Detection method for 3D hard metal processing based on weighted calculation

Abstract

The invention belongs to the technical field of hard metal processing, and particularly relates to a 3D hard metal processing detection method based on weighted calculation, which not only covers the complete process from sample pretreatment to final report compiling, but also introduces a weight factor to adjust the density distribution of each point in virtual mapping to form an optimized internal structure diagram. Through high-resolution scanning and local enhanced scanning, the method focuses on suspected defects to refine the image quality, and in combination with quantitative evaluation, all internal tiny defects can be accurately detected, and the sizes, forms and distribution characteristics of the internal tiny defects are comprehensively analyzed. Finally, a detailed detection report is generated, and a reliable basis is provided for product quality control. According to the method based on weighted calculation, the problem of insufficient resolution in the prior art is effectively solved, finer and more reliable internal defect detection is realized, and the overall quality level of a 3D hard metal processing product can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of hard gold processing, and in particular relates to a detection method for 3D hard gold processing based on weighted calculation. Background Art

[0002] In the field of 3D hard metal processing, ensuring the quality and performance of the finished product is crucial. Traditional inspection methods mainly rely on surface visual inspection or destructive testing such as slice analysis. Although these methods can ensure product quality to a certain extent, they have obvious limitations. For example, surface visual inspection is difficult to find hidden defects inside, while destructive testing will damage the sample itself and is not suitable for widespread application in mass production.

[0003] In recent years, with the development of non-destructive testing technology, especially the technology based on optical imaging and multi-angle light source illumination, it has been gradually applied to the inspection of 3D hard gold products. This type of method can construct a virtual map of the internal structure of the sample by applying different lighting conditions and recording the reflection spectrum characteristics, thereby achieving preliminary identification of internal defects. However, when dealing with complex structures, existing technologies often face the problem of insufficient resolution, especially when it is necessary to accurately reflect tiny defects. Traditional methods are difficult to provide sufficient detailed information, resulting in inaccurate detection results. Summary of the invention

[0004] The purpose of the present invention is to provide a detection method for 3D hard gold processing based on weighted calculation, which effectively solves the problem of insufficient resolution of the prior art, realizes more precise and reliable internal defect detection, and helps to improve the overall quality level of 3D hard gold processed products.

[0005] To achieve the above object, the present invention adopts the following technical solution: a 3D hard gold processing detection method based on weighted calculation, comprising the following steps: Prepare the samples to be tested, pre-treat them, apply multi-angle light sources to the samples, and record the reflection spectrum characteristics under different lighting conditions; Based on the obtained reflection spectrum characteristics, an initial virtual map of the internal structure of the sample is constructed, and the initial virtual map covers all known physical properties; On the basis of the virtual mapping, the density distribution of each point is adjusted by introducing a weight factor to form an optimized internal structure map. According to the generated structure map, several representative areas are selected and high-resolution scanning is performed to obtain detailed data. The obtained detailed data is compared and analyzed with the original virtual mapping, the differences are identified and the defect locations are marked, and the marked defect locations are used to perform local enhanced scanning, focus on the suspected defects, and refine the image quality; Quantitatively evaluate each defect area to determine its size, shape and distribution characteristics, comprehensively analyze various indicators, and compile a complete test report to point out all detected internal micro defects and their impact.

[0006] Preferably, the pretreatment of the sample comprises: Clean the sample surface with a detergent to remove all non-metallic residues, and purge with dry gas to eliminate moisture left after cleaning; The treated samples are measured and recorded with an optical measuring instrument for the reflectivity DR of the sample surface, wherein the reflectivity DR is calculated by the formula (DR=Ir / Ii), where Ir represents the intensity of reflected light and Ii represents the intensity of incident light; Based on the obtained reflectivity DR value, the output intensity F of the light source E is adjusted to a preset threshold G, and the adjustment process follows the formula F=G*DR.

[0007] Preferably, applying multi-angle light source illumination to the sample and recording the reflection spectrum characteristics under different illumination conditions includes: A group of light sources H are selected, each light source having a different wavelength WL1, WL2, ..., WLn, and the pretreated samples are sequentially irradiated; For each illumination condition at wavelength WLi, the corresponding reflected light intensity RIi is recorded to form a series of reflected light intensity data sets {RI1, RI2, ..., RIn}; Based on the obtained reflected light intensity data set, the reflectance spectrum characteristic value SPi at each wavelength WLi is calculated using the formula SPi=RIi / Ii; According to all the SPi values ​​obtained, a reflection spectrum curve C is constructed, and the reflection spectrum curve C is represented by connecting the points (SP1, WL1), (SP2, WL2), ..., (SPn, WLn).

[0008] Preferably, the initial virtual mapping of the internal structure of the sample is constructed based on the obtained reflection spectrum characteristics, and the initial virtual mapping covers all known physical properties, including: According to the recorded reflectance spectrum characteristic value SPi, combined with the geometric shape parameter GP of the sample surface, the initial mapping relationship M1 is established. The relationship is expressed by the formula M1=f(SPi,GP), where f is the process of building the mapping; Using the mapping relationship M1, the interior of the sample is preliminarily divided to determine multiple discrete regions VR1, VR2, ..., VRm; For each divided area Vri, analyze the corresponding reflectance spectrum characteristic value SPi, calculate the estimated value QAi of the physical property QA in the area, and use the formula QAi=g(SPi), where g is the process of converting to the estimated value of the physical property; All the QAi values ​​and the spatial distribution information of each region Vri are integrated to construct a three-dimensional virtual map M2 containing all known physical properties. The three-dimensional virtual map M2 consists of a series of point sets {(QAi, Vri)}.

[0009] Preferably, the density distribution of each point is adjusted by introducing a weight factor based on the formed virtual mapping to form an optimized internal structure diagram, including: For each point Vri in the formed three-dimensional virtual map M2, the reliability of its corresponding physical attribute QAi is evaluated and an initial weight WF0 is assigned. The weight is calculated by the formula WF0=h(QAi), where h is the process of determining the weight; Based on the obtained initial weight WF0, the relative distance DLi,j between each point Vri is calculated, where DLi,j represents the distance between points Vri and Vrj, using the formula: DLi,j=sqrt((xri-xrj)^2+(yri-yrj)^2+(zri-zrj)^2), where (xri,yri,zri) are the coordinates of point Vri, and (xrj,yrj,zrj) are the coordinates of point Vrj; Combined with the calculated distance DLi,j, adjust the new weight WFi,j between each pair of points Vri and Vrj. The new weight value is determined by the formula WFi,j=WF0*exp(-α*DLi,j^2), where α is a predefined proportional coefficient; Using the updated weights WFi,j, the density distribution DDn of all points Vri is recalculated to form the optimized internal structure graph M3. The new density distribution DDn is calculated by the formula DDn(Vri)=Σ(WFi,j) / N, where N is the number of all points connected to the point Vri, and Σ represents the summation operation.

