structure characteristics of organic coatings by capturing fluorescence signals at specific wavelengths; the terahertz thermal imaging channel identifies different metal materials by utilizing the characteristic spectrum generated by metal lattice vibrations; the polarization holography channel analyzes changes in the polarization state of light waves to reconstruct the stress distribution on the glass surface; and the plasmon resonance channel is used to detect color abrupt changes at the nanoscale. For example: the quantum dot fluorescence channel uses a 365nm ultraviolet light source for excitation to collect the fluorescence lifetime of coating molecules, with a measured sample fluorescence lifetime of 10.5ns; the terahertz thermal imaging channel uses a 94GHz terahertz wave scan, with the characteristic frequency of magnesium alloy lattice vibration at 6.2THz, and the measured value for the sample is 6.0THz; the polarization holography channel uses 532nm polarized light incident, with a standard value of 0.2rad for the birefringence phase difference of stress on the glass surface, and the sample measured 0.35rad at the edge of the camera aperture; the plasmon resonance channel uses a gold nanoparticle coating with a resonance wavelength of 560nm, and the sample's gradient region resonates with a blue shift to 555nm. These are just examples and are not intended to be limiting. Furthermore, step S1 also includes: S1-1. Convert the raw signals acquired by each channel into a pulse timing mode; the output of the quantum dot fluorescence channel is characterized as a decay pulse sequence of excited states of coating molecules; the output of the terahertz thermal imaging channel reflects pulse clusters reflecting lattice vibrational energy level transitions; the output of the polarization holographic channel carries pulse wave packets carrying stress birefringence phase information; the output of the plasma resonance channel corresponds to the pulse group modulated by the surface plasmon resonance frequency; for example, convert the lattice vibration signal of the terahertz thermal imaging channel into a 10kHz pulse cluster with a pulse interval standard deviation of 50ns; S1-2. The pulse timing signal is input into the brain-like spiking neural network. Each neuron in the brain-like spiking neural network dynamically adjusts its connection weights according to the precise timing and intensity of the input pulse, strengthening the transmission of channel signals closely related to specific material features and suppressing irrelevant or interfering signals. The formula is as follows: ; Among them, w ij Let represent the connection weight from neuron i to neuron j, and represent the connection strength between the two neurons; η represents the learning rate, which controls the step size for weight updates; t i t represents the pulse firing time of neuron i; j Let K be the pulse firing time of neuron j; K() represents the pulse time-dependent plasticity kernel function. S1-3. During the competitive fusion process, the contribution of each channel is quantified by the pulse emission synchronization index, as shown in the following formula: ; Among them, Γ i Let represent the pulse synchronization index of the i-th channel, indicating the degree of synchronization between the pulse output of this channel and other channels; N represents the total number of channels participating in the competition; C ij (τ) represents the pulse cross-correlation function between channel i and channel j; C ii (0) and C jj (0) represents the autocorrelation function of the channel. The larger the autocorrelation value, the more frequently the pulses are emitted by the channel and the higher the signal strength. τ represents the time delay variable. For example: In magnesium alloy testing, the peak value C of the cross-correlation function between the terahertz channel and the plasma channel. ij (τ) = 0.82; Autocorrelation value C ii (0) = 1.2,;Pulse synchronization index Γ i =0.85; The pulse cross-correlation function between channel i and channel j is given by the following formula: ; Among them, C ij (τ) represents the pulse cross-correlation function between channel i and channel j; T represents the upper limit of the time interval; t represents the time index; τ represents the time delay variable; S i (t) represents the pulse signal of channel i at time t; S j (t+τ) represents the pulse signal of channel j at time t+τ; During the competitive fusion process, the contribution of each channel is quantified by the pulse emission synchronization index Γ. The higher the value of Γ, the greater the weight of the channel in the final decision. S1-4. Based on the pulse synchronization index of each channel, output the material classification results; for example: after fusion calculation, output the material probability: magnesium alloy (98.7%), glass (1.3%), which are judged as qualified substrates.

