Plastic package surface defect detection system for semiconductor T2T process

Abstract

The present application relates to the technical field of plastic package processing, and discloses a plastic package surface defect detection system for a semiconductor T2T process, which collects a plastic package surface initial image, analyzes the surface type of the current plastic package, determines the scene, matches the illumination mode based on the analysis result of the surface characteristics, collects the plastic package surface to generate a complete panoramic image of the plastic package surface, calculates the geometric feature index, the gray scale feature index and the texture feature index, compares them with the corresponding threshold, determines whether the current plastic package is qualified, uploads the detection result to the T2T process MES system for storage, and sends a position signal to the manipulator for unqualified product rejection, breaking through the limitation of traditional detection without differentiated judgment of surface characteristics and contrast scene, completely solving the problem of most detection relying on a single conventional light source without adapting the illumination according to the surface characteristics, and finally realizing the dual improvement of defect detection precision and process automation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of plastic package processing, in particular to a plastic package surface defect detection system for a semiconductor T2T process. BACKGROUND

[0002] The semiconductor package post-test section T2T (Tray to Tray) is an automatic transfer and processing technology developed around the chip tray (Tray) in the package post section (EOL) process. The core is to realize the "tray-in-tray-out" continuous flow of chips between different production links and detection processes. The whole process relies on standardized tray carriers to complete the bearing, protection and transfer connection of the chips. Specifically, the process starts with the chip into the tray after molding. The chips that have completed the plastic packaging, cutting and molding processes are precisely placed into the anti-static special tray (ICTray) by automatic equipment. The groove size of the tray perfectly matches the package specifications of the chips, which can avoid physical damage and electrostatic damage during transportation. Then the tray with the chips will be transferred to the laser marking, electroplating, post-packaging test (FT) and other processes in turn. Each device automatically grabs the tray through a mechanical arm to complete the feeding and discharging without manual intervention in chip transfer. The tested chips will be re-arranged into the tray, and the unqualified products will be automatically sorted out. Finally, the tray loaded with qualified chips is stacked and packaged, directly meeting the customer's transportation and subsequent SMT assembly requirements.

[0003] The existing plastic package surface defect detection lacks targeted illumination schemes and simplifies the image preprocessing process in the defect development link, which makes it difficult to effectively highlight the defect features, directly affecting the subsequent recognition accuracy. Most detections rely on a single conventional light source without adapting to special lighting schemes according to the surface characteristics of the plastic package. SUMMARY

[0004] (I) Technical problems solved

[0005] To overcome the shortcomings of the prior art, the present application provides a plastic package surface defect detection system for a semiconductor T2T process, which breaks through the limitations of traditional detection without distinguishing between surface characteristics and contrast scenes. Different lighting schemes are matched for different surface types and contrast scenes, completely solving the problem of relying on a single conventional light source without adapting to the lighting according to the surface characteristics. At the same time, the defect detection module replaces the traditional simplified image preprocessing process, effectively highlighting various defects, and finally achieving the dual advantages of improving defect detection accuracy and process automation.

[0006] (II) Technical solutions

[0007] To achieve the above object, the application provides the following technical scheme: a plastic package surface defect detection system for a semiconductor T2T process, comprising a surface characteristic identification module, the surface characteristic identification module is used for collecting a plastic package surface initial image through a front auxiliary camera, and based on the image gray scale distribution and texture frequency of the initial image, analyzing whether the surface type of the current plastic package is glossy or textured, and determining whether it is a low-contrast defect scene;

[0008] An illumination mode matching module is established, which is used for matching the illumination mode based on the analysis and determination result of the surface characteristic identification module;

[0009] An image acquisition module is established, which is used for collecting the plastic package surface through a main detection camera to generate a complete plastic package surface panoramic image;

[0010] A defect detection module is established, which is used for calculating geometric feature indexes, gray scale feature indexes and texture feature indexes based on the plastic package surface panoramic image;

[0011] An analysis output module is established, which is used for comparing the geometric feature indexes, the gray scale feature indexes and the texture feature indexes with geometric feature index thresholds, gray scale feature index thresholds and texture feature index thresholds to determine whether the current plastic package is qualified, and uploading the detection result to a T2T process MES system for saving, and sending a position signal to a manipulator for unqualified product rejection for the unqualified plastic package.

