Surface flaw detection device for hollow glass bead thermal insulation material

Abstract

The invention relates to the technical field of machine vision detection, and discloses a hollow glass bead thermal insulation material surface flaw detection device, which comprises a table top, a detection box, a light source, a detection camera and a belt conveyor, and is characterized in that a flaw classification threshold real-time adjustment system is integrated. The system calculates an image quality coefficient, environment interference-image quality collaboration, a threshold suitability coefficient and an equipment attenuation coefficient in real time through an image quality evaluation module, a collaboration evaluation module, a threshold suitability evaluation module and an equipment attenuation evaluation module respectively. And the flaw classification threshold adjustment module dynamically determines and adjusts a target flaw classification threshold through linear mapping based on the product relationship of the coefficients and in combination with preset threshold upper and lower limits. According to the method, the problem that a fixed threshold cannot adapt to the environment change of a complex industrial site is solved, intelligent self-adaptive optimization of the detection threshold is realized, and the accuracy, stability and robustness of flaw detection are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of machine vision inspection technology, and in particular relates to a device for detecting surface defects in hollow glass microsphere thermal insulation materials. Background Technology

[0002] Insulating glass microspheres, with their advantages of being lightweight, high-strength, and having low thermal conductivity, have irreplaceable application value in building energy conservation, aerospace, and special packaging. The surface integrity of these materials directly determines their insulation performance and service life. Therefore, efficient and precise automated detection of microscopic defects such as scratches, dents, bubbles, and uneven coatings during the production process is a core aspect of quality control. In current industrial practice, machine vision-based online inspection systems have become the mainstream solution. A typical architecture includes a fixed light source, an industrial camera, an image processing unit, and a transmission mechanism, using image segmentation algorithms with preset fixed thresholds to identify surface defects. However, the real-world continuous production environment presents multiple severe challenges to the detection system: industrial sites are generally plagued by random fluctuations in ambient lighting and mechanical vibrations of equipment, causing unpredictable time-varying characteristics in the contrast, brightness, and clarity of the acquired images; simultaneously, the light source components inevitably experience light intensity attenuation over time, which, if not calibrated in time, leads to systematic imaging deviations and continuous drift of the detection benchmark; furthermore, the dynamic adjustment of production cycle causing changes in conveyor belt speed, and the inherent differences in background color and the complexity of background textures among different batches of raw materials, further exacerbate the dynamic uncertainty of the detection conditions. The fundamental flaw of existing technologies lies in their widespread use of static classification threshold mechanisms, which cannot adaptively adjust according to real-time operating conditions. This rigid design causes the system to lose its intelligent balance between false alarms and false negatives—when the equipment is operating well and the imaging quality is stable, the system should raise the threshold to suppress false alarms and optimize sorting efficiency, but a fixed threshold cannot capture such opportunities; conversely, under conditions of equipment aging or environmental degradation, the system urgently needs to lower the threshold to ensure defect detection rate and avoid quality leakage, but existing solutions cannot respond to these needs. In essence, traditional detection devices lack the ability to collaboratively perceive the dynamic evolution of image quality, the coupling effect of environmental interference, the degradation of equipment health status, and historical detection feedback from multiple dimensions. Furthermore, they lack a closed-loop threshold adjustment mechanism based on these parameters, making it difficult for the system to maintain long-term stable detection accuracy and reliability in complex industrial scenarios. Therefore, there is an urgent need to develop an intelligent adjustment architecture that can deeply integrate real-time image quality assessment, quantitative analysis of environmental interference, dynamic modeling of equipment degradation, and detection feedback to achieve precise adaptive optimization of defect classification thresholds.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] The purpose of this invention is to provide a device for detecting surface defects in hollow glass microsphere thermal insulation materials, thereby solving the aforementioned problems.

[0005] This invention is implemented as follows: a surface defect detection device for insulating glass microspheres includes a platform and a detection box fixed on the platform. The detection box contains a light source and a detection camera. The device also includes a belt conveyor mounted on the platform, used to move the insulating glass microspheres into and out of the detection box. A real-time defect classification threshold adjustment system, connected to an image processing algorithm, is used to adjust the target defect classification threshold in the image processing algorithm in real time. This real-time defect classification threshold adjustment system includes an image quality assessment module, which calculates and obtains image quality based on global image contrast, average image brightness, image sharpness, and background texture intensity. The system includes: a quantity coefficient; a synergy evaluation module, which calculates the synergy between environmental interference and image quality based on the ambient light interference level and equipment vibration amplitude under the image quality coefficient; a threshold suitability evaluation module, which calculates the threshold suitability coefficient based on the recent false alarm rate, recent false alarm rate, and real-time conveyor belt speed; an equipment attenuation evaluation module, which calculates the equipment attenuation coefficient based on the light source attenuation degree and the time since the last calibration; and a defect classification threshold adjustment module, which calculates the target defect classification threshold in the image processing algorithm based on the environmental interference-image quality synergy, the threshold suitability coefficient, and the equipment attenuation coefficient, and adjusts the current defect classification threshold to the target defect classification threshold.

[0006] A further technical solution is that the method for calculating and obtaining the target defect classification threshold in the image processing algorithm is as follows: based on the product relationship of the environmental interference-image quality synergy, the threshold suitability coefficient and the device attenuation coefficient, and combined with the system's preset lower threshold and upper threshold, the target defect classification threshold is determined by a linear mapping method.

[0007] A further technical solution involves the following steps for calculating the threshold suitability coefficient: obtaining the recent false alarm rate, the recent false alarm rate, and the real-time speed of the conveyor belt; comparing the recent false alarm rate, the recent false alarm rate, and the real-time speed of the conveyor belt with their respective maximum values ​​to obtain the recent false alarm rate index, the recent false alarm rate index, and the real-time speed index of the conveyor belt; and calculating the threshold suitability coefficient, which ranges from 0 to 1, by weighted combination of the recent false alarm rate index, the recent false alarm rate index, and the real-time speed index of the conveyor belt.

[0008] Further technical solutions show that the threshold suitability coefficient is negatively correlated with the recent false alarm rate index and the real-time speed index of the conveyor belt, and positively correlated with the recent false alarm rate index.

