CN109919908A - The method and apparatus of light-emitting diode chip for backlight unit defects detection - Google Patents
The method and apparatus of light-emitting diode chip for backlight unit defects detection Download PDFInfo
- Publication number
- CN109919908A CN109919908A CN201910065080.1A CN201910065080A CN109919908A CN 109919908 A CN109919908 A CN 109919908A CN 201910065080 A CN201910065080 A CN 201910065080A CN 109919908 A CN109919908 A CN 109919908A
- Authority
- CN
- China
- Prior art keywords
- image
- electrode
- defect
- defects
- convolutional neural
- 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
Landscapes
- Image Processing (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method and apparatus of light-emitting diode chip for backlight unit defects detection, belong to technical field of semiconductors.The described method includes: obtaining the image of chip to be detected；The image of chip to be detected is divided into the image of epitaxial part and the image of electrode section；The image of epitaxy defect part is intercepted from the image of epitaxial part, and the image of epitaxy defect part is inputted into the first convolutional neural networks, the defect type of epitaxy defect part is obtained, the parameter of the first convolutional neural networks is trained to obtain by using the epitaxy defect image of multiple defective types of calibration；The image of electrode defects part is intercepted from the image of electrode section, and the image of electrode defects part is inputted into the second convolutional neural networks, the defect type of electrode defects part is obtained, the parameter of the second convolutional neural networks is trained to obtain by using the electrode defects image of multiple defective types of calibration.The present invention especially meets industrial demand.
Description
Technical field
The present invention relates to technical field of semiconductors, in particular to a kind of the method and dress of light-emitting diode chip for backlight unit defects detection
It sets.
Background technique
Light emitting diode (English: Light Emitting Diode, referred to as: LED) it is that one kind can be converted to electric energy
The semiconductor diode of luminous energy.The core component of LED is chip, and chip includes epitaxial wafer and the electrode that extension on piece is arranged in.
Chip may generate various defects, such as hexagonal, micro- thick, scuffing, particle, atomization, green point in process,
Therefore after chip manufacture, defects detection can generally be carried out to chip, and (type including defect, big according to testing result
Small, quantity etc.) divide chip grade.
The defect inspection method of most original is manually by micro- sem observation chip to be detected, and detection efficiency and identification are quasi-
True rate is all unable to satisfy industrial production demand.It is available to chip to be detected later with the development of optical detection apparatus
Image can determine that chip to be detected is by comparing the image of the image of chip to be detected and defect-free chip
No existing defects, detection efficiency and recognition accuracy have all obtained great promotion, but detectability is limited, cannot achieve core
The grade classification of piece.It is now based on being substantially improved for computer process ability, it can be by the image of chip to be detected successively and respectively
The image of kind defective chip compares, using the corresponding defect type of image of the highest defective chip of similarity as to be detected
The defect type of chip, and then according to the defect type divided rank of chip to be detected.
In the implementation of the present invention, the inventor finds that the existing technology has at least the following problems:
The form of expression of same type defect is varied on chip, on hexagonal crystal system, cubic system and orthorhombic system
The shape of planar defect is different.But can not substantially provide in specific implementation the images of all representation defects with
The image of chip to be detected compares, and the accuracy of testing result is not high, and respectively and together by the image of chip to be detected
The image of the various representations of one type flaw compares, and compares substantial amounts, and detection efficiency is lower.And if only selecting one
The image of kind form of expression defect compares, then testing result is likely to inaccuracy, is still unable to satisfy industrial production demand.
Summary of the invention
The embodiment of the invention provides a kind of method and apparatus of light-emitting diode chip for backlight unit defects detection, are able to solve existing
The problem of accuracy of technological deficiency detection is unable to satisfy industrial production demand.The technical solution is as follows:
On the one hand, the embodiment of the invention provides a kind of method of light-emitting diode chip for backlight unit defects detection, the method packets
It includes:
Obtain the image of chip to be detected；
The image of the chip to be detected is divided by the image that the chip to be detected is handled using Model for Edge Detection
The image of epitaxial part and the image of electrode section；
The image of the epitaxial part is compared with the image of zero defect epitaxial wafer, from the image of the epitaxial part
The image of middle interception epitaxy defect part, and the image of the epitaxy defect part is inputted into the first convolutional neural networks, it obtains
The defect type of the epitaxy defect part, the parameter of first convolutional neural networks is by using multiple defective classes of calibration
The epitaxy defect image of type is trained to obtain；
The image of the electrode section is compared with the image of intact collapsible electrode, from the image of the electrode section
The image of electrode defects part is intercepted, and the image of the electrode defects part is inputted into the second convolutional neural networks, obtains institute
The defect type of electrode defects part is stated, the parameter of second convolutional neural networks is by using multiple defective types of calibration
Electrode defects image be trained to obtain.
Optionally, the image by the electrode defects part inputs the second convolutional neural networks, obtains the electrode
The defect type of defect part, comprising:
The image of the electrode defects part is normalized, the electrode image of predetermined dimension is obtained；
The electrode image of the predetermined dimension is inputted into second convolutional neural networks, obtains the electrode defects part
Defect type.
Optionally, the method also includes:
Obtain multiple electrodes defect image；
Receive the defect type of each electrode defects image calibration；
Using the defect type of multiple electrode defects images and each electrode defects image calibration, to described
Two convolutional neural networks are trained.
Further, the defect using multiple electrode defects images and each electrode defects image calibration
Type is trained second convolutional neural networks, comprising:
Multiple electrode defects images are successively inputted into second convolutional neural networks, each electrode is obtained and lacks
The defect type of image is fallen into, and in the obtained defect type of electrode defects image and the defect type difference of calibration, instead
To propagating the parameter for adjusting second convolutional neural networks, until defect type that multiple electrode defects images obtain with
The defect type of calibration is identical.
Optionally, the method also includes:
Count the quantity, size and defect class of the image of the epitaxy defect part intercepted in the image of the epitaxial part
Quantity, size and the defect type of the image of the electrode defects part intercepted in the image of type and the electrode section determine
The credit rating of the chip to be detected.