[0018] S2. Based on the thermal image obtained by the bionic optical sensor array, feature parameters are extracted, the optical features of each mobile phone exterior material are encoded into DNA sequences, and abnormal sites of the DNA sequences are automatically marked at the material interface to record the color difference diffusion trend. Specifically, step S2 includes: S2-1. Based on the thermal images of quantum dot fluorescence channel, terahertz thermal imaging channel, polarization holographic channel and plasma resonance channel, extract the optical feature parameters of the mobile phone appearance material; the feature parameters include fluorescence spectrum peak, fluorescence lifetime, lattice vibration characteristic frequency, thermal diffusivity, stress birefringence phase difference, degree of polarization, surface plasmon resonance wavelength shift and resonance intensity. S2-2. Construct a mapping function between feature parameters and DNA bases to convert continuous feature parameters into discrete DNA sequences D; S2-3. Based on the gradient changes in the heatmap, determine the material boundary line; extract the DNA sequences D from both sides of the boundary line. left and D right The boundary sequence D pre-existing in the standard twin std Perform dynamic time-normalized alignment and calculate the Hamming distance using the following formula: ; Where H represents the Hamming distance; M represents the DNA sequence length; δ() represents the indicator function used to determine whether bases are identical; D det (k) represents the k-th base of the detected sequence; k represents the index of the base; When the Hamming distance is greater than a preset threshold, the site k in the labeled sequence that differs from the standard value is defined as a mutation site and the mutation type is recorded; for example: Hamming distance H=1 (mutation at the 4th base). S2-4. For consecutive samples of the same batch of mobile phones, collect DNA sequences in chronological order of production time, count the frequency of occurrence of each abnormal site in the time series, and construct a mutation hotspot map. A color difference diffusion model is constructed using the Fokker-Planck equation, as shown in the following formula: ; Where f(k,t) represents the mutation frequency at position k in the time sequence at time t; D diff V(k) is represented by the color difference diffusion coefficient, used to describe the propagation speed; V(k) is represented by the drift velocity, used to describe the migration direction of the mutation site. S2-5. Based on the color difference diffusion model, predict the diffusion range and severity of color difference in future production batches and generate early warning thresholds.

[0019] S3. Combining spatial coordinates, wavelength, time, and observation angle, a color gamut model exclusive to the mobile phone appearance is constructed; the material transition zone is accurately located through the plasma resonance channel, and the detection parameters are dynamically adjusted in the color gamut model to enhance the sensitivity to capture subtle color differences, so that the color transition smoothness of the gradient area meets the standard. Specifically, step S3 includes: S3-1. Construct a four-dimensional color gamut model that includes spatial coordinates, wavelength, time, and observation angle, as shown in the following formula: ; Wherein, Ω(x,y,λ,t,θ) represents the four-dimensional color gamut model; B(λ) represents the spectral power distribution of the light source, used to simulate different ambient light; R(x,y,λ,t,θ) represents the reflectivity of the mobile phone appearance under spatial position, wavelength, time, and observation angle, obtained by fusing multi-channel data collected by a bionic optical sensor array; Y(λ) represents the CIE standard photometric function of the human eye, used to describe the sensitivity of the human eye to light of different wavelengths; λ1 and λ2 represent the lower and upper wavelength limits of the visible spectrum, respectively. For example: Spatial coordinates: from the top left corner of the back cover (0,0) to the bottom right corner (80,150) mm; Wavelength range: 400-700 nm, sampling interval 5 nm; S3-2. Obtain the spatial gradient of the plasma resonance channel signal. The gradient abrupt change is defined as the boundary of the material transition zone. The transition zone is represented as a parameterized curve L(α), where α∈[0,1] is the curve parameter, corresponding to the color transition progress. S3-3. Apply a multi-angle adjustable light source to the gradient region, ensuring the light source direction and observation angle satisfy the Brewster angle condition to suppress specular reflection interference. The formula is: θ i =arctan(n2 / n1); where θ i The direction of the light source is indicated by n; n1 and n2 represent the refractive indices of the materials on both sides of the transition zone, respectively. In the material transition zone region, the spectral sampling interval is increased, and the sampling is focused on wavelength regions that are sensitive to the human eye; S3-4. Obtain the color gradient along the transition zone curve L(α), using the following formula: ; Where C represents the color value in the CIE Lab color space; Represented as a color gradient; L * Represented as luminance coordinates in the CIELab color space; a * Represented as red-green axis coordinates in the CIE Lab color space; b * Represented as the yellow and blue axis coordinates in the CIE Lab color space; For example: the maximum gradient value of the sample at α=0.5 is 0.6ΔE / μm > 0.5ΔE; this is marked as an abnormal region; ΔE represents the color difference detection sensitivity. If the maximum value of the color gradient is less than the preset threshold, the color transition smoothness is determined to meet the standard; otherwise, it is marked as an abnormal area.