[0010] Preferably, the method of selecting several representative areas based on the generated structural diagram and performing high-resolution scanning to obtain detailed data includes: According to the optimized internal structure diagram M3, multiple candidate regions CR1, CR2, ..., CRk with significant physical property differences are identified and marked by comparing the property change rate ΔQA between adjacent points Vri, and the formula is ΔQA=|QAi+1-QAi| / DLi,i+1; Based on the determined candidate regions Cri, select standard regions SR1, SR2, ..., SRm that meet the preset attribute change threshold TL, and the selection of standard regions follows the condition TL≤ΔQA; For each selected standard area Sri, a high-resolution scanning path PLi is planned; the path design ensures that the entire area is covered while minimizing the scanning time, and the path length PLi is calculated by the formula PLi=Σ(DLj,j+1); According to the set path PLi, a detailed high-resolution scan is performed on each standard area Sri, and finer structural information FIi in the area is recorded to construct a higher-precision local map MIi.

[0011] Preferably, the step of comparing and analyzing the obtained detailed data with the original virtual mapping, identifying differences and marking defect locations includes: The obtained detailed data set DF5 is combined with the original virtual map VMAP to generate a comparative analysis matrix BM; the comparative analysis matrix BM quantifies the data difference between the corresponding positions of the two into a value ΔDAi,j, and the formula is expressed as ΔDAi,j=|DAi,j(DF5)-DAi,j(VMAP)|; Based on the formed comparative analysis matrix BM, all coordinate points exceeding the preset difference threshold TL are identified to form a set of suspected outlier points PS; For all the marked suspected outlier points PSi, a local refinement algorithm is applied to calculate the neighborhood structure feature vector FVi of each point; the feature vector FVi is determined by the local pattern formed by the current point and its surrounding n closest points; Based on the obtained feature vector FVi, each suspected abnormal point PSi is deeply evaluated to decide whether it is finally marked as a defect location. The confidence score CScore is introduced into the evaluation process, and the calculation method is CScore(PSi)=h(FVi).

[0012] Preferably, the method of using the marked defect position to perform local enhanced scanning, focusing on the suspected defect position, and refining the image quality includes: Based on the marked suspected defect location set DS, determine the coordinate range RSi of each defect location DSi. The coordinate range RSi is defined by extending the defect point by a certain distance δ, and the formula is RSi={P|d(P,DSi)≤δ}; For each determined coordinate range RSi, a high-precision local scanning path PLi is planned; the path design ensures that the entire area is covered while optimizing the scanning path length PLi; According to the set path PLi, an enhanced resolution scan is performed at each suspected defect to record detailed structural information FIi; the structural information FIi is used to construct a local refinement map MIi; The obtained local refinement map MIi is used to evaluate the image quality of each suspected defect position DSi. The evaluation process introduces the clarity score IScore, which is calculated as IScore(MIi)=h(Isharp,Inoise); Isharp is the sharpness index and Inoise is the noise level.

[0013] Preferably, the quantitative evaluation of each defect area to determine its size, shape and distribution characteristics includes: Based on the local refinement map MIi, the area AS of each suspected defect location DSi is calculated; the area is calculated by the formula AS(DSi)=Σ(Pij), where Pij represents the set of all pixels belonging to the defect area DSi; According to the obtained area AS(DSi), the morphological feature FShape of the suspected defect area is analyzed; the morphological feature FShape is determined by the defect boundary shape descriptor SD, and the expression is FShape(DSi)=g(SD); Combined with the morphological feature FShape, the spatial distribution characteristics DDist of each suspected defect area are evaluated; the spatial distribution characteristics are measured by analyzing the distance DLi,j between adjacent defects and the relative angle θi,j between them; The obtained spatial distribution characteristics DDist, together with the area AS and the morphological characteristics FShape, are used to comprehensively and quantitatively evaluate the overall attributes QTotal of each suspected defect area; the overall attributes are calculated by the formula QTotal(DSi)=w1*AS(DSi)+w2*FShape(DSi)+w3*DDist(DSi); w1, w2, w3 are weight factors.

[0014] Preferably, the comprehensive indicators are combined to prepare a complete test report, including: According to the overall property QTotal(DSi) obtained by quantitative evaluation, the impact level GL is determined for each suspected defect location DSi; the impact level GL is calculated by the formula GL(DSi)=f(QTotal(DSi)); Based on the determined impact level GL, all suspected defects are classified and summarized to form a defect list LS; each entry in the list contains the defect location DSi, area AS(DSi), shape feature FShape(DSi) and spatial distribution characteristic DDist(DSi), as well as the corresponding impact level GL(DSi); Using the defect list LS, prepare a preliminary inspection report RDraft; the report structure follows the template TM, which defines how to organize and present defect information; Review the prepared preliminary report RDraft and add the comprehensive conclusion CC; the conclusion summarizes the number, location and impact of all internal minor defects, and proposes recommended measures SM; the final inspection report RFinal is composed of the formula RFinal=RDraft+CC+SM.

[0015] Technical effects and advantages of the present invention: The 3D hard gold processing detection method based on weighted calculation proposed by the present invention has the following advantages compared with the prior art: The present invention not only covers the complete process from sample pretreatment to final report preparation, but more importantly, it introduces weight factors to adjust the density distribution of each point in the virtual mapping to form an optimized internal structure diagram. This method can significantly improve the detection accuracy, especially the ability to identify tiny defects has made a qualitative leap. Through high-resolution scanning and local enhanced scanning, focus on suspected defects to refine the image quality, combined with quantitative evaluation, to ensure that all internal tiny defects can be accurately detected, and conduct a comprehensive analysis of their size, shape and distribution characteristics. Finally, a detailed test report is generated, providing a reliable basis for product quality control. This method based on weighted calculation effectively solves the problem of insufficient resolution of the existing technology, realizes more refined and reliable internal defect detection, and helps to improve the overall quality level of 3D hard gold processing products. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of the detection method for 3D hard gold processing based on weighted calculation. DETAILED DESCRIPTION

[0017] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] The present invention provides Figure 1 The detection method for 3D hard gold processing based on weighted calculation shown includes the following steps: Step 1: Prepare the sample to be tested and pre-treat the sample; further comprising: Use a cleaning agent to clean the sample surface to remove all non-metallic residues, and use dry gas to purge to eliminate moisture left after cleaning; by removing non-metallic residues and moisture on the sample surface, ensure that there is no interference from external substances during the subsequent detection process, so that the reflection spectrum characteristics can truly reflect the internal structure of the sample. Cleaning and drying can ensure the consistency of the sample surface and reduce measurement errors caused by differences in surface conditions.

[0019] The processed samples are measured and recorded with an optical measuring instrument for the reflectivity DR of the sample surface. The reflectivity DR is calculated using the formula (DR=Ir / Ii), where Ir represents the reflected light intensity and Ii represents the incident light intensity. Reflectivity refers to the ratio of the light intensity reflected from the sample surface to the light intensity incident on the sample. It reflects the reflectivity of the sample surface for light of a specific wavelength and is one of the important parameters for evaluating the surface properties of the sample. This formula is used to calculate the reflectivity, where Ir is the reflected light intensity, i.e., the light intensity reflected from the sample surface; Ii is the incident light intensity, i.e., the light intensity irradiated onto the sample surface. The reflectivity DR value range is usually between 0 and 1, and the higher the value, the stronger the reflection.

[0020] Based on the reflectivity DR value obtained, the output intensity F of the light source E is adjusted to the preset threshold value G. The adjustment process follows the formula F=G*DR. This formula is used to determine the final output intensity of the light source. By multiplying the preset threshold value G by the reflectivity DR, the light source output can be dynamically adjusted so that the light intensity received by the sample surface is kept within an ideal range, neither overexposed nor underexposed.

