Industrial visual defect detection system based on atomization development

Abstract

The invention discloses an industrial visual defect detection system based on atomization development. The system comprises an atomization development device, a synchronous imaging device, a differential feature extraction device, a defect recognition device and a self-adaptive control device. According to the system, a fog screen with a controllable particle size is sprayed to the surface of a workpiece, multiple frames of images are collected in a developing period, brightness and texture changes caused by fog drop attachment and evaporation are extracted, and a developing feature set is constructed. The defect identification device identifies the type, the position and the grade of the defect by using a learning network in which a physical constraint model is introduced; the adaptive control device dynamically adjusts developing and imaging parameters according to the recognition result. Furthermore, the system supports fogdrop particle size and temperature control matching, development stage segmented sensing, grade dynamic correction and subsequent workpiece-oriented parameter optimization, realizes a detection-identification-control closed loop, and is suitable for online defect detection in high-precision manufacturing.

Description

Technical Field

[0001] The present invention relates to the field of visual inspection technology, and in particular to an industrial visual defect inspection system based on atomization development. Background Art

[0002] With the development of industries such as precision manufacturing, electronic assembly, and high-end material surface treatment, online detection of workpiece surface defects has placed higher demands on real-time performance, resolution, and automation. Traditional machine vision-based inspection systems rely on direct imaging to extract grayscale, texture, or edge features. Their ability to visualize and identify subtle defects such as tiny scratches, coating peeling, and pores and microcracks remains limited. This is particularly true in scenes with complex natural lighting conditions or severe surface reflections and texture interference, resulting in low detection rates and high false positive rates.

[0003] To improve detection sensitivity, some solutions attempt to introduce developing media for auxiliary imaging, such as using spraying or atomization to create optical interference or adhesion response. However, these developing processes often rely on empirically set parameters and lack precise modeling of the temporal variations in the developing process. This results in poor imaging consistency and unstable feature extraction, making it difficult to meet the continuous inspection requirements of dynamic production lines.

[0004] Furthermore, most existing vision systems employ static sampling strategies and single-scale image analysis methods, making them difficult to adapt to the dynamic responses of different defects during the development process. They also lack detection mechanisms that match the physical evolution of defects, limiting the robustness and versatility of the systems under complex conditions. Therefore, developing an online vision inspection system with high adaptability and sensitivity in dynamic development environments remains a key technical challenge that needs to be addressed in the current industrial inspection field. Summary of the Invention

[0005] The purpose of the present invention is to provide an industrial visual defect detection system based on atomization development, which has the advantages of high automation, response sensitivity and overall recognition accuracy.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] Industrial visual defect inspection system based on atomization development, including:

[0008] The atomization developing device is used to spray a mist curtain of microdroplets with controllable particle size onto the workpiece surface. During the preset atomization drying cycle, the adhesion and evaporation of the microdroplets on the defect area trigger brightness and texture responses, forming a visible physical developing effect.

[0009] a synchronous imaging device for collecting image information of the workpiece surface at a plurality of preset time points during the atomization drying cycle to record the brightness and texture changes during the process of droplet attachment, stabilization and evaporation;

[0010] A differential feature extraction device is used to perform time-series differential processing on multiple frames of image information to extract a development feature set including droplet adhesion features and evaporation rate features;

[0011] A defect recognition device, based on a learning network that introduces a physical constraint model, classifies and identifies the development feature set and outputs a defect recognition result, wherein the defect recognition result includes information on the location, type, and level of the surface defect of the workpiece;

[0012] An adaptive control device is in communication with the defect recognition device, and is used to dynamically adjust the spray parameters of the atomizing development device and the sampling strategy of the synchronous imaging device according to the defect recognition result, and trigger local enhanced development and high-frequency image acquisition.

[0013] By adopting the above technical solution and constructing an integrated system including an atomizing development device, a synchronous imaging device, a differential feature extraction device, a defect recognition device and an adaptive control device, a closed-loop process of workpiece surface defects from development generation, image acquisition, feature processing, defect recognition to control feedback is realized, thereby improving the degree of automation, response sensitivity and overall recognition accuracy of detection.

[0014] Further configuration: the atomization developing device further includes a particle size control unit and a temperature control module, the particle size control unit and the temperature control module jointly adjust the droplet size and the developing temperature based on the defect response model so that the evaporation time of the droplets matches the target developing attachment time window, wherein:

[0015] The droplet size value and the development area temperature value form a pair of matching parameters, which are solved by the evaporation time calculation model so that the calculated droplet evaporation time is equal to the center value of the target development attachment time window.