[0020] S4. In the digital twin, the color states of the standard prototype and the device under test are quantum entangled. The color correlation between the standard prototype and the device under test is compared by quantum measurement. When the result violates the limit of Bell's inequality, an unacceptable color difference is determined, and the problem area is automatically marked.

[0021] Specifically, step S4 includes: S4-1. Encode the color states of the standard prototype and the device under test into color quantum bit states respectively. Prepare the color entangled states of the standard prototype and the device under test through controlled gate operations, as shown in the following formula: ; It is represented as the color entanglement state between the standard prototype and the device under test, used to represent the quantum correlation of the color states between the standard prototype and the device under test; and These represent the quantum state representations of whether the standard prototype is qualified or unqualified, respectively. and These represent the quantum states of the tested machine as qualified and unqualified, respectively. S4-2. Measure the correlation function of entangled states under different operators, as shown in the following formula: ; Where E(c,d) represents the correlation function; c and d represent the measurement direction vectors; and σ represents the Pauli matrix vector. Represented as a tensor product operation, it is used to represent the joint measurement of a standard prototype and a machine under test; S4-3. Verify Bell's inequality, the formula is as follows: ; Where S represents the Bell parameter; c1 and c2 represent two different measurement direction vectors for the standard prototype; d1 and d2 represent two different measurement direction vectors for the device under test. S4-4. When the measurement result |S|>2, it is determined that there is a quantum correlation between the color state of the standard sample and the test sample, that is, the color difference exceeds the detection range, which is an unacceptable anomaly. The density matrix of the entangled state is reconstructed by quantum tomography to obtain the quantum variance of the color feature, as shown in the following formula: ; Where A represents the color feature operator; The quantum variance of color features is used to represent the quantum uncertainty of color difference; ρ represents the density matrix of entangled states reconstructed by quantum tomography. By combining a four-dimensional color gamut model, quantum variance is mapped to spatial coordinates of the phone's appearance, marking out areas of color difference abnormalities.

[0022] S5. By combining mutation sites in the DNA sequence and using a pre-constructed causal relationship model, the abnormal production parameters are inferred in reverse based on the DNA sequence mutation type.

[0023] Specifically, step S5 includes: S5-1. Classify mutation sites in DNA sequences, including base substitutions, base insertions or deletions, and continuous mutation hotspots; quantify the impact of different process parameters on DNA mutations using historical production data to form a causal relationship map; use the causal relationship map to calculate the probability of association between each process step and mutation, and screen out process steps with high probability of association. S5-2. Evaluate the impact of each process parameter on DNA mutation, compare the actual production parameters with the standard values, and identify abnormal parameters. S5-3. Calculate the optimal abnormal process parameters using Bayesian inference, as shown in the following formula: ; in, This represents the optimal abnormal process parameters; Q represents the observed mutation type. This is represented as process parameter v under mutation type Q conditions. z The posterior probability; z represents the index of the process parameter; For example, a mutation at the 4th base of the DNA sequence (G→T) corresponds to an abnormal anodic oxidation voltage. Bayesian reasoning yields the optimal adjustment value: v z =25V (actual measured value 28V, standard value 24V); S5-4. Based on the optimal abnormal process parameters, output the set of process parameters that need to be adjusted, and give the direction and recommended range of parameter adjustment.