[0012] Preferably, the working steps of the surface characteristic identification module are:

[0013] S1.1, the collected initial image is , wherein represents the number of high pixels, represents the number of width pixels, and the pixel gray scale value is ;

[0014]

[0015] S1.2, the gray scale standard deviation of the original image and the texture frequency density of the original image are calculated.

[0016] Preferably, the calculation formula of the gray scale standard deviation of the original image is:

[0017]

[0018]

[0019] In the formula, represents the gray scale standard deviation of the original image, Represents the average gray level;

[0020] when It was determined to be of the glossy type. This represents the lower limit threshold of the grayscale standard deviation;

[0021] when It was determined to be a textured type. This represents the upper limit threshold of the grayscale standard deviation;

[0022] when The results are verified based on the texture frequency density of the original image.

[0023] Preferably, the formula for calculating the texture frequency density of the original image is:

[0024]

[0025] In the formula, The texture frequency density represents the original image. Represents low-frequency energy. Represents high-frequency energy;

[0026] The formula for calculating low-frequency energy is:

[0027]

[0028] The formula for calculating high-frequency energy is:

[0029]

[0030] In the formula, Represents the image The frequency domain image obtained after performing a two-dimensional fast Fourier transform. Represents the low-frequency threshold. , Represents the high-frequency threshold. , Represents the frequency domain radius. ;

[0031] Frequency domain image The calculation formula is:

[0032]

[0033] In the formula, , , Represents frequency domain coordinates, Represents the imaginary unit;

[0034] when It was determined to be of the glossy type. a lower limit threshold of texture frequency density;

[0035] When , the pixel is determined as a textured type, an upper limit threshold of texture frequency density.

[0036] Preferably, the final determination rule of the surface type is:

[0037] a glossy type: ;

[0038] a textured type: .

[0039] Preferably, the determination method of the low-contrast defect scenario is:

[0040] S2.1, calculate the pixel gradient amplitude by using Sobel operator The calculation formula is:

[0041]

[0042] In the formula,

[0043] ;

[0044] ;

[0045] When , the pixel is a gray level abrupt change point, which constitutes a defect potential region , a gray level gradient amplitude threshold;

[0046] S2.2, for each pixel in the defect potential region , calculate the average gray level difference with the neighborhood background ;

[0047]

[0048] In the formula, represents the average gray level of the non-abrupt change points in the 3*3 neighborhood around the pixel The calculation formula is:

[0049]

[0050] In the formula, represents the number of abrupt change points in the neighborhood, represents an indicator function;

[0051] S2.3, calculate the proportion of low-contrast pixels The calculation formula is:

[0052]

[0053] In the formula, represents a low-contrast determination threshold, represents the total number of pixels in the defect potential area, represents the number of low-contrast pixels;

[0054] S2.4, when , it is determined as a low-contrast scene, otherwise as a normal-contrast scene, represents a determination threshold for the proportion of low-contrast pixels.

[0055] Preferably, the matching illumination mode is specifically:

[0056] For a glossy + normal-contrast scene: match the on-axis cold light source to irradiate at 0° vertical light path;

[0057] For a glossy + low-contrast scene: match the on-axis light source + ring-shaped supplementary light combination mode;

[0058] For a textured + normal-contrast scene: match the multi-directional strip-shaped diffuse reflection light source to form a uniform diffuse field in a 45° cross-irradiation manner;

[0059] For a textured + low-contrast scene: increase the polarization filter on the basis of the multi-directional strip-shaped diffuse reflection light source.

[0060] Preferably, the calculation formula of the geometric feature index is:

[0061]

[0062] In the formula, represents the geometric feature index, represents the area of the object, represents the area of the circumscribed rectangle;

[0063] Set the geometric feature index threshold interval, when the calculation result of the geometric feature index exceeds the geometric feature index threshold interval, the current plastic package is unqualified.

[0064] Preferably, the calculation formula of the gray-scale feature index is:

[0065]

[0066] In the formula, represents the gray-scale feature index, represents the average gray scale in the plastic package area, represents the gray scale standard deviation in the plastic package area;

[0067] Set the gray scale feature index threshold interval, when the calculation result of the gray scale feature index exceeds the gray scale feature index threshold interval, the current plastic package is unqualified.