[0016] By adopting the above technical solution, introducing a particle size control unit and a temperature control module, and jointly controlling the droplet size and development temperature based on the defect response model, this system can dynamically match the droplet evaporation time with the target development attachment time window, thereby obtaining high-contrast images of defect areas during the optimal time period. This coupled control strategy significantly improves the accuracy of the development response and its adaptability to different workpiece materials and environmental conditions, effectively enhancing the system's defect recognition capabilities and robustness.

[0017] Further setup: The defect response model is constructed by the following steps:

[0018] Collect a multi-frame image sequence of a workpiece with a known defect area during the development process;

[0019] Analyzing a brightness change curve of the defective area in the image sequence to identify a time segment corresponding to a maximum value of the development contrast change;

[0020] The time segment is defined as a target development and adhesion time window, and a mapping relationship between the time window and workpiece material parameters and environmental conditions is established to predict the optimal development time interval under different scenarios.

[0021] By adopting the above technical solution, the optimal development time segment is extracted based on the brightness change curve of the defect area, and a mapping relationship between it and the material properties and environmental parameters is established to achieve prediction and personalized regulation of the target development window, providing a data foundation and modeling basis for multi-scenario adaptive detection.

[0022] Further configuration: the synchronous imaging device includes:

[0023] a multi-band imaging unit, configured to synchronously collect multi-band image data of the workpiece surface at each preset timing point during the atomization drying cycle to obtain image information, wherein the multi-band image data includes at least one of a visible light band and an infrared band or an ultraviolet band;

[0024] The exposure adaptive adjustment unit is used to set independent exposure time and image gain parameters for the imaging channel of each band image data, and dynamically adjust the exposure settings based on the brightness histogram or brightness contrast information of the collected image data.

[0025] By adopting the above technical solutions, through multi-band synchronous imaging and band independent exposure control, the system's imaging adaptability on different material surfaces and lighting conditions is improved, while the perception ability of different types of defects (such as temperature difference type, reflective type, microstructure difference type) is enhanced, significantly expanding the system's scope of application and recognition robustness.

[0026] Further configuration: the differential feature extraction device further comprises:

[0027] a multi-band inter-frame cross-correlation analysis module, configured to perform cross-correlation calculation between adjacent image frames within a development cycle for each band image sequence of the image information, so as to obtain a temporal correlation distribution diagram reflecting the magnitude of brightness or texture changes during the process of droplet attachment and evaporation;

[0028] a main change trend identification module, configured to determine, based on the temporal correlation distribution graph, image regions and time periods with rapid correlation changes or extreme value reversal trends in the image as main change response regions caused by evaporation;

[0029] The model-driven feature dimension compression module is used to further screen the corresponding high-response channels, spatial sub-regions and key time segments in different bands within the main change response area based on the preset droplet evaporation dynamics model, and retain them as effective differential input features. The remaining dimensional feature information is processed or eliminated to construct a development feature set for subsequent defect identification.

[0030] By adopting the above technical solution, through inter-frame cross-correlation analysis, main change trend identification and physical model-driven dimensional compression, we can effectively focus on the image areas and time periods with the strongest response during the development process, reduce redundant calculations, improve feature sparsity and physical consistency, and enhance the effectiveness of differential features and the recognition accuracy of the model.

[0031] Further configuration: the defect identification device further comprises:

[0032] an evaporation process segmented perception module, configured to divide the development process into multiple physical stages, including a droplet attachment period, a stabilization period, and an evaporation acceleration period, based on the timestamp information of the image frames in the development feature set, development cycle parameters, and a droplet evaporation model, and to calibrate the image frame index range corresponding to each stage;

[0033] a defect evolution trend modeling module, configured to perform evolution feature extraction processing on the defect region in the development feature set in each physical stage, wherein the evolution features include but are not limited to the brightness contrast change rate of the defect region, the edge gradient enhancement rate, the shape contour change trend, and the response duration, so as to construct a trend feature vector of the defect evolution along the development stage;

[0034] The defect level dynamic adjustment module is used to determine the response intensity and evolution pattern of the defect during the development cycle based on the defect level information initially identified by the defect recognition device and the trend characteristic vector, and to increase, maintain or decrease the original defect level according to the preset level adjustment rules to generate a dynamically corrected defect level output.

[0035] By adopting the above technical solutions, segmented perception of the development cycle and modeling of defect evolution trends, the recognition system can not only determine whether a defect exists, but also perceive the development behavior of the defect in the time dimension, enhancing the recognition process's ability to respond to slight changes and early anomalies, and providing dynamic support for grade judgment.

