CN113920096A - Method for detecting metal packaging defects of integrated circuit - Google Patents
Method for detecting metal packaging defects of integrated circuit Download PDFInfo
- Publication number
- CN113920096A CN113920096A CN202111198584.4A CN202111198584A CN113920096A CN 113920096 A CN113920096 A CN 113920096A CN 202111198584 A CN202111198584 A CN 202111198584A CN 113920096 A CN113920096 A CN 113920096A
- Authority
- CN
- China
- Prior art keywords
- image
- defect
- network
- difference
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 116
- 239000002184 metal Substances 0.000 title claims abstract description 77
- 238000004806 packaging method and process Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000003044 adaptive effect Effects 0.000 claims abstract description 34
- 238000001514 detection method Methods 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 29
- 230000011218 segmentation Effects 0.000 claims abstract description 16
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000005457 optimization Methods 0.000 claims abstract description 6
- 238000012805 post-processing Methods 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000035945 sensitivity Effects 0.000 claims description 25
- 230000002159 abnormal effect Effects 0.000 claims description 23
- 238000010606 normalization Methods 0.000 claims description 23
- 238000011156 evaluation Methods 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 15
- 230000014509 gene expression Effects 0.000 claims description 15
- 238000007689 inspection Methods 0.000 claims description 14
- 238000010586 diagram Methods 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 239000000284 extract Substances 0.000 claims description 6
- 230000002950 deficient Effects 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 5
- 238000005286 illumination Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 230000008030 elimination Effects 0.000 claims description 3
- 238000003379 elimination reaction Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000003466 welding Methods 0.000 description 8
- 238000009826 distribution Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 238000002372 labelling Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012536 packaging technology Methods 0.000 description 2
- 238000007789 sealing Methods 0.000 description 2
- 238000013256 Gubra-Amylin NASH model Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005538 encapsulation Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000017525 heat dissipation Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000007769 metal material Substances 0.000 description 1
- 239000005022 packaging material Substances 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 239000004033 plastic Substances 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 229910000679 solder Inorganic materials 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30148—Semiconductor; IC; Wafer
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for detecting integrated circuit metal packaging defects, which comprises the following steps: s1: acquiring an integrated circuit metal packaging image, carrying out image preprocessing, and dividing the preprocessed image into a training set and a test set; s2: constructing a self-adaptive depth generation countermeasure network and initializing weight parameters of the network; s3: performing iterative optimization on the adaptive depth generation countermeasure network according to the loss function; s4: counting difference graphs corresponding to the training set grid samples to construct an average feature graph; s5: and detecting the metal packaging of the integrated circuit by combining the trained adaptive depth generation countermeasure network and a preset image post-processing strategy. The invention solves the problem that the defect contour characteristics lose key information or interference points due to threshold segmentation in the existing detection method, and improves the detection accuracy.
Description
Technical Field
The invention relates to the technical field of integrated circuit metal packaging detection, in particular to a method for detecting defects of integrated circuit metal packaging.
Background
The integrated circuit manufacturing industry is a very important industry in the world today, and most devices exist around integrated circuits in daily life, and become an indispensable part. The integrated circuit must be packaged for use in the system. Integrated circuit packages have environmental protection, thermal management, mechanical stability, and electrical connections. Environmental protection prevents chemical damage to the performance of integrated circuits, and mechanical stability and thermal management generally ensure the reliability and useful life of integrated circuits by preventing physical damage to the surface. Common packaging materials for integrated circuits are metal, glass, plastic or ceramic, which are used to separate the inside of the integrated circuit from the outside of the environment and provide sealing performance, such as no moisture penetration, corrosion inside and outside of the package, or the like, which may reduce or even damage the performance of the integrated circuit during use. In summary, the packaging of integrated circuits plays an important role in circuit support, telecommunication, heat dissipation, sealing, and chemical protection. The metal package is considered to be a full-sealed package form due to the material property, and is widely applied to the military and civil fields of aerospace, aviation, navigation, radar, communication and the like. In the process of packaging and manufacturing the integrated circuit function by the metal material, friction exists between the surface and the related contact surface to different degrees due to the action of pressure, and various defects such as scratches, stains, bubbles, stains and the like can be formed under the influence of a plurality of external environmental factors, while the performance of the integrated circuit can be reduced or even the function of the integrated circuit can be failed due to the existence of any packaging defect, which seriously affects the service performance and the service life of the integrated circuit. Therefore, the detection of the metal package of the integrated circuit is particularly important, whether the defect exists or not is detected, and whether the quality requirement is met or not is detected, so that the quality of the metal package and the reliability of the integrated circuit are ensured. In the early stage, the metal packaging quality is ensured by visual inspection by experienced quality inspectors, but the traditional manual detection has the defects of low precision, poor real-time performance, strong subjectivity, high labor intensity, high labor cost and the like. With the development of miniaturization of integrated circuits, the requirements for packaging technology are higher and higher, and the difficulty of packaging technology is higher and more complex, so that in order to prevent potential loss and potential safety hazards, the requirements for more advanced, reliable and robust detection methods in the task of automatic detection of the metal packaging surface of the integrated circuit are higher and higher.