On the other hand, the embodiment of the invention provides a kind of device of light-emitting diode chip for backlight unit defects detection, described devices
Include:
Chip image obtains module, for obtaining the image of chip to be detected；
Image division module will be described to be checked for handling the image of the chip to be detected using Model for Edge Detection
The image for surveying chip is divided into the image of epitaxial part and the image of electrode section；
Extension image processing module, for comparing the image of the epitaxial part and the image of zero defect epitaxial wafer
Compared with, the image of interception epitaxy defect part from the image of the epitaxial part, and the image of the epitaxy defect part is defeated
Enter the first convolutional neural networks, obtains the defect type of the epitaxy defect part, the parameter of first convolutional neural networks
It is trained to obtain by using the epitaxy defect image of multiple defective types of calibration；
Electrode image processing module, for the image of the electrode section to be compared with the image of intact collapsible electrode,
Intercept the image of electrode defects part from the image of the electrode section, and by the image of electrode defects part input the
Two convolutional neural networks, obtain the defect type of the electrode defects part, and the parameter of second convolutional neural networks passes through
It is trained to obtain using the electrode defects image of multiple defective types of calibration.
Optionally, the electrode image processing module includes:
Electrode image normalization submodule is normalized for the image to the electrode defects part, is made a reservation for
The electrode image of specification；
Electrode defects type determination module, for the electrode image of the predetermined dimension to be inputted the second convolution mind
Through network, the defect type of the electrode defects part is obtained.
Optionally, described device further include:
Electrode defects image collection module, for obtaining multiple electrodes defect image；
Electrode defects type reception module, for receiving the defect type of each electrode defects image calibration；
Electrode defects type training module, for using multiple electrode defects images and each electrode defects figure
As the defect type of calibration, second convolutional neural networks are trained.
Further, the electrode defects type training module is used for,
Multiple electrode defects images are successively inputted into second convolutional neural networks, each electrode is obtained and lacks
The defect type of image is fallen into, and in the obtained defect type of electrode defects image and the defect type difference of calibration, instead
To propagating the parameter for adjusting second convolutional neural networks, until defect type that multiple electrode defects images obtain with
The defect type of calibration is identical.
Optionally, described device further include:
Determining module, the quantity of the image of the epitaxy defect part intercepted in the image for counting the epitaxial part,
The quantity of the image of the electrode defects part intercepted in size and defect type and the image of the electrode section, size and
Defect type determines the credit rating of the chip to be detected.
Technical solution provided in an embodiment of the present invention has the benefit that
By the image of chip to be detected being drawn first with Model for Edge Detection after the image for obtaining chip to be detected
It is divided into the image of epitaxial part and the image of electrode section, can be located respectively for two different defect situations in part
Reason, simplifies, and improves treatment effeciency.Similar place is carried out respectively with the image of electrode section to the image of epitaxial part again
The image of epitaxial part, by taking the image of epitaxial part as an example, is first compared with the image of zero defect epitaxial wafer, therefrom cuts by reason
Take out epitaxial wafer defect part image, can for defect part carry out defect type judgement, exclude non-defective part or
The influence of the other defect parts of person, it is with strong points, be conducive to improve accuracy and reduce operand.Multiple calibration are recycled to have scarce
The first convolutional neural networks that the epitaxy defect image of type trains are fallen into, its defect class is obtained by the image of epitaxy defect part
Type can greatly improve detection efficiency there is no repeatedly comparing.And the parameter of convolutional neural networks is using various
Defect image train come, various defect types, defect the various forms of expression can all be related to, the accuracy of testing result
It can guarantee, especially meet industrial demand.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of flow chart of the method for light-emitting diode chip for backlight unit defects detection provided in an embodiment of the present invention；
Fig. 2 is the schematic diagram of chip image provided in an embodiment of the present invention；
Fig. 3 is the flow chart of the method for another light-emitting diode chip for backlight unit defects detection provided in an embodiment of the present invention；
Fig. 4 is a kind of structural schematic diagram of the device of light-emitting diode chip for backlight unit defects detection provided in an embodiment of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
The embodiment of the invention provides a kind of methods of light-emitting diode chip for backlight unit defects detection.Fig. 1 is the embodiment of the present invention
A kind of flow chart of the method for the light-emitting diode chip for backlight unit defects detection provided.Referring to Fig. 1, this method comprises:
Step 101: obtaining the image of chip to be detected.
In practical applications, can use automatic optics inspection (English: Automated Optical Inspection,
Referred to as: AOI) the equipment image that obtains chip to be detected.AOI equipment is to be set based on optical principle to what workpiece, defect was detected
It is standby.When AOI equipment detects workpiece, camera scanning workpiece can be first passed through, the image of workpiece is obtained；Image is reprocessed,
Check the defect on workpiece.But the defect kind that AOI equipment can identify is reduced, therefore the present invention is set just with AOI
The standby image for getting chip to be detected.
Step 102: handling the image of chip to be detected using Model for Edge Detection, the image of chip to be detected is divided into
The image of epitaxial part and the image of electrode section.
In practical applications, light-emitting diode chip for backlight unit includes epitaxial wafer and the electrode that extension on piece is arranged in.Due to electrode
Laying condition can only see that while the case where epitaxial wafer can also be seen from the front of chip from the front of chip, therefore it is logical
Often by the front of chip, that is, eventually form the image of the image as chip on the surface of electrode.Fig. 2 provides for the embodiment of the present invention
Chip image schematic diagram.Wherein, 10 epitaxial wafer is indicated, 20 indicate electrode.Referring to fig. 2, can see from chip image
The case where electrode at the top of chip is set, it can be seen that the case where the epitaxial wafer of electrode uncovered area exposing.
Since epitaxial wafer is light transmission, electrode is lighttight, and the image of chip is to obtain light directive chip,
Therefore epitaxial part and electrode section differ greatly in the image of chip, directly can be by core to be detected using Model for Edge Detection
The image of piece is divided into the image of epitaxial part and the image of electrode section.
It illustratively, can be using in difference edge detection algorithm, Reborts algorithm, Sobel algorithm, Canny algorithm
A kind of image by chip to be detected is divided into the image of epitaxial part and the image of electrode section.
Step 103: the image of epitaxial part being compared with the image of zero defect epitaxial wafer, from the image of epitaxial part
The image of middle interception epitaxy defect part, and the image of epitaxy defect part is inputted into the first convolutional neural networks, obtain extension
The defect type of defect part, the parameter of the first convolutional neural networks is by using multiple epitaxy defects for demarcating defective types
Image is trained to obtain.
In the present embodiment, the image of zero defect epitaxial wafer is the image for not having defective epitaxial wafer.Do not have defective outer
Prolonging piece can be determined by the way of artificial detection, and the image of epitaxial wafer can use the acquisition of AOI equipment.It should be noted that
Since the image of epitaxial part is to mark off to come from the image of chip to be detected, can be lacked in the image of epitaxial part
The image of electrode section, and the image of zero defect epitaxial wafer generally can uniformly select the image of entire epitaxial wafer, to avoid being directed to
Electrode image of different shapes accordingly replaces extension picture of different shapes, is only needed at this time by the image and nothing of epitaxial part
The image of corresponding portion is compared in defect epitaxial wafer.