[0021] Light source output intensity F: Light source output intensity refers to the ability or brightness of a light source to emit light. The unit is usually watt (W) or candela (cd), depending on the application scenario.

[0022] Preset threshold G: This is a pre-set target value designed to ensure that different samples receive consistent light intensity under the same lighting conditions, thereby ensuring consistency and comparability of test results.

[0023] By adjusting the light source output intensity, each sample is ensured to be under the same lighting conditions when being tested, thus improving the reliability and repeatability of the test results.

[0024] Embodiment 1

[0025] Assume that there is a 3D hard gold processed sample, whose surface is cleaned and dried, and the incident light intensity Ii measured by an optical measuring instrument is 100 units, and the reflected light intensity Ir is 85 units. According to the formula: DR=Ir / IiDR=85 / 100; DR=0.85; Next, knowing that the preset threshold G is 120 units, in order to adjust the output intensity F of the light source E, the formula is applied: F=G*DRF=120*0.85; F = 102; Therefore, the output intensity of light source E should be adjusted to 102 units to ensure that the light intensity received by the sample surface is moderate, neither too bright nor too dark, thereby providing ideal conditions for subsequent multi-angle light source illumination and reflection spectrum feature recording.

[0026] Through the above embodiments, it can be seen how to achieve precise control of the output intensity of the light source through specific measurement and calculation, thereby improving the accuracy and reliability of the entire detection process.

[0027] Step 2: applying multi-angle light source illumination to the sample and recording the reflection spectrum characteristics under different illumination conditions; further comprising: A group of light sources H are selected, each with a different wavelength WL1, WL2, ..., WLn, and the pre-treated samples are irradiated in sequence; by selecting multiple light sources with different wavelengths, a wide spectrum range from ultraviolet to infrared can be covered, ensuring that the characteristics of the sample under various spectral conditions can be detected. Different wavelengths of light have different penetration depths and reflection characteristics on the material. Using a multi-wavelength light source can improve the sensitivity of detection and help discover subtle internal structural changes or defects.

[0028] For each illumination condition at wavelength WLi, the corresponding reflected light intensity RIi is recorded to form a series of reflected light intensity data sets {RI1, RI2, ..., RIn}; reflected light intensity RIi: the light intensity reflected back from the sample surface at a specific wavelength WLi. It reflects the sample's ability to reflect light of that wavelength and is the basic data for constructing the characteristics of the reflective spectrum. By recording the reflected light intensity at different wavelengths, detailed information about the optical properties of the sample can be obtained, providing rich data support for subsequent analysis. Multiple measurements and forming data sets can help reduce the random errors that may be caused by a single measurement and improve the accuracy and reliability of the data.

[0029] Based on the obtained reflected light intensity data set, the reflectance spectrum characteristic value SPi at each wavelength WLi is calculated using the formula SPi=RIi / Ii; Reflection spectrum characteristic value SPi: This refers to the reflectance of the sample at a specific wavelength WLi. It is calculated by dividing the reflected light intensity RIi by the incident light intensity Ii, reflecting the reflectance characteristics of the sample at that wavelength. By calculating the reflectance spectrum characteristic value, the reflected light intensity at different wavelengths can be standardized for easy comparison and analysis. The reflectance spectrum characteristic value directly reflects the optical properties of the sample at different wavelengths, which is very important for understanding the internal structure and composition of the material.

[0030] According to all the SPi values ​​obtained, the reflection spectrum curve C is constructed, and the reflection spectrum curve C is represented by connecting the points (SP1, WL1), (SP2, WL2), ..., (SPn, WLn). The reflection spectrum curve intuitively shows the reflection characteristics of the sample at different wavelengths, allowing researchers to quickly understand the optical behavior of the sample. The reflection spectrum curve provides rich information and can be used to assist in diagnosing whether there are defects, impurities or other abnormalities inside the sample.

[0031] Embodiment 2

[0032] Assume that there is a 3D hard gold processed sample that has been pre-processed. Now select five light sources H with different wavelengths, namely WL1=450nm, WL2=550nm, WL3=650nm, WL4=750nm and WL5=850nm. For each wavelength, the measured incident light intensity Ii is 100 units (to simplify the calculation), and the reflected light intensity RIi is: WL1:RI1=70 units; WL2:RI2=85 units; WL3:RI3=90 units; WL4:RI4=80 units; WL5:RI5=75 units; According to the formula SPi=RIi / Ii, the characteristic value of the reflection spectrum at each wavelength can be calculated: SP1=RI1 / Ii=70 / 100=0.70SP2=RI2 / Ii=85 / 100=0.85; SP3=RI3 / Ii=90 / 100=0.90; SP4=RI4 / Ii=80 / 100=0.80; SP5=RI5 / Ii=75 / 100=0.75; Next, based on these SPi values, a reflectance spectrum curve C is constructed by connecting the points (SP1, WL1), (SP2, WL2), ..., (SP5, WL5). The specific points are as follows: (0.70,450nm); (0.85,550nm); (0.90,650nm); (0.80,750nm); (0.75,850nm); This reflectance spectrum curve shows the reflectance characteristics of the sample at different wavelengths, which can help further analyze the internal structure and possible defects of the sample. For example, if the reflectance spectrum characteristic value at a certain wavelength is significantly lower than that at other wavelengths, this may be due to some absorption defects or composition changes inside the sample.

[0033] Step 3: Based on the obtained reflectance spectrum characteristics, construct an initial virtual map of the internal structure of the sample, which covers all known physical properties; further comprising: According to the recorded reflectance spectrum characteristic value SPi, combined with the geometric shape parameter GP of the sample surface, the initial mapping relationship M1 is established. The relationship is expressed by the formula M1=f(SPi,GP), where f is the process of constructing the mapping; combining the reflectance spectrum characteristic value SPi with the geometric shape parameter GP of the sample surface can more accurately reflect the actual physical state of the sample. Establishing the initial mapping relationship M1 provides a basic framework for subsequent detailed analysis and regional division, ensuring that the optical characteristics of each point correspond to its spatial position.

[0034] Using the mapping relationship M1, the interior of the sample is preliminarily divided to determine multiple discrete regions VR1, VR2, ..., VRm;: Through the preliminary division, the interior of the sample can be divided into multiple discrete regions, which is convenient for detailed physical property evaluation of different regions. Regional division simplifies the subsequent analysis process, allowing computing resources to be concentrated in important or suspicious areas, improving the overall analysis efficiency.

[0035] For each divided area Vri, the corresponding reflection spectrum characteristic value SPi is analyzed, and the estimated value QAi of the physical property QA in the area is calculated, using the formula QAi=g(SPi), where g is the process of converting to the estimated value of the physical property; through the formula QAi=g(SPi), the optical measurement results can be converted into specific physical properties, providing a quantitative basis for subsequent quality assessment. The estimated value of the physical property QAi helps to identify abnormal areas, such as uneven hardness or composition changes, which may be signs of potential defects.

[0036] All the QAi values ​​and the spatial distribution information of each region Vri are combined to construct a three-dimensional virtual map M2 containing all known physical properties. The three-dimensional virtual map M2 consists of a series of point sets {(QAi, Vri)}. The three-dimensional virtual map M2 intuitively displays the physical properties and spatial distribution of each region inside the sample, providing a visualization tool for a comprehensive understanding of the sample. Through three-dimensional mapping, researchers can quickly locate problem areas and support further defect analysis and quality control decisions.