[0009] A further technical solution involves the following steps for calculating and obtaining the device attenuation coefficient: obtaining the current light source attenuation level and the time since the last calibration; comparing the current light source attenuation level and the time since the last calibration with their maximum values ​​to obtain the light source attenuation index and the calibration interval time index; and calculating the device attenuation coefficient, which takes a value between 0 and 1, using a negative exponential function relationship between the light source attenuation index and the calibration interval time index. The device attenuation coefficient decreases as the light source attenuation index and the calibration interval time index increase.

[0010] A further technical solution involves the following steps for calculating and obtaining the environmental interference-image quality synergy: obtaining the image quality coefficient, ambient light interference level, and device vibration amplitude; performing maximum-minimum normalization on both the current ambient light interference level and device vibration amplitude to obtain the ambient light interference level index and device vibration amplitude index; calculating the comprehensive environmental interference index by weighting the image quality coefficient, ambient light interference level index, and device vibration amplitude index; and obtaining the environmental interference-image quality synergy coefficient with a value between 0 and 1 by calculating the ratio deviation between the comprehensive environmental interference index and the image quality deviation, and constructing an exponential mapping relationship based on the absolute value of this deviation.

[0011] A further technical solution is that the value of the environmental interference-image quality coordination coefficient approaches 1, indicating that the system operates in a highly coordinated manner, meaning that the image quality can well cope with or compensate for the current environmental interference. The value of the environmental interference-image quality coordination coefficient approaches 0, indicating that the system coordination is poor, that there is a significant negative impact of environmental interference on image quality, or that the image quality itself is poor.

[0012] A further technical solution involves the following steps for calculating and obtaining the image quality coefficient: obtaining the global contrast ratio, average brightness ratio, image sharpness, and background texture intensity; converting the global contrast ratio into a global contrast ratio index, where the global contrast ratio increases when the global contrast ratio approaches the optimal contrast reference value, and decreases when the global contrast ratio deviates from the reference value; mapping the average brightness ratio to a brightness suitability index using a Gaussian function, where the brightness suitability index reaches its maximum value when the average brightness ratio equals a preset ideal value, and decreases monotonically as the average brightness ratio deviates from the ideal value; performing maximum-minimum normalization on the current image sharpness and background texture intensity to obtain the image sharpness index and background texture intensity index; and calculating the image quality coefficient, with values ​​between 0 and 1, using a geometric mean based on the contrast ratio, brightness deviation index, sharpness index, and texture intensity index.

[0013] Further technical solutions include: the larger the contrast index, brightness deviation index, and sharpness index, the larger the image quality coefficient, indicating better image quality; the texture intensity index participates in the calculation in a negatively correlated manner, the smaller the texture intensity index, the larger the image quality coefficient, indicating better image quality.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] 1. It realizes dynamic closed-loop adjustment of defect classification threshold, which can respond in real time to image quality fluctuations, changes in environmental interference, equipment performance degradation and production process adjustments, fundamentally overcoming the inherent defects of fixed threshold system in complex working conditions where false alarms and false alarms are unbalanced.

[0016] 2. By integrating and quantitatively modeling multi-dimensional parameters (image quality, environmental compatibility, detection performance feedback, and device health status), threshold adjustment has clear physical meaning and directionality, making the adjustment process more intelligent and precise.

[0017] 3. By introducing an equipment attenuation coefficient, long-term factors such as light source aging and calibration cycle are incorporated into the adjustment system, enhancing the system's performance maintenance capability and reliability throughout its entire life cycle.

[0018] 4. The entire adjustment mechanism is based on a clear mathematical model and formula, with a clear algorithm that is easy to implement, debug and optimize in industrial controllers or computers, and has good engineering applicability. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of a surface defect detection device for hollow glass microsphere thermal insulation material provided by the present invention;

[0020] Figure 2 Provided by the present invention Figure 1 Internal structure diagram of the testing box;

[0021] Figure 3 A flowchart of the real-time adjustment system for defect classification thresholds provided by the present invention.

[0022] In the attached diagram: 1. Tabletop; 2. Testing box; 3. Light source; 4. Testing camera; 5. Belt conveyor. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] In traditional methods for detecting surface defects in insulating glass microspheres, existing devices employ image processing algorithms with fixed thresholds, failing to dynamically adjust defect classification thresholds based on real-time operating conditions. This issue stems from fluctuations in image quality and system baseline drift caused by factors such as unstable ambient lighting, equipment vibration, light source attenuation, conveyor belt speed variations, and differences in raw material color and texture. Specifically, the fixed threshold mechanism lacks a comprehensive evaluation capability considering image quality, equipment health status, environmental interference levels, and detection feedback, resulting in an inability to intelligently balance false alarm and false negative rates. Furthermore, image quality coefficients are not incorporated into threshold adjustment, the synergy between environmental interference and image quality is not quantitatively assessed, threshold suitability coefficients are not dynamically calculated, and equipment attenuation coefficients are not monitored in real-time. Consequently, the stability and accuracy of the detection system are difficult to maintain long-term in complex industrial environments.

[0025] For example, in a building energy-saving materials production line, insulating glass microspheres continuously enter the inspection area via a belt conveyor. When the external environment changes from cloudy to sunny, the ambient light intensity changes abruptly, resulting in uneven brightness distribution in the images captured by the industrial camera. Simultaneously, the extended use time of the light source leads to increased attenuation, and the conveyor belt speed increases due to production cycle adjustments. In this scenario, the fixed-threshold detection algorithm cannot adapt to the decline in image quality and speed changes. Specifically, when the image quality coefficient decreases, the system fails to raise the threshold to reduce false alarms, causing qualified products to be incorrectly rejected. When the equipment vibration amplitude increases, leading to a decrease in the synergy between environmental interference and image quality, the system fails to lower the threshold to ensure the detection rate, resulting in missed detection of defective products. Furthermore, recent fluctuations in false alarm and missed alarm rates are not converted into threshold suitability coefficients, and changes in equipment attenuation coefficients caused by extended time since the last calibration are not included in the adjustment criteria. This problem directly affects the sorting efficiency of the production line and the product quality control process.

[0026] If the aforementioned problems are not addressed, the detection system will face continuous performance degradation during long-term operation. The cumulative effects of environmental interference and equipment aging will cause the system baseline to drift continuously, leading to an imbalance between false alarm and false negative rates. Consequently, product quality consistency cannot be guaranteed, potentially triggering batch quality incidents. Simultaneously, system maintenance costs will increase, requiring operators to frequently manually intervene in threshold settings, reducing the practical value of automated detection. This technical deficiency severely restricts the level of intelligence and reliability of surface defect detection in insulating glass microspheres.