If the epitaxial part of fruit chip does not have defect, then corresponding portion in the image of epitaxial part and zero defect epitaxial wafer
Image is the same；Such as the epitaxial part existing defects of fruit chip, then the image of extension defect part can be in the image of epitaxial part
The image of corresponding portion is different in zero defect epitaxial wafer, while in the image of epitaxial part outside the image and zero defect of other parts
The image for prolonging corresponding portion in piece is the same.Therefore, directly the image of epitaxial part and the image of zero defect epitaxial wafer are compared
Compared with, find out part different from zero defect epitaxial wafer in the epitaxial part of chip, can be obtained in the epitaxial part of chip lack
The image of concave portion point.
In specific implementation, successively compare the characteristic value of each pixel and zero defect epitaxial wafer in the image of epitaxial part
The characteristic value of same position pixel on image, selected characteristic value difference or difference are more than all pixels of setting range, and will
Two pixels that mutual distance is no more than setting value in the pixel of selection are classified as the image of the same defect part.
For example, first compare the characteristic value of the 1st column pixel of the 1st row in the image of epitaxial part, the figure with zero defect epitaxial wafer
Whether the characteristic value of same position pixel is identical as in or difference is more than setting range；Compare in the image of epitaxial part again
Whether the characteristic value of same position pixel is identical or poor in the characteristic value of the 2nd column pixel of 1 row, with the image of zero defect epitaxial wafer
Value is more than setting range；……；The characteristic value for comparing in the image of epitaxial part the 1st row last column pixel, outside zero defect
Prolong same position pixel in the image of piece whether characteristic value identical or difference is more than setting range；Compare the figure of epitaxial part
Whether the characteristic value of the 1st column pixel of the 2nd row, identical as the characteristic value of same position pixel in the image of zero defect epitaxial wafer as in
Or difference is more than setting range；……；The characteristic value for comparing last column pixel of last line in the image of epitaxial part, with
Whether the characteristic value of same position pixel is identical in the image of zero defect epitaxial wafer or difference is more than setting range.
If the characteristic value of the i-th row jth column pixel is identical with the image of zero defect epitaxial wafer in the image of epitaxial part
The characteristic value of position pixel is different or difference is more than setting range, while the n-th column of m row pixel in the image of epitaxial part
Characteristic value is different from the characteristic value of same position pixel in the image of zero defect epitaxial wafer or difference is more than setting range, i, j,
M and n is positive integer, i < a, j < b, m < a, n < b.If between the i-th row jth column pixel and the i-th row jth column pixel away from
From setting value is no more than, then the i-th row jth column pixel and i-th row (j+1) column pixel in the image of epitaxial part are classified as same
The image of a epitaxy defect part；If the distance between the i-th row jth column pixel and the i-th row jth column pixel are more than setting value,
The i-th row jth column pixel and i-th row (j+1) column pixel in the image of epitaxial part are then classified as two different epitaxy defects portions
The image divided.
In addition, convolutional neural networks (Convolutional Neural Network, abbreviation CNN) are artificial neural networks
One kind, it has also become current speech analysis and field of image recognition research hotspot.Its weight is shared network structure and is allowed to more
Similar to biological neural network, the complexity of network model is reduced, reduces the quantity of weight.Input of the advantage in network
What is showed when being multidimensional image becomes apparent, and avoids image in tional identification algorithm directly as the input of network
Complicated feature extraction and data reconstruction processes.Convolutional neural networks are one multilayer of special designing for identification two-dimensional shapes
Perceptron, this network structure have height invariance to the deformation of translation, inclination or other forms.
Input is set as the image of epitaxy defect part on the basis of general convolutional neural networks by the present embodiment,
Output is set as the defect type of epitaxy defect part, and using the epitaxy defect image of multiple defective types of calibration to convolution
Parameter in neural network is trained, and the first convolutional neural networks are obtained.
Step 104: the image of electrode section being compared with the image of intact collapsible electrode, from the image of electrode section
The image of electrode defects part is intercepted, and the image of electrode defects part is inputted into the second convolutional neural networks, electrode is obtained and lacks
The defect type of concave portion point, the parameter of the second convolutional neural networks is by using multiple electrode defects figures for demarcating defective types
As being trained to obtain.
In the present embodiment, the image of intact collapsible electrode is the image for not having defective electrode.Do not have defective electrode can
It is determined in a manner of using artificial detection, the image of chip where the image of electrode can obtain electrode first with AOI equipment, then
It is divided and is obtained from the image of chip using Model for Edge Detection.It should be noted that since there may be not for the shape of electrode
Together, than if any only pad, some includes pad and electrode wires, therefore would generally be for the shape of electrode in chip to be detected
Shape, the image of the Yu Xianxuanding intact collapsible electrode being compared.
It is similar with epitaxial part as specific comparison procedure, successively compare the spy of each pixel in the image of electrode section
The characteristic value of same position pixel on the image of value indicative and intact collapsible electrode, selected characteristic value is different or difference is more than setting model
The all pixels enclosed, and two pixels that distance mutual in the pixel of selection is no more than setting value are classified as same lack
The image of concave portion point.
In addition, the second convolutional neural networks are also similar with the first convolutional neural networks, in general convolutional neural networks
On the basis of, input is set as to the image of electrode defects part, output is set as the defect type of electrode defects part, and uses
The electrode defects image of multiple defective types of calibration trains the parameter in convolutional neural networks, obtains the second convolution
Neural network.
It should be noted that the no sequencing of the execution of step 103 and step 104, can first carry out step 103 and hold again
Row step 104 can also first carry out step 104 and execute step 103 again, may also be performed simultaneously step 103 and step 104.
The embodiment of the present invention, will be to be detected first with Model for Edge Detection by after the image for obtaining chip to be detected
The image of chip is divided into the image of epitaxial part and the image of electrode section, can be directed to the different defect situation in two parts
It is respectively processed, simplifies, improve treatment effeciency.The image of the image of epitaxial part and electrode section is carried out respectively again
Similar processing first compares the image of epitaxial part and the image of zero defect epitaxial wafer by taking the image of epitaxial part as an example
Compared with, the image of epitaxial wafer defect part is therefrom intercepted out, can be for the judgement of defect part progress defect type, exclusion is non-to be lacked
The influence of concave portion point or other defect parts, it is with strong points, be conducive to improve accuracy and reduce operand.It recycles multiple
The first convolutional neural networks that the epitaxy defect image of defective type trains are demarcated, are obtained by the image of epitaxy defect part
Its defect type can greatly improve detection efficiency there is no repeatedly comparing.And the parameter of convolutional neural networks is
Using various defect images train come, various defect types, defect the various forms of expression can all be related to, testing result
Accuracy can guarantee, especially meet industrial demand.