[0037] Embodiment 3

[0038] Assume that there is a 3D hard gold processing sample that has been processed in the first two steps and the reflection spectrum characteristic value SPi has been obtained. Now the following specific operations will be performed: Establishing the initial mapping relationship M1: Based on the reflectance spectrum characteristic value SPi and the geometric shape parameter GP of the sample surface, the formula M1=f(SPi,GP) is used to construct the initial mapping relationship. Assuming that the mapping relationship M1 is established by a certain mapping method (such as interpolation or fitting), this step provides a basis for subsequent analysis.

[0039] Preliminary division of the interior of the sample: Based on the mapping relationship M1, the interior of the sample is divided into five discrete regions VR1 to VR5. Each region covers a different part, and each region contains a set of reflectance spectrum characteristic values ​​SPi.

[0040] Calculate the estimated value of physical property QAi: For each region Vri, use the formula QAi=g(SPi) to calculate the estimated value of physical property in the region. Assuming that hardness is selected as the physical property QA, through some conversion method (such as table lookup or model prediction), the estimated value of hardness QAi for each region is obtained as follows: VR1:QA1=80HV; VR2:QA2=75HV; VR3:QA3=82HV; VR4:QA4=78HV; VR5:QA5=81HV, where HV represents the Vickers hardness unit.

[0041] Construct three-dimensional virtual mapping M2: Finally, all QAi values ​​and the spatial distribution information of each region Vri are integrated to construct a three-dimensional virtual mapping M2. Mapping M2 consists of a series of point sets {(QAi, Vri)}, the specific points are as follows: (80HV, VR1); (75HV, VR2); (82HV, VR3); (78HV, VR4); (81HV, VR5); Through the above examples, we can see how to gradually build a three-dimensional virtual mapping covering all known physical properties starting from the reflectance spectrum characteristic values. This method not only improves the accuracy of detection, but also provides strong support for subsequent quality assessment and defect analysis.

[0042] Step 4: Based on the formed virtual mapping, the density distribution of each point is adjusted by introducing a weight factor to form an optimized internal structure diagram; further comprising: For each point Vri in the formed three-dimensional virtual map M2, the reliability of its corresponding physical attribute QAi is evaluated and an initial weight WF0 is assigned. The weight is calculated by the formula WF0=h(QAi), where h is the process of determining the weight. By evaluating the reliability of the physical attribute QAi and assigning the initial weight WF0, it can be ensured that those more reliable data points occupy a larger proportion in subsequent analysis, thereby improving the accuracy of the overall test results. Low reliability points may indicate potential measurement errors or sample defects. Assigning lower weights can help filter out these outliers and reduce the impact on the final results.

[0043] Based on the obtained initial weight WF0, the relative distance DLi,j between each point Vri is calculated, where DLi,j represents the distance between points Vri and Vrj, using the formula: DLi,j=sqrt((xri-xrj)^2+(yri-yrj)^2+(zri-zrj)^2), where (xri,yri,zri) is the coordinate of point Vri, (xrj,yrj,zrj) is the coordinate of point Vrj; by calculating the relative distance DLi,j between each point, the spatial relationship between them can be quantified, which is crucial for understanding the internal structure of the sample. Distance information helps to identify local features, such as similarities and differences between adjacent regions.

[0044] Combined with the calculated distance DLi,j, the new weight WFi,j of the connection between each pair of points Vri and Vrj is adjusted. The new weight value is determined by the formula WFi,j=WF0*exp(-α*DLi,j^2), where α is a predefined proportional coefficient. By introducing the distance factor, the weight can be dynamically adjusted according to the relative positions between the points, so that the connection between neighboring points is closer, while the influence of distant points is weakened.

[0045] Using the updated weights WFi,j, the density distribution DDn of all points Vri is recalculated to form the optimized internal structure diagram M3. The new density distribution DDn is calculated by the formula DDn(Vri)=Σ(WFi,j) / N, where N is the number of all points connected to point Vri, and Σ represents the summation operation. By recalculating the density distribution, the actual structure inside the sample can be more accurately reflected, and the resolution and accuracy of the detection can be improved. The optimized internal structure diagram M3 intuitively shows the density distribution inside the sample, providing strong support for quality assessment.

[0046] Embodiment 4

[0047] Assume that there is a 3D hard gold processing sample, a 3D virtual map M2 has been formed, and the physical properties QAi of each point Vri have been obtained. Now the following specific operations will be performed: Evaluate reliability and assign initial weight WF0: According to the reliability of the physical attribute QAi, use the formula WF0=h(QAi) to assign an initial weight to each point. Assuming a linear mapping method is used (for example, the reliability score is directly used as the weight), the initial weight WF0 of each point is obtained as follows: Vr1:QA1=80HV->WF0_1=0.9; Vr2:QA2=75HV->WF0_2=0.8; Vr3:QA3=82HV->WF0_3=0.95; Vr4:QA4=78HV->WF0_4=0.85; Vr5:QA5=81HV->WF0_5=0.92; Calculate the relative distance DLi,j between each point: For each pair of points Vri and Vrj, use the formula DLi,j=sqrt((xri-xrj)^2+(yri-yrj)^2+(zri-zrj)^2) to calculate the relative distance between them. Assume the following coordinates: Vr1:(1,2,3); Vr2:(4,5,6); Vr3:(7,8,9); Vr4:(10,11,12); Vr5:(13,14,15); The following relative distances are calculated: DL1,2=5.196; DL1,3=10.392; DL1,4=15.588; DL1,5=20.785; ... (and so on); Adjust the new weight WFi,j of the connection between each pair of points: Based on the calculated distance DLi,j, use the formula WFi,j=WF0*exp(-α*DLi,j^2) to update the weight. Assuming the proportionality coefficient α=0.01, the new weight is calculated as follows: WFi,2=WF0_1*exp(-0.01*DL1,2^2)=0.9*exp(-0.01*5.196^2)≈0.82; WFi,3=WF0_1*exp(-0.01*DL1,3^2)=0.9*exp(-0.01*10.392^2)≈0.67; WFi,4=WF0_1*exp(-0.01*DL1,4^2)=0.9*exp(-0.01*15.588^2)≈0.48; WFi,5=WF0_1*exp(-0.01*DL1,5^2)=0.9*exp(-0.01*20.785^2)≈0.33; ... (and so on); Recalculate the density distribution DDn and form the optimized internal structure diagram M3: Finally, use the updated weight WFi,j to recalculate the density distribution of all points through the formula DDn(Vri)=Σ(WFi,j) / N. Assuming that each point is connected to its four nearest neighbors, the following density distribution values ​​are calculated: DDn(Vr1)=(WFi,2+WFi,3+WFi,4+WFi,5) / 4; ... (and so on); Through the above embodiment, we can see how to start from the initial virtual mapping, gradually introduce weight factors to adjust the density distribution of each point, and finally form the optimized internal structure diagram M3.

[0048] Step 5: Based on the generated structure map, select several representative areas and perform high-resolution scanning to obtain detailed data; further including: According to the optimized internal structure diagram M3, multiple candidate regions CR1, CR2, ..., CRk with significant differences in physical properties are identified and marked. This is achieved by comparing the property change rate ΔQA between adjacent points Vri. The formula is ΔQA=|QAi+1-QAi| / DLi,i+1. By calculating the physical property change rate between adjacent points, those areas with obvious differences in physical properties can be accurately located. These areas may be defects, boundaries or other important structures.