[0027] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0028] like Figure 1 , Figure 2 and Figure 3As shown, an embodiment of the present invention provides a surface defect detection device for insulating glass microspheres, comprising a platform 1 and a detection box 2 fixed on the platform 1. The detection box 2 is equipped with a light source 3 and a detection camera 4. The light source 3 provides uniform and stable illumination to the surface of the insulating glass microspheres, enabling the detection camera 4 to clearly capture images of the material surface. The detection camera 4 functions to acquire image data of the material surface illuminated by the light source 3 in real time and transmit these images to a subsequent image processing unit for analysis. The device also includes a belt conveyor 5 installed on the platform 1, which drives the insulating glass microspheres in and out of the detection box 2, realizing material flow on an automated production line. The light source 3 can be a high-brightness LED array or a ring light source, and its brightness can be manually adjusted. The detection camera 4 can be an industrial-grade CCD or CMOS camera with adjustable exposure time, gain, and frame rate. The belt conveyor 5 can be driven by a motor, and its operating speed can be adjusted via a manual knob or control panel. In this configuration, the material is fed into the inspection box 2 by the belt conveyor 5. Under the illumination of the light source 3, the inspection camera 4 captures an image of the material surface, and then the image is transmitted to the image processing unit for defect detection.

[0029] A real-time defect classification threshold adjustment system, connected to an image processing algorithm, is used to adjust the target defect classification threshold in the image processing algorithm in real time. This system can be a standalone embedded controller or an industrial computer running specific software programs. The real-time defect classification threshold adjustment system includes:

[0030] The image quality assessment module calculates and obtains an image quality coefficient based on the image's global contrast, average brightness, sharpness, and background texture intensity. This module can be a software module running on an industrial computer. It can periodically acquire images from the detection camera 4 and calculate the image's global contrast, average brightness, sharpness (e.g., by calculating the image's gradient magnitude or edge intensity), and background texture intensity (e.g., by calculating the gray-level co-occurrence matrix features of the background region).

[0031] The synergy assessment module calculates the synergy between environmental interference and image quality based on the ambient light interference level and equipment vibration amplitude under the image quality coefficient. This module can integrate an ambient light sensor and a vibration sensor. The ambient light sensor can be a photoresistor or photodiode used to measure the ambient light intensity inside or outside the detection box 2, thereby obtaining the ambient light interference level. The vibration sensor can be a simple accelerometer, mounted on the detection box 2 or the platform 1, used to measure the equipment vibration amplitude. This module can perform a simple logical judgment between the image quality coefficient and the ambient light interference level and equipment vibration amplitude.

[0032] The threshold suitability assessment module calculates the threshold suitability coefficient based on the recent false alarm rate, recent false negative rate, and the real-time speed of the conveyor belt. The recent false alarm rate and recent false negative rate can be obtained through statistical analysis of the results of sampling re-inspections of the tested products. The real-time speed of the conveyor belt can be directly obtained from the control system of the belt conveyor 5. This module can use simple logical judgment or table lookup methods to determine the threshold suitability coefficient. For example, when the recent false negative rate is high, the system determines that the detection sensitivity needs to be increased, thereby reducing the threshold suitability coefficient; when the recent false alarm rate is high, the system determines that the detection sensitivity can be decreased, thereby increasing the threshold suitability coefficient.

[0033] The device attenuation assessment module calculates the device attenuation coefficient based on the light source attenuation level and the time since the last calibration. This module records the cumulative operating time of light source 3 and estimates the attenuation level based on data provided by the light source manufacturer. The time since the last calibration can be recorded via system logs or manual input. This module allows setting thresholds for the light source attenuation level and calibration interval.

[0034] The defect classification threshold adjustment module calculates the target defect classification threshold in the image processing algorithm based on environmental interference-image quality synergy, threshold suitability coefficient, and device attenuation coefficient, and adjusts the current defect classification threshold to the target threshold. This module can use a preset lookup table or a simple piecewise function to determine the target threshold based on the three coefficients. For example, when environmental interference-image quality synergy is high, the threshold suitability coefficient is high (indicating low sensitivity requirements), and the device attenuation coefficient is high (indicating good device condition), the target threshold can be set to a higher value; conversely, when any one or more coefficients are low, the target threshold can be lowered accordingly. This adjustment module sends the calculated target threshold to the image processing algorithm to update its current classification threshold.

[0035] In existing technologies, detection systems generally employ preset fixed classification thresholds. This means that regardless of fluctuations in ambient lighting, equipment vibration, light source attenuation, or changes in production cycle, the defect discrimination standard remains unchanged. This fixed threshold scheme cannot effectively respond to dynamic situations such as temporary light source activation leading to increased brightness, slight equipment vibration, slight light source attenuation, and the need to balance false alarms and false negatives, as seen in the examples above.

[0036] In contrast, the device in this embodiment can perceive and evaluate multi-dimensional operating condition information in real time. For example, when the image quality assessment module detects increased image brightness and enhanced background texture, and when the coordination assessment module identifies ambient light fluctuations and equipment vibrations, this information is comprehensively used to adjust the threshold. Furthermore, the sensitivity requirement assessment module can intelligently determine the actual sensitivity requirements of the current detection based on feedback from recent false alarm and false negative rates, avoiding the dilemma of being unable to optimize when the false alarm rate is too high or the false negative rate is too low under a fixed threshold. The attenuation assessment module considers the impact of light source attenuation and calibration interval time on equipment performance, ensuring the long-term reliability of the detection benchmark.

[0037] By comprehensively utilizing these multi-dimensional evaluation results through the defect classification threshold adjustment module, the device can calculate and adjust the target defect classification threshold in real time to adapt to the current operating conditions. This dynamic adaptive adjustment mechanism allows the detection system to move beyond a single, static discrimination standard and intelligently optimize based on actual operating conditions. For example, in the above example, the system can adjust the threshold to 150 based on the comprehensive evaluation results, thereby effectively controlling false alarms while ensuring the detection rate, thus improving detection efficiency and accuracy. This capability is not present in existing fixed threshold systems, fundamentally enhancing the robustness and intelligence of industrial vision inspection systems in complex and ever-changing production environments.