The embodiment of the invention provides the methods of another light-emitting diode chip for backlight unit defects detection, are shown in FIG. 1 shine
A kind of specific implementation of diode chip for backlight unit defects detection.Fig. 3 is another light-emitting diode chip for backlight unit provided in an embodiment of the present invention
The flow chart of the method for defects detection.Referring to Fig. 3, this method comprises:
Step 201: obtaining multiple epitaxy defect images.
Illustratively, which may include:
Obtain the image of multiple epitaxial wafers；
Successively the image of each epitaxial wafer is compared with the image of zero defect epitaxial wafer, and will be in the image of epitaxial wafer
The part different from the image of zero defect epitaxial wafer is as epitaxy defect image.
In practical applications, the image of the epitaxial wafer can use the acquisition of AOI equipment.
In addition, the training effect (output of such as convolutional neural networks is accurate) of the parameter in order to ensure convolutional neural networks,
Need to obtain a large amount of defect image to be trained, such as various defect types epitaxy defect image quantity 1000 with
On, so that the output of the first convolutional neural networks is accurate.
Optionally, after step 201, this method can also include:
Extension defect image is normalized, the epitaxy defect image of predetermined dimension is obtained.
Wherein, predetermined dimension may include the size or format of setting.
In practical applications, in order to guarantee that convolutional neural networks export the accuracy of result, all inputs are subjected to normalizing
Change processing, for example it is normalized to the image of 64 pixel *, 64 pixel, convolution can be caused refreshing to avoid due to the inconsistent of input picture
Error is generated through network output result.
Step 202: receiving the defect type of each epitaxy defect image calibration.
In practical applications, in order to be adjusted when convolutional neural networks export incorrect to the parameter of convolutional neural networks
It is whole, while getting epitaxy defect image, the defect type of the also correspondingly received each epitaxy defect image artificially demarcated
(such as respectively indicating different defect types with different numbers), the defect type artificially demarcated with basis is to convolutional neural networks
Parameter is adjusted, and the defect type for exporting convolutional neural networks is identical as received defect type.
Step 203: using the defect type of multiple electrodes defect image and each electrode defects image calibration, to the first volume
The parameter of product neural network is trained.
Since people is from part to the overall situation to extraneous cognition, and the space relationship of image is also that local pixel is more tight
It is close, it is then weaker apart from farther away pixel interdependence.Therefore, each neuron of neural network is not necessarily in fact to global image
It is perceived, it is only necessary to part be perceived, then get up the informix of part just to have obtained the overall situation in higher
Information.
Convolutional neural networks are using the convolutional layer (alternating being generally arranged at close to network inputs end position
Convolutional layer) it realizes to locally perceiving, being connected using being generally arranged at close to the complete of network output position
Layer is connect to realize the informix of part.
In the concrete realization, convolutional layer does convolution using at least one convolution kernel on the image, extracts each office in image
The feature of portion sensing region, the characteristics of image that different types of convolution kernel extracts are different.Full articulamentum is establish one layer each
The connection of a neuron and next layer of all neurons.All inputs of full articulamentum and any one output meet following public
Formula:
H=f (∑iWi*xi+b)；
Wherein, h is the output of full articulamentum, and b is bias, xiFor each input of full articulamentum, WiFor full articulamentum
Each weight for being input to output, i indicate any one input of full articulamentum, ∑iIndicate all inputs to full articulamentum
Summation, f () representative function relationship, generally sigmoid function or tanh function.
The parameter of convolutional neural networks may include the convolution kernel that convolutional layer uses in the present embodiment, and full articulamentum is adopted
Weight and bias.In addition, convolutional neural networks can also include being generally arranged at volume in addition to convolutional layer and full articulamentum
The pond layer (pooling layer) of lamination output end, for reducing the dimension of characteristics of image.Spy of the pond layer to different location
Sign carries out aggregate statistics.If convolutional neural networks further include pond layer, the parameter of convolutional neural networks can also include pond
Change the unit that layer is divided.
In practical applications, the initial value of the parameter of convolutional neural networks can be randomly provided, and then successively be lacked each
It falls into image and inputs convolutional neural networks, and after each defect image inputs convolutional neural networks, obtain convolutional neural networks
The defect type of output is compared with the defect type of calibration, when the two does not adjust the parameter of convolutional neural networks simultaneously, with
Keep the two identical.I.e. the step 203 may include:
Multiple epitaxy defect images are successively inputted into the first convolutional neural networks, obtain the defect of each epitaxy defect image
Type, and in the defect type difference of defect type and calibration that extension defect image obtains, backpropagation adjusts the first volume
The parameter of product neural network, until the defect type that multiple epitaxy defect images obtain is identical as the defect type of calibration.
For example, the 1st epitaxy defect image is first inputted the first convolutional neural networks, the 1st epitaxy defect image is obtained
Defect type, if the defect type that the 1st epitaxy defect image obtains keeps first as the defect type of calibration
The parameter constant of convolutional neural networks；If the 1st obtained defect type of epitaxy defect image and the defect type of calibration are not
Together, then the parameter that the first convolutional neural networks are adjusted according to the sequence of setting, until the defect that the 1st epitaxy defect image obtains
Type is as the defect type of calibration.The 2nd epitaxy defect image is inputted into the first convolutional neural networks again, obtains the 2nd
The defect type of epitaxy defect image, if the defect type one of the 2nd the obtained defect type of epitaxy defect image and calibration
Sample then keeps the parameter constant of the first convolutional neural networks；If defect type and calibration that the 2nd epitaxy defect image obtains
Defect type it is different, then the parameter of the first convolutional neural networks is adjusted according to the sequence of setting, until the 2nd epitaxy defect figure
The last one epitaxy defect image is finally inputted into the first volume as obtained defect type is as the defect type of calibration ...
Product neural network, obtains the defect type of the last one epitaxy defect image, if what the last one epitaxy defect image obtained
Defect type then keeps the parameter constant of the first convolutional neural networks as the defect type of calibration；If outside the last one
Prolong that the defect type that defect image obtains is different from the defect type of calibration, then adjusts the first convolutional Neural according to the sequence of setting
The parameter of network, the defect type that the last an epitaxy defect image obtains is as the defect type of calibration.Then it weighs
Multiple above-mentioned circulation, the parameter constant until being always maintained at the first convolutional neural networks in a circulation, then the first convolution nerve net
Network training finishes.
