CN111681235B - IC welding spot defect detection method based on learning mechanism - Google Patents
IC welding spot defect detection method based on learning mechanism Download PDFInfo
- Publication number
- CN111681235B CN111681235B CN202010529781.9A CN202010529781A CN111681235B CN 111681235 B CN111681235 B CN 111681235B CN 202010529781 A CN202010529781 A CN 202010529781A CN 111681235 B CN111681235 B CN 111681235B
- Authority
- CN
- China
- Prior art keywords
- sample
- welding spot
- samples
- qualified
- statistical model
- 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.)
- Active
Links
- 238000003466 welding Methods 0.000 title claims abstract description 99
- 238000001514 detection method Methods 0.000 title claims abstract description 55
- 230000007547 defect Effects 0.000 title claims abstract description 44
- 230000007246 mechanism Effects 0.000 title claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 101
- 238000011156 evaluation Methods 0.000 claims abstract description 54
- 238000013179 statistical model Methods 0.000 claims abstract description 45
- 229910000679 solder Inorganic materials 0.000 claims description 18
- 238000000034 method Methods 0.000 abstract description 18
- 230000000694 effects Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005476 soldering Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000004377 microelectronic Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 150000003071 polychlorinated biphenyls Chemical class 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000000758 substrate Substances 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
-
- 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
- G06T2207/30152—Solder
-
- 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)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a learning mechanism-based IC welding spot defect detection method, which comprises the following steps: establishing a local statistical model according to a plurality of qualified IC welding spot samples; collecting an IC welding spot sample, inputting the IC welding spot sample into a local statistical model, and comparing the IC welding spot sample with the local statistical model to obtain a potential defect image as a training set, wherein the training set comprises a qualified sample and a unqualified sample; training the plurality of classifiers with the training subset; evaluating the training set by utilizing the trained multiple classifiers, determining the weight of each classifier, calculating the average disqualification probability of each sample, and calculating an evaluation threshold; and acquiring an IC welding spot sample picture to be detected, obtaining the average disqualification probability of the IC welding spot sample by utilizing the local statistical model and a plurality of classifiers, and comparing the average disqualification probability with an evaluation threshold value to obtain a detection result. According to the method provided by the invention, the classifier weight and the evaluation threshold are adaptively set, the accuracy and the robustness of the detection of the defects of the welding spots of the IC are improved, and the evaluation result is more reasonable.
Description
Technical Field
The invention relates to the field of digital image processing application, in particular to an IC welding spot defect detection method based on a learning mechanism.
Background
The surface mount technology (Surface Mount Technology, SMT) is widely used in the production of printed circuit boards (Printed Circuit Board, PCB), and is a circuit mounting technology for mounting a leadless or short-lead surface mount component (SMC/SMD, chinese called chip component) on the surface of a printed circuit board or the surface of another substrate, and soldering and assembling the components by reflow soldering or dip soldering.
The quality of the integrated circuit IC assembly is critical to the reliability and practicality of the PCB and electronic devices, as the IC pins are bridges of the electronic connection between the IC components and the PCB, by using SMT technology to mount or place the IC assembly directly on the PCB surface. With the rapid development of microelectronic manufacturing, the density of IC devices on PCBs is increasing and the size is decreasing, making it difficult to manually inspect IC pads.
With the development of computer vision technology, automated Optical Inspection (AOI) systems are widely used for product surface defect inspection to evaluate the quality of many industrial products. In order to realize AOI systems of different products, the detection method of the AOI system needs to be adjusted according to the characteristics of the object. Currently, few algorithms are available on AOI systems for detecting IC solder joint defects.
In the existing method for detecting the defects of the welding spots of the IC, modeling is carried out on a qualified sample to obtain a model of the qualified sample, and then the model is compared with other samples to obtain a binary image; the pixel is matched with the template and is 0 in the binary image, and the mismatch is 1; and then, evaluating the sample by combining pixel point weighted summation in the binary image of the sample to be tested.