[0038] This application further proposes the following method for calculating the target defect classification threshold in the image processing algorithm:

[0039] Based on the product relationship between the environmental interference-image quality synergy, the threshold suitability coefficient, and the device attenuation coefficient, and combined with the system's preset lower and upper threshold limits, the target defect classification threshold is determined through a linear mapping method. Specifically, the calculation method involves importing the environmental interference-image quality synergy, the threshold suitability coefficient, and the device attenuation coefficient into the formula. Obtain the target defect classification threshold. The threshold is a key parameter in image processing algorithms used to distinguish between normal and defective regions, and its value directly affects the sensitivity and specificity of defect detection. For example, a higher threshold may lead to an increased false negative rate, while a lower threshold may lead to an increased false positive rate. This threshold can be a pixel grayscale value, a distance from a feature vector, or a probability score output by a classifier. This is the minimum allowable threshold for the system, representing the defect classification threshold that the system must not fall below under any circumstances. This value is set to prevent excessively low thresholds from causing a large number of false alarms, thus affecting detection efficiency and subsequent processing. This minimum threshold can be preset based on historical data, industry standards, or expert experience. For example, by analyzing a large number of defect-free samples, a minimum acceptable threshold that can effectively exclude background noise can be determined. This is the maximum threshold allowed by the system, representing the defect classification threshold that the system must not exceed under any circumstances. This value is set to prevent excessively high thresholds from causing a large number of missed detections, thereby reducing the reliability of detection. This maximum threshold can be determined based on the typical characteristics of known defects or the maximum acceptable defect size. For example, by analyzing typical defect samples, a maximum acceptable threshold that can effectively capture all real defects can be determined. The environmental interference-image quality synergy reflects the degree to which the current detection environment (such as ambient light interference and equipment vibration) affects image quality and the system's ability to adapt to these interferences. A higher value indicates a smaller impact of environmental interference on image quality, or a stronger ability of the system to compensate for this impact. This is the threshold suitability coefficient. The attenuation coefficient reflects the health and reliability of the testing equipment (especially the light source). A higher value indicates better equipment condition and lower attenuation.

[0040] The following is a concrete example to illustrate this. Assume the system has a preset minimum threshold. The maximum threshold is 0.3. The value is 0.8. At a certain moment, the real-time defect classification threshold adjustment system calculates the environmental interference-image quality synergy through various evaluation modules. The threshold suitability coefficient is 0.7. The equipment attenuation coefficient is 0.6. The value is 0.9. At this point, the defect classification threshold adjustment module imports these parameters into the formula and calculates to obtain 0.489. Therefore, the target defect classification threshold is... The threshold was set to 0.489. Subsequently, the defect classification threshold adjustment module will adjust the current defect classification threshold to 0.489 to adapt to the current detection environment and system status.

[0041] By employing the product relationship between environmental interference and image quality synergy, threshold suitability coefficient, and equipment attenuation coefficient, and combining this with a linear mapping between the system's preset lower and upper threshold limits, this application achieves accurate and dynamic calculation of the target defect classification threshold. This method effectively solves the problem of the difficulty in effectively integrating multi-dimensional parameters, ensuring that the defect classification threshold can be reasonably and stably determined under different environmental conditions, detection requirements, and equipment states. Therefore, this scheme significantly improves the accuracy and robustness of surface defect detection in hollow glass microsphere insulation materials, reduces false alarms and false negatives, and thus improves the overall performance and reliability of the detection system.

[0042] This application further proposes the following steps for calculating and obtaining the threshold suitability coefficient:

[0043] The system acquires recent false alarm rate, recent false negative rate, and real-time conveyor belt speed. The recent false alarm rate refers to the proportion of flawless materials incorrectly identified as defective by the system within a recent period. This data can be obtained by manually re-inspecting the tested materials and comparing the results with the system's detection; or by using the system's internal self-learning or expert system to perform secondary verification on a subset of samples. The recent false negative rate refers to the proportion of actual defects that the system failed to detect within a recent period. This data can be obtained by sampling and re-inspecting materials judged as flawless by the system and comparing the results with the system's detection; or by introducing higher-precision auxiliary detection methods to verify a subset of materials. The real-time conveyor belt speed refers to the current operating speed of the belt conveyor 5. This speed can be obtained in real-time by an encoder or photoelectric sensor installed on the belt conveyor 5; or by monitoring the motor speed driving the belt conveyor 5 and calculating it in conjunction with the pulley diameter.

[0044] The recent false alarm rate, recent false negative rate, and real-time conveyor belt speed are each compared to their maximum values ​​to obtain the recent false alarm rate index, recent false negative rate index, and real-time conveyor belt speed index. This aims to map raw data with different dimensions and numerical ranges to a standardized interval of 0 to 1, facilitating subsequent comprehensive calculations. For example, the false alarm rate index can be obtained by dividing the current false alarm rate by the maximum allowed false alarm rate; the false negative rate index can be obtained by dividing the current false negative rate by the maximum allowed false negative rate; and the real-time conveyor belt speed index can be obtained by dividing the current real-time conveyor belt speed by the maximum operating speed set by the system. This normalization process ensures the comparability of various indicators in the formulas and avoids the problem of excessive weighting of any indicator due to differences in dimensions.

