CN109919908B - Method and device for detecting defects of light-emitting diode chip - Google Patents
Method and device for detecting defects of light-emitting diode chip Download PDFInfo
- Publication number
- CN109919908B CN109919908B CN201910065080.1A CN201910065080A CN109919908B CN 109919908 B CN109919908 B CN 109919908B CN 201910065080 A CN201910065080 A CN 201910065080A CN 109919908 B CN109919908 B CN 109919908B
- Authority
- CN
- China
- Prior art keywords
- image
- defect
- electrode
- epitaxial
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a device for detecting defects of a light-emitting diode chip, and belongs to the technical field of semiconductors. The method comprises the following steps: acquiring an image of a chip to be detected; dividing an image of a chip to be detected into an image of an extension part and an image of an electrode part; intercepting an image of an epitaxial defect part from an image of the epitaxial part, inputting the image of the epitaxial defect part into a first convolution neural network to obtain a defect type of the epitaxial defect part, and training parameters of the first convolution neural network by adopting a plurality of epitaxial defect images with calibrated defect types; and intercepting the image of the electrode defect part from the image of the electrode part, inputting the image of the electrode defect part into a second convolutional neural network to obtain the defect type of the electrode defect part, and training the parameters of the second convolutional neural network by adopting a plurality of electrode defect images with calibrated defect types. The invention particularly meets the requirements of industrial production.
Description
Technical Field
The invention relates to the technical field of semiconductors, in particular to a method and a device for detecting defects of a light-emitting diode chip.
Background
A Light Emitting Diode (LED) is a semiconductor Diode that can convert electrical energy into Light energy. The core component of an LED is a chip that includes an epitaxial wafer and electrodes disposed on the epitaxial wafer.
Since various defects such as hexagon, micro-roughness, scratch, particle, fog, green dot, etc. may be generated during the chip processing, the chip is generally subjected to defect detection after the chip processing, and the chip is classified according to the detection result (including the type, size, number, etc. of the defects).
The most original defect detection method is to manually observe a chip to be detected through a microscope, and the detection efficiency and the identification accuracy rate cannot meet the requirements of industrial production. Later along with the development of optical detection equipment, can acquire the image of waiting to detect the chip, through waiting to detect the image of chip and the image of nondefective chip and comparing, can determine and wait to detect whether there is the defect in the chip, detection efficiency and discernment accuracy have all obtained very big promotion, but the detection capability is limited, can't realize the grade division of chip. Based on the improvement of the processing capability of the computer, the image of the chip to be detected can be compared with the images of various defective chips in sequence, the defect type corresponding to the image of the defective chip with the highest similarity is used as the defect type of the chip to be detected, and then the grade is divided according to the defect type of the chip to be detected.
In the process of implementing the invention, the inventor finds that the prior art has at least the following problems:
the same type of defect on a chip can be expressed in various forms, such as the shape of defects on the hexagonal system, the cubic system and the orthorhombic system is different. However, in the specific implementation, it is basically impossible to provide images representing defects in all forms for comparison with the image of the chip to be detected, the accuracy of the detection result is not high, and the comparison between the image of the chip to be detected and the images representing defects in the same type in various forms is huge, and the detection efficiency is low. If only one image showing the form defects is selected for comparison, the detection result is possibly inaccurate, and the industrial production requirements cannot be met.
Disclosure of Invention
The embodiment of the invention provides a method and a device for detecting defects of a light-emitting diode chip, which can solve the problem that the accuracy of defect detection in the prior art cannot meet the requirement of industrial production. The technical scheme is as follows:
in one aspect, an embodiment of the present invention provides a method for detecting defects of a light emitting diode chip, where the method includes:
acquiring an image of a chip to be detected;
processing the image of the chip to be detected by adopting an edge detection model, and dividing the image of the chip to be detected into an image of an extension part and an image of an electrode part;
comparing the image of the epitaxial part with the image of a defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, and inputting the image of the epitaxial defect part into a first convolution neural network to obtain the defect type of the epitaxial defect part, wherein the parameter of the first convolution neural network is obtained by training a plurality of epitaxial defect images with calibrated defect types;
and comparing the image of the electrode part with the image of the non-defective electrode, intercepting the image of the defective electrode part from the image of the electrode part, inputting the image of the defective electrode part into a second convolutional neural network to obtain the defect type of the defective electrode part, wherein the parameters of the second convolutional neural network are obtained by training the electrode defect images with a plurality of calibrated defect types.