In the prior art, only qualified samples are used as training samples, and the establishment of the model and the threshold value are determined based on the qualified samples, so that the methods ignore the patterns of unqualified samples, the detection result can only determine the patterns of the qualified samples learned in the training samples, namely, the detection result can only determine the qualified samples within a certain range, and the qualified samples exceeding a certain range can be easily considered as the unqualified samples.
In the prior art, the weights of the pixel points are difficult to reasonably set, in the process of evaluating the quality of the sample, the pixel points of the binary image of the sample are generally weighted and summed, and the weight is determined in a manner of manually setting according to priori knowledge or determining according to a defect frequency chart. Both the former and the latter are very dependent on the quality of the qualified sample model, and if the training samples of the qualified sample model are few, the detection result is easy to be degraded.
The existing evaluation method in the scheme lacks overall consideration, is too simple, is difficult to evaluate the quality of the sample on the whole, and is difficult to determine the evaluation threshold value, so that more training samples are needed to obtain a reliable detection result.
Disclosure of Invention
The invention aims to provide an IC welding spot defect detection method based on a learning mechanism, which aims to solve the technical problems that the quality of a sample is difficult to evaluate on the whole and an evaluation threshold is difficult to determine in the existing method.
The aim of the invention can be achieved by the following technical scheme:
establishing a local statistical model according to a plurality of qualified IC welding spot samples;
collecting pictures of an IC welding spot sample, inputting the pictures into the local statistical model for comparison, and obtaining a potential defect image of the IC welding spot sample as a training set; the training set comprises a qualified sample and a disqualified sample, wherein the qualified sample is a potential defect image of a qualified IC welding spot sample, and the disqualified sample is a potential defect image of a disqualified IC welding spot sample;
respectively training a plurality of classifiers by utilizing different training subsets; wherein the training subset consists of all unqualified samples in the training set and the same number of qualified samples;
each sample in the training set is evaluated by utilizing a plurality of trained classifiers, the weight of each classifier is determined, and the average disqualification probability of each sample is calculated, so that an evaluation threshold value of the training set sample is obtained;
and acquiring an IC welding spot sample picture to be detected, obtaining the average disqualification probability of the IC welding spot sample by using the local statistical model and the plurality of classifiers, and comparing the average disqualification probability of the IC welding spot sample with the evaluation threshold value to obtain a detection result of the IC welding spot sample.
Optionally, determining the average failure probability of the IC bond pad samples using the local statistical model and the plurality of classifiers further comprises: and inputting the IC welding spot sample picture into the local statistical model to obtain a potential defect image of the IC welding spot sample, and evaluating the potential defect image by utilizing a plurality of classifiers to obtain the average disqualification probability of the IC welding spot sample.
Optionally, comparing the average failure probability of the IC pad sample with the evaluation threshold value to obtain a detection result of the IC pad sample further includes: and judging whether the average disqualification probability of the IC welding spot sample is smaller than the evaluation threshold, if so, the IC welding spot sample is qualified, and if not, the IC welding spot sample is disqualified.
Optionally, determining the weight of each of the classifiers further comprises calculating the weight of each of the classifiers using the following equation:
wherein ,wi Representing the weight of the ith classifier, e i Representing the accuracy of the ith classifier, e max Representing the highest accuracy in the classifier, e min Representing the lowest accuracy in the classifier.
Optionally, the classifier is a KNN classifier.
Optionally, calculating the average failure probability for each sample further comprises calculating the average failure probability for each sample using the following equation:
wherein ,MUPX Represents the average failure probability of sample X, KNN i (X) represents the failure probability of the ith KNN classifier on sample X, and N represents the number of KNN classifiers.
Optionally, the evaluation threshold is calculated using the following formula:
UPT＝αU min +(1+α)Q max ；
wherein alpha represents the proportion of qualified samples in the training set, U min Represents the minimum value, Q, of MUP in the failed sample max Represents the maximum value of MUP in the acceptable samples, Q represents an acceptable sample, U represents an unacceptable sample,num ({ Q }) represents the number of qualifying samples in the training set and num ({ Q } + { U }) represents the number of samples in the training set.