[0045] A threshold suitability coefficient, ranging from 0 to 1, is calculated by weighting the recent false alarm rate index, the recent false alarm rate index, and the real-time conveyor belt speed index. The threshold suitability coefficient shows a negative correlation with both the recent false alarm rate index and the real-time conveyor belt speed index, and a positive correlation with the recent false alarm rate index. The weighting method involves importing the recent false alarm rate index, the recent false alarm rate index, and the real-time conveyor belt speed index into the formula. Obtain the threshold suitability coefficient , The value range is 0-1, when When the value is large, the system may be more inclined to raise the monitoring standards and reduce false alarms due to a higher false alarm rate or slower production speed. In this case, the defect classification threshold should be increased accordingly. Conversely, when... A lower value indicates a higher requirement for system detection sensitivity. For example, a higher false negative rate might mean the system needs to more rigorously identify defects, in which case the defect classification threshold should be lowered. This clear guidance allows the threshold suitability coefficient to be directly used for subsequent adjustment of the target defect classification threshold. This is the recent false alarm rate index. This is the recent underreporting rate index. This refers to the real-time speed index of the conveyor belt. The value is a tradeoff coefficient ranging from 0 to 1. This is used to balance the relative importance of detection accuracy and the real-time speed of the conveyor belt. For example, when more emphasis is placed on detection accuracy, a larger setting can be used. Value; when production efficiency needs to be considered, it can be adjusted appropriately. The value reflects the effect of speed, and is adjusted... This can make the threshold suitability coefficient This better aligns with the needs of real-world application scenarios, and can be achieved through preset strategies or dynamic algorithms for value assignment. The first part of the formula... This reflects the impact of detection accuracy (i.e., false alarms and false negatives) on sensitivity requirements. The higher the false alarm rate and the lower the false negative rate, the larger this value, indicating a higher sensitivity requirement. (The second part of the formula...) This reflects the impact of the real-time speed of the conveyor belt on the sensitivity requirement. It is generally believed that within a certain range, a lower conveyor belt speed helps to improve detection accuracy. Therefore, when the real-time speed index of the conveyor belt is low, the larger the value of this part, the greater the sensitivity requirement.

[0046] As a specific implementation method, the sensitivity requirement assessment module can calculate the threshold suitability coefficient according to the following steps. First, the system can obtain the recent false alarm rate (e.g., 0.05, or 5%), recent false negative rate (e.g., 0.02, or 2%), and the real-time conveyor belt speed (e.g., 1.5 m / s) for the current time period from historical data records or the real-time monitoring interface. Assume the system's preset maximum false alarm rate is 0.1 (10%), maximum false negative rate is 0.05 (5%), and maximum conveyor belt speed is 2.0 m / s. Next, ratio processing is performed to obtain the index: the recent false alarm rate index. The calculated value is 0.5; the recent underreporting rate index. The calculated value is 0.4; Real-time speed index of the conveyor belt. The value is calculated to be 0.75. These indices are then imported into the weighted combination formula. Assume a tradeoff coefficient... Setting it to 0.6 indicates that the system prioritizes false alarm and false negative rates slightly over speed. The threshold suitability coefficient is then calculated using the formula. It is 0.28. Because... A relatively small value (close to 0) indicates that the current system has a high requirement for detection sensitivity, and it may be necessary to lower the defect classification threshold to reduce false negatives. This calculation result will then be passed to the defect classification threshold adjustment module as an important basis for adjusting the target defect classification threshold.

[0047] Through the above technical solution, this application provides a refined and dynamic method for calculating the threshold suitability coefficient. This method comprehensively considers three key factors: recent false alarm rate, recent false negative rate, and real-time conveyor belt speed, quantifying them into a calculable index. Through a scientific weighted combination formula, it can accurately assess the sensitivity requirements of the current detection system for defect classification thresholds. This dynamic evaluation mechanism effectively solves the problem of rigid threshold settings in traditional detection systems, which are difficult to adapt to changes in the production environment. It avoids false negatives caused by excessively high thresholds or false positives caused by excessively low thresholds, thereby significantly improving the accuracy and reliability of surface defect detection in insulating glass microspheres. Simultaneously, by incorporating the real-time conveyor belt speed, threshold adjustment can balance production efficiency, optimizing the smoothness of the production process while ensuring detection quality. This allows the defect classification threshold to respond more intelligently to actual working conditions, improving the intelligence level and adaptability of the entire detection system.

[0048] This application further proposes the following steps for calculating and obtaining the device attenuation coefficient:

[0049] The system acquires the current light source attenuation level and the time since the last calibration. "Light source attenuation level" refers to the degree to which the luminous intensity or spectral characteristics of light source 3 decreases relative to its initial state after prolonged use. This can be obtained in various ways, such as by using a photoelectric sensor to monitor the output brightness of light source 3 in real time and comparing it with a preset initial brightness value; or by indirectly assessing its attenuation state by monitoring changes in the driving current or voltage of light source 3. "Time since the last calibration" refers to the length of time elapsed since the detection device last performed system calibration (e.g., image calibration, light source calibration, etc.). This time information can be recorded and updated in real time by the system's internal timer module, or obtained by reading the calibration timestamp stored in non-volatile memory.

[0050] The light source attenuation level and the time since the last calibration are compared with their maximum values ​​to obtain the light source attenuation index and the calibration interval time index. The purpose of this "ratio processing" is to quantize the raw data of different dimensions and ranges into a dimensionless index between 0 and 1, facilitating subsequent mathematical model calculations. For the light source attenuation level, its current value can be compared with the maximum attenuation the light source may reach within its design life to obtain the "light source attenuation index." This index reflects the relative degree of current attenuation of the light source. Similarly, for the time since the last calibration, its current value can be compared with the system's preset maximum allowable calibration interval to obtain the "calibration interval time index". This index reflects the relative degree to which the system is currently deviating from its latest calibration state.

[0051] The device attenuation coefficient, ranging from 0 to 1, is calculated using a negative exponential function relationship between the light source attenuation index and the calibration interval time index. The device attenuation coefficient decreases as both the light source attenuation index and the calibration interval time index increase. Specifically, the negative exponential function relationship is defined as follows: [The formula is missing from the original text]. Obtain the device attenuation coefficient , , A higher value indicates a better condition and higher device reliability. The light source attenuation index, To calibrate the interval time index, For the light source attenuation rate parameter, To calibrate the decay rate parameter at the interval, and These parameters characterize the sensitivity of light source attenuation and calibration interval time to the overall attenuation of the equipment. These parameters are typically determined through experimental data fitting, expert experience, or analysis of historical operating data to ensure that the model accurately reflects the actual attenuation characteristics of the equipment.