[0036] Further configuration: the level adjustment rules are specifically as follows:

[0037] If the brightness contrast change rate of the defect area during the stable period and the evaporation acceleration period is greater than the first preset threshold, and the edge definition enhancement rate maintains positive growth for three consecutive time points, then the defect level is determined to be increased;

[0038] If the brightness contrast change rate of the defect area during the stable period is less than the second preset threshold, and the response change rate during the evaporation acceleration period is less than the change threshold, the original defect level is maintained;

[0039] If the brightness contrast change rate of the defective area during the evaporation acceleration period drops by more than a third preset threshold, or the development response duration is less than a minimum response time threshold, it is determined that the defect level is lowered.

[0040] By adopting the above technical solution and setting grade adjustment rules based on indicators such as brightness contrast change rate and edge enhancement rate, the defect grade judgment has a clear quantitative basis and physical interpretation capability, thus avoiding misjudgment and missed judgment, and improving the system's accuracy and stability in distinguishing multiple grades of defects.

[0041] Further configuration: the adaptive control device includes:

[0042] A defect result analysis module is used to receive and analyze the defect identification results output by the defect identification device, and extract statistical features related to defect location distribution, type frequency and level trend in the defect identification results;

[0043] A spray parameter optimization module is used to dynamically adjust the spray particle size, atomization flow rate, or fog curtain coverage angle of the atomization developer based on the distribution of defect types and grades to enhance the development response of similar defects in subsequent workpieces;

[0044] The sampling strategy optimization module is used to adjust the sampling frame rate, timing point density or local sampling weighting strategy of the synchronous imaging device according to the spatial distribution characteristics of defects in the previous batch of workpieces.

[0045] By adopting the above technical solution and analyzing the spatial distribution, type frequency and grade trend of defects in the identification results, the atomization parameters and sampling strategies of subsequent workpieces can be automatically optimized, forming a feedback self-regulation mechanism, thereby improving the system detection efficiency, robustness and resource utilization.

[0046] In summary, the present invention has the following beneficial effects: by constructing a closed-loop detection structure encompassing development generation, image acquisition, feature processing, defect recognition, and feedback control, it solves the core issues of traditional visual inspection, such as unstable defect development, redundant image information, weak feature response, and static control. The system has the following comprehensive advantages: the development process is controllable and defect response can be enhanced; differential features are accurately extracted and feature dimensions are automatically compressed; the recognition model embeds physical priors for high recognition accuracy; grade judgments have trend inference capabilities, making classification results more reliable; and the feedback mechanism enables subsequent self-optimization of detection, providing strong adaptability. It is suitable for online, real-time, high-resolution intelligent detection of surface defects on precision workpieces and complex materials, and has broad industrial application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is an overall structural block diagram of the embodiment. DETAILED DESCRIPTION

[0048] The present invention will be further described in detail below with reference to the accompanying drawings.

[0049] Example:

[0050] like Figure 1 As shown, the industrial visual defect detection system based on atomization development includes:

[0051] The atomization developing device is used to spray a mist curtain of microdroplets with controllable particle size onto the workpiece surface. During the preset atomization drying cycle, the adhesion and evaporation of the microdroplets on the defect area trigger brightness and texture responses, forming a visible physical developing effect.

[0052] a synchronous imaging device for collecting image information of the workpiece surface at a plurality of preset time points during the atomization drying cycle to record the brightness and texture changes during the process of droplet attachment, stabilization and evaporation;

[0053] A differential feature extraction device is used to perform time-series differential processing on multiple frames of image information to extract a development feature set including droplet adhesion features and evaporation rate features;

[0054] A defect recognition device, based on a learning network that introduces a physical constraint model, classifies and identifies the development feature set and outputs a defect recognition result, wherein the defect recognition result includes information on the location, type, and level of the surface defect of the workpiece;

[0055] An adaptive control device is in communication with the defect recognition device, and is used to dynamically adjust the spray parameters of the atomizing development device and the sampling strategy of the synchronous imaging device according to the defect recognition result, and trigger local enhanced development and high-frequency image acquisition.

[0056] The atomizing developer is equipped with a particle size control unit and a developer temperature control module, which are linked in real time with the developer timing control signal output by the defect recognition device. The core control logic relies on a pre-trained defect response model. This model predicts the optimal developer adhesion time window based on factors such as workpiece material, ambient temperature and humidity, and defect type, and reversely derives the required droplet size and developer temperature parameters.

[0057] In the present invention, the defect response model is a key model for predicting the optimal development attachment time window t*. The construction process includes the following steps:

[0058] (1) Image timing acquisition and brightness modeling

[0059] In a standard experimental environment, a workpiece with a known defect area is selected, and a synchronous imaging system is used to collect image sequences during the development process at a high frame rate. Assume that the image frame number is t∈[t0, t n ], definition: the average brightness of the defect area is I d (t), the average brightness of the background area is I b (t);

[0060] The defect contrast function is defined as:

[0061] C(t)=| I d (t)-I b (t)|

[0062] (2) Time window extraction

[0063] By taking the derivative of dC / dt, we can get the slope of the contrast change curve and find the interval of its local maximum point:

[0064]

[0065] The interval corresponding to this peak is set as [t1, t2], which is called the optimal development and attachment time window, that is, the time period when the defect development is clearest during the droplet evaporation process.