Optionally, building the local statistical model from the plurality of qualified IC bond pad samples further comprises: and establishing a local statistical model, initializing the local statistical model by using a first qualified IC solder joint sample, and updating the local statistical model according to a subsequent qualified IC solder joint sample.
Optionally, the local statistical model is a VIBE model.
Optionally, the training subset consisting of all of the failed samples in the training set and the same number of failed samples further comprises: the qualifying samples are randomly selected.
The invention provides a learning mechanism-based IC welding spot defect detection method, which comprises the following steps: establishing a local statistical model according to a plurality of qualified IC welding spot samples; collecting pictures of an IC welding spot sample, inputting the pictures into the local statistical model for comparison, and obtaining a potential defect image of the IC welding spot sample as a training set; the training set comprises a qualified sample and a disqualified sample, wherein the qualified sample is a potential defect image of a qualified IC welding spot sample, and the disqualified sample is a potential defect image of a disqualified IC welding spot sample; respectively training a plurality of classifiers by utilizing different training subsets; wherein the training subset consists of all unqualified samples in the training set and the same number of qualified samples; each sample in the training set is evaluated by utilizing a plurality of trained classifiers, the weight of each classifier is determined, and the average disqualification probability of each sample is calculated, so that an evaluation threshold value of the training set sample is obtained; and acquiring an IC welding spot sample picture to be detected, obtaining the average disqualification probability of the IC welding spot sample by using the local statistical model and the plurality of classifiers, and comparing the average disqualification probability of the IC welding spot sample with the evaluation threshold value to obtain a detection result of the IC welding spot sample.
The IC welding spot defect detection method based on the learning mechanism provided by the invention has the following beneficial effects:
(1) Based on a mechanism from local statistical modeling to global evaluation detection, not only characteristic information in training set samples can be well grasped, but also the relation between the training set samples can be revealed, and the quality of the IC welding spot samples is globally evaluated, so that the method can complete the detection task of the IC welding spot samples under the condition that the number of the training samples is insufficient and unbalanced.
(2) The weight of the classifier is set in a self-adaptive mode, the influence of the classifier with good detection effect can be improved by the self-adaptive weight, the influence of the classifier with poor detection effect on global evaluation is reduced, and therefore the robustness of the whole detection scheme and the detection accuracy are improved.
(3) By setting the evaluation threshold in an adaptive manner, the limit of the qualified samples and the limit of the unqualified samples can be considered simultaneously due to the addition of the unqualified samples to the training set, so that the evaluation is more reasonable. Meanwhile, according to the proportion of the qualified samples in the training set, the proportion of the qualified sample limit and the unqualified sample limit in the evaluation threshold is adjusted, and the evaluation threshold is changed along with the composition proportion of the qualified samples and the unqualified samples in the training set, so that the evaluation result is more reasonable.
Drawings
FIG. 1 is a schematic diagram of a detection flow of a method for detecting defects of IC solder joints based on learning according to the present invention.
Detailed Description
The embodiment of the invention provides an IC welding spot defect detection method based on a learning mechanism, which aims to solve the technical problems that the quality of a sample is difficult to evaluate on the whole and an evaluation threshold is difficult to determine in the existing method.
In order that the invention may be readily understood, a more complete description of the invention will be rendered by reference to the appended drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.
