Production quality evaluation method of gold bonding wire

Abstract

The invention discloses a production quality evaluation method for a gold bonding wire, and relates to the technical field of microelectronic packaging, and the method comprises the following steps: obtaining a surface image of the gold bonding wire through a high-resolution image collection system; performing preprocessing on the image data based on the acquired image to eliminate noise and enhance contrast; through the image data, extracting surface topography gradient features by using morphological operation, and quantifying microstructure changes of the surface of the gold wire; extracting texture energy characteristics through wavelet multi-scale decomposition, and analyzing distribution characteristics of surface textures from different scales; self-adaptive fusion is carried out on the morphological gradient features and the texture energy features, and the feature expression of the defect area is enhanced by dynamically adjusting the weight; dividing the boundary of a defect region by adopting an adaptive threshold segmentation method, and separating a normal region from an abnormal region; and extracting geometric and textural features of the defect region obtained by segmentation, realizing automatic discrimination of defect types through a classification model, and outputting a quality evaluation result.

Description

Technical Field

[0001] The invention belongs to the technical field of microelectronic packaging, and in particular relates to a production quality assessment method for gold bonding wires. Background Art

[0002] Gold bonding wire is a key material in the field of microelectronics packaging. Its surface quality directly affects the reliability and performance of the device. Currently, the industry generally adopts manual visual inspection combined with traditional image processing technology for quality assessment, but there are significant limitations. Manual inspection relies on operator experience, is inefficient and easily affected by subjective factors, and cannot meet the needs of high-precision and large-scale production. Traditional automated methods are mostly based on single feature extraction, and have insufficient sensitivity for detecting micron-level surface defects. In particular, they are susceptible to noise interference in complex texture backgrounds, resulting in high false detection and missed detection rates. Although wavelet transform has multi-resolution characteristics, it is difficult to take into account the coordinated representation of morphological gradient and energy distribution when used alone, resulting in insufficient utilization of feature information and insufficient accuracy in defect boundary positioning. Therefore, we propose a production quality assessment method for gold bonding wire. Summary of the Invention

[0003] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0004] The present invention provides a method for evaluating the production quality of a gold bonding wire, comprising the following steps:

[0005] Step S1: Acquire a surface image of the bonding wire using a high-resolution image acquisition system;

[0006] Step S2: Based on the acquired image, preprocess the image data to eliminate noise and enhance contrast;

[0007] Step S3: using morphological operations to extract surface morphology gradient features from the pre-processed image data and quantify the microstructural changes on the gold wire surface;

[0008] Step S4: Based on morphological analysis, texture energy features are extracted through wavelet multi-scale decomposition, and the distribution characteristics of surface texture are analyzed at different scales;

[0009] Step S5: Adaptively fuse the morphological gradient features and texture energy features, and enhance the feature representation of the defect area by dynamically adjusting the weights;

[0010] Step S6: Based on the fused feature map, an adaptive threshold segmentation method is used to divide the defect area boundary and separate the normal area from the abnormal area;

[0011] Step S7: Extract geometric and texture features from the segmented defect area, automatically identify the defect type through a classification model, and finally output the quality assessment result.

[0012] Furthermore, the step S1 includes the following steps:

[0013] Step S11: Capture micron-level surface defects by vertically installing a line scan camera 30 cm above the production line conveyor belt and setting the scanning resolution to 10 μm;

[0014] Step S12: A coaxial ring light source is configured with an incident angle of 15° and a light intensity of 8000 lux ± 5% to eliminate reflection interference and enhance texture contrast;

[0015] Step S13, the transmission speed of 50 mm / s is synchronously triggered to collect data, and a single frame image covers 50 mm of the gold wire length, so that the image is collected continuously.

[0016] Furthermore, step S2 includes the following steps:

[0017] Step S21: suppress image noise and retain edge details, and raw (x,y) is Gaussian filtered, and the filtering formula is as follows:

[0018]

[0019] Step S22: perform CLAHE contrast enhancement, set the grid partition to 8×8, and the contrast limit threshold to 2.0;

[0020] Step S23: automatically determine the segmentation threshold T using the Otsu algorithm, and extract the ROI region by binarization.