[0068] Preferably, the formula for calculating the texture feature index is:

[0069]

[0070] In the formula, represents the texture feature index, represents the gray level co-occurrence matrix, , represents the gray value of the pixel; represents the probability of the pixel with a gray value of and the pixel with a gray value of appearing in a specific adjacent relationship; represents the square of the gray difference between the two pixels;

[0071] Set the texture feature index threshold interval, when the calculation result of the texture feature index exceeds the texture feature index threshold interval, the current plastic package is unqualified.

[0072] Compared with the prior art, the present application provides a plastic package surface defect detection system for a semiconductor T2T process, which has the following beneficial effects:

[0073] The present application first passes through the surface characteristic identification module, based on the dual quantization analysis of gray standard deviation and texture frequency density, accurately determines that the surface of the plastic package is glossy or textured, and combines Sobel operator gradient analysis and neighborhood gray difference calculation to identify low-contrast defect scenarios, breaking through the limitations of traditional detection without differentiated judgment of surface characteristics and contrast scenarios. For different surface types and contrast scenarios, the illumination mode matching module customizes adaptive solutions, completely solving the problem of relying on a single conventional light source for most detections and not adapting the illumination according to the surface characteristics. At the same time, the defect detection module extracts the geometric feature index, the gray scale feature index, and the texture feature index, multi-dimensionally quantifies the defect features, replaces the traditional simplified image preprocessing process, effectively highlights various defects, provides a solid foundation for subsequent accurate identification, and finally outputs the detection results by comparing with the threshold interval, links the MES system and the mechanical hand, and realizes the dual improvement of defect detection accuracy and process automation. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 The present application is a system schematic diagram. DETAILED DESCRIPTION

[0075] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0076] Please refer to Figure 1 , a plastic package surface defect detection system for semiconductor T2T process, comprising a surface characteristic identification module, the surface characteristic identification module is used for collecting the initial image of the plastic package surface through the front auxiliary camera, and based on the image gray scale distribution and texture frequency of the initial image, the surface type of the current plastic package is analyzed to be glossy type or textured type, and whether it is a low-contrast defect scene is determined, and the working steps are:

[0077] S1.1, set the collected initial image as , wherein represents the number of high pixels, represents the number of width pixels, and the pixel gray scale value is ;

[0078]

[0079] S1.2, calculate the gray scale standard deviation of the original image and the texture frequency density of the original image;

[0080] The calculation formula of the gray scale standard deviation of the original image is:

[0081]

[0082]

[0083] In the formula, represents the gray scale standard deviation of the original image, represents the gray scale mean value;

[0084] When , it is determined to be glossy type, represents the lower limit threshold of the gray scale standard deviation;

[0085] When , it is determined to be textured type, represents the upper limit threshold of the gray scale standard deviation;

[0086] When , the judgment result is verified according to the texture frequency density of the original image;

[0087] The mean and standard deviation of image grayscale are calculated using formulas. The standard deviation reflects the dispersion of image grayscale values. Glossy surfaces have small grayscale fluctuations due to uniform reflection, while textured surfaces have large grayscale fluctuations due to texture undulations. For the fuzzy range between the two, further verification is needed by combining texture frequency density.

[0088] The formula for calculating the texture frequency density of the original image is:

[0089]

[0090] In the formula, The texture frequency density represents the original image. Represents low-frequency energy. Represents high-frequency energy;

[0091] The formula for calculating low-frequency energy is:

[0092]

[0093] The formula for calculating high-frequency energy is:

[0094]

[0095] In the formula, Represents the image The frequency domain image obtained after performing a two-dimensional fast Fourier transform. Represents the low-frequency threshold. , Represents the high-frequency threshold. , Represents the frequency domain radius. ;

[0096] Frequency domain image The calculation formula is:

[0097]

[0098] In the formula, , , Represents frequency domain coordinates, Represents the imaginary unit;

[0099] when It was determined to be of the glossy type. This represents the lower limit threshold of texture frequency density;

[0100] when It was determined to be a textured type. This represents the upper limit threshold of texture frequency density;

[0101] The final rule for determining the surface type is:

[0102] Glossy type: ;

[0103] Textured type: ;