[0066] (3) Multivariate regression modeling

[0067] Through multiple experiments, we recorded the mapping relationship between t* and material and environmental variables under different conditions and established the following model:

[0068] t * =f(λ, θ, T env , H, M)

[0069] Where, λ: thermal conductivity of the workpiece, θ: surface contact angle, T env : Ambient temperature, H: Relative humidity, M:

[0070] Material category (embedded as a discrete variable).

[0071] This function can be implemented through methods such as neural network regression, support vector machine or polynomial fitting.

[0072] In order to make the droplet evaporation behavior match the target development and attachment time window t* output by the aforementioned defect response model, this system predicts the droplet evaporation time t by establishing a physical-thermodynamic model. evap , and accordingly reversely control the droplet size D and the developing temperature T.

[0073] In a standard windless environment, the fog droplets are approximately regarded as spherical. Based on the heat-mass diffusion coupling model, the fog droplet evaporation time is estimated as follows:

[0074]

[0075] Where, ρ: developer density (kg / m 3 ), D: average droplet diameter (m), k e : Evaporation mass transfer coefficient, related to air flow rate and material properties, P v (T): saturated vapor pressure at temperature T, which can be obtained from the Antoine equation, P a : Ambient water vapor partial pressure, which can be derived from relative humidity RH.

[0076] Find the combination of particle size D and temperature T so that the evaporation time t evap (D, T) is closest to the target time t* predicted by the defect response model. Let the optimization objective function be:

[0077] J(D, T) = |t evap (D, T)-t * | 2

[0078] The joint optimization problem is then:

[0079]

[0080] This problem can be solved by numerical optimization methods such as gradient descent, grid search or Bayesian parameter tuning.

[0081] The complete joint control process is:

[0082] The system receives the inspection task and identifies the current workpiece number;

[0083] Call the defect response model, input the current material parameters, ambient temperature and humidity, and predict the development time window t*;

[0084] According to the predicted t * , the optimal particle size D is calculated by joint optimization * and temperature T * ;

[0085] The particle size control unit sets the output particle size of the nebulizer (by adjusting the atomization frequency, air pressure, etc.);

[0086] Control the temperature control module to set the temperature of the developing air curtain or developing area to T * ;

[0087] This control process ensures the time domain accuracy and maximum response of the physical development process, enabling the subsequent image processing module to obtain high-quality input features, significantly improving the accuracy and robustness of defect recognition.

[0088] The synchronous imaging device comprises:

[0089] a multi-band imaging unit, configured to synchronously collect multi-band image data of the workpiece surface at each preset timing point during the atomization drying cycle to obtain image information, wherein the multi-band image data includes at least one of a visible light band and an infrared band or an ultraviolet band;

[0090] The exposure adaptive adjustment unit is used to set independent exposure time and image gain parameters for the imaging channel of each band image data, and dynamically adjust the exposure settings based on the brightness histogram or brightness contrast information of the collected image data.

[0091] The synchronous imaging device not only includes a standard visible light imaging unit for collecting image information during the development cycle, but also includes a set of multi-band imaging units and a set of exposure adaptive adjustment modules to improve the imaging contrast and multi-dimensional texture expression capabilities in each stage of droplet adhesion and evaporation.

[0092] The structural configuration of the multi-band imaging unit includes: a visible light imaging module for collecting images of brightness changes in standard surface textures; an infrared imaging module for enhancing the differences in the material's response to the heat conduction and evaporation rates of droplets; or an ultraviolet imaging module (UV) for highlighting the ultraviolet scattering characteristics of high-reflectivity surface defects.

[0093] The multi-band acquisition is performed synchronously at each preset timing point within the development cycle, ensuring that image information in different bands has a consistent time reference.

[0094] To adapt to the rapid changes in light intensity during the evaporation process and avoid overexposure or underexposure of the image, the system introduces an adaptive exposure adjustment mechanism. The process is as follows:

[0095] Step 1: Brightness histogram statistics

[0096] For each band image I i (x, y, t), calculate its grayscale histogram H i (L), where L is the brightness level, which defines the image brightness center value:

[0097]

[0098] Calculate the standard deviation of brightness (a measure of contrast):

[0099]

[0100] Step 2: Exposure time and gain dynamic adjustment strategy

[0101] Assume the initial exposure time is E i (0), the current exposure time is E i (t), then the control target is:

[0102]

[0103] Where: target : target brightness contrast, α: adjust the sensitivity coefficient, the value range is 0.5-1.5 (experience setting), when σ i When (t) is significantly smaller than the target value, the system will automatically increase the exposure time or image gain to enhance the dynamic range of image brightness.