Metal package defect detection currently faces a number of difficulties and challenges. On the one hand, most current methods rely on negative examples and manual labeling. However, in an actual industrial scene, the number of defect images is small, the defect images are not easy to collect, the problem of unbalance of positive and negative samples exists, and a large amount of manpower and material resources are consumed for manual labeling. On the other hand, because the defect characteristics of the metal packaging surface are very weak and have low contrast, the metal packaging surface is extremely similar to qualified background pixels, and has many types of metal packaging defects, the representation diversity is more urgent to the generation quality of the template, and the template generated to be closer to the background pixels can better extract various defect types; and some metal packaging images have the phenomenon of uneven illumination in some areas due to the reflection influence of light, different illumination brightness backgrounds put higher requirements on threshold segmentation defects, the traditional threshold method can cause the problems that key information is lost in the defect outline characteristics or interference points appear, and the like, and the difficulty is brought to defect detection.
In the prior art, the publication numbers are: CN110880175A, Chinese invention patent in 3/13/2020 discloses a solder joint defect detection method, system and device. The invention provides an IC welding spot detection method based on a countermeasure generation network, which can adaptively extract the position of a welding spot region of a welding spot image according to a complete welding spot image through a joint supervision network and an unsupervised network, and generate a local countermeasure generation template at the welding spot position based on the appearance characteristics of the welding spot. And finally, using the trained generation network G and the region-of-interest classification network CNN for the welding spot image detection process. The surface of an image of integrated circuit metal packaging does not have texture features with clear and regular welding spot outlines, so that the generation of a qualified template is challenged, the fine texture features of the metal packaging are continuously reduced or lose fine textures in the feature extraction process, the qualified background generated by the template is only the appearance of the qualified background, the ideal template generation effect cannot be achieved, the problems of random abnormal interference, background noise point interference and the like of a difference image between the template image and an input image are caused, and the misjudgment and the missing judgment of a detection result are caused.
Disclosure of Invention
The invention provides a method for detecting the metal package defects of the integrated circuit, aiming at solving the problems that the prior art metal package detection causes the loss of key information or interference points of defect outline characteristics due to threshold segmentation, easily causes the loss of real defective pixels when processing background noise, has low detection accuracy and the like.
The primary objective of the present invention is to solve the above technical problems, and the technical solution of the present invention is as follows:
a method for detecting integrated circuit metal packaging defects comprises the following steps:
s1: acquiring and preprocessing an integrated circuit metal packaging image, and dividing the preprocessed image into a training set and a test set;
s2: constructing a self-adaptive depth generation countermeasure network and initializing weight parameters of the network;
s3: performing iterative optimization on the adaptive depth generation countermeasure network according to the loss function;
s4: counting difference graphs corresponding to the training set grid samples to construct an average feature graph;
s5: and detecting the metal packaging of the integrated circuit by combining the trained adaptive depth generation countermeasure network and a preset image post-processing strategy.
Further, the specific process of step S1 is:
s101: acquiring an integrated circuit metal packaging image, positioning an ROI (region of interest) of the image, and cutting out an ROI image;
s102: manually dividing the cut ROI area image into a qualified sample image and an unqualified sample image;
s103: randomly selecting a plurality of qualified sample images as a training set, using the remaining qualified sample images and the remaining unqualified sample images as a test set, and simultaneously performing pixel normalization on the images in the training set and the test set before inputting the images into a network to enable the pixel value to be in a preset range.
Further, the adaptive deep-generation countermeasure network includes: the encoder comprises nine network layers, wherein the first seven network layers comprise a convolutional layer, a batch normalization layer and a LeaklyRelu activation function, the eighth network layer comprises a hole convolutional layer, a batch normalization layer and a LeaklyRelu activation function, and the ninth network layer comprises a 1x1 convolutional layer; the decoder is provided with seven network layers, wherein the first six layers all comprise an deconvolution layer, batch normalization and a LeaklyRelu activation function, and the seventh network layer adopts the deconvolution layer, the batch normalization and a tanh activation function; the discriminator comprises eight network layers, the first seven network layers comprise a convolutional layer, a batch normalization layer, a LeaklyReLU activation function and a maximum pooling layer, and the eighth network layer is a 1x1 convolutional layer and is used for dimension reduction output; the encoder extracts the high-dimensional features of the input image, and then the qualified template of the deep image features is reconstructed from the high-dimensional features through deconvolution of the decoder.
Further, in step S3, the adaptive depth generation countermeasure network is iteratively optimized according to a loss function, where the network loss function includes: a generator loss function and a discriminator loss function, wherein the generator loss function comprises: content loss lcon=||x-G(x)||1To combat the loss ladv=||D(7)(x)-D(7)(G(x))||1Characteristic loss lenc=||En1(x)-En2(G(x))||1Generator loss function LGDiscriminator loss function LDThe expressions are respectively:
LG=wcon*lcon+wenc*lenc+wadv*ladv (1)
wherein wcon,wenc、wadvThe parameter is a hyper-parameter, which adjusts the influence of each loss on the overall loss function, D (x) represents the probability of judging whether the network judges the real picture as real, and D (G (x)) is the probability of judging whether the picture generated by G is the real picture or not.
Further, the specific process of step S4 is:
inputting the images in the training set into the trained model, and outputting corresponding template images;
performing difference operation on the output template image and the corresponding input image to obtain a difference image set of the training set;
accumulating and averaging the difference images to obtain an average characteristic diagram, wherein a calculation formula of the average characteristic diagram is as follows:
wherein d isn(i, j, k) is the pixel value of the difference image, and each constructed qualified sample is superposed and averaged to obtain a pixel which can describe random noise possibly occurring in the generation process, namely an abnormal pixel point;
and performing reciprocal normalization operation on the average feature map, and applying the average feature map subjected to the normalization operation to the elimination of random abnormal values of the detection image difference map in the later test set.