[0049] Based on the determined candidate regions Cri, standard regions SR1, SR2, ..., SRm that meet the preset attribute change threshold TL are selected. The selection of standard regions follows the condition TL≤ΔQA. By setting an attribute change threshold TL, standard regions that are truly worthy of attention can be selected from many candidate regions, ensuring that only those regions of great significance are further analyzed.

[0050] For each selected standard area Sri, a high-resolution scanning path PLi is planned; the path design ensures that the entire area is covered while minimizing the scanning time, and the path length PLi is calculated by the formula PLi=Σ(DLj,j+1); reasonable path planning can not only ensure that all areas of interest are fully covered, but also minimize the scanning time and improve work efficiency.

[0051] According to the set path PLi, a detailed high-resolution scan is performed on each standard area Sri, and the finer structural information FIi in the area is recorded to construct a higher-precision local map MIi. High-resolution scanning provides more detailed data than conventional scanning, which helps to find subtle structural features or defects. Refined local mapping provides a solid foundation for subsequent quality assessment, fault diagnosis, etc., and helps to make more accurate and reliable judgments.

[0052] Embodiment 5

[0053] Assume that you have obtained an optimized internal structure map M3 of a hard gold sample and want to select some areas with significant differences in physical properties for high-resolution scanning. You will follow the following steps: Identify candidate regions CR: First, use the formula ΔQA=|QAi+1-QAi| / DLi,i+1 to calculate the attribute change rate between adjacent points. For example, in a specific direction, the following results may be obtained: ΔQA1=|80HV-75HV| / 5.196≈0.96; ΔQA2=|75HV-82HV| / 10.392≈0.67; ΔQA3=|82HV-78HV| / 15.588≈0.26; ΔQA4=|78HV-81HV| / 20.785≈0.14; Assume that several locations with high ΔQA are found and marked as candidate regions CR1, CR2, ...

[0054] Select standard region SR: Next, set an attribute change threshold TL, such as 0.5. Then, check which candidate regions have ΔQA greater than or equal to this threshold. If only ΔQA1 exceeds the threshold in the above example, CR1 is selected as standard region SR1.

[0055] Planning scanning path PL: For the selected standard area SR1, it is necessary to plan a path that covers the entire area and minimizes the scanning time. Assuming that SR1 consists of five key points, the path length can be calculated based on the distance between points: PL1=DL1,2+DL2,3+DL3,4+DL4,5; If these distances are 5.196, 10.392, 15.588, 20.785, then PL1 = 5.196 + 10.392 + 15.588 + 20.785 = 51.961.

[0056] Perform high-resolution scan: Finally, a high-resolution scan is performed on the standard area SR1 along the planned path PL1 to record finer structural information FI1. This may involve collecting additional images, spectra or other forms of data, which are ultimately used to construct a higher-precision local map MI1.

[0057] Through the above embodiments, it can be seen how to start from the optimized internal structure diagram, gradually identify, select and scan areas with significant physical property differences, so as to obtain more detailed data and provide strong support for subsequent analysis.

[0058] Step 6: Compare and analyze the obtained detailed data with the original virtual map, identify the differences and mark the defect locations; further including: The acquired detailed data set DF5 is combined with the original virtual map VMAP to generate a comparative analysis matrix BM. The comparative analysis matrix BM quantifies the data difference between the corresponding positions of the two into a value ΔDAi,j, and the formula is expressed as ΔDAi,j=|DAi,j(DF5)-DAi,j(VMAP)|. Through the comparative analysis matrix BM, the difference between the detailed data after high-resolution scanning and the original virtual map can be accurately quantified, so that any potential changes or anomalies can be clearly seen.

[0059] Based on the formed comparative analysis matrix BM, all coordinate points exceeding the preset difference threshold TL are identified to form a set of suspected abnormal points PS; by setting the difference threshold TL, those coordinate points with significant differences can be screened out from the comparative analysis matrix, which may be defects or other abnormal conditions inside the sample.

[0060] For all marked suspected outlier points PSi, a local refinement algorithm is applied to calculate the neighborhood structure feature vector FVi of each point; the feature vector FVi is determined by the local pattern formed by the current point and its surrounding n closest points; by calculating the feature vector FVi, more detailed structural information about the suspected outlier point PSi and its surrounding areas can be obtained, which helps to further evaluate the properties of these points.

[0061] Based on the obtained feature vector FVi, each suspected abnormal point PSi is deeply evaluated to determine whether it is finally marked as a defect location. The confidence score CScore is introduced into the evaluation process, and the calculation method is CScore(PSi)=h(FVi). By introducing the confidence score CScore, it is possible to more accurately determine whether the suspected abnormal point is indeed a defect, reducing the possibility of incorrect labeling.

[0062] Embodiment 6

[0063] Assume that a detailed dataset DF5 of a standard region Sri has been obtained and the original virtual map VMAP is available. The following steps will be followed: Generate the comparative analysis matrix BM: First, use the formula ΔDAi,j=|DAi,j(DF5)-DAi,j(VMAP)| to calculate the data difference of each corresponding position between the detailed data set DF5 and the original virtual map VMAP. For example, in a specific area, the following results may be obtained: ΔDA1,1=|80HV-79HV|=1; ΔDA1,2=|75HV-76HV|=1; ΔDA2,1=|82HV-83HV|=1; ΔDA2,2=|78HV-77HV|=1; These difference values ​​form part of the comparative analysis matrix BM.

[0064] Identify the suspected outlier set PS: Next, set a difference threshold TL, such as 2. Then, check which points in the comparative analysis matrix BM have difference values ​​exceeding this threshold. If all difference values ​​in the above example are less than 2, no point will be selected into the suspected outlier set PS. But suppose there is ΔDAx,y=3 at another position, then this point will be marked as suspected outlier PS1.

[0065] Calculate the feature vector FVi: For each suspected outlier point PSi (such as PS1), apply the local refinement algorithm to calculate its neighborhood structure feature vector FVi. Assuming that the three nearest points around PS1 are P1, P2, and P3, the feature vector FV1 can be constructed based on their physical properties and spatial relationships.

[0066] Deep evaluation and marking of defect locations: Finally, the formula CScore(PSi)=h(FVi) is used to deeply evaluate each suspected anomaly point PSi. Assuming that a machine learning-based classification method is used to determine the confidence score, the following results are obtained: PS1: CScore(PS1)=h(FV1)=0.85 (high confidence); If the confidence score is higher than a preset threshold (e.g. 0.8), the current point is officially marked as a defect location.

[0067] Through the above examples, we can see how to start from the comparative analysis matrix, gradually identify, evaluate and finally mark the defect location, thereby providing a scientific basis and technical support for quality control and defect management. This method not only improves the accuracy of detection, but also provides guidance for subsequent repair and improvement measures.

[0068] Step 7: Using the marked defect locations, perform local enhanced scanning, focus on the suspected defects, and refine the image quality; further including: Based on the marked suspected defect position set DS, the coordinate range RSi of each defect position DSi is determined. The coordinate range RSi is defined by extending the defect point by a certain distance δ, and the formula is RSi={P|d(P,DSi)≤δ}; by extending the defect point by a certain distance δ, it can be ensured that the scan covers the defect and the surrounding areas that may be affected, avoiding missing any potential problems.