[0104] The differential feature extraction device further comprises:

[0105] a multi-band inter-frame cross-correlation analysis module, configured to perform cross-correlation calculation between adjacent image frames within a development cycle for each band image sequence of the image information, so as to obtain a temporal correlation distribution diagram reflecting the magnitude of brightness or texture changes during the process of droplet attachment and evaporation;

[0106] a main change trend identification module, configured to determine, based on the temporal correlation distribution graph, image regions and time periods with rapid correlation changes or extreme value reversal trends in the image as main change response regions caused by evaporation;

[0107] The model-driven feature dimension compression module is used to further screen the corresponding high-response channels, spatial sub-regions and key time segments in different bands within the main change response area based on the preset droplet evaporation dynamics model, and retain them as effective differential input features. The remaining dimensional feature information is processed or eliminated to construct a development feature set for subsequent defect identification.

[0108] The three work together to extract the main response areas and time segments reflecting the droplet adhesion and evaporation behavior from multi-band image information, and construct a development feature set with sparse structure and strong physical correlation for subsequent analysis by the defect recognition device.

[0109] Multi-band inter-frame cross-correlation analysis module, which pre-processes the image information collected by the synchronous imaging device and calculates the inter-frame cross-correlation of the continuous image frames of each band image sequence to reflect the trend of brightness and texture changes during the droplet attachment and evaporation process. Processing process: Assume that the image sequence obtained under a certain band λ during the development cycle is {I λ (x, y, t1), I λ (x, y, t2),…,I λ (x, y, t n )}, the system performs cross-correlation calculation on adjacent frames and obtains the correlation coefficient R at each time t λ (t), which is used to reflect the trend of image similarity changes during the droplet development process.

[0110] For the image sequence in band λ, the normalized cross-correlation coefficient R between frame t and frame t+1 is λ (t) is calculated as follows:

[0111]

[0112] Where: μ t is the average pixel brightness of frame t.

[0113] Inter-frame cross-correlation change sequence {R λ (t)} is constructed into a time series correlation distribution diagram of the band for subsequent main trend judgment.

[0114] The main change trend identification module identifies the image areas and time periods with significant changes during the development cycle based on the time series correlation distribution diagram, that is, the spatial and temporal dimensions of the main response of the evaporation process to the surface features. Processing logic: The system extracts the main change response based on the following indicators:

[0115] 1. Time direction identification: Find the position with a sudden change (such as a sudden drop in slope or a minimum value point) in the mutual correlation coefficient sequence to determine whether the time period is likely to be a period of rapid evaporation of droplets;

[0116] 2. Spatial direction recognition: Construct a time series change trajectory for each pixel in the image and calculate its correlation change amplitude or maximum change rate;

[0117] 3. Joint construction: Mark the area with the above change characteristics as the main change response area, and record its corresponding time index range.

[0118] Calculate the cross-correlation change amplitude of each pixel position (x, y) in the image during the entire development cycle:

[0119]

[0120] Set the threshold τ and extract the main change response area:

[0121] Ω * ={(x, y)|C λ (x, y) ≥ τ}

[0122] The system also analyzes the changing trends in the time direction (such as minimum value points, slope mutations) to determine the main evaporation response time window.

[0123] The output is: main response space sub-region + main response time segment index.

[0124] The model-driven feature dimension compression module, combined with the preset droplet evaporation dynamics model, further screens the channels, spatial locations, and time segments that are sensitive to physical development behavior within the main change response area, and reduces or eliminates other dimensions to construct an effective differential input feature set. Model screening process:

[0125] The system is based on the following inputs: material surface parameters (thermal conductivity, contact angle, etc.); atomization development parameters (particle size, temperature and humidity); and identified main response areas and time periods.

[0126] Using the evaporation model, we estimate the response intensity of droplets in different channels. We screen: Band channels: retain bands with strong physical responses (such as infrared, which is sensitive to temperature differences); Spatial location: focus on pixel areas with significant evaporation residue contrast; Time segment: retain frame segments close to the "strongest evaporation response window." Finally, we construct a development feature set:

[0127]

[0128] Where: Λ*: the effective band set after model screening; Ω*: the main change response area; Main evaporation response time window.

[0129] The development feature set is structurally organized into a multi-band and multi-dimensional response tensor for the defect recognition device to further extract advanced discriminant features, such as defect contours, dynamic behavior of evaporation residues, texture mutations, etc., to improve overall recognition accuracy and classification robustness.