Further, the specific process of step S5 is:
s501: normalizing the metal packaging image to be detected, inputting the normalized image into a trained adaptive depth generation countermeasure network, and outputting a template image;
s502: carrying out differential operation on the metal packaging image to be detected and the output model image to obtain a difference image;
s503: eliminating random abnormal values in the difference value graph by using the average characteristic graph to obtain a difference value graph with the abnormal values removed;
s504: extracting the defect contour of each difference image processed by the average characteristic image by using a self-adaptive threshold value to obtain a defect contour image;
s505: and further obtaining a defect segmentation graph and a defect evaluation score which meet the quality inspection sensitivity requirement by using a local defect verification strategy for the defect contour image with the background noise.
Further, the template image in step S501 is a template image similar to the background appearance feature of the metal package image to be detected, but does not contain defective pixels.
Further, according to the difference image pixel distribution of the qualified integrated circuit metal packaging and defect samples and the mean value and variance conditions thereof, selecting a coefficient in front of the mean value with stable change and setting a super parameter to macroscopically regulate and control a threshold value, so that defect extraction is carried out on each difference image by self-adapting the threshold value, and the self-adapting threshold expression is as follows:
wherein the content of the first and second substances,indicating a difference map threshold after removal of the outlier,representing the mean of the difference map after removal of the outliers,the variance of the difference image after removing the abnormal value, alpha is a hyper-parameter and is lightened according to different lightsAnd (4) a control constant determined by the range of the difference image pixel values corresponding to the qualified samples in the training set.
Further, a defect segmentation graph and a defect evaluation score meeting the quality inspection sensitivity requirement are obtained for the defect contour image with background noise by using a local defect verification strategy, and the specific process is as follows:
recording the result image obtained by the adaptive threshold method as a setWherein b isnThe image processing method comprises the steps of obtaining a difference image corresponding to an nth output image to be detected through an average characteristic image and a result image after adaptive threshold processing, wherein the result image is a large-scale image, and the scale of the result image is H multiplied by W;
inputting a large-scale result graph, and sliding by using the size of a sliding window as f multiplied by f and the step length as stride to obtain Row multiplied by Col local image blocks in the large-scale result graph, wherein Wherein pad is the padding number;
the set of partial images after sliding window can be described as
Respectively calculating the defect probability of each local image, wherein the calculation formula is as follows:
whereinThe defect probability value corresponding to the local image of the nth input large-scale image at the row and column positions is shown, the size of the local image is the sliding window size f multiplied by f, wherein the number of channels is C,outputting local images of a row number and a col column number corresponding to the image for the self-adaptive threshold value;
setting the sensitivity requirement of quality inspection as tau, carrying out probability analysis on the defect probability calculated by each local image, and recording the probability within the sensitivity as a numerical value of 0; otherwise, the probability except the sensitivity is recorded as a numerical value 1; the local defect verification is marked as DVP, and the mathematical expression is as follows:
whereinThe mark values corresponding to the local images at the row and column positions of the nth image are represented, wherein the sensitivity calculation formula isMarking the mark lower than the quality inspection sensitivity tau as 0, otherwise, marking the mark higher than the sensitivity tau as 1 to indicate that the local block is not in the acceptable range of quality inspection, marking the mark through 0 or 1 of the image block which identifies the block to obtain a mask image which identifies the defect, wherein the mask image represents a mask image of the position of the defect occurrence region, and performing dot-product fusion operation on the mask image and the input image in the fifth step to obtain a defect segmentation image;
and accumulating the mark values of all local images in the result graph to be used as an image defect evaluation score for evaluating the large-scale result graph, wherein the expression is as follows:
wherein ESnAnd expressing the defect evaluation score of the nth image to be detected.
Further, comparing the obtained defect evaluation score with a preset threshold value to obtain a final detection value result, wherein the expression is as follows:
and th is the maximum defect score threshold value for all qualified samples in the training set to be qualified, if the defect evaluation score of the to-be-detected metal packaging image is greater than or equal to the threshold value, the corresponding metal packaging is judged to be unqualified, and if the defect evaluation score of the to-be-detected metal packaging image is less than the threshold value, the corresponding metal packaging is judged to be qualified.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
according to the invention, the image defect area can be extracted according to the complete metal packaging image through the adaptive depth generation countermeasure network, the countermeasure generation template with qualified data distribution similar to the background appearance of the input metal packaging image is generated in a self-adaptive mode, and meanwhile, the preset image post-processing strategy is combined, so that the problem that the defect outline characteristics lose key information or interference points due to threshold segmentation in the existing detection method is solved, the background noise in the characteristic diagram is eliminated, the problem that the real defect pixels are lost while the background noise is processed in the traditional method is solved, and the detection accuracy is improved.
Drawings
FIG. 1 is a flow chart of a method for detecting defects in a metal package of an integrated circuit according to the present invention.
FIG. 2 is a comparison graph of the processing effect of the average feature map according to the embodiment of the present invention.
FIG. 3 is a schematic diagram of a qualified integrated circuit metal package image defect detection according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of detecting defects in an IC metal package image that fails to meet the requirements of the embodiments of the present invention.
Detailed Description
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described herein, and therefore the scope of the present invention is not limited by the specific embodiments disclosed below.