[0069] For each determined coordinate range RSi, a high-precision local scanning path PLi is planned; the path design ensures coverage of the entire area while optimizing the scanning path length PLi; reasonable path planning can ensure that the scanning device effectively covers the entire specified area while minimizing unnecessary movement, saving time and resources.

[0070] According to the set path PLi, an enhanced resolution scan is performed at each suspected defect to record detailed structural information FIi; the structural information FIi is used to construct a local refinement map MIi; high-resolution scanning provides more detailed data than conventional scanning, which helps to discover subtle structural features or defects. The detailed structural information FIi provides a solid foundation for subsequent in-depth analysis and the construction of local refinement maps MIi.

[0071] The obtained local refinement map MIi is used to evaluate the image quality of each suspected defect position DSi. The sharpness score IScore is introduced into the evaluation process, and the calculation method is IScore(MIi)=h(Isharp,Inoise); Isharp is the sharpness index, and Inoise is the noise level. By introducing the sharpness score IScore, the quality of the local refinement map can be objectively evaluated to ensure that the final result meets the expected standards.

[0072] Embodiment 7

[0073] Assume that several suspected defect locations have been marked and you want to perform local enhanced scanning on them. You will follow the steps below: Determine the coordinate range RSi: First, use the formula RSi={P|d(P,DSi)≤δ} to define the coordinate range of each defect location DSi. For example, if a defect location DS1 is located at coordinates (5,6,7) and the expansion distance δ=2 is set, the coordinate range RS1 will include all points P that are no more than 2 units away from DS1.

[0074] RS1={P|d(P,(5,6,7))≤2}; Planning high-precision local scanning path PLi: Next, plan a high-precision local scanning path PLi for each coordinate range RSi that covers the entire area and minimizes the path length. Assuming that RS1 consists of several key points, the optimal path can be designed based on the positions of these points. For example, by connecting adjacent key points, ensure that the path is as short as possible and covers the area comprehensively.

[0075] Perform enhanced resolution scans and record structural information FIi: Perform high-resolution scans at each suspected defect according to the set path PLi, and record detailed structural information FIi. This may involve collecting additional images, spectra, or other forms of data, which are ultimately used to construct a local refinement map MIi.

[0076] For RS1, detailed structural information FI1 is obtained, and the local refinement map MI1 is constructed based on it.

[0077] Image quality assessment and calculation of clarity score IScore: Finally, the image quality of each suspected defect position DSi is assessed using the formula IScore(MIi)=h(Isharp,Inoise). Assuming that the sharpness index Isharp of MI1 is 0.9 (high sharpness) and the noise level Inoise is 0.1 (low noise), the clarity score is as follows: IScore(MI1)=h(0.9,0.1); If a comprehensive assessment method, such as a weighted average or a machine learning model, is used to calculate the clarity score, the following results may be obtained: IScore(MI1)=0.85 (high definition); Through the above embodiments, it can be seen how to start from the marked defect position and gradually perform local enhanced scanning, path planning, high-resolution data acquisition and image quality assessment, so as to obtain high-quality local refinement mapping, which provides a scientific basis and technical support for subsequent quality assessment and defect management.

[0078] Step 8: Quantitatively evaluate each defect area to determine its size, shape and distribution characteristics; further including: Based on the local refinement map MIi, the area AS of each suspected defect location DSi is calculated; the area is calculated by the formula AS(DSi)=Σ(Pij), where Pij represents the set of all pixel points belonging to the defect area DSi; by calculating the area AS, the actual size of each defect area can be intuitively quantified, providing basic data for subsequent analysis.

[0079] According to the obtained area AS(DSi), the morphological feature FShape of the suspected defect area is analyzed; the morphological feature FShape is determined by the defect boundary shape descriptor SD, and the expression is FShape(DSi)=g(SD); the morphological feature FShape provides information about the geometric shape of the defect area, such as circular, elliptical or irregular shape, which helps to understand the cause of the defect.

[0080] Combined with the morphological features FShape, the spatial distribution characteristics DDist of each suspected defect area are evaluated; the spatial distribution characteristics are measured by analyzing the distance DLi,j between adjacent defects and the relative angle θi,j between them; by analyzing the spatial distribution characteristics DDist, the relationship between defect areas and their distribution pattern in the entire sample can be revealed, thereby better understanding the defect generation mechanism.

[0081] The obtained spatial distribution characteristics DDist, together with the area AS and morphological characteristics FShape, are used to comprehensively and quantitatively evaluate the overall attribute QTotal of each suspected defect area; the overall attribute is calculated by the formula QTotal(DSi)=w1*AS(DSi)+w2*FShape(DSi)+w3*DDist(DSi); w1, w2, w3 are weight factors. By comprehensively considering the area, morphology and distribution characteristics, a comprehensive quantitative evaluation can be performed on each suspected defect area to ensure the integrity and accuracy of the evaluation results.

[0082] Embodiment 8

[0083] Assume that a local refinement map MI1 of a suspected defect location DS1 has been obtained and it is desired to be quantitatively evaluated. The following steps will be followed: Calculate the defect area AS: First, use the formula AS(DSi)=Σ(Pij) to calculate the area of ​​the defect area. Assuming that DS1 contains 50 pixels, then: AS(DS1)=50; Analyze the morphological features FShape: Next, analyze the morphological features FShape of the suspected defect area based on the area AS(DS1). Assume that the morphological features of DS1 are determined as follows through the defect boundary shape descriptor SD (such as perimeter, convex hull, main axis direction, etc.): SD(DS1)={perimeter: 25, convex hull area: 60, main axis direction: 45 degrees}; The morphological feature values ​​are calculated using a method (such as a machine learning model or an empirical formula): FShape(DS1)=g(SD(DS1))=0.7 (high morphological complexity); Evaluate the spatial distribution characteristics DDist: Combined with the morphological features FShape, evaluate the spatial distribution characteristics DDist of each suspected defect area. Assuming that there are two other defects DS2 and DS3 near DS1, calculate the distance DLi,j and relative angle θi,j between them as follows: DL1,2=10,θ1,2=30 degrees; DL1,3=15,θ1,3=60 degrees; Use some method (such as statistical analysis or clustering algorithm) to calculate the spatial distribution characteristic value: DDist(DS1)=h(DLi,j,θi,j)=0.6 (medium degree of aggregation); Comprehensive quantitative evaluation of the overall attribute QTotal: Finally, the formula QTotal(DSi)=w1*AS(DSi)+w2*FShape(DSi)+w3*DDist(DSi) is used to conduct a comprehensive quantitative evaluation of the suspected defect area. Assuming that the weight factors w1=0.5, w2=0.3, and w3=0.2 are set, then: QTotal(DS1)=0.5*50+0.3*0.7+0.2*0.6=25+0.21+0.12=25.33; Through the above embodiments, it can be seen how to start from local refinement mapping, gradually calculate the area, analyze the morphological characteristics, evaluate the spatial distribution characteristics, and finally comprehensively and quantitatively evaluate the overall properties of each suspected defect area.