[0021] Furthermore, step S3 includes the following steps:

[0022] Step S31: Enhance the local contrast of the defect edge by using a 3×3 circular structure element B to perform an erosion operation. The calculation formula is as follows:

[0023]

[0024] Where E(x,y) is the pixel value at (x,y) after corrosion, B is a 3×3 circular structure element, To obtain the minimum pixel value in the neighborhood covered by the structural element B;

[0025] Step S32: highlight the morphological features of the defect area and perform a dilation operation. The calculation formula is as follows:

[0026]

[0027] Where D(x,y) is the pixel value of the expanded image at (x,y), To obtain the maximum value of pixels in the neighborhood covered by the structural element B;

[0028] Step S33: quantify the surface gradient change intensity and calculate the morphological gradient amplitude. The calculation formula is as follows:

[0029] Mgrad(x,y)=D(x,y)-E(x,y);

[0030] Where Mgrad(x,y) is the morphological gradient amplitude, which is used to reflect the intensity of local brightness change, and D(x,y)-E(x,y) is the difference between the dilation and erosion results, which represents the edge width.

[0031] Furthermore, step S4 includes the following steps:

[0032] Step S41: Capture texture features of different scales, use Daubechies4 wavelet basis to perform three-level decomposition on the image, and obtain horizontal, vertical and diagonal detail coefficients.

[0033] Step S42: Characterize the texture energy distribution and calculate the energy map of each scale. The calculation formula is as follows:

[0034]

[0035] Where, E k (x,y) is the value of the k-th scale wavelet energy map at (x,y), and are the wavelet detail coefficients in the horizontal, vertical and diagonal directions at the kth scale, To integrate the energy in three directions and characterize the local texture intensity;

[0036] Step S43: quantify the local texture complexity and calculate the energy variance σ in a 5×5 window. W (x,y), the calculation formula is as follows:

[0037]

[0038] Where, σ W (x,y) is the wavelet energy variance at coordinate (x,y), μ E E in the window k The mean of (x,y).

[0039] Furthermore, step S5 includes the following steps:

[0040] Step S51: Adaptively fuse the morphological gradient and texture features to propose a defect sensitivity weighting factor. The calculation formula is as follows:

[0041]

[0042] Where α(x,y) is the adaptive weighting factor, ranging from [0,1], ||M grad (x,y)||2 is the absolute value of the morphological gradient amplitude at the coordinate (x,y), max(||M grad ||) is the global maximum value of the morphological gradient amplitude in the entire image, max(σ W ) is the global maximum of the wavelet energy variance in the entire image;

[0043] Step S52: Generate a fusion feature map, combine the weighting factor with the normalized feature, and the formula is as follows:

[0044]

[0045] In the formula, the value of the feature map after F(x,y) fusion at (x,y) ranges from [0,1], is the normalized result of the morphological gradient amplitude, is the normalized result of the wavelet energy variance, and α is the weighting factor;

[0046] Step S53: To unify the feature dimensions, F(x,y) is normalized and mapped to the interval [0,1].

[0047] Furthermore, step S6 includes the following steps:

[0048] Step S61: Adopt the adaptive double threshold method to segment the defect area and distinguish between real defects and noise. The threshold is calculated as follows:

[0049] High threshold: T h =0.7μ F +0.3max(F);

[0050] Low threshold: T l =0.4μ F ;

[0051] Where μ F The mean value of the fused feature map F(x,y) in the defect area, max(F) is the maximum value in the fused feature map F(x,y);

[0052] Step S62: Remove pseudo-defects, perform connected domain analysis, and remove noise areas with an area less than 50 pixels;

[0053] Step S63: extract the defect geometric features and calculate the coordinates of the vertices of the defect outline polygon.

[0054] Furthermore, step S7 includes the following steps:

[0055] Step 71: Characterize the multi-dimensional characteristics of the defect and construct the defect feature vector. The formula is as follows:

[0056]

[0057] Where A is the pixel area of the defect area, σ F The standard deviation of the fused feature map F(x,y) within the defect area, E1 is the energy of the first-level wavelet decomposition, the finest scale, E2 is the energy of the second-level wavelet decomposition, and E3 is the energy of the third-level wavelet decomposition, the coarsest scale;

[0058] Step S72: Defect type identification is implemented, and a pre-trained XGBoost classification model is loaded for online reasoning.

[0059] Step S73: Output the inspection results and push the defect type code and confidence level to the quality management system.