[0104] The initial image is subjected to two-dimensional fast Fourier transform to obtain a frequency domain image, the low frequency region and the high frequency region are divided by the frequency domain radius, the low frequency energy and the high frequency energy are calculated respectively, and then the texture frequency density is obtained, the lower the texture frequency density value represents the less obvious the image texture, the higher the texture frequency density value represents the more dense the texture, and finally the two-dimensional results are integrated through logical rules to ensure the accuracy of the type determination;

[0105] The determination method of the low contrast defect scene is:

[0106] S2.1, the Sobel operator is used to calculate the pixel gradient amplitude , and the calculation formula is:

[0107]

[0108] In the formula,

[0109] ;

[0110] ;

[0111] When , the pixel is a gray scale mutation point, which constitutes a defect potential area , represents the gray scale gradient amplitude threshold value;

[0112] S2.2, for each pixel in the defect potential area , calculate the average gray scale difference with the neighborhood background ;

[0113]

[0114] In the formula, represents the average gray scale of the non-mutation points in the 3*3 neighborhood around the pixel , and the calculation formula is:

[0115]

[0116] In the formula, represents the number of mutation points in the neighborhood, represents the indicator function;

[0117] S2.3, calculate the proportion of low contrast pixels , and the calculation formula is:

[0118]

[0119] In the formula, represents a low-contrast determination threshold, represents the total number of pixels in the defect potential area, represents the number of low-contrast pixels;

[0120] S2.4, when , it is determined as a low-contrast scene, otherwise as a normal-contrast scene, represents a determination threshold for the proportion of low-contrast pixels;

[0121] The Sobel operator is used to calculate the gradient amplitude of each pixel, and then the degree of gray scale mutation is quantified, the defect potential area is divided, and the area where the defect may exist is focused. Then, for each pixel in the defect potential area, the average gray scale difference with the non-mutation points in the 3*3 neighborhood is calculated, the gray scale contrast between the defect and the background is accurately quantified, and finally the proportion of low-contrast pixels is calculated to determine the low-contrast scene, improve the illumination mode pertinence, and avoid defect missed detection;

[0122] The surface characteristic recognition module determines the surface type and contrast state through quantitative analysis, considers the visual presentation difference of defects in different scenes, provides accurate input for the subsequent illumination mode matching module, avoids the problem of weakening of defect signal and enhancement of background noise caused by mismatching of illumination and surface characteristics, guarantees detection accuracy, and realizes comprehensive recognition of complex surface characteristics through multi-dimensional index determination, ensures that the subsequent detection process can respond to different scenes, and significantly reduces the missed detection rate and the false detection rate;

[0123] An illumination mode matching module is established, which is used to match the illumination mode based on the result determined by the surface characteristic recognition module, specifically:

[0124] For glossy + normal contrast scene: match coaxial cold light source (wavelength 500-550nm), irradiate with 0° vertical light path, use the matching of vertical incident light and surface reflection characteristics to maximize the suppression of specular reflection of glossy surface, and highlight the edge profile of geometric defects such as scratches and depressions;

[0125] For glossy + low contrast scene: use coaxial light source + ring light combination mode, the main coaxial light source brightness is improved to 1000-1500lux, and the ring light is irradiated at an angle of 30°. Through the superposition of double light paths, the gray scale difference between defects and background is enhanced, the problem of defect invisibility under low contrast is solved, and geometric defects can be effectively suppressed;

[0126] For scenes with texture + normal contrast: adapt multi-directional strip diffuse light source, form uniform diffuse field in 45° cross-illumination mode, weaken periodic gray fluctuation of normal texture, focus on texture mutation caused by defects;

[0127] For scenes with texture + low contrast: increase polarization filter on the basis of multi-directional strip diffuse light source, filter stray reflection light on the texture surface, adjust brightness to 1200-1800 lux, improve normal texture suppression rate, break through texture occlusion, develop defects such as pinhole and microcrack;

[0128] An image acquisition module is established, and the image acquisition module is configured to acquire the surface of the plastic package body through a main detection camera to generate a complete panoramic image of the surface of the plastic package body;

[0129] A defect detection module is established, and the defect detection module is configured to calculate a geometric feature index, a gray feature index and a texture feature index based on the panoramic image of the surface of the plastic package body;

[0130] The calculation formula of the geometric feature index is:

[0131]