[0084] Step 9: Comprehensively analyze various indicators and compile a complete test report, indicating all detected internal minor defects and their impact; further including: Based on the overall property QTotal(DSi) obtained by quantitative evaluation, the impact level GL is determined for each suspected defect location DSi; the impact level GL is calculated using the formula GL(DSi)=f(QTotal(DSi)); by calculating the impact level GL, the impact of each defect on the overall performance of the sample can be intuitively evaluated, which helps to prioritize high-risk defects.

[0085] Based on the determined impact level GL, all suspected defects are classified and summarized to form a defect list LS; each entry in the list contains the defect location DSi, area AS(DSi), morphological characteristics FShape(DSi) and spatial distribution characteristics DDist(DSi), as well as the corresponding impact level GL(DSi); through classification and summary, all suspected defects can be organized into an ordered list, which is convenient for subsequent analysis and report preparation.

[0086] Using the defect list LS, a preliminary test report RDraft is prepared. The report structure follows the template TM, which defines how to organize and present defect information. By following the template TM, all test reports are ensured to have a consistent structure and content, which improves the professionalism and readability of the report.

[0087] Review the prepared preliminary report RDraft and add the comprehensive conclusion CC; the conclusion summarizes the number, location and impact of all internal minor defects, and proposes recommended measures SM; the final test report RFinal is composed of the formula RFinal=RDraft+CC+SM. By adding the comprehensive conclusion CC, the test results can be summarized more comprehensively and provide a deep insight into the overall condition of the sample. The recommended measures SM provide specific guidance for subsequent work, helping to take appropriate repair or improvement measures to reduce potential risks.

[0088] Embodiment 9

[0089] Assume that you have completed a quantitative assessment of several suspected defect locations and wish to compile a complete inspection report. You will follow these steps: Determine the impact level GL: First, use the formula GL(DSi)=f(QTotal(DSi)) to calculate the impact level GL of each suspected defect location DSi. Assume the following overall attribute QTotal value: QTotal(DS1)=25.33; QTotal(DS2)=18.75; QTotal(DS3)=30.20; Using some method (such as threshold classification or machine learning model), the impact level GL can be determined as follows: GL(DS1)=f(25.33)=high; GL(DS2)=f(18.75)=medium; GL(DS3)=f(30.20)=high; Forming a defect list LS: Next, based on the determined impact level GL, all suspected defects are classified and summarized to form a defect list LS. Each entry in the list contains the defect location DSi, area AS(DSi), morphological features FShape(DSi), spatial distribution characteristics DDist(DSi), and the corresponding impact level GL(DSi). For example: DS1: area = 50, morphological characteristics = 0.7, spatial distribution characteristics = 0.6, impact level = high; DS2: area = 40, morphological characteristics = 0.5, spatial distribution characteristics = 0.4, impact level = medium; DS3: area = 60, morphological characteristics = 0.8, spatial distribution characteristics = 0.7, impact level = high; Prepare preliminary inspection report RDraft: Use the defect list LS to prepare the preliminary inspection report RDraft according to the requirements of the template TM. Assume that the template TM requires the report to include the following parts: Overview: Provides an overview of the inspection process and major issues found.

[0090] Detailed Description: Itemize information about each defect.

[0091] Graphic display: Use graphics to help illustrate the defect situation.

[0092] Review and add comprehensive conclusions CC: Finally, review the prepared preliminary report RDraft and add comprehensive conclusions CC. The conclusion summarizes the number, location and impact of all internal minor defects and proposes recommended measures SM. For example: Comprehensive conclusion CC: A total of three internal micro-defects were detected, two of which were located in the middle of the sample and one at the edge.

[0093] The defects are mainly manifested in highly complex morphological features and a high degree of spatial aggregation.

[0094] All defects were rated as medium to high impact and require prompt attention.

[0095] Suggested ActionsSM: For defects DS1 and DS3 with high impact levels, the repair process is initiated immediately to prevent further deterioration.

[0096] For defects DS2 of medium impact level, their development is monitored regularly and remedial measures are taken when necessary.

[0097] Strengthen the quality control process to reduce the probability of similar defects.

[0098] Form the final test report RFinal: The final test report RFinal consists of the preliminary report RDraft, the comprehensive conclusion CC and the recommended measures SM to ensure that the report content is complete and has practical operational value.

[0099] Through the above examples, we can see how to start from quantitative assessment, gradually determine the impact level, compile a defect list, write a preliminary report, and finally add comprehensive conclusions and recommended measures to form a detailed inspection report. This method not only improves the professionalism and accuracy of the report, but also provides valuable guidance and support for subsequent quality control and defect management.

[0100] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A 3D hard gold processing detection method based on weighted calculation, characterized in that: The following steps are involved: Prepare the samples to be tested, pre-treat them, apply multi-angle light sources to the samples, and record the reflection spectrum characteristics under different lighting conditions; Based on the obtained reflection spectrum characteristics, an initial virtual map of the internal structure of the sample is constructed, and the initial virtual map covers all known physical properties; On the basis of the virtual mapping, the density distribution of each point is adjusted by introducing a weight factor to form an optimized internal structure map. According to the generated structure map, several representative areas are selected and high-resolution scanning is performed to obtain detailed data. The obtained detailed data is compared and analyzed with the original virtual mapping, the differences are identified and the defect locations are marked, and the marked defect locations are used to perform local enhanced scanning, focus on the suspected defects, and refine the image quality; Conduct quantitative assessment of each defect area to determine its size, shape and distribution characteristics, and comprehensively analyze various indicators to compile a complete test report, pointing out all detected internal micro defects and their impact.

2. The 3D hard gold processing detection method based on weighted calculation according to claim 1, characterized in that: The pretreatment of the sample comprises: Clean the sample surface with a detergent to remove all non-metallic residues, and purge with dry gas to eliminate moisture left after cleaning; The treated samples are measured and recorded with an optical measuring instrument for the reflectivity DR of the sample surface, wherein the reflectivity DR is calculated by the formula (DR=Ir / Ii), where Ir represents the intensity of reflected light and Ii represents the intensity of incident light; Based on the obtained reflectivity DR value, the output intensity F of the light source E is adjusted to a preset threshold value G, and the adjustment process follows the formula F=G*DR.

3. The 3D hard gold processing detection method based on weighted calculation according to claim 2, characterized in that: The step of applying multi-angle light source illumination to the sample and recording the reflection spectrum characteristics under different illumination conditions includes: A group of light sources H are selected, each light source having a different wavelength WL1, WL2, ..., WLn, and the pretreated samples are sequentially irradiated; For each illumination condition at wavelength WLi, the corresponding reflected light intensity RIi is recorded to form a series of reflected light intensity data sets {RI1, RI2, ..., RIn}; Based on the obtained reflected light intensity data set, the reflectance spectrum characteristic value SPi at each wavelength WLi is calculated using the formula SPi=RIi / Ii; According to all the SPi values ​​obtained, a reflection spectrum curve C is constructed, and the reflection spectrum curve C is represented by connecting the points (SP1, WL1), (SP2, WL2), ..., (SPn, WLn).