[0060] The present invention has the following beneficial effects:

[0061] 1. This invention effectively improves the detection sensitivity of micron-level surface defects through a high-resolution image acquisition system and multi-scale morphological gradient analysis. Combined with wavelet multi-scale decomposition technology, it accurately extracts texture energy features in a complex texture background, overcoming the limitations of single feature extraction in traditional methods. The adaptive fusion algorithm dynamically adjusts feature weights, significantly enhancing the performance of defect areas and reducing false detection and missed detection rates caused by noise interference.

[0062] 2. The present invention adopts adaptive dual-threshold segmentation and connected domain analysis technology to accurately divide defect boundaries and eliminate pseudo-defect areas, solving the problem of blurred boundary positioning in traditional methods. Through the multi-dimensional representation of defect geometry and texture features, a high-discrimination feature vector is constructed, and combined with a pre-trained classification model, automatic identification of defect types is achieved, reducing manual intervention and improving detection efficiency and consistency.

[0063] 3. Integrate the collaborative analysis of morphological gradients and wavelet energy features to break through the bottleneck of insufficient utilization of single feature information, comprehensively cover the surface microstructure and texture distribution characteristics, and meet the needs of high-precision, large-scale production through full-process automated processing and online reasoning. It significantly reduces the subjective dependence of manual visual inspection and provides an efficient and reliable quality assessment solution for the field of microelectronics packaging.

[0064] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0066] Figure 1 The figure is a flow chart of a method for evaluating the production quality of gold bonding wire according to the present invention. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0068] See also Figure 1 As shown, the present invention is a method for evaluating the production quality of gold bonding wire, comprising the following steps:

[0069] Step S1: Acquire a surface image of the bonding wire using a high-resolution image acquisition system;

[0070] Step S2: Based on the acquired image, preprocess the image data to eliminate noise and enhance contrast;

[0071] Step S3: using morphological operations to extract surface morphology gradient features from the pre-processed image data and quantify the microstructural changes on the gold wire surface;

[0072] Step S4: Based on morphological analysis, texture energy features are extracted through wavelet multi-scale decomposition, and the distribution characteristics of surface texture are analyzed at different scales;

[0073] Step S5: Adaptively fuse the morphological gradient features and texture energy features, and enhance the feature representation of the defect area by dynamically adjusting the weights;

[0074] Step S6: Based on the fused feature map, an adaptive threshold segmentation method is used to divide the defect area boundary and separate the normal area from the abnormal area;

[0075] Step S7: Extract geometric and texture features from the segmented defect area, automatically identify the defect type through a classification model, and finally output the quality assessment result.

[0076] Step S1 includes the following steps:

[0077] Step S11: Capture micron-level surface defects by vertically installing a line scan camera 30 cm above the production line conveyor belt and setting the scanning resolution to 10 μm;

[0078] Step S12: A coaxial ring light source is configured with an incident angle of 15° and a light intensity of 8000 lux ± 5% to eliminate reflection interference and enhance texture contrast;

[0079] Step S13, the transmission speed of 50 mm / s is synchronously triggered to collect data, and a single frame image covers 50 mm of the gold wire length, so that the image is collected continuously.

[0080] Step S2 includes the following steps:

[0081] Step S21: suppress image noise and retain edge details, and raw (x,y) is Gaussian filtered, and the filtering formula is as follows:

[0082]

[0083] Step S22: perform CLAHE contrast enhancement, set the grid partition to 8×8, and the contrast limit threshold to 2.0;

[0084] Step S23: automatically determine the segmentation threshold T using the Otsu algorithm, and extract the ROI region by binarization.

[0085] Step S3 includes the following steps:

[0086] Step S31: Enhance the local contrast of the defect edge by using a 3×3 circular structure element B to perform an erosion operation. The calculation formula is as follows:

[0087]

[0088] Where E(x,y) is the pixel value at (x,y) after corrosion, B is a 3×3 circular structure element, To obtain the minimum pixel value in the neighborhood covered by the structural element B;

[0089] Step S32: highlight the morphological features of the defect area and perform a dilation operation. The calculation formula is as follows:

[0090]

[0091] Where D(x,y) is the pixel value of the expanded image at (x,y), To obtain the maximum value of pixels in the neighborhood covered by the structural element B;

[0092] Step S33: quantify the surface gradient change intensity and calculate the morphological gradient amplitude. The calculation formula is as follows:

[0093] Mgrad(x,y)=D(x,y)-E(x,y);

[0094] Where Mgrad(x,y) is the morphological gradient amplitude, which is used to reflect the intensity of local brightness change, and D(x,y)-E(x,y) is the difference between the dilation and erosion results, which represents the edge width.