[0024] Furthermore, causal relationship diagrams are used to represent the causal relationship between process parameters and defects, as shown in the following formula: ; Where G represents the causal relationship graph, which is a directed acyclic graph; W represents the set of nodes used to represent production process parameters; and F represents the edit sum used to represent process parameter v. g For defect v h The causal relationship; The strength of a causal effect is expressed by the following formula: ; in, Represented as process parameter v g For defect v h The strength of the causal effect; This is represented as the process parameter v g Defect v under conditions h The probability of occurrence.

[0025] This invention provides a technical solution: a method for full inspection of mobile phone appearance based on visual twins, involving the encoding conversion of optical feature parameters to DNA sequences; this is only an example and is not intended to be limiting. Four optical characteristic parameters of the mobile phone glass back cover coating were encoded into DNA sequences; Based on the thermal images of quantum dot fluorescence channel, terahertz thermal imaging channel, polarization holographic channel and plasma resonance channel, optical characteristic parameters of mobile phone appearance material are extracted; the characteristic parameters include fluorescence spectrum peak, fluorescence lifetime, lattice vibration characteristic frequency, thermal diffusivity, stress birefringence phase difference, degree of polarization, surface plasmon resonance wavelength shift and resonance intensity. The optical characteristic parameters of the mobile phone's exterior material are normalized. For example, the fluorescence lifetime quantization value is 2.3 ns; the minimum value is 1.0 ns; the maximum value is 3.0 ns; the normalization is (2.3-1.0) / (3.0-1.0)=0.65. The normalized values ​​are divided into four intervals, corresponding to the bases {A,T,C,G}, namely: [0,0.25)-A; [0.25,0.5)-T; [0.5,0.75)-C; [0.75,1]-G; The fluorescence lifetime quantization value of 0.65 corresponds to the base [0.5, 0.75)-C; Weighted fusion generates DNA sequences, with channel weights dynamically determined by pulse synchronization indicators; For example: Channel: F1, Γ value: 0.92, weight w m 0.35; Channel: F2, Γ value: 0.85, weight w m : 0.30; Channel: F3, Γ value: 0.78, weight w m : 0.20; Channel: F4, Γ value: 0.88, weight w m 0.15; Weighted quantization value: F = 0.35 × 0.65 + 0.30 × 0.50 + 0.20 × 0.56 + 0.15 × 0.40 = 0.54; Final base mapping: 0.54 ∈ [0.5, 0.75) - C; Output DNA sequence: Single-point sequence: D=C; Extended to length L=6 (6 consecutive detection points): D=CCCTAC; Alignment with standard sequence: Detection sequence: CCCTAC; Standard sequence: CCCCTC; Hamming distance H=1 (site k=5: AC mutation); Mutation type: Base substitution (AC), associated with uneven coating thickness.

[0026] This invention provides a technical solution: a method for full inspection of mobile phone appearance based on visual twins, specifically for detecting the smoothness of color difference in gradient areas; this is merely an example and not intended to be limiting. By combining spatial coordinates, wavelength, time, and observation angle, a color gamut model specifically designed for mobile phone appearance is constructed. For example: Spatial coordinates: x∈[0,50]mm (gradient direction), y∈[0,5]mm (width); Wavelength: λ∈[450,490]nm (blue gradient interval); Observation angle: θ=45° (simulating human eye perspective); Time: t=0; Plasma signal gradient abrupt change points: x=10mm (dark blue to medium blue); x=40mm (medium blue to light blue); A multi-angle adjustable light source is applied to the gradient area. The direction of the light source and the observation angle satisfy the Brewster angle condition. Brewster angle light source: material refractive index: glass refractive index = 1.52, coating refractive index = 1.48; light source angle: light source direction = arctan(1.48 / 1.52) = 44.2; effect: specular reflection intensity is reduced by 70%, highlighting the true color difference; The color gradient is obtained along the transition zone curve L(α): α=0.2-0.5: dC / dα=8.2; α=0.5-0.8 segment: dC / dα=7.6; The preset threshold dC / dα < 10 (industry standard); the maximum color gradient value is 8.2 < 10, and the gradient smoothness is qualified.