[0130] The defect recognition device further comprises:

[0131] an evaporation process segmented perception module, configured to divide the development process into multiple physical stages, including a droplet attachment period, a stabilization period, and an evaporation acceleration period, based on the timestamp information of the image frames in the development feature set, development cycle parameters, and a droplet evaporation model, and to calibrate the image frame index range corresponding to each stage;

[0132] a defect evolution trend modeling module, configured to perform evolution feature extraction processing on the defect region in the development feature set in each physical stage, wherein the evolution features include but are not limited to the brightness contrast change rate of the defect region, the edge gradient enhancement rate, the shape contour change trend, and the response duration, so as to construct a trend feature vector of the defect evolution along the development stage;

[0133] The defect level dynamic adjustment module is used to determine the response intensity and evolution pattern of the defect during the development cycle based on the defect level information initially identified by the defect recognition device and the trend characteristic vector, and to increase, maintain or decrease the original defect level according to the preset level adjustment rules to generate a dynamically corrected defect level output.

[0134] The specific level adjustment rules are as follows:

[0135] If the brightness contrast change rate of the defect area during the stable period and the evaporation acceleration period is greater than the first preset threshold, and the edge definition enhancement rate maintains positive growth for three consecutive time points, then the defect level is determined to be increased;

[0136] If the brightness contrast change rate of the defect area during the stable period is less than the second preset threshold, and the response change rate during the evaporation acceleration period is less than the change threshold, the original defect level is maintained;

[0137] If the brightness contrast change rate of the defective area during the evaporation acceleration period drops by more than a third preset threshold, or the development response duration is less than a minimum response time threshold, it is determined that the defect level is lowered.

[0138] This structure extracts the response trends of defects in different stages based on the physical change stages of droplet adhesion and evaporation during the development cycle, and automatically determines whether the original defect grade results should be corrected based on quantitative rules.

[0139] According to the time series and droplet physical model within the development cycle, the entire development process is divided into three physical stages: 1. Droplet adhesion period (initial rapid wetting stage); 2. Stable period (droplets basically cover the surface and evaporate slowly); 3. Evaporation acceleration period (droplet mass decreases rapidly and surface features are most obvious). Division basis: The system combines the current detection environment parameters (temperature T, humidity RH), the physical parameters of the developer (particle size D, vapor pressure P v ) and the workpiece material properties (thermal conductivity κ), the total evaporation time is calculated and estimated:

[0140]

[0141] The development period [0, t evap ] is divided into three stages: attachment period: [0, t1], stable period: [t1, t2], evaporation acceleration period: [t2, t evap ], t1 and t2 can be determined by the inflection point or change rate characteristics of the development response curve.

[0142] The dynamic changes of the defect area in the above three stages are modeled to form a trend vector. Assuming that the average brightness of the defect area at time t is I d (t), the non-defective area is Ib (t), defines the brightness contrast change rate within stage k:

[0143]

[0144] Where k = 1, 2, and 3 correspond to the adhesion period, stable period, and accelerated evaporation period, respectively. In addition, the system extracts the following features: edge definition change rate (e.g., Sobel gradient mean change); defect response area change rate; and defect development duration Td, defined as the total length of time in which the differential contrast in consecutive frames exceeds threshold T. These features are combined into a trend feature vector:

[0145] T d =[ΔC1,ΔC2,ΔC3,G1,G2,G3,T d ]

[0146] Combining preliminary recognition results with trend vectors, the need for defect grade correction is determined based on pre-defined quantitative rules. This embodiment effectively addresses the inaccurate recognition of initial minor defects, dynamically evolving defects, and defects with blurred boundaries in traditional single-frame analysis by introducing time-scale modeling and trend analysis during the development process. Dynamic adjustment of defect grades based on time-segment analysis and trend analysis enhances the recognition system's adaptability and accuracy across multiple scenarios and materials.

[0147] The adaptive control device comprises:

[0148] A defect result analysis module is used to receive and analyze the defect identification results output by the defect identification device, and extract statistical features related to defect location distribution, type frequency and level trend in the defect identification results;

[0149] A spray parameter optimization module is used to dynamically adjust the spray particle size, atomization flow rate, or fog curtain coverage angle of the atomization developer based on the distribution of defect types and grades to enhance the development response of similar defects in subsequent workpieces;

[0150] The sampling strategy optimization module is used to adjust the sampling frame rate, timing point density or local sampling weighting strategy of the synchronous imaging device according to the spatial distribution characteristics of defects in the previous batch of workpieces, increase the sampling density in possible high-risk areas, and implement sparse sampling in stable areas to optimize resource utilization and recognition accuracy.