Optionally, the inputting the image of the electrode defect portion into a second convolutional neural network to obtain a defect type of the electrode defect portion includes:
normalizing the image of the electrode defect part to obtain an electrode image with a preset specification;
and inputting the electrode image with the preset specification into the second convolutional neural network to obtain the defect type of the defective part of the electrode.
Optionally, the method further comprises:
acquiring a plurality of electrode defect images;
receiving the defect type calibrated by each electrode defect image;
and training the second convolutional neural network by adopting a plurality of electrode defect images and the defect types calibrated by the electrode defect images.
Further, the training of the second convolutional neural network by using the plurality of electrode defect images and the defect types calibrated by the electrode defect images includes:
and when the defect type obtained by the electrode defect image is different from the calibrated defect type, the parameters of the second convolutional neural network are adjusted in a back propagation mode until the defect type obtained by the electrode defect image is the same as the calibrated defect type.
Optionally, the method further comprises:
and counting the number, the size and the defect type of the images of the epitaxial defect parts intercepted in the images of the epitaxial parts, and the number, the size and the defect type of the images of the electrode defect parts intercepted in the images of the electrode parts, and determining the quality grade of the chip to be detected.
In another aspect, an embodiment of the present invention provides an apparatus for detecting defects of a light emitting diode chip, where the apparatus includes:
the chip image acquisition module is used for acquiring an image of a chip to be detected;
the image dividing module is used for processing the image of the chip to be detected by adopting an edge detection model and dividing the image of the chip to be detected into an image of an extension part and an image of an electrode part;
the epitaxial image processing module is used for comparing the image of the epitaxial part with the image of a defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, inputting the image of the epitaxial defect part into a first convolution neural network to obtain the defect type of the epitaxial defect part, and training the parameters of the first convolution neural network by adopting a plurality of epitaxial defect images with calibrated defect types;
and the electrode image processing module is used for comparing the image of the electrode part with the image of a non-defective electrode, intercepting the image of the defective electrode part from the image of the electrode part, inputting the image of the defective electrode part into a second convolutional neural network to obtain the defect type of the defective electrode part, and training the parameters of the second convolutional neural network by adopting a plurality of electrode defect images with calibrated defect types.
Optionally, the electrode image processing module includes:
the electrode image normalization submodule is used for normalizing the image of the electrode defect part to obtain an electrode image with a preset specification;
and the electrode defect type determining submodule is used for inputting the electrode image with the preset specification into the second convolutional neural network to obtain the defect type of the electrode defect part.
Optionally, the apparatus further comprises:
the electrode defect image acquisition module is used for acquiring a plurality of electrode defect images;
the electrode defect type receiving module is used for receiving the defect type calibrated by each electrode defect image;
and the electrode defect type training module is used for training the second convolutional neural network by adopting a plurality of electrode defect images and the defect types calibrated by the electrode defect images.
Further, the electrode defect type training module is configured to,
and when the defect type obtained by the electrode defect image is different from the calibrated defect type, the parameters of the second convolutional neural network are adjusted in a back propagation mode until the defect type obtained by the electrode defect image is the same as the calibrated defect type.