[0095] Step S4 includes the following steps:

[0096] Step S41: Capture texture features of different scales, use Daubechies4 wavelet basis to perform three-level decomposition on the image, and obtain horizontal, vertical and diagonal detail coefficients.

[0097] Step S42: Characterize the texture energy distribution and calculate the energy map of each scale. The calculation formula is as follows:

[0098]

[0099] Where, E k (x,y) is the value of the k-th scale wavelet energy map at (x,y), and are the wavelet detail coefficients in the horizontal, vertical and diagonal directions at the kth scale, To integrate the energy in three directions and characterize the local texture intensity;

[0100] Step S43: quantify the local texture complexity and calculate the energy variance σ in a 5×5 window. W (x,y), the calculation formula is as follows:

[0101]

[0102] Where, σ W (x,y) is the wavelet energy variance at coordinate (x,y), μ E E in the window k The mean of (x,y).

[0103] Step S5 includes the following steps:

[0104] Step S51: Adaptively fuse the morphological gradient and texture features to propose a defect sensitivity weighting factor. The calculation formula is as follows:

[0105]

[0106] Where α(x,y) is the adaptive weighting factor, ranging from [0,1], ||M grad (x,y)||2 is the absolute value of the morphological gradient amplitude at the coordinate (x,y), max(||Mgrad ||) is the global maximum value of the morphological gradient amplitude in the entire image, max(σ W ) is the global maximum of the wavelet energy variance in the entire image;

[0107] Step S52: Generate a fusion feature map, combine the weighting factor with the normalized feature, and the formula is as follows:

[0108]

[0109] In the formula, the value of the feature map after F(x,y) fusion at (x,y) ranges from [0,1], is the normalized result of the morphological gradient amplitude, is the normalized result of the wavelet energy variance, and α is the weighting factor;

[0110] Step S53: To unify the feature dimensions, F(x,y) is normalized and mapped to the interval [0,1].

[0111] Step S6 includes the following steps:

[0112] Step S61: Adopt the adaptive double threshold method to segment the defect area and distinguish between real defects and noise. The threshold is calculated as follows:

[0113] High threshold: T h =0.7μ F +0.3max(F);

[0114] Low threshold: T l =0.4μ F ;

[0115] Where μ F The mean value of the fused feature map F(x,y) in the defect area, max(F) is the maximum value in the fused feature map F(x,y);

[0116] Step S62: Remove pseudo-defects, perform connected domain analysis, and remove noise areas with an area less than 50 pixels;

[0117] Step S63: extract the defect geometric features and calculate the coordinates of the vertices of the defect outline polygon.

[0118] Step S7 includes the following steps:

[0119] Step 71: Characterize the multi-dimensional characteristics of the defect and construct the defect feature vector. The formula is as follows:

[0120]

[0121] Where A is the pixel area of the defect area, σ FThe standard deviation of the fused feature map F(x,y) within the defect area, E1 is the energy of the first-level wavelet decomposition, the finest scale, E2 is the energy of the second-level wavelet decomposition, and E3 is the energy of the third-level wavelet decomposition, the coarsest scale;

[0122] Step S72: Defect type identification is implemented, and a pre-trained XGBoost classification model is loaded for online reasoning.

[0123] Step S73: Output the inspection results and push the defect type code and confidence level to the quality management system.