[0052] As a specific implementation method, when calculating the device attenuation coefficient, the brightness output of the light source 3 can first be monitored in real time by a photoelectric sensor installed in the detection box 2, and compared with the initial brightness value of the light source 3 at the time of its manufacture, thereby obtaining the current attenuation level of the light source 3. Simultaneously, the real-time clock module inside the system records and updates the time interval since the last overall calibration of the detection camera 4 and the light source 3, i.e., the time since the last calibration. For example, if the initial brightness of the light source 3 is 10,000 lux and the current brightness is 8,000 lux, then the attenuation level is 2,000 lux. If the maximum allowable calibration interval of the system is 30 days, and 15 days have passed, then the light source attenuation level (e.g., 2,000 lux) is compared with the maximum attenuation that the light source may reach within its design life (e.g., 5,000 lux) to obtain the light source attenuation index. Simultaneously, the ratio of the time since the last calibration (e.g., 15 days) to the system's preset maximum allowable calibration interval (e.g., 30 days) is used to obtain the calibration interval index. Then, the calculated and Import into the preset exponential decay formula In the middle. Among them, and After long-term operational testing and data analysis of similar equipment, parameters such as 0.5 and 0.3 can be empirically determined. Finally, the processing unit in the defect classification threshold real-time adjustment system performs this calculation to obtain the attenuation coefficient of the current equipment. .

[0053] Through the above technical solution, this application can evaluate the equipment attenuation status of the surface defect detection device for insulating glass microspheres in real time and quantitatively. By obtaining the light source attenuation level and the time since the last calibration, and converting them into a light source attenuation index and a calibration interval time index, the equipment attenuation coefficient is calculated using an exponential attenuation model. This coefficient directly reflects the health status of the equipment and the reliability of the test results. This equipment attenuation coefficient... By incorporating this into the calculation of the target defect classification threshold, the system can adaptively adjust based on the aging and maintenance status of the equipment. This effectively solves the negative impact of light source aging and sensor performance degradation caused by long-term equipment operation on detection accuracy and stability, avoids false alarms or missed alarms caused by equipment degradation, and thus significantly improves the reliability and robustness of defect detection, ensuring that a high level of detection performance is maintained throughout the entire life cycle of the equipment.

[0054] This application further proposes the following steps for calculating and obtaining the environmental interference-image quality synergy:

[0055] The image quality coefficient (Q) reflects the intrinsic quality of the image captured by camera 4, such as sharpness and contrast. The ambient light interference level quantifies the potential impact of external ambient light on the detection process; excessively strong or weak ambient light can lead to image acquisition distortion. The device vibration amplitude measures the mechanical stability of the detection device itself; excessive vibration can cause image blurring. These parameters can be obtained in various ways, such as the image quality coefficient. The image quality assessment module can calculate the image global contrast, average brightness, sharpness, and background texture intensity. The ambient light interference level can be monitored in real time by setting an ambient light sensor (such as a photoresistor or photodiode) outside or inside the detection box 2. The vibration amplitude of the equipment can be measured in real time by installing an acceleration sensor or vibration sensor on the table 1 or the detection box 2.

[0056] Both the current ambient light interference level and equipment vibration amplitude are subjected to maximum-minimum normalization to obtain the ambient light interference level index and equipment vibration amplitude index. This aims to unify physical quantities with different dimensions and ranges into a standardized range of 0 to 1. This processing method allows these parameters to be effectively compared and integrated in subsequent calculation formulas, avoiding calculation deviations caused by differences in dimensions. The normalization process can use the standard maximum-minimum normalization method, where minimum and maximum represent the minimum and maximum values ​​that the parameter may encounter in actual operation, respectively. These extreme values ​​can be preset based on the equipment's operating environment and historical data.

[0057] Based on the image quality coefficient, ambient light interference level index, and equipment vibration amplitude index, the ambient light interference level index and equipment vibration amplitude index are first weighted and averaged to calculate the comprehensive environmental interference index. Then, by calculating the ratio deviation between the comprehensive environmental interference index and the image quality deviation, an exponential mapping relationship is constructed based on the absolute value of this deviation, thereby obtaining the environmental interference-image quality synergy coefficient, which ranges from 0 to 1. A larger environmental interference-image quality synergy coefficient indicates more coordinated system operation. Specifically, the calculation method involves importing the image quality coefficient, ambient light interference level index, and equipment vibration amplitude index into the formula... Acquiring the synergy between environmental interference and image quality , A larger value indicates a more coordinated system operation. This is an index representing the level of ambient light interference. The vibration amplitude index of the equipment. This is the image quality coefficient. Assigning an ambient light interference weight allows the system to adjust the relative importance of ambient light interference and equipment vibration to synergy based on the actual application scenario. For example, in light-sensitive detection tasks, the weight can be appropriately increased. The value can be assigned through a preset strategy or a dynamic algorithm. To prevent division by zero by small constants, the numerical stability of the formula is ensured. The formula is designed so that when the weighted environmental disturbance term... Image quality item When the ratio is close to 1, A value close to 1 indicates that the system is operating in a highly coordinated manner, meaning that the image quality can effectively cope with or compensate for current environmental interference. Conversely, when the ratio deviates from 1, A decrease in the value indicates poor system coordination, which may be due to significant negative impacts of environmental interference on image quality, or poor image quality itself.

[0058] The following is a concrete example to illustrate this. Suppose that during a certain detection process, the image captured by detection camera 4 is processed by the image quality assessment module to obtain an image quality coefficient. The value is 0.85 (indicating good image quality). Meanwhile, the ambient light sensor detected a moderate to high level of ambient light interference, which, after normalization, yielded the ambient light interference level index. The value is 0.6. The accelerometer detected a slight vibration in the equipment, and the vibration amplitude index was obtained after normalization. The value is 0.2. Assume ambient light interference weight. Set to 0.7 to prevent small constants from being divided by zero. Set it to 0.001. Substituting these values ​​into the formula yields 0.113. This is relatively low. The value (0.113) indicates that, despite the image quality coefficient... It is acceptable, but the current ambient light interference and equipment vibration have a significant potential impact on image quality, and the system's coordination is poor. It may be necessary to make a more conservative adjustment to the defect classification threshold to avoid missed detections.

[0059] Through the above technical solution, this application provides a method for quantitatively and dynamically evaluating the synergy between environmental interference and image quality. This method can accurately reflect the operating status of the detection system under different environmental conditions, overcoming the subjectivity and inaccuracy of traditional methods in assessing environmental factors. By integrating multiple key factors such as image quality, ambient light interference, and equipment vibration into a unified synergy index, the adjustment of the target defect classification threshold can be made more intelligent and precise, thereby significantly improving the accuracy, stability, and robustness of surface defect detection in hollow glass microsphere insulation materials, and effectively reducing the false alarm rate and false negative rate caused by environmental changes or equipment status fluctuations.