Optionally, the apparatus further comprises:
and the determining module is used for counting the number, the size and the defect type of the images of the epitaxial defect parts intercepted in the images of the epitaxial parts, and the number, the size and the defect type of the images of the electrode defect parts intercepted in the images of the electrode parts, and determining the quality grade of the chip to be detected.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
after the image of the chip to be detected is obtained, the image of the chip to be detected is divided into the image of the extension part and the image of the electrode part by using the edge detection model, so that different defect conditions of the two parts can be respectively processed, the complexity is reduced, and the processing efficiency is improved. And then, respectively carrying out similar processing on the image of the epitaxial part and the image of the electrode part, taking the image of the epitaxial part as an example, comparing the image of the epitaxial part with the image of the defect-free epitaxial wafer, and intercepting and taking out the image of the defect part of the epitaxial wafer, so that the defect type can be judged according to the defect part, the influence of the non-defect part or other defect parts is eliminated, the pertinence is strong, and the accuracy is improved and the operand is reduced. And then, the defect type of the first convolutional neural network is obtained from the images of the epitaxial defect part by utilizing the first convolutional neural network trained by the epitaxial defect images with the defect types calibrated, so that the problem of multiple comparison is solved, and the detection efficiency can be greatly improved. And the parameters of the convolutional neural network are trained by adopting various defect images, various defect types and various expression forms of the defects can be related, the accuracy of the detection result can be ensured, and the requirements of industrial production are particularly met.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a method for detecting defects of an led chip according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a chip image provided by an embodiment of the invention;
fig. 3 is a flowchart of another method for detecting defects of an led chip according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an apparatus for detecting defects of a light emitting diode chip according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The embodiment of the invention provides a method for detecting defects of a light-emitting diode chip. Fig. 1 is a flowchart of a method for detecting defects of a light emitting diode chip according to an embodiment of the present invention. Referring to fig. 1, the method includes:
step 101: and acquiring an image of the chip to be detected.
In practical application, an image of a chip to be detected can be obtained by using Automatic Optical Inspection (AOI). The AOI equipment is equipment for detecting defects of workpieces based on an optical principle. When the AOI equipment detects a workpiece, the workpiece is scanned by a camera to obtain an image of the workpiece; and processing the image to detect the defects on the workpiece. However, the defect types which can be identified by the AOI equipment are reduced, so that the invention only acquires the image of the chip to be detected by using the AOI equipment.
Step 102: and processing the image of the chip to be detected by adopting an edge detection model, and dividing the image of the chip to be detected into an image of the extension part and an image of the electrode part.
In practical applications, the light emitting diode chip includes an epitaxial wafer and an electrode disposed on the epitaxial wafer. Since the electrode layout can only be seen from the front side of the chip, and the epitaxial wafer can also be seen from the front side of the chip, the image of the front side of the chip, i.e., the surface on which the electrodes are finally formed, is usually taken as the image of the chip. Fig. 2 is a schematic diagram of a chip image according to an embodiment of the present invention. In which 10 denotes an epitaxial wafer and 20 denotes an electrode. Referring to fig. 2, from the chip image, both the case of the electrode disposed on the top of the chip and the case of the epitaxial wafer with the uncovered area of the electrode exposed can be seen.
Because the epitaxial wafer is light-transmitting, the electrode is light-tight, and the image of the chip is obtained by emitting light to the chip, the difference between the epitaxial part and the electrode part in the image of the chip is large, and the image of the chip to be detected can be divided into the image of the epitaxial part and the image of the electrode part by directly utilizing the edge detection model.
Illustratively, one of a differential edge detection algorithm, a Reborts algorithm, a Sobel algorithm, and a Canny algorithm may be adopted to divide the image of the chip to be detected into an image of the extension portion and an image of the electrode portion.
Step 103: and comparing the image of the epitaxial part with the image of the defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, inputting the image of the epitaxial defect part into a first convolution neural network to obtain the defect type of the epitaxial defect part, and training the parameters of the first convolution neural network by adopting a plurality of epitaxial defect images for calibrating the defect type.
In the present embodiment, the image of a defect-free epitaxial wafer is an image of a defect-free epitaxial wafer. The epitaxial wafer without defects can be determined by adopting a manual detection mode, and the image of the epitaxial wafer can be acquired by utilizing an AOI device. It should be noted that, since the image of the epitaxial portion is divided from the image of the chip to be detected, the image of the electrode portion is missing from the image of the epitaxial portion, and the image of the defect-free epitaxial wafer generally selects the image of the entire epitaxial wafer in a unified manner, so as to avoid replacing the images of the epitaxial wafers of different shapes correspondingly for the images of the electrodes of different shapes, and at this time, only the image of the epitaxial portion needs to be compared with the image of the corresponding portion in the defect-free epitaxial wafer.