[0124] A specific application of this embodiment is:

[0125] Step S1: Use a line scan camera, model: XYZ-2000, vertically installed 30 cm above the conveyor belt, and set the scanning resolution to 10 μm;

[0126] Equipped with a coaxial ring light source, model: LightPro-8000, with an incident angle of 15° and light intensity stably controlled at 8000 lux ± 5%;

[0127] The conveyor belt speed was set to 50 mm / s, and the camera was triggered synchronously to continuously acquire images, with a single frame covering a length of 50 mm for the gold wire;

[0128] Step S2: For the original image I raw (x,y) performs Gaussian filtering to remove noise and retain edge details;

[0129] The CLAHE algorithm was used to enhance contrast, with the grid partition set to 8 × 8 and the contrast limit threshold set to 2.0;

[0130] The binarization threshold T is automatically determined by the Otsu algorithm to extract the ROI area on the gold wire surface;

[0131] Step S3: Use the 3×3 circular structure element B to perform corrosion and expansion operations on the preprocessed image and calculate the morphological gradient amplitude M grad (x,y)=D(x,y)-E(x,y);

[0132] The gradient amplitude quantifies the brightness change of the surface microstructure. If the gradient amplitude of a certain area is detected to be 120 (0-255 range), which is significantly higher than the average of 50 in the normal area, it is preliminarily determined to be a potential defect.

[0133] Step S4: Use Daubechies4 wavelet basis to perform three-level decomposition on the image and calculate the wavelet energy E in the horizontal, vertical and diagonal directions at each scale k (x,y);

[0134] Calculate the energy variance σ within a 5×5 windowW (x,y), when the energy variance of the defective area is 0.8, while the mean variance of the normal area is 0.2, it indicates that the texture complexity is significantly abnormal;

[0135] Step S5: Dynamically fuse the morphological gradient and texture features according to the defect sensitivity weighting factor α(x,y). For a certain defect area, α = 0.9, indicating that the morphological gradient is dominant; for another area, α = 0.3, focusing on texture variance;

[0136] Generate a normalized fusion feature map F(x,y)1 with a value range of [0,1];

[0137] Step S6: Adopting the adaptive double threshold method: μ F =0.6, max(F)=0.95, calculated T h =0.7×0.6+0.3×0.95=0.705, T l =0.4×0.6=0.24, segmenting the high confidence defect area;

[0138] Connected domain analysis removes pseudo defects with an area less than 50 pixels and retains the true defect outline;

[0139] Step S7: construct a defect feature vector V = [150, 0.65, 0.12, 3.2, 1.8];

[0140] The pre-trained XGBoost classification model identified it as “surface crack” with a confidence level of 96.7%;

[0141] The results are pushed to the quality management system, triggering an automatic sorting mechanism to remove unqualified gold wire.

[0142] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0143] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for evaluating the production quality of gold bonding wire, characterized by: The following steps are involved: Step S1: Acquire a surface image of the bonding wire using a high-resolution image acquisition system; Step S2: Based on the acquired image, preprocess the image data to eliminate noise and enhance contrast; Step S3: using morphological operations to extract surface morphology gradient features from the pre-processed image data and quantify the microstructural changes on the gold wire surface; Step S4: Based on morphological analysis, texture energy features are extracted through wavelet multi-scale decomposition, and the distribution characteristics of surface texture are analyzed at different scales; Step S5: Adaptively fuse the morphological gradient features and texture energy features, and enhance the feature representation of the defect area by dynamically adjusting the weights; Step S6: Based on the fused feature map, an adaptive threshold segmentation method is used to divide the defect area boundary and separate the normal area from the abnormal area; Step S7: Extract geometric and texture features from the segmented defect area, automatically identify the defect type through a classification model, and finally output the quality assessment result.

2. A method for evaluating the production quality of a gold bonding wire according to claim 1, characterized in that: The step S1 includes the following steps: Step S11: Capture micron-level surface defects by vertically installing a line scan camera 30 cm above the production line conveyor belt and setting the scanning resolution to 10 μm; Step S12: A coaxial ring light source is configured with an incident angle of 15° and a light intensity of 8000 lux ± 5% to eliminate reflection interference and enhance texture contrast; Step S13, the transmission speed of 50 mm / s is synchronously triggered to collect data, and a single frame image covers 50 mm of the gold wire length, so that the image is collected continuously.

3. The method for evaluating the production quality of a gold bonding wire according to claim 1, wherein: The step S2 includes the following steps: Step S21: Suppress image noise and retain edge details, and perform Gaussian filtering on the original image Iraw(x,y). The filtering formula is as follows: Step S22: perform CLAHE contrast enhancement, set the grid partition to 8×8, and the contrast limit threshold to 2.0; Step S23: automatically determine the segmentation threshold T using the Otsu algorithm, and extract the ROI region by binarization.