[0060] This application further proposes the following steps for calculating and obtaining the image quality coefficient:

[0061] The system acquires global contrast, average brightness, sharpness, and background texture intensity; these parameters are fundamental indicators for measuring the visual quality and information content of an image. Global contrast reflects the difference between bright and dark areas in an image, affecting the ability to distinguish blemishes from the background. Average brightness represents the overall brightness of the image; excessive brightness or darkness can affect detection performance. Sharpness measures the sharpness of image details and is crucial for identifying minor blemishes. Background texture intensity reflects the complexity of the background region; excessively strong background texture may interfere with blemish identification. These parameters can be extracted from the images captured by the detection camera using image processing algorithms. For example, global contrast can be obtained by calculating the variance or entropy of the image's gray-level histogram; average brightness can be obtained by calculating the average of all pixel gray-level values; image sharpness can be evaluated using gradient magnitude, Laplacian operator, or frequency domain analysis; and background texture intensity can be quantified using methods such as gray-level co-occurrence matrix or local binary pattern.

[0062] The global image contrast is converted into a global image contrast index. When the global image contrast approaches the optimal contrast reference value, the global image contrast index increases; when the global image contrast deviates from this reference value, the global image contrast index decreases. Specifically, the calculation method involves importing the global image contrast into the formula... Obtain the global contrast index of the image. ,in, For global image contrast, For optimal contrast ratio reference value, The range of contrast variation is sensitive; this non-linear mapping method can more accurately reflect the impact of contrast on image quality in different ranges, so that the exponential change is gradual near the optimal contrast, while the exponential change is more significant when deviating from the optimal contrast, thus quantifying the contribution of contrast to detection performance.

[0063] Based on the average image brightness, it is mapped to a brightness suitability index using a Gaussian function. The brightness suitability index reaches its maximum value when the average image brightness equals a preset ideal value, and decreases monotonically as the average image brightness deviates from this ideal value. The specific calculation method is as follows: The average image brightness is imported into the formula... Obtain the brightness suitability index ,in, The average brightness of the image. The optimal brightness value for detection. This refers to the allowable range of brightness deviation; the closer the average image brightness is to... hour, The closer the value is to 1, the more ideal the brightness conditions; conversely, when the brightness deviates from... When it is larger, The smaller the value, the less ideal the brightness conditions. This processing method can accurately measure the impact of brightness on image quality, ensuring that detection is performed under optimal brightness conditions.

[0064] The image sharpness and background texture intensity are both subjected to min-max normalization to obtain the image sharpness index and background texture intensity index. Min-max normalization is a common data processing method that linearly scales the original data to a preset range (e.g., 0 to 1). For image sharpness, a standardized sharpness index is obtained by subtracting the minimum value from the original value and then dividing by the difference between the maximum and minimum values, reflecting its relative level within an acceptable range. Similarly, the background texture intensity undergoes the same normalization process to obtain the background texture intensity index. This process ensures that image features of different dimensions can be uniformly included in subsequent calculations, avoiding the problem of excessive weighting of a certain feature due to differences in dimensions.

[0065] Based on the contrast index, brightness deviation index, sharpness index, and texture intensity index, an image quality coefficient with values ​​between 0 and 1 is calculated using a geometric mean. The texture intensity index participates in the calculation in a negatively correlated manner. Specifically, the calculation method involves combining the global contrast index and brightness suitability index of the image. Import formulas for image sharpness index and background texture intensity index. Obtain the image quality coefficient , Output range 0-1, The larger the value, the better the image quality. The global contrast index of the image. The brightness suitability index, Image sharpness index This represents the background texture intensity index. , and As a positive indicator, a higher value indicates better image quality; while the background texture intensity index... Then The form in which the calculation is performed means that the weaker the background texture (i.e., the more...), the more... The smaller the value, the greater its contribution to image quality, because complex background textures can interfere with defect detection. Through calculation using the fourth root, this formula smoothly integrates these metrics, ultimately outputting an image quality coefficient between 0 and 1. ,in A higher value indicates better image quality, providing a comprehensive and quantitative basis for image quality assessment for subsequent defect classification threshold adjustment.

[0066] As a specific implementation, after receiving the image data captured by the detection camera 4, the image quality assessment module initiates a series of image processing algorithms to calculate various indicators. For example, for global image contrast, the entropy value of the gray-level histogram can be used as its quantification indicator, assuming a calculated global image contrast of 0.65. For average image brightness, the average gray value of all pixels in the image can be calculated, assuming a brightness suitability of 135. Image sharpness can be obtained by calculating the Tenengrad gradient value of the image, assuming a value of 0.75. Background texture intensity can be quantified by the variance of the Local Binary Pattern (LBP) features, assuming a value of 0.3. After obtaining these raw indicators, the image quality assessment module performs exponential transformation. A preset optimal contrast reference value is assumed. The value is 0.5, indicating a sensitive range for contrast changes. If the value is 0.2, then the global contrast index of the image is... The value will be calculated using the formula to obtain 0.68. For the average image brightness, assume the optimal brightness value for detection. The value is 128, and the allowable deviation range for brightness is... If the value is 30, then the brightness suitability index is... The value will be calculated to be 0.947 using the formula. The image sharpness index and background texture intensity index are then processed using maximum-minimum normalization. Assuming the maximum value of image sharpness is 1.0 and the minimum value is 0.0, the image sharpness index... The value is 0.75. Assuming the maximum background texture intensity is 1.0 and the minimum is 0.0, then the background texture intensity index is... The value is 0.3. Finally, these indices are substituted into the image quality coefficient. The calculation formula yields 0.76. Through the above calculation, the image quality assessment module can obtain a quantified image quality coefficient. For example, 0.76, this value will be used as input to the subsequent synergy evaluation module to further adjust the defect classification threshold.