Further, backpropagation adjusts the parameter of convolutional neural networks, may include:
Using the parameter of gradient descent method adjustment convolutional neural networks.
In practical applications, can gradually adjust and (be gradually increased or be gradually reduced) parameter of convolutional neural networks, such as after
The continuous parameter for increasing or reducing convolutional neural networks, until obtained defect type is as the defect type of calibration.
It should be noted that step 201- step 203 is optional step, may be implemented pair by step 201- step 203
The training of the parameter of first convolutional neural networks.
Step 204: obtaining multiple electrodes defect image.
Illustratively, which may include:
Obtain the image of multiple chips；
The image of multiple chips is handled using Model for Edge Detection, and the image of each chip is divided into epitaxial part respectively
Image and electrode section image；
Successively the image for the electrode section that the image of each chip marks off is compared with the image of intact collapsible electrode,
And using part different from the image of intact collapsible electrode in the image of electrode section as electrode defects image.
In practical applications, the image of the chip also can use the acquisition of AOI equipment.
In addition, as previously mentioned, parameter in order to ensure convolutional neural networks training effect (such as convolutional neural networks it is defeated
It is accurate out), it needs to obtain a large amount of defect image and is trained.For example, the quantity of the electrode defects image of various defect types
At 1000 or more, so that the output of the second convolutional neural networks is accurate.
Optionally, after step 204, this method can also include:
Electrode defects image is normalized, the electrode defects image of predetermined dimension is obtained.
Wherein, predetermined dimension may include the size or format of setting.
In practical applications, in order to guarantee that convolutional neural networks export the accuracy of result, all inputs are subjected to normalizing
Change processing, for example it is normalized to the image of 64 pixel *, 64 pixel, convolution can be caused refreshing to avoid due to the inconsistent of input picture
Error is generated through network output result.
Step 205: receiving the defect type of each electrode defects image calibration.
Optionally, which can be similar with step 202, and this will not be detailed here.
Step 206: using the defect type of multiple electrodes defect image and each electrode defects image calibration, to volume Two
Product neural network is trained.
Optionally, which can be similar with step 203, i.e., the step 206 may include:
Multiple electrodes defect image is successively inputted into the second convolutional neural networks, obtains the defect of each electrode defects image
Type, and in the defect type difference of defect type and calibration that electrode defects image obtains, backpropagation adjusts volume Two
The parameter of product neural network, until the defect type that multiple electrodes defect image obtains is identical as the defect type of calibration.
It should be noted that step 204- step 206 is optional step, may be implemented pair by step 204- step 206
The training of the parameter of second convolutional neural networks.In addition, the execution of step 201- step 203 and step 204- step 206 does not have
Sequencing can first carry out step 201- step 203 and execute step 204- step 206 again, can also first carry out step 204-
Step 206 executes step 201- step 203 again, may also be performed simultaneously step 201- step 203 and step 204- step 206.
Step 207: obtaining the image of chip to be detected.
Illustratively, which can be identical as step 101, and this will not be detailed here.
Step 208: handling the image of chip to be detected using Model for Edge Detection, the image of chip to be detected is divided into
The image of epitaxial part and the image of electrode section.
Illustratively, which can be identical as step 102, and this will not be detailed here.
Step 209: the image of epitaxial part being compared with the image of zero defect epitaxial wafer, from the image of epitaxial part
The image of middle interception epitaxy defect part, and the image of epitaxy defect part is inputted into the first convolutional neural networks, obtain extension
The defect type of defect part, the parameter of the first convolutional neural networks is by using multiple epitaxy defects for demarcating defective types
Image is trained to obtain.
Optionally, the image of epitaxy defect part is inputted into the first convolutional neural networks, obtains lacking for epitaxy defect part
Type is fallen into, may include:
The image of extension defect part is normalized, the epitaxy defect image of predetermined dimension is obtained；
The epitaxy defect image of predetermined dimension is inputted into the first convolutional neural networks, obtains the defect class of epitaxy defect part
Type.
It is corresponding with step 201, image is normalized and inputs convolutional neural networks again, it can be to avoid due to input
The inconsistent of image causes convolutional neural networks output result to generate error.
Step 210: the image of electrode section being compared with the image of intact collapsible electrode, from the image of electrode section
The image of electrode defects part is intercepted, and the image of electrode defects part is inputted into the second convolutional neural networks, electrode is obtained and lacks
The defect type of concave portion point, the parameter of the second convolutional neural networks is by using multiple electrode defects figures for demarcating defective types
As being trained to obtain.
Optionally, the image of electrode defects part is inputted into the second convolutional neural networks, obtains lacking for electrode defects part
Type is fallen into, may include:
The image of electrode defects part is normalized, the electrode defects image of predetermined dimension is obtained；
The electrode defects image of predetermined dimension is inputted into the second convolutional neural networks, obtains the defect class of electrode defects part
Type.
It is corresponding with step 204, image is normalized and inputs convolutional neural networks again, it can be to avoid due to input
The inconsistent of image causes convolutional neural networks output result to generate error.
Step 211: counting quantity, size and the defect of the image of the epitaxy defect part intercepted in the image of epitaxial part
Quantity, size and the defect type of the image of the electrode defects part intercepted in the image of type and electrode section, determine to
The credit rating of detection chip.
In practical applications, can according in the target level of product quality of industry universal to defects count, size and type etc.
Chip to be detected is divided into corresponding grade by the demand of aspect according to statistical result.
It should be noted that step 211 is optional step, the screening of chip may be implemented, by step 211 so as to core
Piece carries out different processing.For example, for have good quality (such as defective proportion be lower than 10%) chip and up-to-standard (such as defect
Ratio is between 20%~30%) chip carry out different packing, for it is off quality (such as defective proportion 40% with
On) chip abandon etc..