Example 1
As shown in fig. 1, a method for detecting defects of an integrated circuit metal package includes the following steps:
s1: acquiring and preprocessing an integrated circuit metal packaging image, and dividing the preprocessed image into a training set and a test set;
it should be noted that, the acquired metal package image of the integrated circuit is raw data, and the raw data needs to be preprocessed, and the specific processing process is as follows:
s101: acquiring an integrated circuit metal packaging image, positioning an ROI (region of interest) of the image, and cutting out an ROI image;
s102: manually dividing the cut ROI area image into a qualified sample image and an unqualified sample image;
s103: randomly selecting a plurality of qualified sample images as a training set, using the remaining qualified sample images and the remaining unqualified sample images as a test set, and simultaneously performing pixel normalization on the images in the training set and the test set before inputting the images into a network to enable the pixel value to be in a preset range.
It should be noted that in step S103, the image pixels are normalized so that the image pixel values are between [ -1,1 ].
S2: constructing a self-adaptive depth generation countermeasure network and initializing weight parameters of the network;
it should be noted that, in the present invention, a countermeasure network is generated by constructing an adaptive depth, and an adaptive template image similar to the background appearance of an input image is output by using a trained network, where the adaptive depth generation countermeasure network includes: the encoder comprises nine network layers, wherein the first seven network layers comprise a convolutional layer, a batch normalization layer and a LeaklyRelu activation function, the eighth network layer comprises a hole convolutional layer, a batch normalization layer and a LeaklyRelu activation function, and the ninth network layer comprises a 1x1 convolutional layer; the decoder is provided with seven network layers, wherein the first six layers all comprise an deconvolution layer, batch normalization and a LeaklyRelu activation function, and the seventh network layer adopts the deconvolution layer, the batch normalization and a tanh activation function; the discriminator comprises eight network layers, the first seven network layers comprise a convolutional layer, a batch normalization layer, a LeaklyReLU activation function and a maximum pooling layer, and the eighth network layer is a 1x1 convolutional layer and is used for dimension reduction output; the encoder extracts the high-dimensional features of the input image, and then the qualified template of the deep image features is reconstructed from the high-dimensional features through deconvolution of the decoder.
S3: performing iterative optimization on the adaptive depth generation countermeasure network according to the loss function;
it should be noted that, an adaptive deep generation countermeasure network is constructed, and iterative optimization needs to be performed on the adaptive deep generation countermeasure network according to a network loss function, so as to obtain a trained network model, where the network loss function includes: a generator loss function and a discriminator loss function, wherein the generator loss function comprises: content loss lcon=||x-G(x)||1To combat the loss ladv=||D(7)(x)-D(7)(G(x))||1Characteristic loss lenc=||En1(x)-En2(G(x))||1(ii) a Wherein, the content loss is obtained by calculating L1 norm of the input image and the output image, the constraint output image and the actual sample tend to be equal in the low-dimensional feature distribution, and the content loss function is Lcon=||x-G(x)||1. The countermeasure loss is a generated template image which restricts the similar appearance of the generated template and the input image, and the training sample and the output image are input into the judgment moduleAnd D, extracting the seventh-layer high-dimensional features of the discriminator D, and obtaining the antagonism loss by calculating the L1 distance between the high-dimensional features. The resistance loss not only reduces the instability of the GAN model training, but also limits the generator from producing images that match the distribution of the training sample features, thereby ensuring that the generated images are qualified images. The formula of the antagonistic loss is ladv=||D(7)(x)-D(7)(G(x))||1. The feature loss calculates the L1 norm between the high-dimensional features of the encoder En1 and the encoder En2, with a feature loss of Lenc=||En1(x)-En2(G(x))||1The constraint output template is equal to the actual input sample in the high-dimensional feature.
Generator loss function LGDiscriminator loss function LDThe expressions are respectively:
LG=wcon*lcon+wenc*lenc+wadv*ladv (1)
wherein wcon,wenc、wadvThe parameter is a hyper-parameter, which adjusts the influence of each loss on the overall loss function, D (x) represents the probability of judging whether the network judges the real picture as real, and D (G (x)) is the probability of judging whether the picture generated by G is the real picture or not.
S4: counting difference graphs corresponding to the training set grid samples to construct an average feature graph;
the specific steps for constructing the average characteristic graph are as follows:
will train the setThe images in (1) are input into a trained model, corresponding template images are output, and the set of the template images is recorded with sound
Performing difference operation on the output template image and the corresponding input image to obtain a difference image set of the training set
Accumulating and averaging the difference images to obtain an average characteristic diagram, wherein a calculation formula of the average characteristic diagram is as follows:
wherein d isnAnd (i, j, k) is the pixel value of the difference image, and each constructed qualified sample is superposed and averaged to obtain a pixel which can describe random noise possibly occurring in the generation process, namely an abnormal pixel point. In order to reduce the influence of abnormal pixel interference points in the template generation process, the average feature map is further subjected to reciprocal normalization operation, so that smaller weight is given to larger pixel points, larger weight is given to smaller pixel points, and the constructed average feature map is applied to random abnormal value elimination of the detection image difference map in the later test set.
S5: and detecting the metal packaging of the integrated circuit by combining the trained adaptive depth generation countermeasure network and a preset image post-processing strategy.
The specific process of step S5 is:
s501: normalizing the metal packaging image to be detected, inputting the normalized image into a trained adaptive depth generation countermeasure network, and outputting a template image;
it should be noted that the metal package image in the test set can be used as the metal package image to be detected, the normalized image is input to the trained adaptive depth generation countermeasure network, the convolution layer in the encoder of the network is used to gradually extract the appearance image features in the image, and then the image high-dimensional features obtained by the encoder are input to the decoder, so that the template image which is similar to the background appearance features of the metal package image to be detected but does not contain the defective pixels is generated.