[0027] This invention provides a technical solution: a method for full inspection of mobile phone appearance based on visual twins, which reversely deduces abnormalities in spraying process parameters based on DNA sequence mutations; this is only an example and is not intended to be limiting. During the testing of a batch of gradient glass back covers, an abnormal color difference was found at the material interface, indicating that the Hamming distance exceeded the standard. Detection of sequences on both sides of the junction, left side (glass): D left =ATCGAT; Right side (coating): D right =TAGCGA; Standard sequence: D std =ATCGTA; The calculated Hamming distance H=1 exceeds the threshold H. max =0; Relevant content is obtained based on causal relationship graphs: Mutation type: base substitution (G→T); associated process parameter: spraying speed (v1); causality strength: 0.85; Mutation type: base insertion; associated process parameter: baking temperature (v2); causal strength: 0.72; The highest correlated process parameters were selected: spraying speed (v1) and causality strength 0.85. Standard spraying speed: 50±2cm / s; Actual speed of abnormal batches: 43cm / s; The optimal abnormal process parameters are calculated using Bayesian inference, as shown in the following formula: ; in, This represents the optimal abnormal process parameters; Q represents the observed mutation type. This is represented as process parameter v under mutation type Q conditions. z The posterior probability; z represents the index of the process parameter; Observed mutation Q: base substitution (G→T); prior probability of parameters (historical data): P(v1 is low) = 0.75; P(v2 is high) = 0.25; the optimal outlier parameter is the spraying speed. Based on the optimal abnormal process parameter of spraying speed, output the set of process parameters that need to be adjusted, and give the direction of parameter adjustment and recommended range; for example: increase the speed to 52cm / s, and check the nozzle wear condition at the same time.

[0028] This invention provides another technical solution: a mobile phone appearance full inspection system based on visual twins. The system includes: a data acquisition module, a material classification module, a visual twin modeling module, a color difference detection module, and a process optimization module. The data acquisition module includes a biomimetic optical sensor module and a signal preprocessing module. The biomimetic optical sensor module deploys four detection channels: quantum dot fluorescence, terahertz thermal imaging, polarization holography, and plasmon resonance, to acquire mobile phone appearance data. The signal preprocessing module converts the raw signals from each channel into pulse timing patterns. The material classification module includes a spiking neural network and a channel competition module. The spiking neural network uses the pulse time-dependent plasticity mechanism to dynamically adjust the connection weights of neurons and strengthen the channel signals related to material features. The channel competition module quantifies the contribution of each channel through the pulse synchronization index and outputs the material classification results. The visual twin modeling module includes a DNA encoding module and a four-dimensional color gamut module. The DNA encoding module maps optical feature parameters to DNA sequences and marks mutation sites at material boundaries. The four-dimensional color gamut module integrates spatial coordinates, wavelength, time, and observation angle to construct a four-dimensional color gamut model, simulates color performance under different environments, and enhances the sensitivity of color difference detection in gradient areas. The color difference detection module includes a quantum state preparation module and a quantum decision module. The quantum state preparation module encodes the color parameters of the standard sample and the test machine into quantum bit states, prepares Bell entangled states, and establishes quantum correlations of color states. The quantum decision module verifies and determines color difference through Bell's inequality and marks abnormal areas. The process optimization module includes a causal relationship mapping module and an anomaly reasoning module. The causal relationship mapping module uses historical production data to build a causal relationship mapping and quantifies the causal effect strength between process parameters and DNA mutations. The anomaly reasoning module, based on the causal relationship mapping, obtains the optimal anomaly parameters and outputs adjustment schemes.