The embodiment of the invention provides a kind of device of light-emitting diode chip for backlight unit defects detection, be adapted to carry out Fig. 1 or
The method of light-emitting diode chip for backlight unit defects detection shown in Fig. 3.Fig. 4 is a kind of light-emitting diodes tube core provided in an embodiment of the present invention
The structural schematic diagram of the device of piece defects detection.Referring to fig. 4, which includes:
Chip image obtains module 301, for obtaining the image of chip to be detected；
Image division module 302, for handling the image of chip to be detected using Model for Edge Detection, by chip to be detected
Image be divided into the image of epitaxial part and the image of electrode section；
Extension image processing module 303, for the image of epitaxial part to be compared with the image of zero defect epitaxial wafer,
The image of epitaxy defect part is intercepted from the image of epitaxial part, and the image of epitaxy defect part is inputted into the first convolution mind
Through network, the defect type of epitaxy defect part is obtained, the parameter of the first convolutional neural networks has scarce by using multiple calibration
The epitaxy defect image for falling into type is trained to obtain；
Electrode image processing module 304, for the image of electrode section to be compared with the image of intact collapsible electrode, from
The image of electrode defects part is intercepted in the image of electrode section, and the image of electrode defects part is inputted into the second convolutional Neural
Network, obtains the defect type of electrode defects part, and the parameter of the second convolutional neural networks is defective by using multiple calibration
The electrode defects image of type is trained to obtain.
Optionally, extension image processing module 303 may include:
Extension image normalization submodule, is normalized for the image to extension defect part, obtains predetermined dimension
Extension image；
Epitaxy defect type determination module, for the extension image of predetermined dimension to be inputted the first convolutional neural networks,
Obtain the defect type of epitaxy defect part.
Correspondingly, electrode image processing module 304 may include:
Electrode image normalization submodule, is normalized for the image to electrode defects part, obtains predetermined dimension
Electrode image；
Electrode defects type determination module, for the electrode image of predetermined dimension to be inputted the second convolutional neural networks,
Obtain the defect type of electrode defects part.
Optionally, which can also include:
Epitaxy defect image collection module, for obtaining multiple epitaxy defect images；
Epitaxy defect type reception module, for receiving the defect type of each epitaxy defect image calibration；
Epitaxy defect type training module, for using multiple epitaxy defect images and each epitaxy defect image calibration
Defect type is trained the first convolutional neural networks.
Further, epitaxy defect type training module can be used for,
Multiple epitaxy defect images are successively inputted into the first convolutional neural networks, obtain the defect of each epitaxy defect image
Type, and in the defect type difference of defect type and calibration that extension defect image obtains, backpropagation adjusts the first volume
The parameter of product neural network, until the defect type that multiple epitaxy defect images obtain is identical as the defect type of calibration.
Optionally, which can also include:
Electrode defects image collection module, for obtaining multiple electrodes defect image；
Electrode defects type reception module, for receiving the defect type of each electrode defects image calibration；
Electrode defects type training module, for using multiple electrodes defect image and each electrode defects image calibration
Defect type is trained the second convolutional neural networks.
Further, electrode defects type training module can be used for,
Multiple electrodes defect image is successively inputted into the second convolutional neural networks, obtains the defect of each electrode defects image
Type, and in the defect type difference of defect type and calibration that electrode defects image obtains, backpropagation adjusts volume Two
The parameter of product neural network, until the defect type that multiple electrodes defect image obtains is identical as the defect type of calibration.
Optionally, which can also include:
Determining module, quantity, the size of the image of the epitaxy defect part intercepted in the image for counting epitaxial part
And quantity, size and the defect type of the image of the electrode defects part intercepted in defect type and the image of electrode section,
Determine the credit rating of chip to be detected.
It should be understood that the device of light-emitting diode chip for backlight unit defects detection provided by the above embodiment is in detection luminous two
When pole pipe chip defect, only the example of the division of the above functional modules, in practical application, can according to need and
Above-mentioned function distribution is completed by different functional modules, i.e., the internal structure of device is divided into different functional modules, with
Complete all or part of function described above.In addition, light-emitting diode chip for backlight unit defects detection provided by the above embodiment
The embodiment of the method for device and light-emitting diode chip for backlight unit defects detection belongs to same design, and specific implementation process is detailed in method reality
Example is applied, which is not described herein again.
Those of ordinary skill in the art will appreciate that realizing that all or part of the steps of above-described embodiment can pass through hardware
It completes, relevant hardware can also be instructed to complete by program, the program can store in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only memory, disk or CD etc..
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of method of light-emitting diode chip for backlight unit defects detection, which is characterized in that the described method includes:
Obtain the image of chip to be detected；
The image of the chip to be detected is divided into extension by the image that the chip to be detected is handled using Model for Edge Detection
The image of partial image and electrode section；
The image of the epitaxial part is compared with the image of zero defect epitaxial wafer, is cut from the image of the epitaxial part
The image of epitaxy defect part is taken, and the image of the epitaxy defect part is inputted into the first convolutional neural networks, is obtained described
The defect type of epitaxy defect part, the parameter of first convolutional neural networks is by using multiple defective types of calibration
Epitaxy defect image is trained to obtain；
The image of the electrode section is compared with the image of intact collapsible electrode, is intercepted from the image of the electrode section
The image of electrode defects part, and the image of the electrode defects part is inputted into the second convolutional neural networks, obtain the electricity
The defect type of pole defect part, the parameter of second convolutional neural networks is by using multiple electricity for demarcating defective types
Pole defect image is trained to obtain.
2. the method according to claim 1, wherein the image by the electrode defects part inputs second
Convolutional neural networks obtain the defect type of the electrode defects part, comprising:
The image of the electrode defects part is normalized, the electrode image of predetermined dimension is obtained；
The electrode image of the predetermined dimension is inputted into second convolutional neural networks, obtains lacking for the electrode defects part
Fall into type.
3. method according to claim 1 or 2, which is characterized in that the method also includes:
Obtain multiple electrodes defect image；
Receive the defect type of each electrode defects image calibration；
Using the defect type of multiple electrode defects images and each electrode defects image calibration, to the volume Two
Product neural network is trained.
4. according to the method described in claim 3, it is characterized in that, described use multiple electrode defects images and each institute
The defect type for stating electrode defects image calibration is trained second convolutional neural networks, comprising:
Multiple electrode defects images are successively inputted into second convolutional neural networks, obtain each electrode defects figure
The defect type of picture, and in the obtained defect type of electrode defects image and the defect type difference of calibration, it is reversed to pass
The parameter for adjusting second convolutional neural networks is broadcast, the defect type obtained until multiple electrode defects images and calibration
Defect type it is identical.
5. method according to claim 1 or 2, which is characterized in that the method also includes:
Quantity, size and the defect type of the image of the epitaxy defect part intercepted in the image of the epitaxial part are counted, with
And quantity, size and the defect type of the image of the electrode defects part intercepted in the image of the electrode section, determine described in
The credit rating of chip to be detected.