S502: carrying out differential operation on the metal packaging image to be detected and the output model image to obtain a difference image;
it should be noted that, because the ideal state of the network model is to generate a qualified template that is the same as the background information of the metal package image of the integrated circuit to be detected, but the generated network cannot reach the ideal state in practice, an appearance similar to the detected image in the aspect of texture details is usually generated, that is, some abnormal interference points will be randomly generated in the difference map, and therefore, a strategy needs to be adopted to eliminate the interference points to avoid affecting the final detection result.
S503: eliminating random abnormal values in the difference value graph by using the average characteristic graph to obtain a difference value graph with the abnormal values removed;
it should be noted that, because the generation of the template image is difficult to reach an ideal qualified metal encapsulation template that is the same as the background of the input image, in the process of network model parameter training and optimization, the generated template may have interference of random noise, and some abnormal values exist, and if the abnormal values are not eliminated, the abnormal values generated randomly will be wrongly determined as defective pixel points in the subsequent defect determination. Therefore, for the abnormal values existing in the difference map, the average feature map is introduced to solve the influence of random abnormal noise generated in the template generation process on the later detection.
More specifically, after the average feature map is normalized, the normalized average feature map is convolved with a difference map of the image to be detected to obtain a difference map set with abnormal values removedWherein N is2The total number of images to be detected in the test set. FIG. 2 is a graph showing the comparison of the processing effect of the average feature map, and FIG. 2 (a) shows that the average feature map is not processedThe positive and negative sample difference image pixel distribution processed by the average feature map is shown in fig. 2 (b).
It should be noted that the present invention is advantageous to solve the problem of generating random outliers for templates through the average feature map construction.
S504: extracting the defect contour of each difference image processed by the average characteristic image by using a self-adaptive threshold value to obtain a defect contour image;
it should be noted that, according to the difference image pixel distribution of the qualified integrated circuit metal package and the defect sample, and the mean and variance conditions thereof, a coefficient in front of the mean with stable variation is selected to set a hyper-parameter to macroscopically regulate and control the threshold, so as to achieve the purpose of performing defect extraction on each difference image by using the adaptive threshold, wherein the adaptive threshold expression is as follows:
wherein the content of the first and second substances,indicating a difference map threshold after removal of the outlier,representing the mean of the difference map after removal of the outliers,the method is a control constant determined according to the pixel value range of the difference image corresponding to the qualified samples in the training set with different illumination brightness, wherein the variance of the difference image is obtained after the abnormal value is removed, and alpha is a hyper-parameter.
It should be noted that the invention designs the adaptive threshold strategy according to the integrated circuit metal package difference diagram data analysis, which accords with the defect pixel extraction, effectively realizes the pixel-level segmentation of different backgrounds and various characterization defect features of the integrated circuit metal package image, and solves the problem that the defect contour features lose key information or interfere points due to the threshold segmentation in the related art.
S505: and further obtaining a defect segmentation graph and a defect evaluation score which meet the quality inspection sensitivity requirement by using a local defect verification strategy for the defect contour image with the background noise.
For a defect outline image with background noise, obtaining a defect segmentation graph and a defect evaluation score meeting the quality inspection sensitivity requirement by using a local defect verification strategy, wherein the specific process comprises the following steps of:
recording the result image obtained by the adaptive threshold method as a setWherein b isnThe image processing method comprises the steps of obtaining a difference image corresponding to an nth output image to be detected through an average characteristic image and a result image after adaptive threshold processing, wherein the result image is a large-scale image, and the scale of the result image is H multiplied by W;
inputting a large-scale result graph, and sliding by using the size of a sliding window as f multiplied by f and the step length as stride to obtain Row multiplied by Col local image blocks in the large-scale result graph, whereinCol=Wherein pad is the padding number;
the set of partial images after sliding window can be described as
The Defect probability (DPP) of each local image is calculated as follows:
whereinThe defect probability value corresponding to the local image of the nth input large-scale image at the row and column positions is shown, the size of the local image is the sliding window size f multiplied by f, wherein the number of channels is C,outputting local images of a row number and a col column number corresponding to the image for the self-adaptive threshold value;
setting the sensitivity requirement of quality inspection as tau, carrying out probability analysis on the defect probability calculated by each local image, and recording the probability within the sensitivity as a numerical value of 0; otherwise, the probability except the sensitivity is recorded as a numerical value 1;
local Defect Verification (DVP), the mathematical expression is:
whereinThe mark values corresponding to the local images at the row and column positions of the nth image are represented, wherein the sensitivity calculation formula isThe mark lower than the quality inspection sensitivity tau is 0, otherwise, the mark higher than the sensitivity tau is 1, the local block is not in the acceptable range of quality inspection, the mark is used for marking the 0 or 1 of the image block which is identified as the block to obtain a mask graph which identifies the defect, the mask graph represents the mask graph of the position of the defect occurrence region, and the mask graph and the input image in the previous step are subjected to dot-product fusion to obtain a defect segmentation image;
in order to judge the quality of the metal packaging image more intuitively, a quality evaluation strategy is designed on the basis to judge whether the metal packaging image is qualified.
The mark values of all local images of each result image in the result image set are accumulated to be used as an image defect Evaluation Score (ES) for evaluating the large-scale result image, and the expression is as follows:
wherein ESnAnd expressing the defect evaluation score of the nth image to be detected.