[0067] Through the above technical solution, this application can comprehensively and precisely evaluate the quality of detected images. By comprehensively considering four key dimensions—global contrast, average brightness, sharpness, and background texture intensity—and using nonlinear functions and normalization to transform them into a unified index, the image quality coefficient is finally calculated using a fusion formula. This multi-factor, quantitative evaluation method overcomes the limitations of single-index evaluation, resulting in a more comprehensive image quality coefficient. It can more accurately reflect the true detection suitability of the image. Therefore, when this precise image quality coefficient... The synergy evaluation module is used to calculate the synergy between environmental interference and image quality. At that time, it can significantly improve This improves the accuracy of the image quality, enabling the defect classification threshold adjustment module to set more reasonable and adaptive target defect classification thresholds. This effectively avoids false alarms and missed alarms caused by fluctuations in image quality, greatly improving the detection accuracy and stability of the surface defect detection device for hollow glass microsphere insulation materials in complex industrial environments, and ensuring the reliability of the detection results.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A device for detecting surface defects in hollow glass microsphere thermal insulation material, comprising a table and a detection box fixed on the table, wherein a light source and a detection camera are installed inside the detection box, characterized in that, Also includes: A belt conveyor installed on the tabletop is used to drive the hollow glass microsphere thermal insulation material into and out of the testing box; A real-time defect classification threshold adjustment system, connected to an image processing algorithm, is used to adjust the target defect classification threshold in the image processing algorithm in real time. The real-time defect classification threshold adjustment system includes: The image quality assessment module calculates and obtains the image quality coefficient based on the image's global contrast, average brightness, sharpness, and background texture intensity. The synergy assessment module calculates and obtains the synergy between environmental interference and image quality based on the ambient light interference level and equipment vibration amplitude under the image quality coefficient. The threshold suitability assessment module calculates and obtains the threshold suitability coefficient based on the recent false alarm rate, the recent false alarm rate, and the real-time speed of the conveyor belt. The attenuation assessment module calculates and obtains the device attenuation coefficient based on the degree of light source attenuation and the time since the last calibration. The defect classification threshold adjustment module calculates and obtains the target defect classification threshold in the image processing algorithm based on the environmental interference-image quality synergy, threshold suitability coefficient, and device attenuation coefficient, and adjusts the current defect classification threshold to the target defect classification threshold.

2. The device for detecting surface defects in hollow glass microsphere thermal insulation material according to claim 1, characterized in that, The method for calculating and obtaining the target defect classification threshold in the image processing algorithm is as follows: Based on the product relationship of the environmental interference-image quality synergy, the threshold suitability coefficient, and the device attenuation coefficient, and combined with the system's preset lower and upper threshold limits, the target defect classification threshold is determined through a linear mapping method.

3. The surface defect detection device for hollow glass microsphere thermal insulation material according to claim 2, characterized in that, The steps for calculating and obtaining the threshold suitability coefficient are as follows: Obtain recent false alarm rate, recent false negative rate, and real-time conveyor belt speed; The recent false alarm rate, recent false alarm rate, and real-time speed of the conveyor belt are each compared with their maximum values ​​to obtain the recent false alarm rate index, the recent false alarm rate index, and the real-time speed index of the conveyor belt. The recent false alarm rate index, the recent false alarm rate index, and the real-time speed index of the conveyor belt are weighted and combined to obtain the threshold suitability coefficient, which ranges from 0 to 1.

4. The surface defect detection device for hollow glass microsphere thermal insulation material according to claim 3, characterized in that, The threshold suitability coefficient is negatively correlated with the recent false alarm rate index and the real-time speed index of the conveyor belt, and positively correlated with the recent false alarm rate index.

5. The surface defect detection device for hollow glass microsphere thermal insulation material according to claim 2, characterized in that, The steps for calculating and obtaining the device attenuation coefficient are as follows: Obtain the current light source attenuation level and the time since the last calibration; The current light source attenuation level and the time since the last calibration are compared with their maximum values ​​to obtain the light source attenuation index and the calibration interval time index. The device attenuation coefficient, which takes values ​​between 0 and 1, is calculated by using a negative exponential function relationship between the light source attenuation index and the calibration interval time index. The device attenuation coefficient decreases as the light source attenuation index and the calibration interval time index increase.

6. The device for detecting surface defects in hollow glass microsphere thermal insulation material according to claim 2, characterized in that, The steps for calculating and obtaining the environmental interference-image quality synergy are as follows: Acquire image quality index, ambient light interference level, and device vibration amplitude; The current ambient light interference level and equipment vibration amplitude are both processed by maximum-minimum normalization to obtain the ambient light interference level index and equipment vibration amplitude index. Based on the image quality coefficient, ambient light interference level index, and equipment vibration amplitude index, the ambient light interference level index and equipment vibration amplitude index are first weighted and averaged to calculate the comprehensive environmental interference index. By calculating the ratio deviation between the comprehensive environmental interference index and the image quality deviation, and constructing an exponential mapping relationship based on the absolute value of the deviation, the environmental interference-image quality synergy coefficient with a value between 0 and 1 is obtained.

7. The surface defect detection device for hollow glass microsphere thermal insulation material according to claim 6, characterized in that, An environmental interference-image quality coherence coefficient close to 1 indicates that the system operates in a highly coordinated manner, meaning that the image quality can effectively cope with or compensate for current environmental interference. An environmental interference-image quality coherence coefficient close to 0 indicates poor system coherence, meaning that environmental interference has a significant negative impact on image quality, or that the image quality itself is poor.

8. The surface defect detection device for hollow glass microsphere thermal insulation material according to claim 6, characterized in that, The steps for calculating and obtaining the image quality coefficient are as follows: Obtain the global contrast, average brightness, sharpness, and background texture intensity of the image; The global contrast of the image is converted into a global contrast index. When the global contrast of the image is close to the optimal contrast reference value, the global contrast index increases. When the global contrast of the image deviates from the reference value, the global contrast index decreases. Based on the average brightness of the image, it is mapped to a brightness suitability index through a Gaussian function. The brightness suitability index reaches its maximum value when the average brightness of the image is equal to the preset ideal value, and decreases monotonically as the average brightness of the image deviates from the ideal value. The image sharpness and background texture intensity are both subjected to maximum-min normalization to obtain the image sharpness index and background texture intensity index. Based on the contrast index, luminance deviation index, sharpness index, and texture intensity index, the image quality coefficient with a value between 0 and 1 is calculated by geometric mean.

9. The surface defect detection device for hollow glass microsphere thermal insulation material according to claim 8, characterized in that, The higher the contrast index, brightness deviation index, and sharpness index, the higher the image quality coefficient, indicating better image quality. The texture intensity index participates in the calculation in a negatively correlated manner; the lower the texture intensity index, the higher the image quality coefficient, indicating better image quality.

Images