[0132] In the formula, represents the geometric feature index, represents the area of the object, represents the area of the circumscribed rectangle;

[0133] The geometric feature index formula quantifies the geometric shape of the target area by the ratio of the area of the object to the area of the circumscribed rectangle, can accurately distinguish the regularity and integrity of the defects on the surface of the plastic package body, and effectively identify defects with abnormal geometric shape;

[0134] The calculation formula of the gray feature index is:

[0135]

[0136] In the formula, represents the gray feature index, represents the average gray value in the plastic package area, represents the gray standard deviation in the plastic package area;

[0137] The gray feature index formula constructs a quantitative index of the gray dimension by summing the average gray value and the gray standard deviation, can capture the overall gray level and local gray fluctuation of the surface of the plastic package body, and efficiently identify gray abnormal defects;

[0138] The calculation formula of the texture feature index is:

[0139]

[0140] In the formula, represents a texture feature index, represents a gray level co-occurrence matrix, , represents a gray value of a pixel; represents a pixel with a gray value of and a pixel with a gray value of occur in a specific adjacent relationship; represents a square of a gray difference between two pixels;

[0141] The texture feature index formula accurately quantifies the uniformity and regularity of the surface texture of the plastic package by weighted summation of the square of the pixel gray difference and the corresponding probability in the gray level co-occurrence matrix, and effectively identifies the texture abnormality class defects;

[0142] An analysis output module is established, which is used to compare the geometric feature index, the gray feature index, the texture feature index, and the geometric feature index threshold, the gray feature index threshold, and the texture feature index threshold. The geometric feature index threshold interval is set. When the calculation result of the geometric feature index exceeds the geometric feature index threshold interval, the current plastic package is unqualified. The gray feature index threshold interval is set. When the calculation result of the gray feature index exceeds the gray feature index threshold interval, the current plastic package is unqualified. The texture feature index threshold interval is set. When the calculation result of the texture feature index exceeds the texture feature index threshold interval, the current plastic package is unqualified.

[0143] The analysis output module uploads the detection result to the T2T process MES system for saving, and sends a position signal to the manipulator for unqualified product rejection for the unqualified plastic package;

[0144] The analysis output module realizes accurate defect determination through multi-feature threshold interval comparison. The preset threshold interval of the geometric feature index, the gray feature index, and the texture feature index is that exceeding the interval is the clear rule for unqualified. This avoids subjective errors in manual determination, ensures that the defect determination standard of each plastic package is unified and the result is reliable, controls product quality from the source, realizes standardized management and traceability of detection data, uploads the detection result to the T2T process MES system, completes real-time archiving of data, and provides searchable and traceable data support for subsequent quality analysis. In addition, a position signal is sent to the manipulator in real time for the unqualified plastic package, realizing automatic and accurate rejection of unqualified products without human intervention, improving production efficiency, avoiding unqualified products flowing into the next link, ensuring the continuity and stability of the T2T process, and finally realizing multiple goals from quality control to data traceability to production efficiency improvement.

[0145] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A system for detecting surface defects in molded packages used in semiconductor T2T processes, characterized in that, This includes establishing a surface characteristic recognition module, which is used to acquire an initial image of the surface of the encapsulated body through a front-facing auxiliary camera, and analyze the surface type of the current encapsulated body based on the image grayscale distribution and texture frequency of the initial image to determine whether it is a glossy or textured surface, and at the same time determine whether it is a low-contrast defect scene. A lighting pattern matching module is established, which is used to match lighting patterns based on the analysis and judgment results of the surface characteristic identification module. The specific matching lighting mode is as follows: For scenes with glossy surfaces and normal contrast: Match a coaxial cold light source and illuminate with a 0° vertical light path; For scenes with glossy surfaces and low contrast: Match the coaxial light source + ring fill light combination mode; For textured scenes with normal contrast: Match multi-directional strip diffuse light sources to form a uniform diffuse field with 45° cross illumination. For textured and low-contrast scenes: add a polarizing filter to the multi-directional strip diffuse light source; An image acquisition module is established, which is used to acquire a full-coverage image of the molded body surface through the main detection camera, and generate a complete panoramic image of the molded body surface. A defect detection module is established, which is used to calculate geometric feature index, grayscale feature index, and texture feature index based on a panoramic image of the molded body surface. An analysis output module is established. This module is used to compare the geometric feature index, grayscale feature index, and texture feature index with the threshold values ​​of the geometric feature index, grayscale feature index, and texture feature index to determine whether the current encapsulated body is qualified. The detection results are then uploaded to the T2T process MES system for storage. Simultaneously, for unqualified encapsulated bodies, a position signal is sent to the robot arm to remove the unqualified products.