Further, comparing the obtained defect evaluation score with a preset threshold to obtain a final detection result (IR), where the expression is as follows:
wherein th is the maximum defect score threshold value for all qualified samples in the training set to be qualified, if the defect evaluation score of the metal packaging image to be detected is greater than or equal to the threshold value, the corresponding metal packaging is judged to be unqualified, and IR is used as the samplenIs 1; if the defect evaluation score of the metal packaging image to be detected is smaller than the threshold value, the corresponding metal packaging is a qualified sample, and IRnIs 0.
It should be noted that the defect segmentation requirement of the quality inspection sensitivity requirement is met by adopting local defect verification, the background noise in the feature map is favorably eliminated, the problem that real defect pixels are lost when the background noise is processed by the traditional method is solved, the defect evaluation standard is defined, the image quality of the surface of the metal package is judged, and the detection accuracy is improved.
FIG. 3 is a schematic diagram showing defect detection of a qualified metal package image of an integrated circuit, in which (a) in FIG. 3 shows a qualified metal package image of an integrated circuit, (b) shows a difference map corresponding to (a), and (c) shows a result map corresponding to (a) after only adaptive threshold is used; (d) and (e) a Mask graph obtained by verifying the corresponding local defects of the (a). FIG. 4 is a schematic diagram showing defect detection of an image of a failed metal package of an integrated circuit, wherein (f) in FIG. 4 shows an image of a failed metal package of an integrated circuit, (g) shows a difference map corresponding to (f), and (h) shows a result map corresponding to (f) after only adaptive threshold is used; (j) and (f) showing a corresponding result graph after all post-processing strategies, and (k) showing a Mask graph obtained by local defect verification corresponding to the (f).
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A method for detecting defects of metal package of integrated circuit is characterized by comprising the following steps:
s1: acquiring and preprocessing an integrated circuit metal packaging image, and dividing the preprocessed image into a training set and a test set;
s2: constructing a self-adaptive depth generation countermeasure network and initializing weight parameters of the network;
s3: performing iterative optimization on the adaptive depth generation countermeasure network according to the loss function;
s4: counting difference graphs corresponding to the training set grid samples to construct an average feature graph;
s5: and detecting the metal packaging of the integrated circuit by combining the trained adaptive depth generation countermeasure network and a preset image post-processing strategy.
2. The method as claimed in claim 1, wherein the step S1 comprises the following steps:
s101: acquiring an integrated circuit metal packaging image, positioning an ROI (region of interest) of the image, and cutting out an ROI image;
s102: manually dividing the cut ROI area image into a qualified sample image and an unqualified sample image;
s103: randomly selecting a plurality of qualified sample images as a training set, using the remaining qualified sample images and the remaining unqualified sample images as a test set, and simultaneously performing pixel normalization on the images in the training set and the test set before inputting the images into a network to enable the pixel value to be in a preset range.
3. The method of claim 1, wherein the adaptive deep generation countermeasure network comprises: the encoder comprises nine network layers, wherein the first seven network layers comprise a convolutional layer, a batch normalization layer and a LeaklyRelu activation function, the eighth network layer comprises a hole convolutional layer, a batch normalization layer and a LeaklyRelu activation function, and the ninth network layer comprises a 1x1 convolutional layer; the decoder is provided with seven network layers, wherein the first six layers all comprise an deconvolution layer, batch normalization and a LeaklyRelu activation function, and the seventh network layer adopts the deconvolution layer, the batch normalization and a tanh activation function; the discriminator comprises eight network layers, the first seven network layers comprise a convolutional layer, a batch normalization layer, a LeaklyReLU activation function and a maximum pooling layer, and the eighth network layer is a 1x1 convolutional layer and is used for dimension reduction output; the encoder extracts the high-dimensional features of the input image, and then the qualified template of the deep image features is reconstructed from the high-dimensional features through deconvolution of the decoder.
4. The method of claim 1, wherein in step S3, the adaptive depth generation network is iteratively optimized according to a loss function. The network loss function includes: a generator loss function and a discriminator loss function, wherein the generator loss function comprises: content loss lcon=||x-G(x)||1To combat the loss ladv=||D(7)(x)-D(7)(G(x))||1Characteristic loss lenc=||En1(x)-En2(G(x))||1Generator loss function LGDiscriminator loss function LDThe expressions are respectively:
LG=wcon*lcon+wenc*lenc+wadv*ladv (1)
wherein wcon,wenc、wadvThe parameter is a hyper-parameter, which adjusts the influence of each loss on the overall loss function, D (x) represents the probability of judging whether the network judges the real picture as real, and D (G (x)) is the probability of judging whether the picture generated by G is the real picture or not.
5. The method of claim 1, wherein the step S4 of constructing the average feature map comprises the following steps:
inputting the images in the training set into the trained model, and outputting corresponding template images;
performing difference operation on the output template image and the corresponding input image to obtain a difference image set of the training set;
accumulating and averaging the difference images to obtain an average characteristic diagram, wherein a calculation formula of the average characteristic diagram is as follows:
wherein d isn(i, j, k) is the pixel value of the difference image, and each constructed qualified sample is superposed and averaged to obtain a pixel which can describe random noise possibly occurring in the generation process, namely an abnormal pixel point;
and performing reciprocal normalization operation on the average feature map, and applying the average feature map subjected to the normalization operation to the elimination of random abnormal values of the detection image difference map in the later test set.