[0029] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for full inspection of mobile phone appearance based on visual twins, characterized in that: The method includes the following steps: S1. Deploy a biomimetic optical sensor array to collect mobile phone appearance data; the array includes a quantum dot fluorescence channel, a terahertz thermal imaging channel, a polarization holographic channel, and a plasma resonance channel; based on the output signals of each detection channel, competitive fusion is performed through a brain-like spiking neural network to output the mobile phone appearance material classification result; S2. Based on the thermal image obtained by the bionic optical sensor array, extract feature parameters, encode the optical features of each mobile phone exterior material into a DNA sequence, and automatically mark abnormal sites of the DNA sequence at the material interface to record the color difference diffusion trend. S3. Combining spatial coordinates, wavelength, time, and observation angle, a color gamut model exclusive to the mobile phone appearance is constructed; the material transition zone is accurately located through the plasma resonance channel, and the detection parameters are dynamically adjusted in the color gamut model to enhance the sensitivity to capture subtle color differences, so that the color transition smoothness of the gradient area meets the standard. S4. In the digital twin, the color states of the standard prototype and the test machine are quantum entangled. The color correlation between the standard prototype and the test machine is compared by quantum measurement. When the result violates the limit of Bell's inequality, it is determined that there is an unacceptable color difference and the problem area is automatically marked. S5. By combining mutation sites in the DNA sequence and using a pre-constructed causal relationship model, the abnormal production parameters are inferred in reverse based on the DNA sequence mutation type.

2. The method for full inspection of mobile phone appearance based on visual twins according to claim 1, characterized in that: The ultraviolet fluorescence channel identifies the molecular structural features of the organic coating by capturing fluorescence signals at specific wavelengths. The terahertz thermal imaging channel uses the characteristic spectrum generated by the vibration of metal lattice to identify different metal materials; the polarization holographic channel analyzes the changes in the polarization state of light waves and reconstructs the stress distribution on the glass surface; the plasma resonance channel is used to detect color abrupt change signals at the nanoscale.

3. The method for full inspection of mobile phone appearance based on visual twins according to claim 1, characterized in that: Step S1 includes: S1-1. Convert the raw signals acquired by each channel into a pulse timing mode; the output of the quantum dot fluorescence channel is characterized as a decay pulse sequence of excited states of the coating molecules; the output of the terahertz thermal imaging channel reflects the pulse clusters of lattice vibrational energy level transitions; the output of the polarization holographic channel carries pulse wave packets carrying stress birefringence phase information; the output of the plasma resonance channel corresponds to the pulse group modulated by the surface plasmon resonance frequency. S1-2. The pulse timing signal is input into the brain-like spiking neural network. Each neuron in the brain-like spiking neural network dynamically adjusts its connection weights according to the precise timing and intensity of the input pulse, strengthening the transmission of channel signals closely related to specific material characteristics and suppressing irrelevant or interfering signals, as shown in the following formula: ； Among them, w ij Let represent the connection weight from neuron i to neuron j, and represent the connection strength between the two neurons; η represents the learning rate, which controls the step size for weight updates; t i t represents the pulse firing time of neuron i; j Let K be the pulse firing time of neuron j; K() represents the pulse time-dependent plasticity kernel function. S1-3. In the competitive fusion process, the contribution of each channel is quantified by the pulse emission synchronization index. The channel with the higher the synchronization index has a greater weight in the final decision. S1-4. Output the material classification results based on the pulse synchronization index of each channel.