2. The surface defect detection system for molding compounds in semiconductor T2T processes according to claim 1, characterized in that, The working steps of the surface characteristic identification module are as follows: S1.1, Let the initial image acquired be... ,in Represents the height in pixels. Represents the width in pixels, and the pixel grayscale value is... ; S1.2 Calculate the grayscale standard deviation and texture frequency density of the original image.

3. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 2, characterized in that: The formula for calculating the grayscale standard deviation of the original image is: In the formula, The grayscale standard deviation of the original image. Represents the average gray level; when It was determined to be of the glossy type. This represents the lower limit threshold of the grayscale standard deviation; when It was determined to be a textured type. This represents the upper limit threshold of the grayscale standard deviation; when The results are verified based on the texture frequency density of the original image.

4. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 2, characterized in that: The formula for calculating the texture frequency density of the original image is: In the formula, The texture frequency density represents the original image. Represents low-frequency energy. Represents high-frequency energy; The formula for calculating low-frequency energy is: The formula for calculating high-frequency energy is: In the formula, Represents the image The frequency domain image obtained after performing a two-dimensional fast Fourier transform. Represents the low-frequency threshold. , Represents the high-frequency threshold. , Represents the frequency domain radius. ; Frequency domain image The calculation formula is: In the formula, , , Represents frequency domain coordinates, Represents the imaginary unit; when It was determined to be of the glossy type. This represents the lower limit threshold of texture frequency density; when It was determined to be a textured type. This represents the upper limit threshold of texture frequency density.

5. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 4, characterized in that: The final rule for determining the surface type is as follows: Glossy type: ; Textured type: .

6. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 5, characterized in that: The method for determining the low-contrast defect scene is as follows: S2.1 Calculate pixel gradient magnitude using the Sobel operator. The calculation formula is: In the formula: ； ； when Then this pixel is a grayscale abrupt change point, constituting a potential defect region. , This represents the threshold value for the grayscale gradient. S2.2, For potential defect areas For each pixel, calculate the average grayscale difference between it and the surrounding background. ; In the formula, Representing pixels The average gray value of non-mutation points within a 3×3 neighborhood is calculated using the following formula: In the formula, Represents the number of mutation points in the neighborhood. Represents an indicator function; S2.3 Calculate the proportion of low-contrast pixels The calculation formula is: In the formula, This represents the threshold for determining low contrast. This represents the total number of pixels in the potential defect area. This represents the number of low-contrast pixels; S2.4, when If the image is not in a low-contrast scene, it is considered a low-contrast scene; otherwise, it is considered a normal-contrast scene. The threshold representing the proportion of low-contrast pixels.

7. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 1, characterized in that, The formula for calculating the geometric feature index is as follows: In the formula, Represents geometric characteristic index, Represents the area of ​​an object. Represents the area of ​​the circumscribed rectangle; A threshold range for the geometric feature index is set. If the calculated result of the geometric feature index exceeds the threshold range, the current encapsulated body is considered unqualified.

8. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 1, characterized in that, The formula for calculating the grayscale feature index is as follows: In the formula, Represents the grayscale feature index. This represents the average gray level within the encapsulated area. This represents the standard deviation of grayscale within the encapsulated area; Set a grayscale feature index threshold range. If the calculated result of the grayscale feature index exceeds the grayscale feature index threshold range, the current encapsulated body is unqualified.

9. The surface defect detection system for a molding compound in a semiconductor T2T process according to claim 1, characterized in that, The formula for calculating the texture feature index is as follows: In the formula, Represents the texture feature index. Represents the gray-level co-occurrence matrix. , The grayscale value representing a pixel; The grayscale value represents The pixel and grayscale values The probability of a pixel appearing in a specific adjacent relationship; Represents the square of the grayscale difference between two pixels; Set a threshold range for the texture feature index. If the calculated result of the texture feature index exceeds the threshold range, the current encapsulated body is unqualified.

Images