6. a kind of device of light-emitting diode chip for backlight unit defects detection, which is characterized in that described device includes:
Chip image obtains module, for obtaining the image of chip to be detected；
Image division module, for handling the image of the chip to be detected using Model for Edge Detection, by the core to be detected
The image of piece is divided into the image of epitaxial part and the image of electrode section；
Extension image processing module, for the image of the epitaxial part to be compared with the image of zero defect epitaxial wafer, from
The image of epitaxy defect part is intercepted in the image of the epitaxial part, and the image of the epitaxy defect part is inputted first
Convolutional neural networks obtain the defect type of the epitaxy defect part, and the parameter of first convolutional neural networks is by adopting
It is trained to obtain with the epitaxy defect image of multiple defective types of calibration；
Electrode image processing module, for the image of the electrode section to be compared with the image of intact collapsible electrode, from institute
The image for intercepting electrode defects part in the image of electrode section is stated, and the image of the electrode defects part is inputted into volume Two
Product neural network, obtains the defect type of the electrode defects part, the parameters of second convolutional neural networks by using
The electrode defects image of multiple defective types of calibration is trained to obtain.
7. device according to claim 6, which is characterized in that the electrode image processing module includes:
Electrode image normalization submodule is normalized for the image to the electrode defects part, obtains predetermined dimension
Electrode image；
Electrode defects type determination module, for the electrode image of the predetermined dimension to be inputted the second convolution nerve net
Network obtains the defect type of the electrode defects part.
8. device according to claim 6 or 7, which is characterized in that described device further include:
Electrode defects image collection module, for obtaining multiple electrodes defect image；
Electrode defects type reception module, for receiving the defect type of each electrode defects image calibration；
Electrode defects type training module, for using multiple electrode defects images and each electrode defects image mark
Fixed defect type is trained second convolutional neural networks.
9. device according to claim 8, which is characterized in that the electrode defects type training module is used for,
Multiple electrode defects images are successively inputted into second convolutional neural networks, obtain each electrode defects figure
The defect type of picture, and in the obtained defect type of electrode defects image and the defect type difference of calibration, it is reversed to pass
The parameter for adjusting second convolutional neural networks is broadcast, the defect type obtained until multiple electrode defects images and calibration
Defect type it is identical.
10. device according to claim 6 or 7, which is characterized in that described device further include:
Determining module, quantity, the size of the image of the epitaxy defect part intercepted in the image for counting the epitaxial part
And quantity, size and the defect of the image of the electrode defects part intercepted in defect type and the image of the electrode section
Type determines the credit rating of the chip to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910065080.1A CN109919908B (en) | 2019-01-23 | 2019-01-23 | Method and device for detecting defects of light-emitting diode chip |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910065080.1A CN109919908B (en) | 2019-01-23 | 2019-01-23 | Method and device for detecting defects of light-emitting diode chip |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109919908A true CN109919908A (en) | 2019-06-21 |
| CN109919908B CN109919908B (en) | 2020-11-10 |
Family
ID=66960697
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910065080.1A Active CN109919908B (en) | 2019-01-23 | 2019-01-23 | Method and device for detecting defects of light-emitting diode chip |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919908B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110930390A (en) * | 2019-11-22 | 2020-03-27 | 郑州智利信信息技术有限公司 | Chip pin missing detection method based on semi-supervised deep learning |
| CN111080615A (en) * | 2019-12-12 | 2020-04-28 | 创新奇智(重庆)科技有限公司 | PCB defect detection system and detection method based on convolutional neural network |
| CN111145164A (en) * | 2019-12-30 | 2020-05-12 | 上海感图网络科技有限公司 | IC chip defect detection method based on artificial intelligence |
| CN112304960A (en) * | 2020-12-30 | 2021-02-02 | 中国人民解放军国防科技大学 | High-resolution image object surface defect detection method based on deep learning |
| CN113870202A (en) * | 2021-09-18 | 2021-12-31 | 中科海拓(无锡)科技有限公司 | Far-end chip defect detection system based on deep learning technology |
| CN114066890A (en) * | 2022-01-14 | 2022-02-18 | 成都数联云算科技有限公司 | Gear defect detection method and device, computer equipment and storage medium |
| CN114397306A (en) * | 2022-03-25 | 2022-04-26 | 南方电网数字电网研究院有限公司 | Power grid grading ring hypercomplex category defect multi-stage model joint detection method |
| US11656181B2 (en) | 2018-12-21 | 2023-05-23 | Industrial Technology Research Institute | Inspection apparatus and inspection method for inspecting light-emitting diodes |
| CN116777894A (en) * | 2023-07-05 | 2023-09-19 | 山东睿芯半导体科技有限公司 | Method and device for detecting surface defects of chip |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009236593A (en) * | 2008-03-26 | 2009-10-15 | Mitsubishi Electric Corp | Visual inspection supporting device |
| CN101666923A (en) * | 2008-09-02 | 2010-03-10 | 北京京东方光电科技有限公司 | Processing method, processing system and maintenance equipment for defects of liquid crystal display panel |
| CN103278511A (en) * | 2013-05-17 | 2013-09-04 | 南京大学 | Wafer defect detection method based on multi-scale corner feature extraction |
| CN104597057A (en) * | 2015-02-02 | 2015-05-06 | 东华大学 | Columnar diode surface defect detection device based on machine vision |
| CN106504238A (en) * | 2016-10-31 | 2017-03-15 | 成都交大光芒科技股份有限公司 | Railway contact line defect inspection method based on image procossing and convolutional neural networks |
| CN107408209A (en) * | 2014-12-03 | 2017-11-28 | 科磊股份有限公司 | Without the classification of the automatic defect of sampling and feature selecting |
| CN107563999A (en) * | 2017-09-05 | 2018-01-09 | 华中科技大学 | A kind of chip defect recognition methods based on convolutional neural networks |
| CN107607554A (en) * | 2017-09-26 | 2018-01-19 | 天津工业大学 | A kind of Defect Detection and sorting technique of the zinc-plated stamping parts based on full convolutional neural networks |
| CN107611046A (en) * | 2017-07-28 | 2018-01-19 | 华灿光电（浙江）有限公司 | Inner layer defect detection device and method |