4. The method for full inspection of mobile phone appearance based on visual twins according to claim 1, characterized in that: Step S2 includes: S2-1. Based on the thermal images of quantum dot fluorescence channel, terahertz thermal imaging channel, polarization holographic channel and plasma resonance channel, extract the optical characteristic parameters of the mobile phone appearance material; the characteristic parameters include fluorescence spectrum peak, fluorescence lifetime, lattice vibration characteristic frequency, thermal diffusion coefficient, stress birefringence phase difference, degree of polarization, surface plasmon resonance wavelength shift and resonance intensity. S2-2. Construct a mapping function between feature parameters and DNA bases to convert continuous feature parameters into discrete DNA sequences; Each feature parameter is divided into 4 quantization intervals, corresponding to the 4 bases of DNA; multi-channel features of the same material are weighted according to preset weights and then quantized to generate a DNA sequence of length L, as shown in the following formula: ； Where D represents the DNA sequence; F m Represented as the quantization feature value of the m-th channel; w m This is represented as channel weights, dynamically determined by the synchronization index; Encode() represents the feature-base mapping function. S2-3. Based on the gradient changes in the heatmap, determine the material boundary line; extract the DNA sequences D from both sides of the boundary line. left and D right The boundary sequence D pre-existing in the standard twin std Perform dynamic time-normalized alignment and calculate Hamming distance; when the Hamming distance is greater than a preset threshold, mark the site k in the sequence that differs from the standard value, define it as a mutation site, and record the mutation type; S2-4. For continuous samples of the same batch of mobile phones, DNA sequences were collected in chronological order of production time. The frequency of occurrence of each abnormal site in the time series was counted, and a color difference diffusion model was constructed using the Fokker-Planck equation. S2-5. Based on the color difference diffusion model, predict the diffusion range and severity of color difference in future production batches and generate early warning thresholds.

5. The method for full inspection of mobile phone appearance based on visual twins according to claim 1, characterized in that: Step S3 includes: S3-1. Construct a four-dimensional color gamut model that includes spatial coordinates, wavelength, time, and observation angle, as shown in the following formula: ； Wherein, Ω(x,y,λ,t,θ) represents the four-dimensional color gamut model; B(λ) represents the spectral power distribution of the light source, used to simulate different ambient light; R(x,y,λ,t,θ) represents the reflectivity of the mobile phone appearance under spatial position, wavelength, time, and observation angle, obtained by fusing multi-channel data collected by a bionic optical sensor array; Y(λ) represents the CIE standard photometric function of the human eye, used to describe the sensitivity of the human eye to light of different wavelengths; λ1 and λ2 represent the lower and upper wavelength limits of the visible spectrum, respectively. S3-2. Obtain the spatial gradient of the plasma resonance channel signal. The gradient abrupt change is defined as the boundary of the material transition zone. The transition zone is represented as a parameterized curve L(α), where α∈[0,1] is the curve parameter, corresponding to the color transition progress. S3-3. Apply a multi-angle adjustable light source to the gradient region, ensuring that the direction of the light source and the observation angle satisfy the Brewster angle condition to suppress specular reflection interference; in the material transition zone region, increase the spectral sampling interval and focus on sampling wavelength regions that are sensitive to the human eye. S3-4. Obtain the color gradient along the transition zone curve. When the maximum value of the color gradient is less than the preset threshold, the color transition smoothness is determined to meet the standard; otherwise, it is marked as an abnormal area.

6. The method for full inspection of mobile phone appearance based on visual twins according to claim 1, characterized in that: Step S4 includes: S4-1. Encode the color states of the standard prototype and the test machine into color quantum bit states respectively, and prepare the color entangled states of the standard prototype and the test machine through controlled gate operation; S4-2. Measure the correlation function of entangled states under different operators, as shown in the following formula: ； Where E(c,d) represents the correlation function; c and d represent the measurement direction vectors; and σ represents the Pauli matrix vector. Represented as a tensor product operation, it is used to represent the joint measurement of a standard prototype and a machine under test; It is represented as the color entanglement state between the standard prototype and the device under test, used to represent the quantum correlation of the color states between the standard prototype and the device under test; S4-3. Verify Bell's inequality, the formula is as follows: ； Where S represents the Bell parameter; c1 and c2 represent two different measurement direction vectors for the standard prototype; d1 and d2 represent two different measurement direction vectors for the device under test. S4-4. When the measurement result |S|>2, it is determined that there is a quantum correlation between the color state of the standard sample and the test machine, that is, the color difference exceeds the detection range and is an unacceptable anomaly. The density matrix of the entangled state is reconstructed by quantum tomography to obtain the quantum variance of the color features. Combined with the four-dimensional color gamut model, the quantum variance is mapped to the spatial coordinates of the mobile phone appearance to mark the color difference anomaly area.

7. The method for full inspection of mobile phone appearance based on visual twins according to claim 1, characterized in that: Step S5 includes: S5-1. Classify mutation sites in DNA sequences, including base substitutions, base insertions or deletions, and continuous mutation hotspots; quantify the impact of different process parameters on DNA mutations using historical production data to form a causal relationship map; use the causal relationship map to calculate the probability of association between each process step and mutation, and screen out process steps with high probability of association. S5-2. Evaluate the impact of each process parameter on DNA mutation, compare the actual production parameters with the standard values, and identify abnormal parameters. S5-3. Calculate the optimal abnormal process parameters using Bayesian inference, as shown in the following formula: ； in, This represents the optimal abnormal process parameters; Q represents the observed mutation type. This is represented as process parameter v under mutation type Q conditions. z The posterior probability; z represents the index of the process parameter; S5-4. Based on the optimal abnormal process parameters, output the set of process parameters that need to be adjusted, and give the direction and recommended range of parameter adjustment.

8. The method for full inspection of mobile phone appearance based on visual twins according to claim 7, characterized in that: The causal relationship map includes: The causal relationship diagram is used to represent the causal relationship between process parameters and defects, and the formula is as follows: ； Where G represents the causal relationship graph, which is a directed acyclic graph; W represents the set of nodes used to represent production process parameters; and F represents the edit sum used to represent process parameter v. g For defect v h The causal relationship.

9. A mobile phone appearance full inspection system based on visual twins, characterized in that: The system includes: a data acquisition module, a material classification module, a visual twin modeling module, a color difference detection module, and a process optimization module; The data acquisition module includes a biomimetic optical sensor module and a signal preprocessing module; the biomimetic optical sensor module deploys four detection channels: quantum dot fluorescence, terahertz thermal imaging, polarization holography, and plasma resonance, to acquire mobile phone appearance data; the signal preprocessing module converts the raw signals from each channel into pulse timing patterns. The material classification module includes a spiking neural network and a channel competition module. The spiking neural network uses the pulse time-dependent plasticity mechanism to dynamically adjust the connection weights of neurons and strengthen the channel signals related to material features. The channel competition module quantifies the contribution of each channel through the pulse synchronization index and outputs the material classification result. The visual twin modeling module includes a DNA encoding module and a four-dimensional color gamut module. The DNA encoding module maps optical feature parameters to DNA sequences and marks mutation sites at material boundaries. The four-dimensional color gamut module integrates spatial coordinates, wavelength, time, and observation angle to construct a four-dimensional color gamut model, simulates color performance under different environments, and enhances the sensitivity of color difference detection in gradient areas. The color difference detection module includes a quantum state preparation module and a quantum decision module. The quantum state preparation module encodes the color parameters of the standard sample and the test machine into quantum bit states, prepares Bell entangled states, and establishes quantum correlations of color states. The quantum decision module verifies and determines color differences through Bell inequality and marks abnormal areas. The process optimization module includes a causal relationship mapping module and an anomaly reasoning module. The causal relationship mapping module uses historical production data to establish a causal relationship mapping and quantifies the causal effect strength between process parameters and DNA mutations. The anomaly reasoning module obtains the optimal anomaly parameters based on the causal relationship mapping and outputs an adjustment plan.

Images