The embodiment of the invention provides an IC welding spot defect detection method based on a learning mechanism, which comprises the following steps:
establishing a local statistical model according to a plurality of qualified IC welding spot samples;
collecting pictures of an IC welding spot sample, inputting the pictures into the local statistical model for comparison, and obtaining a potential defect image of the IC welding spot sample as a training set; the training set comprises a qualified sample and a disqualified sample, wherein the qualified sample is a potential defect image of a qualified IC welding spot sample, and the disqualified sample is a potential defect image of a disqualified IC welding spot sample;
respectively training a plurality of classifiers by utilizing different training subsets; wherein the training subset consists of all unqualified samples in the training set and the same number of qualified samples;
each sample in the training set is evaluated by utilizing a plurality of trained classifiers, the weight of each classifier is determined, and the average disqualification probability of each sample is calculated, so that an evaluation threshold value of the training set sample is obtained;
and acquiring an IC welding spot sample picture to be detected, obtaining the average disqualification probability of the IC welding spot sample by using the local statistical model and the plurality of classifiers, and comparing the average disqualification probability of the IC welding spot sample with the evaluation threshold value to obtain a detection result of the IC welding spot sample.
Referring to fig. 1, a detection flow of an embodiment of the present invention is shown in fig. 1, and the embodiment of the present invention provides an IC solder joint defect detection method based on a learning mechanism, based on a detection mechanism from local to global, by detecting potential defective pixel points in an IC solder joint sample through a local statistical model, and then evaluating quality of the IC solder joint sample through a plurality of KNN classifiers. Based on the mechanism, the method provided by the embodiment of the invention not only can grasp the local characteristic information of the IC welding spot sample, and does not put any detail on the detection of the IC welding spot sample, but also can grasp the relation among the IC welding spot samples, and integrally evaluate the quality of the IC welding spot sample.
The IC welding spot defect detection method based on the learning mechanism provided by the embodiment of the invention mainly comprises two parts: (1) local statistical modeling; (2) a global assessment model based on ensemble learning.
The embodiment of the invention obtains the local statistical model by modeling a series of qualified IC welding spot samples, and the obtained statistical model is a qualified sample mode because the IC welding spot samples used for training the local statistical model are all qualified IC welding spot samples. The detected IC solder joint sample is input into a local statistical model, and a potential defect image (Potential Defect Image, PDI) of the IC solder joint sample is obtained through comparison with the local statistical model.
In the global evaluation stage based on ensemble learning, each KNN classifier is trained by using a training subset, wherein each KNN classifier is trained by one training subset, and the training subset consists of balanced unqualified and qualified samples. Unbalanced samples can cause classification results to be biased toward a greater number of classes, with balanced training sets enabling the KNN classifier to be trained well. The trained KNN classifier can be used to detect IC pad samples.
In a global evaluation stage based on ensemble learning, the PDI of the IC pad samples is globally evaluated by a plurality of weighted K-nearest neighbor (KNN) classifiers, wherein the KNN classifiers are trained from PDI images of different IC pad samples. And (3) carrying out weighted average on the detection result of each KNN classifier to obtain the average disqualification probability (Mean Unqualified Probability, MUP) of the input IC welding spot samples, comparing the MUP with the evaluation threshold value of the training set, wherein if the MUP is larger than the evaluation threshold value, the IC welding spot samples are disqualified, and otherwise, the IC welding spot samples are qualified.
Notably, the evaluation threshold herein is determined based on the boundaries of failed samples and qualified samples in the training set (minimum failed, maximum qualified).
In the embodiment of the invention, the local statistical modeling process is as follows:
the local statistical model is built according to a series of input qualified IC welding spot samples, and the embodiment of the invention adopts a VIBE model for modeling: firstly, initializing a local statistical model according to a first input qualified IC welding spot sample; the model is then continuously updated based on the subsequently entered qualified IC pad samples. The VIBE model is adopted because the VIBE modeling speed is high and the accuracy is high. Since the local model is subjected to global evaluation after processing, the local model is not required to be high, and a local statistical model based on the VIBE model is sufficient.
Initializing a local statistical model: six templates with the same size as the qualified IC welding spot sample are extracted from a qualified sample for training, wherein the pixel point of each template is randomly selected from the pixel points at the corresponding position in the qualified IC welding spot sample or the neighborhood pixel points thereof.