4. The method for evaluating the production quality of a gold bonding wire according to claim 1, wherein: The step S3 includes the following steps: Step S31: Enhance the local contrast of the defect edge by using a 3×3 circular structure element B to perform an erosion operation. The calculation formula is as follows: Where E(x,y) is the pixel value at (x,y) after corrosion, B is a 3×3 circular structure element, To obtain the minimum pixel value in the neighborhood covered by the structural element B; Step S32: highlight the morphological features of the defect area and perform a dilation operation. The calculation formula is as follows: Where D(x,y) is the pixel value of the expanded image at (x,y), To obtain the maximum value of pixels in the neighborhood covered by the structural element B; Step S33: quantify the surface gradient change intensity and calculate the morphological gradient amplitude. The calculation formula is as follows: Mgrad(x,y)=D(x,y)-E(x,y); Where Mgrad(x,y) is the morphological gradient amplitude, which is used to reflect the intensity of local brightness change, and D(x,y)-E(x,y) is the difference between the dilation and erosion results, which represents the edge width.

5. The method for evaluating the production quality of a gold bonding wire according to claim 1, wherein: The step S4 includes the following steps: Step S41: Capture texture features of different scales, use Daubechies4 wavelet basis to perform three-level decomposition on the image, and obtain horizontal, vertical and diagonal detail coefficients. Step S42: Characterize the texture energy distribution and calculate the energy map of each scale. The calculation formula is as follows: Where Ek(x,y) is the value of the k-th scale wavelet energy map at (x,y), and are the wavelet detail coefficients in the horizontal, vertical and diagonal directions at the kth scale, To integrate the energy in three directions and characterize the local texture intensity; Step S43: quantify the local texture complexity and calculate the energy variance σW(x,y) in a 5×5 window. The calculation formula is as follows: Where σW(x,y) is the wavelet energy variance at the coordinate (x,y), and μE is the mean of Ek(x,y) within the window.

6. The method for evaluating the production quality of a gold bonding wire according to claim 1, wherein: The step S5 includes the following steps: Step S51: Adaptively fuse the morphological gradient and texture features to propose a defect sensitivity weighting factor. The calculation formula is as follows: Where α(x,y) is an adaptive weighting factor with a value range of [0,1], ||Mgrad(x,y)||2 is the absolute value of the morphological gradient amplitude at the coordinate (x,y), max(||Mgrad||) is the global maximum value of the morphological gradient amplitude in the entire image, and max(σW) is the global maximum value of the wavelet energy variance in the entire image. Step S52: Generate a fusion feature map, combine the weighting factor with the normalized feature, and the formula is as follows: Where, the value of the feature map after F(x,y) fusion at (x,y) ranges from [0,1], is the normalized result of the morphological gradient amplitude, is the normalized result of the wavelet energy variance, and α is the weighting factor; Step S53: To unify the feature dimensions, F(x,y) is normalized and mapped to the interval [0,1].

7. The method for evaluating the production quality of a gold bonding wire according to claim 1, wherein: The step S6 includes the following steps: Step S61: Adopt the adaptive double threshold method to segment the defect area and distinguish between real defects and noise. The threshold is calculated as follows: High threshold: Th = 0.7 μF + 0.3 max (F); Low threshold: Tl = 0.4μF; Wherein, the mean value of the fused feature map F(x,y) in the μF defect area, max(F) is the maximum value in the fused feature map F(x,y); Step S62: Remove pseudo-defects, perform connected domain analysis, and remove noise areas with an area less than 50 pixels; Step S63: extract the defect geometric features and calculate the coordinates of the vertices of the defect outline polygon.

8. The method for evaluating the production quality of gold bonding wire according to claim 1, wherein: The step S7 includes the following steps: Step 71: Characterize the multi-dimensional characteristics of the defect and construct the defect feature vector. The formula is as follows: Where A is the pixel area of the defect area, σF is the standard deviation of the fused feature map F(x,y) in the defect area, E1 is the energy of the first-level wavelet decomposition, the finest scale, E2 is the energy of the second-level wavelet decomposition, and E3 is the energy of the third-level wavelet decomposition, the coarsest scale; Step S72: Defect type identification is implemented, and a pre-trained XGBoost classification model is loaded for online reasoning. Step S73: Output the inspection results and push the defect type code and confidence level to the quality management system.

Images