4. The 3D hard gold processing detection method based on weighted calculation according to claim 3 is characterized in that: Based on the obtained reflection spectrum characteristics, an initial virtual map of the internal structure of the sample is constructed, and the initial virtual map covers all known physical properties, including: According to the recorded reflectance spectrum characteristic value SPi, combined with the geometric shape parameter GP of the sample surface, the initial mapping relationship M1 is established. The relationship is expressed by the formula M1=f(SPi,GP), where f is the process of constructing the mapping; Using the mapping relationship M1, the interior of the sample is preliminarily divided to determine multiple discrete regions VR1, VR2, ..., VRm; For each divided area Vri, analyze the corresponding reflectance spectrum characteristic value SPi, calculate the estimated value QAi of the physical property QA in the area, and use the formula QAi=g(SPi), where g is the process of converting to the estimated value of the physical property; All the QAi values ​​and the spatial distribution information of each region Vri are integrated to construct a three-dimensional virtual map M2 containing all known physical properties. The three-dimensional virtual map M2 consists of a series of point sets {(QAi, Vri)}.

5. The 3D hard gold processing detection method based on weighted calculation according to claim 4, characterized in that: The method of adjusting the density distribution of each point by introducing a weight factor based on the formed virtual mapping to form an optimized internal structure diagram includes: For each point Vri in the formed three-dimensional virtual map M2, the reliability of its corresponding physical attribute QAi is evaluated and an initial weight WF0 is assigned. The weight is calculated by the formula WF0=h(QAi), where h is the process of determining the weight; Based on the obtained initial weight WF0, the relative distance DLi,j between each point Vri is calculated, where DLi,j represents the distance between points Vri and Vrj, using the formula: DLi,j=sqrt((xri-xrj)^2+(yri-yrj)^2+(zri-zrj)^2), where (xri,yri,zri) are the coordinates of point Vri, and (xrj,yrj,zrj) are the coordinates of point Vrj; Combined with the calculated distance DLi,j, adjust the new weight WFi,j between each pair of points Vri and Vrj. The new weight value is determined by the formula WFi,j=WF0*exp(-α*DLi,j^2), where α is a predefined proportional coefficient; Using the updated weights WFi,j, the density distribution DDn of all points Vri is recalculated to form the optimized internal structure graph M3. The new density distribution DDn is calculated by the formula DDn(Vri)=Σ(WFi,j) / N, where N is the number of all points connected to the point Vri, and Σ represents the summation operation.

6. The 3D hard gold processing detection method based on weighted calculation according to claim 5, characterized in that: According to the generated structural map, several representative areas are selected and high-resolution scanning is performed to obtain detailed data, including: According to the optimized internal structure diagram M3, multiple candidate regions CR1, CR2, ..., CRk with significant physical property differences are identified and marked by comparing the property change rate ΔQA between adjacent points Vri, and the formula is ΔQA=|QAi+1-QAi| / DLi,i+1; Based on the determined candidate regions Cri, select standard regions SR1, SR2, ..., SRm that meet the preset attribute change threshold TL, and the selection of standard regions follows the condition TL≤ΔQA; For each selected standard area Sri, a high-resolution scanning path PLi is planned; the path design ensures that the entire area is covered while minimizing the scanning time, and the path length PLi is calculated by the formula PLi=Σ(DLj,j+1); According to the set path PLi, a detailed high-resolution scan is performed on each standard area Sri, and finer structural information FIi in the area is recorded to construct a higher-precision local map MIi.

7. The 3D hard gold processing detection method based on weighted calculation according to claim 6, characterized in that: The detailed data obtained is compared and analyzed with the original virtual map to identify differences and mark defect locations, including: The obtained detailed data set DF5 is combined with the original virtual map VMAP to generate a comparative analysis matrix BM; the comparative analysis matrix BM quantifies the data difference between the corresponding positions of the two into a value ΔDAi,j, and the formula is expressed as ΔDAi,j=|DAi,j(DF5)-DAi,j(VMAP)|; Based on the formed comparative analysis matrix BM, all coordinate points exceeding the preset difference threshold TL are identified to form a set of suspected outlier points PS; For all the marked suspected outlier points PSi, a local refinement algorithm is applied to calculate the neighborhood structure feature vector FVi of each point; the feature vector FVi is determined by the local pattern formed by the current point and its surrounding n closest points; Based on the obtained feature vector FVi, each suspected abnormal point PSi is deeply evaluated to decide whether it is finally marked as a defect location. The confidence score CScore is introduced into the evaluation process, and the calculation method is CScore(PSi)=h(FVi).

8. The 3D hard gold processing detection method based on weighted calculation according to claim 7, characterized in that: The method uses the marked defect position to perform local enhanced scanning, focuses on the suspected defect, and refines the image quality, including: Based on the marked suspected defect location set DS, determine the coordinate range RSi of each defect location DSi. The coordinate range RSi is defined by extending the defect point by a certain distance δ. The formula is RSi={P|d(P,DSi)≤δ}; For each determined coordinate range RSi, a high-precision local scanning path PLi is planned; the path design ensures that the entire area is covered while optimizing the scanning path length PLi; According to the set path PLi, an enhanced resolution scan is performed at each suspected defect to record detailed structural information FIi; the structural information FIi is used to construct a local refinement map MIi; The obtained local refinement map MIi is used to evaluate the image quality of each suspected defect position DSi. The evaluation process introduces the clarity score IScore, which is calculated as IScore(MIi)=h(Isharp,Inoise); Isharp is the sharpness index and Inoise is the noise level.

9. The 3D hard gold processing detection method based on weighted calculation according to claim 8, characterized in that: The quantitative evaluation of each defect area is performed to determine its size, shape and distribution characteristics, including: Based on the local refinement map MIi, the area AS of each suspected defect location DSi is calculated; the area is calculated by the formula AS(DSi)=Σ(Pij), where Pij represents the set of all pixels belonging to the defect area DSi; According to the obtained area AS(DSi), the morphological feature FShape of the suspected defect area is analyzed; the morphological feature FShape is determined by the defect boundary shape descriptor SD, and the expression is FShape(DSi)=g(SD); Combined with the morphological feature FShape, the spatial distribution characteristics DDist of each suspected defect area are evaluated; the spatial distribution characteristics are measured by analyzing the distance DLi,j between adjacent defects and the relative angle θi,j between them; The obtained spatial distribution characteristics DDist, together with the area AS and the morphological characteristics FShape, are used to comprehensively and quantitatively evaluate the overall attributes QTotal of each suspected defect area; the overall attributes are calculated by the formula QTotal(DSi)=w1*AS(DSi)+w2*FShape(DSi)+w3*DDist(DSi); w1, w2, w3 are weight factors.

10. The 3D hard gold processing detection method based on weighted calculation according to claim 9, characterized in that: The above comprehensive indicators are used to prepare a complete test report, including: According to the overall property QTotal(DSi) obtained by quantitative evaluation, the impact level GL is determined for each suspected defect location DSi; the impact level GL is calculated by the formula GL(DSi)=f(QTotal(DSi)); Based on the determined impact level GL, all suspected defects are classified and summarized to form a defect list LS; each entry in the list contains the defect location DSi, area AS(DSi), shape feature FShape(DSi) and spatial distribution characteristic DDist(DSi), as well as the corresponding impact level GL(DSi); Using the defect list LS, prepare a preliminary inspection report RDraft; the report structure follows the template TM, which defines how to organize and present defect information; Review the prepared preliminary report RDraft and add the comprehensive conclusion CC; the conclusion summarizes the number, location and impact of all internal minor defects, and proposes recommended measures SM; the final inspection report RFinal is composed of the formula RFinal=RDraft+CC+SM.

Images