If the epitaxial part of the chip has no defect, the image of the epitaxial part is the same as that of the corresponding part in the defect-free epitaxial wafer; if the epitaxial portion of the chip has a defect, the image of the epitaxial defect portion in the image of the epitaxial portion may be different from the image of the corresponding portion in the defect-free epitaxial wafer, while the image of the other portion in the image of the epitaxial portion is the same as the image of the corresponding portion in the defect-free epitaxial wafer. Therefore, the image of the defect part in the epitaxial part of the chip can be obtained by directly comparing the image of the epitaxial part with the image of the defect-free epitaxial wafer and finding out the part of the epitaxial part of the chip, which is different from the part of the defect-free epitaxial wafer.
In the specific implementation, the characteristic value of each pixel in the image of the epitaxial part is sequentially compared with the characteristic value of the pixel at the same position on the image of the defect-free epitaxial wafer, all pixels with different characteristic values or difference values exceeding a set range are selected, and two pixels, the distance between which does not exceed a set value, in the selected pixels are classified into the image of the same defect part.
For example, firstly comparing the characteristic value of the 1 st row and 1 st column pixel in the image of the epitaxial part, and whether the characteristic value of the same position pixel in the image of the defect-free epitaxial wafer is the same or the difference value exceeds a set range; comparing the characteristic values of the pixels in the 1 st row and the 2 nd column in the image of the epitaxial part, and judging whether the characteristic values of the pixels at the same positions in the image of the defect-free epitaxial wafer are the same or whether the difference value exceeds a set range; … …, respectively; comparing the characteristic values of the pixels in the last column of the 1 st row in the image of the epitaxial part, and judging whether the characteristic values of the pixels at the same position in the image of the defect-free epitaxial wafer are the same or whether the difference value exceeds a set range; comparing the characteristic values of the pixels in the 2 nd row and the 1 st column in the image of the epitaxial part, and judging whether the characteristic values of the pixels at the same positions in the image of the defect-free epitaxial wafer are the same or whether the difference value exceeds a set range; … …, respectively; and comparing the characteristic values of the last row and the last column of pixels in the image of the epitaxial part, and judging whether the characteristic values of the pixels at the same positions in the image of the defect-free epitaxial wafer are the same or whether the difference value exceeds a set range.
If the characteristic value of the ith row and jth column pixel in the image of the epitaxial part is different from the characteristic value of the pixel at the same position in the image of the non-defective epitaxial wafer or the difference value exceeds a set range, and the characteristic value of the mth row and nth column pixel in the image of the epitaxial part is different from the characteristic value of the pixel at the same position in the image of the non-defective epitaxial wafer or the difference value exceeds the set range, i, j, m and n are positive integers, i is less than a, j is less than b, m is less than a, and n is less than b. If the distance between the jth pixel in the ith row and the jth pixel in the ith row does not exceed a set value, grouping the jth pixel in the ith row and the (j +1) th pixel in the ith row in the image of the epitaxial part into the same image of the epitaxial defect part; and if the distance between the ith row and jth column pixel and the ith row and jth column pixel exceeds a set value, classifying the ith row and jth column pixel and the ith row and (j +1) column pixel in the image of the epitaxial part into two images of different epitaxial defect parts.
In addition, a Convolutional Neural Network (CNN) is a kind of artificial Neural Network, and has become a research hotspot in the field of current speech analysis and image recognition. The weight sharing network structure of the system is more similar to a biological neural network, the complexity of a network model is reduced, and the number of weights is reduced. The advantage is more obvious when the input of the network is a multi-dimensional image, so that the image can be directly used as the input of the network, and the complex characteristic extraction and data reconstruction process in the traditional recognition algorithm is avoided. Convolutional neural networks are multi-layered perceptrons specifically designed to recognize two-dimensional shapes, and the network structure is highly invariant to translation, tilt, or other forms of deformation.
In this embodiment, on the basis of a general convolutional neural network, an image with an input set as an epitaxial defect portion is output as a defect type of the epitaxial defect portion, and a plurality of epitaxial defect images with a calibrated defect type are used to train parameters in the convolutional neural network, so as to obtain a first convolutional neural network.