| CN108061735A (en) * | 2017-12-01 | 2018-05-22 | 工业互联网创新中心（上海）有限公司 | The recognition methods of component surface defect and device |
| CN108982508A (en) * | 2018-05-23 | 2018-12-11 | 江苏农林职业技术学院 | A kind of plastic-sealed body IC chip defect inspection method based on feature templates matching and deep learning |
| CN109242033A (en) * | 2018-09-21 | 2019-01-18 | 长鑫存储技术有限公司 | Wafer defect method for classifying modes and device, storage medium, electronic equipment |
-
2019
- 2019-01-23 CN CN201910065080.1A patent/CN109919908B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009236593A (en) * | 2008-03-26 | 2009-10-15 | Mitsubishi Electric Corp | Visual inspection supporting device |
| CN101666923A (en) * | 2008-09-02 | 2010-03-10 | 北京京东方光电科技有限公司 | Processing method, processing system and maintenance equipment for defects of liquid crystal display panel |
| CN103278511A (en) * | 2013-05-17 | 2013-09-04 | 南京大学 | Wafer defect detection method based on multi-scale corner feature extraction |
| CN107408209A (en) * | 2014-12-03 | 2017-11-28 | 科磊股份有限公司 | Without the classification of the automatic defect of sampling and feature selecting |
| CN104597057A (en) * | 2015-02-02 | 2015-05-06 | 东华大学 | Columnar diode surface defect detection device based on machine vision |
| CN106504238A (en) * | 2016-10-31 | 2017-03-15 | 成都交大光芒科技股份有限公司 | Railway contact line defect inspection method based on image procossing and convolutional neural networks |
| CN107611046A (en) * | 2017-07-28 | 2018-01-19 | 华灿光电（浙江）有限公司 | Inner layer defect detection device and method |
| CN107563999A (en) * | 2017-09-05 | 2018-01-09 | 华中科技大学 | A kind of chip defect recognition methods based on convolutional neural networks |
| CN107607554A (en) * | 2017-09-26 | 2018-01-19 | 天津工业大学 | A kind of Defect Detection and sorting technique of the zinc-plated stamping parts based on full convolutional neural networks |
| CN108061735A (en) * | 2017-12-01 | 2018-05-22 | 工业互联网创新中心（上海）有限公司 | The recognition methods of component surface defect and device |
| CN108982508A (en) * | 2018-05-23 | 2018-12-11 | 江苏农林职业技术学院 | A kind of plastic-sealed body IC chip defect inspection method based on feature templates matching and deep learning |
| CN109242033A (en) * | 2018-09-21 | 2019-01-18 | 长鑫存储技术有限公司 | Wafer defect method for classifying modes and device, storage medium, electronic equipment |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11656181B2 (en) | 2018-12-21 | 2023-05-23 | Industrial Technology Research Institute | Inspection apparatus and inspection method for inspecting light-emitting diodes |
| CN110930390A (en) * | 2019-11-22 | 2020-03-27 | 郑州智利信信息技术有限公司 | Chip pin missing detection method based on semi-supervised deep learning |
| CN110930390B (en) * | 2019-11-22 | 2020-09-22 | 深圳市海芯微迅半导体有限公司 | Chip pin missing detection method based on semi-supervised deep learning |
| CN111080615A (en) * | 2019-12-12 | 2020-04-28 | 创新奇智(重庆)科技有限公司 | PCB defect detection system and detection method based on convolutional neural network |
| CN111080615B (en) * | 2019-12-12 | 2023-06-16 | 创新奇智(重庆)科技有限公司 | PCB defect detection system and detection method based on convolutional neural network |
| CN111145164A (en) * | 2019-12-30 | 2020-05-12 | 上海感图网络科技有限公司 | IC chip defect detection method based on artificial intelligence |
| CN112304960A (en) * | 2020-12-30 | 2021-02-02 | 中国人民解放军国防科技大学 | High-resolution image object surface defect detection method based on deep learning |
| CN113870202A (en) * | 2021-09-18 | 2021-12-31 | 中科海拓(无锡)科技有限公司 | Far-end chip defect detection system based on deep learning technology |
| CN114066890A (en) * | 2022-01-14 | 2022-02-18 | 成都数联云算科技有限公司 | Gear defect detection method and device, computer equipment and storage medium |
| CN114397306A (en) * | 2022-03-25 | 2022-04-26 | 南方电网数字电网研究院有限公司 | Power grid grading ring hypercomplex category defect multi-stage model joint detection method |
| CN116777894A (en) * | 2023-07-05 | 2023-09-19 | 山东睿芯半导体科技有限公司 | Method and device for detecting surface defects of chip |
| CN116777894B (en) * | 2023-07-05 | 2023-12-29 | 山东睿芯半导体科技有限公司 | Method and device for detecting surface defects of chip |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109919908B (en) | 2020-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919908A (en) | The method and apparatus of light-emitting diode chip for backlight unit defects detection | |
| US12020417B2 (en) | Method and system for classifying defects in wafer using wafer-defect images, based on deep learning | |
| CN111310756B (en) | Damaged corn particle detection and classification method based on deep learning | |
| US20130322733A1 (en) | System for detection of non-uniformities in web-based materials | |
| TWI834005B (en) | Deep learning networks for nuisance filtering | |
| CN115294473A (en) | Insulator fault identification method and system based on target detection and instance segmentation | |
| CN115439458A (en) | Industrial image defect target detection algorithm based on depth map attention | |
| CN116091505B (en) | Automatic defect detection and classification method and system for sapphire substrate | |
| CN113420619A (en) | Remote sensing image building extraction method | |
| CN114359245A (en) | Method for detecting surface defects of products in industrial scene | |
| CN115861170A (en) | Surface defect detection method based on improved YOLO V4 algorithm | |
| CN116385430A (en) | Machine vision flaw detection method, device, medium and equipment | |
| CN113962929A (en) | Photovoltaic cell assembly defect detection method and system and photovoltaic cell assembly production line | |
| CN117474863A (en) | Chip surface defect detection method for compressed multi-head self-attention neural network | |
| CN114549391A (en) | Circuit board surface defect detection method based on polarization prior | |
| CN109493315A (en) | A kind of detection method of the flaw of polysilicon solar cell | |
| CN114627090A (en) | Convolutional neural network optical lens defect detection method based on attention mechanism | |
| CN109919907A (en) | The method and apparatus of LED epitaxial slice defects detection | |
| CN115830302A (en) | Multi-scale feature extraction and fusion power distribution network equipment positioning identification method | |
| CN116152169A (en) | Printed circuit board defect detection method based on improved YOLOv5s model | |
| CN115471494A (en) | Wo citrus quality inspection method, device, equipment and storage medium based on image processing | |
| CN113469984B (en) | Method for detecting appearance of display panel based on YOLO structure | |
| CN113256576B (en) | Automatic optical element detection system and method based on polarization imaging and machine learning | |
| CN114861771A (en) | Industrial CT image defect classification method based on feature extraction and deep learning | |
| CN114092441A (en) | Product surface defect detection method and system based on dual neural network |
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 |