[0151] By performing batch trend analysis on defect recognition results, the system can dynamically adjust the development and imaging parameters of subsequent workpieces, allowing the equipment to focus more on high-risk areas, defect types that are easily missed, or fuzzy locations of structural boundaries. It can also achieve parameter self-optimization based on historical recognition results, achieving higher efficiency and robustness in online inspection with limited resources.

[0152] Defect result analysis module: The task of this module is to extract statistical information from the defect recognition results of a batch or several continuous workpieces, including but not limited to: defect type frequency distribution; defect spatial location heat distribution (high frequency area); defect level change trend. Defect frequency statistical model: Let each defect type be D i , whose frequency of appearance in the last N artifacts is f i , define the relative frequency:

[0153]

[0154] Where M is the total number of defect types. The system will use high frequency P(D i )>θ f The defect type is used as the focus of parameter tuning.

[0155] Spray parameter optimization module: This module dynamically adjusts the following parameters of the atomization developing device based on the statistical distribution of defect types: average droplet size D; spray angle φ; atomization flow rate Q.

[0156] For high frequency defect type D k , if the expected development window duration is t k * , the system adjusts the spray particle size D so that the evaporation time tends to the target response:

[0157] t evap (D)≈t k *

[0158] Combined with the simplified evaporation model, the required particle size is inferred:

[0159]

[0160] The system adjusts the sprayer output parameters based on this.

[0161] Sampling strategy optimization module: This module analyzes the spatial distribution of defects of multiple workpieces, constructs a heat map, and configures the image frame sampling strategy in a differentiated manner.

[0162] The workpiece surface is divided into grid cells G(x, y), and the frequency of defects in each cell is accumulated:

[0163]

[0164] in Indicates that a defect is found at this position in the nth workpiece. The system detects that the heat is greater than the threshold τ H Assign a denser sampling frame rate or multiple time-series sampling points to the area to improve the time-series image resolution of the area.

[0165] Before each batch of detection tasks begins, the system outputs a set of parameter configuration schemes Θ={D * , Q * ,φ * , fs(x, y)}, used to adjust: the nozzle drive control parameters of the atomizing developing device; the frame rate table, exposure time distribution and local sampling density plan of the synchronous imaging device.

[0166] Through periodic analysis and feedback modeling of defect recognition results, this implementation can: make the development process more aligned with target defect response requirements; focus system resources on areas with higher recognition difficulty or dense distribution; and establish a control closed loop in which "recognition results feed back detection parameters," demonstrating good scalability and industrial adaptability.

[0167] In summary, the present invention has the following beneficial effects: by constructing a closed-loop detection structure encompassing development generation, image acquisition, feature processing, defect recognition, and feedback control, it solves the core issues of traditional visual inspection, such as unstable defect development, redundant image information, weak feature response, and static control. The system has the following comprehensive advantages: the development process is controllable and defect response can be enhanced; differential features are accurately extracted and feature dimensions are automatically compressed; the recognition model embeds physical priors for high recognition accuracy; grade judgments have trend inference capabilities, making classification results more reliable; and the feedback mechanism enables subsequent self-optimization of detection, providing strong adaptability. It is suitable for online, real-time, high-resolution intelligent detection of surface defects on precision workpieces and complex materials, and has broad industrial application value.

[0168] The above-described embodiments do not constitute a limitation on the scope of protection of this technical solution. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the above-described embodiments shall be included in the scope of protection of this technical solution.

Claims

1. The industrial visual defect detection system based on atomization development is characterized by: include: The atomization developing device is used to spray a mist curtain of microdroplets with controllable particle size onto the workpiece surface. During the preset atomization drying cycle, the adhesion and evaporation of the microdroplets on the defect area trigger brightness and texture responses, forming a visible physical developing effect. a synchronous imaging device for collecting image information of the workpiece surface at a plurality of preset time points during the atomization drying cycle to record the brightness and texture changes during the process of droplet attachment, stabilization and evaporation; A differential feature extraction device is used to perform time-series differential processing on multiple frames of image information to extract a development feature set including droplet adhesion features and evaporation rate features; A defect recognition device, based on a learning network that introduces a physical constraint model, classifies and identifies the development feature set and outputs a defect recognition result, wherein the defect recognition result includes information on the location, type, and level of the surface defect of the workpiece; An adaptive control device is in communication with the defect recognition device and is used to dynamically adjust the spray parameters of the atomizing development device and the sampling strategy of the synchronous imaging device according to the defect recognition result.