6. The method as claimed in claim 1, wherein the step S5 comprises the following steps:
s501: normalizing the metal packaging image to be detected, inputting the normalized image into a trained adaptive depth generation countermeasure network, and outputting a template image;
s502: carrying out differential operation on the metal packaging image to be detected and the output model image to obtain a difference image;
s503: eliminating random abnormal values in the difference value graph by using the average characteristic graph to obtain a difference value graph with the abnormal values removed;
s504: extracting the defect contour of each difference image processed by the average characteristic image by using a self-adaptive threshold value to obtain a defect contour image;
s505: and obtaining a defect segmentation graph and a defect evaluation score which meet the quality inspection sensitivity requirement by using a local defect verification strategy for the defect contour image with the background noise.
7. The method as claimed in claim 6, wherein the template image of step S501 is similar to the background appearance of the metal package image to be detected, but does not contain defective pixels.
8. The method of claim 6, wherein the adaptive threshold expression is:
wherein the content of the first and second substances,indicating a difference map threshold after removal of the outlier,representing the mean of the difference map after removal of the outliers,the method is a control constant determined according to the pixel value range of the difference image corresponding to the qualified samples in the training set with different illumination brightness, wherein a is a hyper-parameter.
9. The method for detecting the defects of the metal package of the integrated circuit according to claim 6, wherein a defect segmentation graph and a defect evaluation score meeting quality inspection sensitivity requirements are obtained by using a local defect verification strategy for a defect contour image with background noise, and the specific process is as follows:
recording the result image obtained by the adaptive threshold method as a setWherein b isnThe image processing method comprises the steps of obtaining a difference image corresponding to an nth output image to be detected through an average characteristic image and a result image after adaptive threshold processing, wherein the result image is a large-scale image, and the scale of the result image is H multiplied by W;
inputting a large-scale result graph, and sliding by using the size of a sliding window as f multiplied by f and the step length as stride to obtain Row multiplied by Col local image blocks in the large-scale result graph, wherein Wherein pad is the padding number;
the set of partial images after sliding window can be described as
Respectively calculating the defect probability of each local image, wherein the calculation formula is as follows:
whereinThe defect probability value corresponding to the local image of the nth input large-scale image at the row and column positions is shown, the size of the local image is the sliding window size f multiplied by f, wherein the number of channels is C,outputting local images of a row number and a col column number corresponding to the image for the self-adaptive threshold value;
setting the sensitivity requirement of quality inspection as tau, carrying out probability analysis on the defect probability calculated by each local image, and recording the probability within the sensitivity as a numerical value of 0; otherwise, the probability except the sensitivity is recorded as a numerical value 1; the local defect verification is marked as DVP, and the mathematical expression is as follows:
whereinThe mark values corresponding to the local images at the row and column positions of the nth image are represented, wherein the sensitivity calculation formula isThe mark below the quality detection sensitivity tau is 0, whereas the mark above the sensitivity tau is 1, which indicates that the local block is not in the acceptable range of quality detection, and the mark is used for marking the 0 or 1 of the block identified to the block to obtain a mask graph which identifies the defect, and the graph shows that the defect appearsAt the current region position, performing point-by-point fusion operation on the mask image and the input image to obtain a defect segmentation image;
and accumulating the mark values of all local images of the result image set result image to be used as an image defect evaluation score for evaluating the large-scale result image, wherein the expression is as follows:
wherein ESnAnd expressing the defect evaluation score of the nth image to be detected.
10. The method of claim 9, wherein the obtained defect evaluation score is compared with a preset threshold to obtain a final detection result, and the expression is as follows:
and th is the maximum defect score threshold value for all qualified samples in the training set to be qualified, if the defect evaluation score of the to-be-detected metal packaging image is greater than or equal to the threshold value, the corresponding metal packaging is judged to be unqualified, and if the defect evaluation score of the to-be-detected metal packaging image is less than the threshold value, the corresponding metal packaging is judged to be qualified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111198584.4A CN113920096B (en) | 2021-10-14 | 2021-10-14 | Integrated circuit metal packaging defect detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111198584.4A CN113920096B (en) | 2021-10-14 | 2021-10-14 | Integrated circuit metal packaging defect detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113920096A true CN113920096A (en) | 2022-01-11 |
CN113920096B CN113920096B (en) | 2024-07-02 |
Family
ID=79240619
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111198584.4A Active CN113920096B (en) | 2021-10-14 | 2021-10-14 | Integrated circuit metal packaging defect detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113920096B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114359269A (en) * | 2022-03-09 | 2022-04-15 | 广东工业大学 | Virtual food box defect generation method and system based on neural network |