The number of templates is a parameter and is determined empirically. When the template is extracted, the first qualified IC welding spot sample is extracted, the pixel point is selected from each pixel point or the neighborhood thereof, and assigned to the pixel point at the corresponding position in the template, and the extracted template is also an image, and the size of the extracted template is the same as that of the qualified IC welding spot sample.
Model updating: when a certain pixel point in the input qualified IC welding spot sample is matched with enough templates, the pixel point is a qualified pixel point, and the pixel point or a neighborhood pixel point thereof is randomly selected and used for updating the pixel point at a corresponding position in one template or the neighborhood pixel point thereof; if the pixel point is not the qualified pixel point, updating the pixel point at the corresponding position in a non-matching template by using the pixel point.
It should be noted that the number of templates matched is also a parameter, and is specifically set empirically, if a pixel matches 2 templates out of 6 templates, it can be considered that the pixel matches enough templates.
Matching strategies: taking the pixel value x of a certain pixel point of the input IC pad sample as an example, taking the pixel value of this pixel point as the center, and R as the range, the range can be set empirically and experimentally, for example, r=30 is taken. If the pixel point of the corresponding position of a certain template is in the range, the template is considered to be matched with the pixel point. When the number of matched templates reaches a matching threshold, then the pixel is considered to be matched with the local statistical model and is considered to be a qualified pixel. After the input IC welding spot sample is compared with the local statistical model, a binary image called Potential Defect Image (PDI) of the IC welding spot sample can be obtained, wherein the pixel value of the qualified pixel point is 0, and the pixel value of the unqualified pixel point is 1.
For example, if the pixel value of a gray pixel is 10 and r=5, and the pixel value of the pixel at the corresponding position of the template is 1, 10, 15, 20, 25, 30, respectively, the template of 10, 15 is matched with the pixel, and the pixel can be considered as a qualified pixel.
In the embodiment of the invention, the global evaluation process based on ensemble learning is as follows:
in the global evaluation stage, the classifier is trained or detected, and the PDI diagram of the IC welding spot sample is obtained by comparing the classifier with the local statistical model, wherein the PDI diagram of the IC welding spot sample is a result of detecting pixel points in the image of the IC welding spot sample by utilizing the local statistical model, namely a result of processing detail characteristics of the IC welding spot sample.
Firstly, inputting an IC welding spot sample picture into a local statistical model to obtain a PDI picture of an IC welding spot sample as a training set, wherein the training set comprises a PDI picture (called a qualified sample) of a qualified IC welding spot sample and a PDI picture (called a disqualified sample) of a disqualified IC welding spot sample, and obtaining a balanced training subset consisting of a plurality of qualified samples and disqualified samples through downsampling; then, training the KNN classifier by utilizing each training subset, verifying the KNN classifier obtained by training in the whole training set, and calculating the weight of each KNN classifier; finally, the MUP of each sample in the training set is calculated, so that the evaluation threshold of the training set is obtained.
Downsampling: because the number of unqualified samples in the training set is far smaller than the number of qualified samples, training the KNN classifier using such a training set can result in a bias in detection results towards the qualified samples, which are difficult to detect. Thus, the selection of the qualified samples from the training set at random, and the composition of the unqualified samples into balanced training subsets, each resulting balanced training subset may be different, and the number of qualified samples and the number of unqualified samples in each balanced training subset may be the same.
Training a KNN classifier: and training each KNN classifier by utilizing a balance training subset obtained in the downsampling process, wherein each KNN classifier obtained by the training is different. The KNN classifier trained by the balance training subset obtained in the downsampling process can be used for well screening out unqualified samples, and therefore the problem of sample unbalance is solved.
Adaptively determining the weight of each KNN classifier: since each KNN classifier is different, it is necessary to evaluate their detection effect, and a KNN classifier with a better detection capability is given a relatively higher weight, and a KNN classifier with a worse detection capability is given a lower weight. In this way, the evaluation of the IC pad samples is more reasonable and robust. And evaluating each sample in the whole training set once by using the trained KNN classifier, and counting the accuracy of the evaluation. The weight of each KNN classifier is calculated according to equation (1):
wherein ,wi Representing the weight of the ith KNN classifier, e i Represents the accuracy of the ith KNN classifier, e max Represents the highest accuracy in the KNN classifier, e min Representing the lowest accuracy in the KNN classifier.
By adaptively determining the weight of each KNN classifier, a relatively higher weight is given to a KNN classifier with better detection capability, and a lower weight is given to a KNN classifier with poorer detection capability, so that adverse effects on the evaluation of the accuracy of the IC solder joint sample are reduced.
Global evaluation: after each sample in the training set is detected by using a plurality of KNN classifiers, the average failure probability (MUP) of each sample can be calculated according to formula (2):
wherein ,MUPX Represents the average failure probability of sample X, KNN i (X) represents the detection result of the ith KNN classifier on the sample X, and N represents the number of KNN classifiers.
Evaluation threshold: each sample in the training set is evaluated, and a failure probability threshold (Unqualified Probability Threshold, UPT) of the training set is calculated according to the formula (3), namely the evaluation threshold:
UPT＝αU min +(1-α)Q max ；(3)
wherein alpha represents the proportion of qualified samples in the training set, and U is calculated by using a formula (4) min Represents the minimum value, Q, of MUP in the failed sample max Representing the maximum value of MUP in the qualified sample, so that the limit of the qualified sample and the unqualified sample can be comprehensively considered, and the evaluation threshold is more robust.
Q in equation (4) represents a pass sample, U represents a fail sample, num ({ Q }) represents the number of pass samples in the training set, and num ({ Q } + { U }) represents the number of samples in the training set. And according to the proportion alpha of the qualified samples in the training set, the proportion of the qualified sample limit and the unqualified sample limit in the evaluation threshold is adjusted, and the evaluation threshold is changed along with the composition proportion of the qualified samples and the unqualified samples in the training set, so that the evaluation result is more reasonable.
According to the IC welding spot defect detection method based on the learning mechanism, based on the mechanism from local statistical modeling to global evaluation detection, not only can feature information in training set samples be well grasped, but also the relation between the training set samples can be revealed, and the quality of the IC welding spot samples can be globally evaluated, so that the method can complete the detection task of the IC welding spot samples under the condition that the number of the training samples is insufficient and unbalanced.
In the embodiment of the invention, the weight of the KNN classifier is set in a self-adaptive manner, the influence of the KNN classifier with good detection effect can be improved by the self-adaptive weight, and the influence of the KNN classifier with poor detection effect on global evaluation is reduced, so that the robustness of the whole detection scheme can be improved, and the detection accuracy is also improved.
In the embodiment of the invention, the evaluation threshold is set in a self-adaptive mode, and the limit of the qualified sample and the limit of the unqualified sample can be simultaneously considered by adding the unqualified sample to the training set, so that the evaluation is more reasonable. Meanwhile, according to the proportion of the qualified samples in the training set, the proportion of the qualified sample limit and the unqualified sample limit in the evaluation threshold is adjusted, and the evaluation threshold is changed along with the composition proportion of the qualified samples and the unqualified samples in the training set, so that the evaluation result is more reasonable.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (6)
1. The IC welding spot defect detection method based on the learning mechanism is characterized by comprising the following steps of:
establishing a local statistical model according to a plurality of qualified IC welding spot samples;
collecting pictures of an IC welding spot sample, inputting the pictures into the local statistical model for comparison, and obtaining a potential defect image of the IC welding spot sample as a training set; the training set comprises a qualified sample and a disqualified sample, wherein the qualified sample is a potential defect image of a qualified IC welding spot sample, and the disqualified sample is a potential defect image of a disqualified IC welding spot sample;
respectively training a plurality of classifiers by utilizing different training subsets; wherein the training subset consists of all unqualified samples in the training set and the same number of qualified samples; the classifier is a KNN classifier;
each sample in the training set is evaluated by utilizing a plurality of trained classifiers, the weight of each classifier is determined, and the average disqualification probability of each sample is calculated, so that an evaluation threshold value of the training set sample is obtained;
calculating the average failure probability for each sample includes:
the average failure probability for each sample was calculated using the following formula:
wherein ,MUPX Represents the average failure probability of sample X, KNN i (X) represents the failure probability of the ith KNN classifier on sample X, N represents the number of KNN classifiers;
the evaluation threshold is calculated using the following formula:
wherein ,
representing the proportion of qualified samples in the training set, U min Represents the minimum value, Q, of MUP in the failed sample max Representing the maximum value of MUP in the qualified samples, Q represents the qualified samples, U represents the unqualified samples, num ({ Q }) represents the number of qualified samples in the training set, and num ({ Q } + { U }) represents the number of samples in the training set;
collecting an IC welding spot sample picture to be detected, obtaining average disqualification probability of the IC welding spot sample by using the local statistical model and a plurality of classifiers, and comparing the average disqualification probability of the IC welding spot sample with the evaluation threshold value to obtain a detection result of the IC welding spot sample;
obtaining an average failure probability of the IC bond pad samples using the local statistical model and the plurality of classifiers includes:
and inputting the IC welding spot sample picture into the local statistical model to obtain a potential defect image of the IC welding spot sample, and evaluating the potential defect image by utilizing a plurality of classifiers to obtain the average disqualification probability of the IC welding spot sample.
2. The learning mechanism-based IC solder joint defect detection method of claim 1, wherein comparing the average failure probability of the IC solder joint sample with the evaluation threshold value to obtain a detection result of the IC solder joint sample further comprises: and judging whether the average disqualification probability of the IC welding spot sample is smaller than the evaluation threshold, if so, the IC welding spot sample is qualified, and if not, the IC welding spot sample is disqualified.
3. The learning mechanism based IC pad defect detection method of claim 1 or 2 wherein determining the weight of each of the classifiers further comprises calculating the weight of each of the classifiers using the following equation:
wherein ,wi Representing the weight of the ith classifier, e i Representing the accuracy of the ith classifier, e max Representing the highest accuracy in the classifier, e min Representing the lowest accuracy in the classifier.
4. The learning mechanism based IC solder joint defect detection method of claim 1, wherein building a local statistical model from a plurality of qualified IC solder joint samples further comprises: and establishing a local statistical model, initializing the local statistical model by using a first qualified IC solder joint sample, and updating the local statistical model according to a subsequent qualified IC solder joint sample.
5. The learning mechanism-based IC solder joint defect detection method of claim 1, wherein the local statistical model is a VIBE model.
6. The learning mechanism based IC pad defect detection method of claim 1 wherein the training subset is comprised of all failed samples and the same number of failed samples in the training set further comprises: the qualifying samples are randomly selected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010529781.9A CN111681235B (en) | 2020-06-11 | 2020-06-11 | IC welding spot defect detection method based on learning mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010529781.9A CN111681235B (en) | 2020-06-11 | 2020-06-11 | IC welding spot defect detection method based on learning mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111681235A CN111681235A (en) | 2020-09-18 |
| CN111681235B true CN111681235B (en) | 2023-05-09 |
Family
ID=72454550
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010529781.9A Active CN111681235B (en) | 2020-06-11 | 2020-06-11 | IC welding spot defect detection method based on learning mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111681235B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113255778A (en) * | 2021-05-28 | 2021-08-13 | 广汽本田汽车有限公司 | Welding spot quality detection method and device based on multi-model fusion and storage medium |
| CN113313704B (en) * | 2021-06-18 | 2022-12-30 | 广东工业大学 | Integrated circuit pin detection method, device, terminal and storage medium |
| CN113378957B (en) * | 2021-06-23 | 2023-04-25 | 广东工业大学 | Self-adaptive statistical model training method, welding spot defect detection method and system |
| CN115017862A (en) * | 2022-07-01 | 2022-09-06 | 广东赛昉科技有限公司 | Integrated circuit image matching method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104867145A (en) * | 2015-05-15 | 2015-08-26 | 广东工业大学 | IC element solder joint defect detection method based on VIBE model |
| CN107123117A (en) * | 2017-04-26 | 2017-09-01 | 广东工业大学 | A kind of IC pin quality of welding spot detection method and device based on deep learning |
| CN109014544A (en) * | 2018-08-17 | 2018-12-18 | 龙岩学院 | Miniature resistance spot welding quality on-line monitoring method |
| CN109615609A (en) * | 2018-11-15 | 2019-04-12 | 北京航天自动控制研究所 | A kind of solder joint flaw detection method based on deep learning |
| CN111145175A (en) * | 2020-01-10 | 2020-05-12 | 惠州光弘科技股份有限公司 | SMT welding spot defect detection method based on iForest model verification |
-
2020
- 2020-06-11 CN CN202010529781.9A patent/CN111681235B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104867145A (en) * | 2015-05-15 | 2015-08-26 | 广东工业大学 | IC element solder joint defect detection method based on VIBE model |
| CN107123117A (en) * | 2017-04-26 | 2017-09-01 | 广东工业大学 | A kind of IC pin quality of welding spot detection method and device based on deep learning |
| CN109014544A (en) * | 2018-08-17 | 2018-12-18 | 龙岩学院 | Miniature resistance spot welding quality on-line monitoring method |
| CN109615609A (en) * | 2018-11-15 | 2019-04-12 | 北京航天自动控制研究所 | A kind of solder joint flaw detection method based on deep learning |
| CN111145175A (en) * | 2020-01-10 | 2020-05-12 | 惠州光弘科技股份有限公司 | SMT welding spot defect detection method based on iForest model verification |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111681235A (en) | 2020-09-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111681235B (en) | IC welding spot defect detection method based on learning mechanism | |
| CN111145175B (en) | SMT welding spot defect detection method based on iForest model verification | |
| CN115601364B (en) | Golden finger circuit board detection method based on image analysis | |
| CN112053318B (en) | Two-dimensional PCB defect real-time automatic detection and classification device based on deep learning | |
| Hou et al. | An empirical study of hybrid inspection systems and allocation of inspection functions | |
| CN111899241B (en) | Quantitative on-line detection method and system for defects of PCB (printed Circuit Board) patches in front of furnace | |
| CN105510348A (en) | Flaw detection method and device of printed circuit board and detection equipment | |
| CN114926463A (en) | Production quality detection method suitable for chip circuit board | |
| CN113313704B (en) | Integrated circuit pin detection method, device, terminal and storage medium | |
| CN113971654A (en) | Welding spot detection model training method, welding spot detection method and welding spot detection device | |
| CN111882547A (en) | PCB missing part detection method based on neural network | |
| CN113674260A (en) | SMT welding spot defect detection method | |
| KR20200006028A (en) | Printed circuit board inspecting apparatus, method for determining fault type of screen printer and computer readable recording medium | |
| CN114387230A (en) | PCB defect detection method based on re-verification detection | |
| Lin et al. | A visual inspection system for surface mounted devices on printed circuit board | |
| TWI732296B (en) | System for generating detection model according to standard data to confirm soldering state and method thereof | |
| CN112308816B (en) | Image recognition device, image recognition method and storage medium thereof | |
| CN113176275A (en) | Method, device and system for retesting display panel | |
| CN117589770A (en) | PCB patch board detection method, device, equipment and medium | |
| JP4814116B2 (en) | Mounting board appearance inspection method | |
| CN115170476A (en) | Printed circuit board defect detection method based on image processing | |
| US12007377B2 (en) | System and method for building detection model based on standard value to confirm soldering status | |
| TWI758134B (en) | System for using image features corresponding to component identification for secondary inspection and method thereof | |
| KR20200022835A (en) | Method And Apparatus for Improving Inspection Parameters | |
| CN111539933B (en) | Direct-insert element detection method and system |
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 | ||
| GR01 | Patent grant |