2. The industrial visual defect detection system based on atomization development according to claim 1, characterized in that: The atomization developing device further includes a particle size control unit and a temperature control module, which jointly adjust the droplet size and the developing temperature based on the defect response model to match the evaporation time of the droplets with the target developing attachment time window, wherein: The droplet size value and the development area temperature value form a pair of matching parameters, which are solved by the evaporation time calculation model so that the calculated droplet evaporation time is equal to the center value of the target development attachment time window.

3. The industrial visual defect detection system based on atomization development according to claim 2, characterized in that: The defect response model is constructed by the following steps: Collect a multi-frame image sequence of a workpiece with a known defect area during the development process; Analyzing a brightness change curve of the defective area in the image sequence to identify a time segment corresponding to a maximum value of the development contrast change; The time segment is defined as a target development and adhesion time window, and a mapping relationship between the time window and workpiece material parameters and environmental conditions is established to predict the optimal development time interval under different scenarios.

4. The industrial visual defect detection system based on atomization development according to claim 1, characterized in that: The synchronous imaging device comprises: a multi-band imaging unit, configured to synchronously collect multi-band image data of the workpiece surface at each preset timing point during the atomization drying cycle to obtain image information, wherein the multi-band image data includes at least one of a visible light band and an infrared band or an ultraviolet band; The exposure adaptive adjustment unit is used to set independent exposure time and image gain parameters for the imaging channel of each band image data, and dynamically adjust the exposure settings based on the brightness histogram or brightness contrast information of the collected image data.

5. The industrial visual defect detection system based on atomization development according to claim 4 is characterized in that: The differential feature extraction device further comprises: a multi-band inter-frame cross-correlation analysis module, configured to perform cross-correlation calculation between adjacent image frames within a development cycle for each band image sequence of the image information, so as to obtain a temporal correlation distribution diagram reflecting the magnitude of brightness or texture changes during the process of droplet attachment and evaporation; a main change trend identification module, configured to determine, based on the temporal correlation distribution graph, image regions and time periods with rapid correlation changes or extreme value reversal trends in the image as main change response regions caused by evaporation; The model-driven feature dimension compression module is used to further screen the corresponding high-response channels, spatial sub-regions and key time segments in different bands within the main change response area based on the preset droplet evaporation dynamics model, and retain them as effective differential input features. The remaining dimensional feature information is processed or eliminated to construct a development feature set for subsequent defect identification.

6. The industrial visual defect detection system based on atomization development according to claim 1, characterized in that: The defect recognition device further comprises: an evaporation process segmented perception module, configured to divide the development process into multiple physical stages, including a droplet attachment period, a stabilization period, and an evaporation acceleration period, based on the timestamp information of the image frames in the development feature set, development cycle parameters, and a droplet evaporation model, and to calibrate the image frame index range corresponding to each stage; a defect evolution trend modeling module, configured to perform evolution feature extraction processing on the defect region in the development feature set in each physical stage, wherein the evolution features include but are not limited to the brightness contrast change rate of the defect region, the edge gradient enhancement rate, the shape contour change trend, and the response duration, so as to construct a trend feature vector of the defect evolution along the development stage; The defect level dynamic adjustment module is used to determine the response intensity and evolution pattern of the defect during the development cycle based on the defect level information initially identified by the defect recognition device and the trend characteristic vector, and to increase, maintain or decrease the original defect level according to the preset level adjustment rules to generate a dynamically corrected defect level output.

7. The industrial visual defect detection system based on atomization development according to claim 6, characterized in that: The specific level adjustment rules are as follows: If the brightness contrast change rate of the defect area during the stable period and the evaporation acceleration period is greater than the first preset threshold, and the edge definition enhancement rate maintains positive growth for three consecutive time points, then the defect level is determined to be increased; If the brightness contrast change rate of the defect area during the stable period is less than the second preset threshold, and the response change rate during the evaporation acceleration period is less than the change threshold, the original defect level is maintained; If the brightness contrast change rate of the defective area during the evaporation acceleration period drops by more than a third preset threshold, or the development response duration is less than a minimum response time threshold, it is determined that the defect level is lowered.

8. The industrial visual defect detection system based on atomization development according to claim 1, characterized in that: The adaptive control device comprises: A defect result analysis module is used to receive and analyze the defect identification results output by the defect identification device, and extract statistical features related to defect location distribution, type frequency and level trend in the defect identification results; A spray parameter optimization module is used to dynamically adjust the spray particle size, atomization flow rate, or fog curtain coverage angle of the atomization developer based on the distribution of defect types and grades to enhance the development response of similar defects in subsequent workpieces; The sampling strategy optimization module is used to adjust the sampling frame rate, timing point density or local sampling weighting strategy of the synchronous imaging device according to the spatial distribution characteristics of defects in the previous batch of workpieces.

Images