CN117011650A (en) * | 2023-10-07 | 2023-11-07 | 腾讯科技(深圳)有限公司 | Method and related device for determining image encoder |
CN117115161A (en) * | 2023-10-24 | 2023-11-24 | 四川新康意众申新材料有限公司 | Plastic defect inspection method |
CN117252840A (en) * | 2023-09-26 | 2023-12-19 | 西安交通大学 | Photovoltaic array defect elimination evaluation method and device and computer equipment |
CN117853876A (en) * | 2024-03-08 | 2024-04-09 | 合肥晶合集成电路股份有限公司 | Training method and system for wafer defect detection model |
CN117952983A (en) * | 2024-03-27 | 2024-04-30 | 中电科大数据研究院有限公司 | Intelligent manufacturing production process monitoring method and system based on artificial intelligence |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190303717A1 (en) * | 2018-03-28 | 2019-10-03 | Kla-Tencor Corporation | Training a neural network for defect detection in low resolution images |
CN110880175A (en) * | 2019-11-15 | 2020-03-13 | 广东工业大学 | Welding spot defect detection method, system and equipment |
CN110992315A (en) * | 2019-11-18 | 2020-04-10 | 上海集成电路研发中心有限公司 | Chip surface defect classification device and method based on generative countermeasure network |
CN111223062A (en) * | 2020-01-08 | 2020-06-02 | 西安电子科技大学 | Image deblurring method based on generation countermeasure network |
US20200175352A1 (en) * | 2017-03-14 | 2020-06-04 | University Of Manitoba | Structure defect detection using machine learning algorithms |
CN111340791A (en) * | 2020-03-02 | 2020-06-26 | 浙江浙能技术研究院有限公司 | Photovoltaic module unsupervised defect detection method based on GAN improved algorithm |
-
2021
- 2021-10-14 CN CN202111198584.4A patent/CN113920096B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200175352A1 (en) * | 2017-03-14 | 2020-06-04 | University Of Manitoba | Structure defect detection using machine learning algorithms |
US20190303717A1 (en) * | 2018-03-28 | 2019-10-03 | Kla-Tencor Corporation | Training a neural network for defect detection in low resolution images |
CN110880175A (en) * | 2019-11-15 | 2020-03-13 | 广东工业大学 | Welding spot defect detection method, system and equipment |
CN110992315A (en) * | 2019-11-18 | 2020-04-10 | 上海集成电路研发中心有限公司 | Chip surface defect classification device and method based on generative countermeasure network |
CN111223062A (en) * | 2020-01-08 | 2020-06-02 | 西安电子科技大学 | Image deblurring method based on generation countermeasure network |
CN111340791A (en) * | 2020-03-02 | 2020-06-26 | 浙江浙能技术研究院有限公司 | Photovoltaic module unsupervised defect detection method based on GAN improved algorithm |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114359269A (en) * | 2022-03-09 | 2022-04-15 | 广东工业大学 | Virtual food box defect generation method and system based on neural network |
CN117252840A (en) * | 2023-09-26 | 2023-12-19 | 西安交通大学 | Photovoltaic array defect elimination evaluation method and device and computer equipment |
CN117252840B (en) * | 2023-09-26 | 2024-04-05 | 西安交通大学 | Photovoltaic array defect elimination evaluation method and device and computer equipment |
CN117011650A (en) * | 2023-10-07 | 2023-11-07 | 腾讯科技(深圳)有限公司 | Method and related device for determining image encoder |
CN117011650B (en) * | 2023-10-07 | 2023-12-08 | 腾讯科技(深圳)有限公司 | Method and related device for determining image encoder |
CN117115161A (en) * | 2023-10-24 | 2023-11-24 | 四川新康意众申新材料有限公司 | Plastic defect inspection method |
CN117115161B (en) * | 2023-10-24 | 2024-01-02 | 四川新康意众申新材料有限公司 | Plastic defect inspection method |
CN117853876A (en) * | 2024-03-08 | 2024-04-09 | 合肥晶合集成电路股份有限公司 | Training method and system for wafer defect detection model |
CN117853876B (en) * | 2024-03-08 | 2024-06-11 | 合肥晶合集成电路股份有限公司 | Training method and system for wafer defect detection model |
CN117952983A (en) * | 2024-03-27 | 2024-04-30 | 中电科大数据研究院有限公司 | Intelligent manufacturing production process monitoring method and system based on artificial intelligence |
Also Published As
Publication number | Publication date |
---|---|
CN113920096B (en) | 2024-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113920096B (en) | Integrated circuit metal packaging defect detection method | |
CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
CN111582294B (en) | Method for constructing convolutional neural network model for surface defect detection and application thereof | |
CN112131936B (en) | Inspection robot image recognition method and inspection robot | |
CN110930390B (en) | Chip pin missing detection method based on semi-supervised deep learning | |
CN111833306A (en) | Defect detection method and model training method for defect detection | |
CN110880175B (en) | Welding spot defect detection method, system and equipment | |
CN108827969A (en) | Metal parts surface defects detection and recognition methods and device | |
KR100868884B1 (en) | Flat glass defect information system and classification method | |
CN107369176B (en) | System and method for detecting oxidation area of flexible IC substrate | |
CN115020267A (en) | Semiconductor surface defect detection method | |
CN113781415B (en) | Defect detection method, device, equipment and medium for X-ray image | |
CN113538392B (en) | Wafer detection method, wafer detection equipment and storage medium | |
CN113409245A (en) | Automatic identification method for X-ray inspection defects of electronic components | |
CN114445348B (en) | New material water pump defect detection method and system based on optical means | |
CN115170804B (en) | Surface defect detection method, device, system and medium based on deep learning | |
CN114255212A (en) | FPC surface defect detection method and system based on CNN | |
CN115471476A (en) | Method, device, equipment and medium for detecting component defects | |
CN117011260A (en) | Automatic chip appearance defect detection method, electronic equipment and storage medium | |
CN110660048B (en) | Leather surface defect detection method based on shape characteristics | |
CN112200790A (en) | Cloth defect detection method, device and medium | |
CN115330796A (en) | Copper wire tinning defect identification method | |
CN109978916B (en) | Vibe moving target detection method based on gray level image feature matching | |
CN117437238B (en) | Visual inspection method for surface defects of packaged IC | |
CN115713480A (en) | Integrated circuit metal packaging deep learning defect detection method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant |