A Defect Detection Method Based on Machine Vision

Abstract

This invention relates to the field of precision inspection and discloses a defect detection method based on machine vision. The method includes: an illumination device projecting a modulated light field containing multiple modulation periods, wherein the spatial light intensity distribution is shifted at a preset speed between adjacent periods; a photoelectric detection array acquiring a time-varying light intensity sequence reflected from the material surface, the sequence containing the reflected light intensity values ​​of multiple sampling points within the corresponding period; extracting the transient change gradient of the reflected light intensity within adjacent periods for sampling points at the same physical coordinate position on the corresponding material surface; determining an intrinsic light intensity change threshold based on the shift speed and sampling frequency; and identifying sampling points whose transient change gradient exceeds the threshold as physical breakage defects. This invention utilizes the constraint law of the geometric continuity of the material surface on the optical response to achieve the physical separation of texture interference signals and defect abrupt change signals, thereby solving the problem of misjudgment in defect measurement under complex topological surfaces.

Description

Technical Field

[0001] This invention belongs to the field of precision inspection technology, and in particular relates to a defect detection method based on machine vision. Background Technology

[0002] Currently, in precision manufacturing processes, measuring the physical continuity of material surfaces through optical feedback is a routine step in ensuring product quality. The mainstream measurement method uses a static, uniform light source to illuminate the surface of the material under test, and uses a detection array to collect the spatial distribution of reflected light intensity. Defects are then identified based on the difference in amplitude of the reflected light intensity. However, for materials with complex intrinsic physical textures, the microscopic geometric undulations of the material surface produce significant optical scattering effects. Since existing measurement methods are generally based on the physical assumption of amplitude response differences, that is, distinguishing surface states by measuring the absolute amplitude of reflected light intensity, when testing precision machined surfaces, the amplitude of reflected light intensity generated by the intrinsic texture of the material and the scattering amplitude generated by minute physical fractures overlap by an order of magnitude in physical detection. This method, which relies on comparing spatially adjacent sampling points based on static amplitude characteristics, severs the connection between the evolution of optical response and the origin of the material's physical structure, leading to systematic optical misjudgments when the test system faces anisotropic physical topological surfaces.

[0003] Although the industry has attempted to increase the physical resolution of the detection array or introduce complex algorithm models to perform post-processing on the acquired data, these improvement paths have not changed the current situation where intrinsic texture signals and defect signals are confused in the amplitude dimension during the physical acquisition stage. Simply increasing the sampling density or computational complexity not only increases the hardware burden and data processing latency of the testing system, but also fails to decouple background noise and effective signals at the physical mechanism level. This makes the accuracy of measuring microscopic physical damage in materials limited by the scattering intensity of surface physical texture. Existing technologies are not only limited by the aforementioned hardware bottlenecks in detection, but also suffer from insufficient underlying mechanisms when exploring breakthroughs through software and control methods. For example, Chinese invention patent application CN121784135A discloses a method based on multi-sensor phase signals. The intelligent quantitative detection method for composite material defects relies heavily on complex data models such as convolutional neural networks and support vector machines to separate background interference after multi-sensor signal acquisition. The decoupling path at the pure software algorithm level implicitly assumes that the scattering generated by the intrinsic texture of the material and the defect features remain linearly separable in the discrete mathematical space. However, in real precision machining scenarios, the micro-undulations and local fractures of anisotropic surfaces exhibit highly nonlinear spatiotemporal aliasing in the physical response at the moment of acoustic or electromagnetic excitation. Without the dynamic intervention of the energy distribution by the physical excitation source, simply relying on the back-end algorithm to forcibly decouple not only significantly increases the computational load of the system and induces measurement lag, but also fails to recover the true boundary information annihilated by strong background scattering in the front-end discrete detection stage, thus causing the fidelity of defect feature extraction to encounter a fundamental limitation.

[0004] Therefore, the technical problem to be solved by this invention is how to extract dynamic features reflecting the topological continuity of materials from the source of physical excitation without changing the basic detection hardware, and to construct a measurement mechanism that can decouple intrinsic texture from microscopic physical fracture material. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A defect detection method based on machine vision, comprising the following steps: Step S101: The illumination device projects a modulated light field onto the surface of the material under test. The modulated light field contains multiple modulation cycles, and between adjacent modulation cycles, the spatial light intensity distribution of the modulated light field is controlled to undergo phase shift along a preset direction at a preset phase shift velocity, so as to construct a dynamically changing physical excitation environment on the surface of the material under test. Step S102: The photoelectric detection array is used to collect the time-varying light intensity sequence generated by the reflection of the surface of the material under test at a preset sampling frequency. The time-varying light intensity sequence includes the reflected light intensity values ​​collected by multiple sampling points on the photoelectric detection array in multiple modulation periods, which are used to characterize the time evolution characteristics of the optical response of the material surface with phase shift. Step S103: For sampling points at the same physical coordinate position on the surface of the material under test, extract the transient change gradient of reflected light intensity between adjacent modulation cycles; the transient change gradient of light intensity reflects the degree of optical response jump of the physical area covered by the sampling point during the spatial light intensity distribution movement. Step S104: Based on the geometric continuity constraint of the surface of the material under test, and combined with the preset phase shift velocity and preset sampling frequency, calculate the intrinsic light intensity change threshold of the sampling point during the modulation cycle switching process; the intrinsic light intensity change threshold characterizes the theoretical maximum light intensity change rate that the normal vector of the surface of the material under test can generate in the continuous transition state; by comparison and judgment, the sampling point whose transient light intensity change gradient exceeds the intrinsic light intensity change threshold is judged as a physical fracture defect point on the surface of the material under test, thereby realizing the physical isolation and accurate measurement of the intrinsic texture signal and the defect abrupt signal.

[0006] Preferably, before step S101, the method further includes the following steps: obtaining the intrinsic physical texture distribution of the surface of the material under test and extracting the principal feature direction of the intrinsic texture. The intrinsic physical texture distribution refers to the physical texture formed by the micro-geometric undulations on the surface of the material under test and the optical response field generated therefrom. The principal feature direction of the intrinsic texture is the core vector describing the macroscopic direction of the intrinsic physical texture distribution. The distribution direction of the spatial phase gradient of the modulation light field is adjusted according to the principal feature direction of the intrinsic texture so that the distribution direction of the spatial phase gradient is orthogonal to the principal feature direction of the intrinsic texture, thereby enhancing the nonlinear scattering contrast generated by the physical fracture defect and suppressing the background occlusion of the detection signal by the intrinsic texture.

[0007] Preferably, step S103 further includes the following sub-steps: step S1031: obtaining the relative displacement vector between the material under test and the photoelectric detection array; step S1032: calculating the physical coordinate offset of the sampling point in different modulation cycles based on the relative displacement vector; step S1033: using the physical coordinate offset to perform displacement compensation, accurately aligning the reflected light intensity values ​​collected in different modulation cycles to the same fixed physical area on the surface of the material under test, so as to eliminate the spatial positioning deviation caused by mechanical displacement.

[0008] Preferably, in step S101, a zero-excitation gap is inserted between adjacent modulation cycles. The zero-excitation gap refers to a time-domain interval without active excitation formed by temporarily interrupting the energy output of the lighting device during the physical gap between adjacent modulation cycles. In step S102, the ambient background light signal is synchronously acquired within the zero-excitation gap. In step S103, the acquired ambient background light signal is introduced as a background noise parameter into the time-series differential operation process to offset the common-mode interference generated by stray light sources in the industrial field, improve the signal-to-noise ratio of the extracted transient light intensity gradient, and ensure the stability of the physical feature parameter extraction.

[0009] Preferably, the method for determining the intrinsic light intensity change threshold in step S104 includes: determining the continuous reflection envelope of the normal vector of the surface of the material under test during the phase displacement process; calculating the theoretical light intensity change rate corresponding to the continuous reflection envelope based on the phase step corresponding to the preset phase displacement velocity and the law of reflection; and using the theoretical light intensity change rate to quantify and calibrate the intrinsic light intensity change threshold, thereby providing a kinematic judgment benchmark for identifying physical continuity breaks on the surface of the material under test.

[0010] Preferably, the modulated light field projected in step S101 includes a structured light field with a controlled wavelength and a specific polarization state; the phase change step corresponding to the preset phase shift velocity is set to match the microscopic topological size of the physical fracture defect, so as to ensure that the sampling point can capture the sub-pixel level reflection energy fluctuation caused by the physical fracture defect during the modulation cycle switching process.

[0011] Preferably, the reflected light intensity values ​​of each sampling point in the time-varying light intensity sequence correspond to the spatial discrete sampling results of the photoelectric detection array; the transient change gradient of light intensity is obtained by calculating the first derivative of the reflected light intensity values ​​of the sampling points in different modulation periods; by comparing the calculated first derivative with the intrinsic light intensity change threshold, the transient abrupt change features characterizing the physical continuity break of the material surface are extracted, thereby achieving accurate stripping of defect signals.

[0012] Preferably, in step S101, the excitation frequency of the modulated light field is kept in phase-locked synchronization with the preset sampling frequency to maintain the phase continuity of the time-varying light intensity sequence in the time domain; by establishing a precise temporal coupling relationship between the excitation source and the detector, a judgment model with a consistent spatiotemporal reference is provided to distinguish between the regular reflection of the intrinsic texture of the material and the irregular scattering of physical fracture defects, thereby eliminating the phase jitter error introduced by asynchronous sampling.

[0013] Preferably, after step S104, the following steps are also included: Step S1041: Perform spatial clustering processing on the identified multiple physical fracture defect points to extract the geometric boundary features of the defect area; Step S1042: Based on the amplitude of the transient change gradient of light intensity corresponding to the location of the physical fracture defect point, combined with the optical attenuation model of the material surface, the optical attenuation model is used to establish a quantitative mapping function between the amplitude of the transient change gradient of light intensity and the physical depth of the defect, calculate and invert the physical damage depth parameters of the tested material in the defect area, and realize the quantitative assessment of the degree of defect.

[0014] Compared with existing technologies, the defect detection method based on machine vision of this invention has the following advantages: 1. In defect detection using machine vision, by establishing a mapping logic between the temporal response of reflected light intensity and the physical continuity of materials, the judgment criterion is transformed from the absolute amplitude response of optical physical quantities to the transient gradient of optical physical quantities. This measurement path based on the rate of change in the time domain eliminates the scattering interference amplitude generated by the intrinsic texture of highly reflective anisotropic materials, avoids the masking of the scattering center of microscopic physical defects by the texture signal, and enables the measurement process for material physical properties to have the characteristic of being immune to surface geometric feature fluctuations, thereby optimizing the signal-to-noise ratio of physical state analysis under complex topological surfaces.

[0015] 2. The spatial phase shift velocity of the temporal optical excitation field and the sampling frequency of the detection array form a physical linkage, providing an objective kinematic judgment standard for distinguishing between continuous surfaces of materials and local physical fracture points. Since the surface normal vector of the intrinsic continuous surface of the material exhibits a continuous transition state, its reflected light intensity gradient is rigidly physically constrained by the displacement velocity of the excitation field. By identifying transient jumps that break through the theoretical optical response gradient limit, this invention achieves physical isolation of intrinsic structural interference and microscopic geometric faults, and accurately inverts the true physical damage state of the material.

[0016] 3. By utilizing a dynamic compensation mechanism for sampling coordinate offset, the system precisely anchors the detection sampling points under high-speed motion to a fixed physical region on the material surface. Combined with the environmental background optical response collected by the zero-excitation gap inserted in the time-series excitation sequence, this invention introduces a negative feedback parameter in the time-series differential processing. This mechanism achieves physical offsetting of common-mode interference caused by uncontrollable stray light sources in industrial sites and spatial coordinate errors caused by pipeline displacement, ensuring the reliability of physical feature parameter extraction under dynamic environments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the defect detection method based on temporal gradient extraction of the present invention; Figure 2 This is a diagram of the detection decision logic architecture of the present invention, which features multi-dimensional parameter collaboration. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0019] A defect detection method based on machine vision includes the following steps: Step S101: The illumination device projects a modulated light field onto the surface of the material under test. The modulated light field contains multiple modulation cycles, and between adjacent modulation cycles, the spatial light intensity distribution of the modulated light field is controlled to undergo phase shift along a preset direction at a preset phase shift velocity, so as to construct a dynamically changing physical excitation environment on the surface of the material under test. Step S102: The photoelectric detection array is used to collect the time-varying light intensity sequence generated by the reflection of the surface of the material under test at a preset sampling frequency. The time-varying light intensity sequence includes the reflected light intensity values ​​collected by multiple sampling points on the photoelectric detection array in multiple modulation periods, which are used to characterize the time evolution characteristics of the optical response of the material surface with phase shift. Step S103: For sampling points at the same physical coordinate position on the surface of the material under test, extract the transient change gradient of reflected light intensity between adjacent modulation cycles; the transient change gradient of light intensity reflects the degree of optical response jump of the physical area covered by the sampling point during the spatial light intensity distribution movement. Step S104: Based on the geometric continuity constraint of the surface of the material under test, and combined with the preset phase shift velocity and preset sampling frequency, calculate the intrinsic light intensity change threshold of the sampling point during the modulation cycle switching process; the intrinsic light intensity change threshold characterizes the theoretical maximum light intensity change rate that the normal vector of the surface of the material under test can generate in the continuous transition state; by comparison and judgment, the sampling point whose transient light intensity change gradient exceeds the intrinsic light intensity change threshold is judged as a physical fracture defect point on the surface of the material under test, thereby realizing the physical isolation and accurate measurement of the intrinsic texture signal and the defect abrupt signal.

[0020] Preferably, before step S101, the method further includes the following steps: obtaining the intrinsic physical texture distribution of the surface of the material under test and extracting the principal feature direction of the intrinsic texture. The intrinsic physical texture distribution refers to the physical texture formed by the micro-geometric undulations on the surface of the material under test and the optical response field generated therefrom. The principal feature direction of the intrinsic texture is the core vector describing the macroscopic direction of the intrinsic physical texture distribution. The distribution direction of the spatial phase gradient of the modulation light field is adjusted according to the principal feature direction of the intrinsic texture so that the distribution direction of the spatial phase gradient is orthogonal to the principal feature direction of the intrinsic texture, thereby enhancing the nonlinear scattering contrast generated by the physical fracture defect and suppressing the background occlusion of the detection signal by the intrinsic texture.

[0021] Preferably, step S103 further includes the following sub-steps: step S1031: obtaining the relative displacement vector between the material under test and the photoelectric detection array; step S1032: calculating the physical coordinate offset of the sampling point in different modulation cycles based on the relative displacement vector; step S1033: using the physical coordinate offset to perform displacement compensation, accurately aligning the reflected light intensity values ​​collected in different modulation cycles to the same fixed physical area on the surface of the material under test, so as to eliminate the spatial positioning deviation caused by mechanical displacement.

[0022] Preferably, in step S101, a zero-excitation gap is inserted between adjacent modulation cycles. The zero-excitation gap refers to a time-domain interval without active excitation formed by temporarily interrupting the energy output of the lighting device during the physical gap between adjacent modulation cycles. In step S102, the ambient background light signal is synchronously acquired within the zero-excitation gap. In step S103, the acquired ambient background light signal is introduced as a background noise parameter into the time-series differential operation process to offset the common-mode interference generated by stray light sources in the industrial field, improve the signal-to-noise ratio of the extracted transient light intensity gradient, and ensure the stability of the physical feature parameter extraction.

[0023] Preferably, the method for determining the intrinsic light intensity change threshold in step S104 includes: determining the continuous reflection envelope of the normal vector of the surface of the material under test during the phase displacement process; calculating the theoretical light intensity change rate corresponding to the continuous reflection envelope based on the phase step corresponding to the preset phase displacement velocity and the law of reflection; and using the theoretical light intensity change rate to quantify and calibrate the intrinsic light intensity change threshold, thereby providing a kinematic judgment benchmark for identifying physical continuity breaks on the surface of the material under test.

[0024] Preferably, the modulated light field projected in step S101 includes a structured light field with a controlled wavelength and a specific polarization state; the phase change step corresponding to the preset phase shift velocity is set to match the microscopic topological size of the physical fracture defect, so as to ensure that the sampling point can capture the sub-pixel level reflection energy fluctuation caused by the physical fracture defect during the modulation cycle switching process.

[0025] Preferably, the reflected light intensity values ​​of each sampling point in the time-varying light intensity sequence correspond to the spatial discrete sampling results of the photoelectric detection array; the transient change gradient of light intensity is obtained by calculating the first derivative of the reflected light intensity values ​​of the sampling points in different modulation periods; by comparing the calculated first derivative with the intrinsic light intensity change threshold, the transient abrupt change features characterizing the physical continuity break of the material surface are extracted, thereby achieving accurate stripping of defect signals.

[0026] Preferably, in step S101, the excitation frequency of the modulated light field is kept in phase-locked synchronization with the preset sampling frequency to maintain the phase continuity of the time-varying light intensity sequence in the time domain; by establishing a precise temporal coupling relationship between the excitation source and the detector, a judgment model with a consistent spatiotemporal reference is provided to distinguish between the regular reflection of the intrinsic texture of the material and the irregular scattering of physical fracture defects, thereby eliminating the phase jitter error introduced by asynchronous sampling.

[0027] Preferably, after step S104, the following steps are also included: Step S1041: Perform spatial clustering processing on the identified multiple physical fracture defect points to extract the geometric boundary features of the defect area; Step S1042: Based on the amplitude of the transient change gradient of light intensity corresponding to the location of the physical fracture defect point, combined with the optical attenuation model of the material surface, the optical attenuation model is used to establish a quantitative mapping function between the amplitude of the transient change gradient of light intensity and the physical depth of the defect, calculate and invert the physical damage depth parameters of the tested material in the defect area, and realize the quantitative assessment of the degree of defect.

[0028] Example 1: In a scenario where the surface quality of brushed aluminum alloy sheets with anisotropic physical texture is being inspected, the physical grooves distributed on the material surface generate an optical response that evolves with the phase under the projection of a modulated light field. Since the brushed texture has geometric continuity in space, when the spatial light intensity distribution of the modulated light field undergoes phase shift along a preset direction at a preset phase shift velocity, the reflected light intensity values ​​collected by the photoelectric detection array at a preset sampling frequency exhibit smooth fluctuation characteristics. When the system processes the time-varying light intensity sequence obtained from the sampling, it extracts the transient change gradient of light intensity at the same physical coordinate position on the surface of the tested material at adjacent sampling times. The calculation formula is as follows: ,in, The gradient represents the transient change in light intensity. In order to be in The value of reflected light intensity collected at any time. To be in adjacent sampling time The collected reflected light intensity values.

[0029] The system combines phase shift velocity and sampling frequency to determine the intrinsic light intensity variation threshold. This threshold is determined by the product of the phase shift velocity and the derivative of the spatial light intensity distribution function of the modulated light field. It is used to characterize the maximum theoretical response change rate of the intrinsic texture of the material surface under dynamic excitation. Due to the abrupt change in the surface normal vector at the actual physical defect point, when the phase edge of the modulated light field sweeps over the defect point, the reflected light intensity value undergoes a jump that exceeds the geometric continuity constraint, causing the calculated transient light intensity gradient to change. The intensity change is greater than the intrinsic light intensity change threshold; the system will then display the transient light intensity change gradient. Sampling points with intensity greater than the intrinsic light intensity change threshold are identified as physical fracture defects, thus separating the material's intrinsic physical texture interference signal from the sudden change in response signal caused by the defect. This measurement method based on the transient gradient in the time domain enables accurate extraction of surface topological damage, ensuring the reliability of material physical property analysis under strong scattering background.

[0030] Example 2: In the experimental scenario of determining the physical damage depth of a brushed stainless steel sheet with anisotropic scattering characteristics, the experimental platform includes a projection device with programmable spatial light modulation function and a photodetector array with a sampling frequency of 1000Hz and a pixel size of 5.5μm. The dynamic range of this photodetector array is not less than 60dB, used to capture the sequence of reflected light intensity changes on the sheet surface after being excited by the modulated light field in real time. The obtained data comes from the offline measurement process of industrial samples with different roughness levels by this experimental platform. The intrinsic physical texture distribution specifically refers to the micro-grooves formed on the surface of the stainless steel sheet by the brushing process and the resulting optical response field. This distribution determines the material's continuous... The evolution law of the normal vector under the morphology; the principal feature direction of the intrinsic texture serves as the core vector describing the macroscopic direction of the distribution, providing a definite geometric alignment reference for the phase step of the modulated light field, ensuring that the system can eliminate background interference from the brushed texture, and thus accurately capture the transient jump signal caused by the defect point; the setting of the key parameter sampling frequency is controlled by the phase shift velocity of the modulated light field and the spatial distribution density of the surface texture of the material under test. By balancing the capture integrity of the transient jump at the phase boundary with the system computational load, when the phase shift velocity is set to 100 rad / s and the average spatial period of the material surface texture is 50 μm, in order to satisfy the sampling law and ensure the extraction of the transient change gradient of light intensity. With sufficient signal resolution, the system set the sampling frequency to 500Hz. To simulate an industrial production environment, Gaussian white noise with a signal-to-noise ratio of 24.5dB and background light interference from the power frequency environment at a frequency of 50Hz were superimposed on the signal to verify the ability to suppress non-stationary background interference.

[0031] For the sample group of this invention with a physical crack of 5 μm depth on the surface, the original time-varying light intensity sequence output by the photoelectric detection array shows that its light intensity amplitude fluctuates within the range of 135.4 to 168.2. However, the intrinsic scattered light intensity amplitude generated by the brushed grooves on the stainless steel surface also falls within the overlapping range of 130.5 to 172.6, causing the control group, which is compared only based on amplitude characteristics, to be unable to identify the defect location in the time-domain waveform. The system extracts the time-domain gradient. Since the surface normal vector of the intrinsic texture exhibits a smooth change along the spatial distribution, the calculated intrinsic light intensity change threshold is 32.5. At the defect location, due to the abrupt change in the optical interface caused by the physical topological structure fracture, the extracted transient light intensity change gradient... The value jumped to 112.8 and exhibited pulse characteristics. The system compared the extracted transient gradient value with the intrinsic light intensity change threshold. The results showed that the signal extraction contrast of the sample group at the defect point was more than 3.5 times higher than that of the background texture, while the false detection rate of the control group due to amplitude overlap was 12.5%, and the measurement accuracy of the sample group of the present invention was 99.2%.

[0032] By setting up gradient experimental groups with defect depths of 2μm, 5μm, and 10μm, the measured data showed that as the core variable of defect depth increased, the measured transient change gradient of light intensity also increased. The peak values ​​were 58.4, 112.8, and 235.1, respectively, showing a monotonically increasing trend, proving that this physical characteristic parameter can quantify the degree of damage to the physical continuity of the material surface. In the quantification and inversion process, the optical attenuation model of the material surface established a quantitative mapping function between the transient change gradient of light intensity and the physical depth of the defect. Based on the optical reflectivity and spatial scattering distribution characteristics of a specific material, this model converts the observed nonlinear light intensity jump amplitude into the corresponding microscopic topological damage depth, thus providing a physical basis for the classification of defect severity. The nonlinear effect of phase shift velocity on detection reliability was further investigated. As the phase displacement velocity increases from 50 rad / s to 400 rad / s, the system's sensitivity to micro-fractures steadily improves. However, when the displacement velocity exceeds the performance inflection point of 500 rad / s, the extracted transient gradient signal exhibits motion blur and amplitude attenuation due to limitations in the integration time of the detection array and pixel response delay, resulting in a decrease in the signal-to-noise ratio from 18.6 dB to 6.2 dB. The above data confirms that by reconstructing the detection dimension into a temporal transient gradient constrained by physical geometry, the background occlusion of the material's intrinsic texture on defect measurement can be eliminated, achieving accurate characterization of surface micro-physical property damage.

[0033] Example 3: This example combines Figures 1 to 2 This paper describes a defect detection method based on machine vision, such as... Figure 1As shown, step S101 involves the illumination device projecting a modulated light field containing multiple modulation periods onto the surface of the material under test. Between adjacent modulation periods, the spatial intensity distribution of the modulated light field is controlled to undergo phase shift along a preset direction at a preset phase shift velocity, thereby constructing a dynamically changing physical excitation environment on the surface of the material under test. Step S102 involves using a photodetector array to collect a time-varying light intensity sequence generated by reflection from the surface of the material under test at a preset sampling frequency. This time-varying light intensity sequence includes the reflected light intensity values ​​collected by multiple sampling points on the photodetector array within multiple modulation periods, used to characterize the time evolution characteristics of the material surface's optical response with phase shift. Step S103 involves extracting the reflected light intensity at the sampling points corresponding to the same physical coordinate position on the surface of the material under test. The transient change gradient of light intensity between adjacent modulation cycles reflects the degree of optical response jump in the physical area covered by the sampling point during the spatial light intensity distribution movement. Step S104 is to calculate the intrinsic light intensity change threshold of the sampling point during the modulation cycle switching process based on the geometric continuity constraint of the surface of the material under test, combined with the preset phase shift velocity and the preset sampling frequency. The intrinsic light intensity change threshold characterizes the theoretical maximum light intensity change rate that the normal vector of the surface of the material under test can generate in the continuous transition state. By comparison and judgment, the sampling point whose transient change gradient of light intensity exceeds the intrinsic light intensity change threshold is judged as a physical fracture defect point on the surface of the material under test, thereby realizing the physical isolation and accurate measurement of the intrinsic texture signal and the defect abrupt signal.

[0034] like Figure 2 As shown, the architecture uses the defect detection decision core of machine vision as the top-level control center. Its lower-level parallel operations include a time-series dynamic physical excitation branch, an environmental background light signal offset logic, and a spatial displacement compensation branch. The time-series dynamic physical excitation branch directly acts on the time-varying light intensity sequence acquisition stage, while the spatial displacement compensation branch is responsible for performing the same physical coordinate resampling alignment to eliminate mechanical motion errors. The acquired light intensity sequence data, after introducing the negative feedback parameter generated by the environmental background light signal offset and undergoing coordinate resampling alignment calibration, converges into the calculation module that extracts the transient change gradient of light intensity. The calculated gradient value enters the core judgment stage, which judges whether the transient change gradient of light intensity exceeds the intrinsic light intensity change threshold. If the judgment condition is not met, the conclusion is output as: continuous surface intrinsic texture signal. If the judgment condition is met, the conclusion is output as: physical fracture defect point. For the identified defect points, spatial clustering of physical fracture defect points is performed to lock the geometric boundary, and finally, the degree of damage to the material surface is accurately assessed by inverting and quantifying the physical damage depth parameter.

[0035] Example 4: In the application scenario of structural integrity measurement of single-crystal silicon wafers with high-frequency concentric circular textures, the intrinsic physical texture of the single-crystal silicon wafer surface exhibits extremely high directional sensitivity in spatial distribution. Furthermore, uncontrolled external stray light sources interfere with the industrial measurement site. The system performs defect extraction according to established engineering procedures, acquires the distribution of intrinsic physical textures on the single-crystal silicon wafer surface, uses a photoelectric detection array to acquire texture images of the single-crystal silicon wafer surface under static lighting conditions, and uses the structural tensor algorithm to extract the principal feature directions of the intrinsic textures. During this process, the system calculates the gray-level gradient vector at each pixel in the image and constructs a local structure tensor matrix. By performing eigenvalue decomposition on this local structure tensor matrix, the direction of the eigenvector corresponding to the largest eigenvalue is determined as the principal feature direction of the intrinsic texture of the region. The system is based on the direction of the principal features of the intrinsic texture. Adjusting the spatial intensity distribution of the modulated light field, and controlling the fringe orientation of the modulated light field to influence the distribution direction of the spatial phase gradient. With the direction of the principal features of the intrinsic texture Maintain orthogonality and satisfy the formula °; This geometric arrangement minimizes the excitation response of the modulated light field to the smooth surface texture and suppresses the background occlusion of the probe signal by the intrinsic texture.

[0036] To eliminate common-mode interference caused by stray ambient light, the system inserts a 2ms zero-excitation gap between adjacent modulation periods of the temporal optical excitation field. The zero-excitation gap refers to a time-domain interval without active excitation formed by temporarily interrupting energy output through the lighting device during the physical gap between adjacent modulation periods. Its purpose is to extract the ambient background signal without excitation response characteristics from the detector, thus providing an objective physical reference for subsequent time-differential computation to cancel stray light interference in the industrial environment. During this zero-excitation gap, the excitation source stops outputting, and the photoelectric detection array synchronously acquires the ambient background light signal. In subsequent data processing, the system obtains the relative displacement vector between the monocrystalline silicon wafer and the photodetector array through the encoder feedback pulse of the transmission mechanism. Based on relative displacement vector Calculate the physical coordinate offset of the sampling point in different modulation periods It utilizes this to perform displacement compensation and resampling alignment on the reflected light intensity values ​​collected at different sampling times, ensuring that the light intensity data within different modulation periods correspond to the same fixed physical region on the surface of the single-crystal silicon wafer; the resampled and aligned reflected light intensity values ​​are then compared with the ambient background light signal. Performing temporal difference operations yields the net response sequence after filtering out background noise, thereby enhancing the gradient of transient light intensity changes. Extraction signal-to-noise ratio.

[0037] For the determination process, the system uses the following calculation model to determine the intrinsic light intensity change threshold. To establish a closed-loop reference for the detection logic; the system obtains the maximum derivative of the preset modulation light field spatial light intensity distribution function. And combined with the preset phase displacement velocity set in the measurement task According to the formula The intrinsic light intensity variation threshold was calculated. ;in, The threshold for intrinsic light intensity variation. The angular velocity of the phase shift. The sensitivity factor is the spatial light intensity distribution function as a function of phase. Specifically, the sensitivity factor... The quantitative calibration process is deeply coupled with the geometric continuity constraints of the surface of the material under test; during the system initialization phase, a pre-scan is performed on a standard smooth template (i.e., a defect-free surface with a continuous transition of normal vectors) to establish the spatial phase of the modulated optical field. With the light intensity received by the detector array pixels mapping function The maximum value of the derivative of this function is defined as Essentially, it represents the upper limit of the physical response rate induced by the spatial displacement of the excitation field under the ideal state of completely continuous geometric morphology on the material surface; this kinematic benchmark determined by the intrinsic reflection properties of the material is used to... With real-time phase displacement velocity Multiplication transforms the abstract geometric continuity principle into a directly comparable temporal gradient threshold. This model establishes a quantitative mapping relationship between the intrinsic physical property response of the material surface and the dynamic parameters of the excitation field, enabling the system to interpret the transient gradient of light intensity based on the calculation results. Sampling points exceeding this threshold were accurately identified as physical fracture defects. Measurement data showed that after introducing orthogonal excitation and background offsetting mechanisms, the false alarm signal intensity caused by concentric circle textures on the single-crystal silicon wafer surface decreased from 45.2 to 5.8, and the signal-to-noise ratio of the measurement system improved by 7.8 times. For micro-scratches with a depth of 1 μm, the extracted transient light intensity gradient... The value stabilized at around 88.6, which is much greater than the calculated intrinsic light intensity change threshold of 12.4. This measurement method based on physical mechanism analysis achieved stable capture of surface damage, ensuring the reliability of the evaluation of the physical continuity of single-crystal silicon wafers in complex industrial contexts.

[0038] Example 5: Before different batches of materials are connected to the detection system, the system establishes a baseline for pre-parameter parameters using a standard reflector; a plane mirror without physical texture is placed at the center of the effective field of view of the detection array, and the illumination device projects a spatial modulation period onto the plane mirror. The system modulates the optical field, controls the spatial light intensity distribution to shift in a step-by-step phase, records the voltage signal sequence output by each pixel of the detector array, reconstructs the envelope features of the spatial light intensity distribution curve, calculates the slope of the spatial light intensity distribution curve at different spatial coordinates through numerical difference operations, and extracts the maximum value of the derivative of the spatial light intensity distribution function. For the current material sample to be tested, the system acquires texture images and constructs a gray-level co-occurrence matrix under diffuse illumination with constant light intensity. By calculating the entropy distribution of the gray-level co-occurrence matrix in different offset directions, the texture direction features are determined, and the geometric angle with the minimum entropy value is determined as the intrinsic texture principal feature direction. .

[0039] When the system faces background drift caused by fluctuations in ambient light radiation, the judgment parameters are corrected by introducing physical offset compensation; the system obtains the dark current noise reference from the output of the occluded pixels in the detection array. and the ambient noise component fed back by the sensor The raw values ​​of the collected reflected light intensity Revised to The calculation formula is as follows: ,in, This is the corrected value for the reflected light intensity. This is the original value of the collected reflected light intensity. As a reference for pixel dark current noise, For environmental noise components; the system incorporates phase shift velocity. Calculate the threshold for intrinsic light intensity variation and the gradient of transient changes in light intensity With intrinsic light intensity change threshold The ratio is determined as the damage quantification index. When damage quantification indicators Greater than the preset judgment coefficient At that time, the system outputs the physical fracture defect judgment result, so that the measurement standard is dynamically correlated with the detector noise and environmental intensity.

[0040] Example 6: In a scenario where real-time surface physical damage measurement is performed on a high-speed, continuously operating strip steel production line, the system executes a spatiotemporal synchronization calibration procedure to establish a unique mapping relationship between sampling points and material physical coordinates; the real-time linear velocity of the material under test is determined by receiving the synchronization pulse signal output from the rotary encoder equipped on the production line conveyor mechanism. And based on the sampling frequency Calculate the physical displacement increment between adjacent sampling times The system utilizes physical displacement increments Spatial translation compensation is performed on the time-varying light intensity sequence captured by the detection array to align the reflected light intensity values ​​collected at different times to the same physical coordinate point on the strip surface, ensuring the calculated transient gradient of light intensity. This system accurately reflects the optical response jump sensitivity of the physical coordinate position under the evolution of the modulated light field. Considering the observation angle drift that may be caused by the micro-texture of the strip surface during high-speed displacement, the optical front end of this system is equipped with a telecentric imaging lens. Its object-side telecentricity ensures that the geometric projection relationship on the image plane does not change with the object distance fluctuation within a certain depth of field. Combined with the stability constraints of pipeline operation, the change in projection angle of the micro-topology within an extremely short sampling interval is controlled below the pixel-level resolution, thus allowing the displacement of the three-dimensional shape to be simplified to a precise coordinate translation of the two-dimensional image plane, maintaining the physical consistency of the temporal gradient calculation. The system uses an interpolation algorithm to resample the discrete pixel energy after displacement compensation, and the resampled reflected light intensity value is then used to... As input for subsequent gradient extraction steps, this eliminates pixel blurring and coordinate misalignment caused by high-speed strip movement. To maintain the edge sharpness of the light intensity gradient in the sub-pixel dimension, this embodiment preferably uses a bicubic interpolation algorithm. This algorithm considers the energy weights of pixels in the 4×4 neighborhood around the point to be resampled and uses a cubic polynomial to construct a continuous light intensity spatial distribution surface. Compared with linear interpolation, bicubic interpolation can more effectively suppress high-frequency aliasing noise introduced by non-integer coordinate translation, ensuring that the time-varying light intensity sequence after resampling can truly reflect the transient response of a fixed physical region and preventing the interpolation smoothing effect from masking subtle defect jump features.

[0041] When the system operates under metallurgical conditions with fluctuating ambient thermal radiation, the measurement reference is corrected by executing a pre-context suppression process. The system controls the lighting device to output a static calibration light field of a preset duration at the beginning of each detection cycle, and the photoelectric detection array synchronously acquires the reference response intensity fed back by the standard reflector. The system calculates The local variance in the spatial domain is used to evaluate the transmittance attenuation of the current optical path and generate a compensation operator. ;in, This is the standard light intensity response value under the factory calibration state of the equipment; the system multiplies the real-time collected reflected light intensity value by this compensation operator. Then, gradient calculation is performed to obtain the gradient of the transient change in light intensity. To eliminate common-mode amplitude attenuation interference caused by lens condensation or environmental dust, the determination results of the measurement system for physical fracture defects are always anchored to the physical constraints of the geometric continuity of the material surface.

[0042] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A method of machine vision-based defect detection, characterized in that, The method comprises the following steps: Step S101, the illumination device projects a modulated light field to the surface of the measured material; The modulated light field comprises a plurality of modulation periods, and between adjacent modulation periods, the spatial light intensity distribution of the modulated light field is phase-shifted at a preset phase shift speed along a preset direction to construct a dynamically changing physical excitation environment on the surface of the measured material; Step S102, a time-varying light intensity sequence generated by reflection of the surface of the measured material is collected by using a photoelectric detection array at a preset sampling frequency; the time-varying light intensity sequence comprises reflection light intensity values collected by a plurality of sampling points on the photoelectric detection array within a plurality of modulation periods, and is used to represent the time evolution characteristics of the optical response of the material surface with the phase shift; Step S103, for the sampling points corresponding to the same physical coordinate position of the surface of the measured material, the light intensity transient change gradient of the reflection light intensity between adjacent modulation periods is extracted; the light intensity transient change gradient reflects the degree of jump of the optical response of the physical region covered by the sampling point in the movement process of the spatial light intensity distribution; Step S104, based on the geometric continuity constraint of the surface of the measured material, the preset phase shift speed and the preset sampling frequency are combined to calculate the intrinsic light intensity change threshold of the sampling point in the modulation period switching process; the intrinsic light intensity change threshold represents the theoretical maximum light intensity change rate that can be generated by the normal vector of the surface of the measured material in the continuous transition state; by comparison and determination, the sampling points whose light intensity transient change gradient exceeds the intrinsic light intensity change threshold are determined as the physical fracture defect points of the surface of the measured material. 2.The machine vision-based defect detection method of claim 1, wherein, Before step S101, the following steps are further included: obtaining the intrinsic physical texture distribution of the surface of the measured material and extracting the intrinsic texture main feature direction; the intrinsic physical texture distribution refers to the physical texture formed by the micro-geometric undulation of the surface of the measured material and the optical response field generated thereby; the intrinsic texture main feature direction is a core vector describing the macro trend of the intrinsic physical texture distribution; the distribution direction of the spatial phase gradient of the modulated light field is adjusted according to the intrinsic texture main feature direction, so that the distribution direction of the spatial phase gradient is orthogonal to the intrinsic texture main feature direction, thereby enhancing the nonlinear scattering contrast generated by the physical fracture defect points and suppressing the background shielding of the intrinsic texture to the detection signal. 3.The machine vision-based defect detection method of claim 1, wherein, Step S103 further comprises the following sub-steps: step S1031, obtaining the relative displacement vector between the measured material and the photoelectric detection array; step S1032, calculating the physical coordinate offset of the sampling point within different modulation periods according to the relative displacement vector; step S1033, performing displacement compensation by using the physical coordinate offset to accurately align the reflection light intensity values collected within different modulation periods to the same fixed physical region of the surface of the measured material, so as to eliminate the spatial positioning deviation caused by mechanical displacement. 4.The machine vision-based defect detection method of claim 1, wherein, In step S101, a zero-excitation gap is inserted between adjacent modulation cycles. The zero-excitation gap refers to the time domain interval without active excitation formed by temporarily interrupting the energy output of the lighting device during the physical gap between adjacent modulation cycles. In step S102, the ambient background light signal is synchronously acquired within the zero-excitation gap. In step S103, the acquired ambient background light signal is introduced as a background noise parameter into the time-series differential operation process to offset the common-mode interference generated by stray light sources in the industrial field, improve the signal-to-noise ratio of the extracted transient light intensity gradient, and ensure the stability of the physical feature parameter extraction. 5.The machine vision-based defect detection method of claim 1, wherein, The method for determining the intrinsic light intensity change threshold in step S104 includes: determining the continuous reflection envelope of the normal vector of the surface of the material under test during the phase displacement process; calculating the theoretical light intensity change rate corresponding to the continuous reflection envelope based on the phase step corresponding to the preset phase displacement velocity and the law of reflection; and using the theoretical light intensity change rate to quantify and calibrate the intrinsic light intensity change threshold, thereby providing a kinematic judgment benchmark for identifying physical continuity breaks on the surface of the material under test. 6.The machine vision-based defect detection method of claim 1, wherein, The modulated light field projected in step S101 includes a structured light field with a controlled wavelength and a specific polarization state; the phase change step corresponding to the preset phase shift velocity is set to match the microscopic topological size of the physical fracture defect to ensure that the sampling point can capture the sub-pixel level reflection energy fluctuation caused by the physical fracture defect during the modulation cycle switching process. 7.The machine vision-based defect detection method of claim 1, wherein, The reflected light intensity values ​​at each sampling point in the time-varying light intensity sequence correspond to the spatial discrete sampling results of the photoelectric detection array; the transient change gradient of light intensity is obtained by calculating the first derivative of the reflected light intensity values ​​at the sampling points in different modulation periods; by comparing the calculated first derivative with the intrinsic light intensity change threshold, the transient abrupt change features characterizing the physical continuity break of the material surface are extracted, thereby achieving accurate stripping of defect signals. 8.The machine vision-based defect detection method of claim 1, wherein, In step S101, the excitation frequency of the modulated light field is kept in phase-locked synchronization with the preset sampling frequency to maintain the phase continuity of the time-varying light intensity sequence in the time domain; by establishing a precise temporal coupling relationship between the excitation source and the detector, a judgment model with a consistent spatiotemporal reference is provided to distinguish between the regular reflection of the intrinsic texture of the material and the irregular scattering of physical fracture defects, thereby eliminating the phase jitter error introduced by asynchronous sampling. 9.The machine vision-based defect detection method of claim 1, wherein, The steps following step S104 are as follows: Step S1041: Spatial clustering is performed on the identified multiple physical fracture defects to extract the geometric boundary features of the defect area; Step S1042: Based on the amplitude of the transient change gradient of light intensity corresponding to the location of the physical fracture defect, combined with the optical attenuation model of the material surface, the optical attenuation model is used to establish a quantitative mapping function between the amplitude of the transient change gradient of light intensity and the physical depth of the defect, calculate and invert the physical damage depth parameters of the tested material in the defect area, and realize the quantitative assessment of the degree of defect.

Images