Step 104: and comparing the image of the electrode part with the image of the non-defective electrode, intercepting the image of the defective electrode part from the image of the electrode part, inputting the image of the defective electrode part into a second convolutional neural network to obtain the defect type of the defective electrode part, and training the parameters of the second convolutional neural network by adopting a plurality of electrode defect images with calibrated defect types.
In the present embodiment, the image of a non-defective electrode is an image of a non-defective electrode. The electrode without defects can be determined in a manual detection mode, the image of the electrode can be obtained by firstly using AOI equipment to obtain the image of the chip where the electrode is located, and then, the image of the chip is obtained by dividing the image of the chip by using an edge detection model. It should be noted that, since the shapes of the electrodes may be different, for example, there is only a pad, and there is a pad and an electrode line included in some of the electrodes, images of non-defective electrodes to be compared are usually pre-selected according to the shapes of the electrodes in the chip to be detected.
As for the specific comparison process, similar to the epitaxial portion, the characteristic values of the pixels in the image of the electrode portion and the characteristic values of the pixels at the same positions on the image of the non-defective electrode are sequentially compared, all the pixels with different characteristic values or the difference value exceeding the set range are selected, and two pixels of which the mutual distance does not exceed the set value in the selected pixels are classified as the image of the same defective portion.
In addition, the second convolutional neural network is similar to the first convolutional neural network, on the basis of the general convolutional neural network, an image with an input set as an electrode defect part is output and set as a defect type of the electrode defect part, and parameters in the convolutional neural network are trained by adopting a plurality of electrode defect images with calibrated defect types to obtain the second convolutional neural network.
It should be noted that, the execution of step 103 and step 104 is not in a sequential order, and step 103 may be executed first and then step 104 is executed, or step 104 may be executed first and then step 103 is executed, or step 103 and step 104 may be executed simultaneously.
According to the embodiment of the invention, after the image of the chip to be detected is obtained, the image of the chip to be detected is divided into the image of the extension part and the image of the electrode part by using the edge detection model, so that different defect conditions of the two parts can be respectively processed, the processing is simplified, and the processing efficiency is improved. And then, respectively carrying out similar processing on the image of the epitaxial part and the image of the electrode part, taking the image of the epitaxial part as an example, comparing the image of the epitaxial part with the image of the defect-free epitaxial wafer, and intercepting and taking out the image of the defect part of the epitaxial wafer, so that the defect type can be judged according to the defect part, the influence of the non-defect part or other defect parts is eliminated, the pertinence is strong, and the accuracy is improved and the operand is reduced. And then, the defect type of the first convolutional neural network is obtained from the images of the epitaxial defect part by utilizing the first convolutional neural network trained by the epitaxial defect images with the defect types calibrated, so that the problem of multiple comparison is solved, and the detection efficiency can be greatly improved. And the parameters of the convolutional neural network are trained by adopting various defect images, various defect types and various expression forms of the defects can be related, the accuracy of the detection result can be ensured, and the requirements of industrial production are particularly met.
The embodiment of the invention provides another method for detecting defects of a light emitting diode chip, which is a specific implementation of the method for detecting defects of the light emitting diode chip shown in fig. 1. Fig. 3 is a flowchart of another method for detecting defects of an led chip according to an embodiment of the present invention. Referring to fig. 3, the method includes:
step 201: a plurality of epitaxial defect images are acquired.
Illustratively, this step 201 may include:
acquiring images of a plurality of epitaxial wafers;
and sequentially comparing the image of each epitaxial wafer with the image of the defect-free epitaxial wafer, and taking the part of the image of the epitaxial wafer, which is different from the image of the defect-free epitaxial wafer, as an epitaxial defect image.
In practical applications, an image of the epitaxial wafer may be acquired using an AOI device.
In addition, in order to ensure the training effect of the parameters of the convolutional neural network (e.g., the output of the convolutional neural network is accurate), a large number of defect images need to be acquired for training, for example, the number of epitaxial defect images of each defect type is more than 1000, so that the output of the first convolutional neural network is accurate.
Optionally, after step 201, the method may further include:
and normalizing the epitaxial defect image to obtain an epitaxial defect image with a preset specification.
The predetermined specification may include a set size or format, among others.
In practical applications, in order to ensure the accuracy of the output result of the convolutional neural network, normalization processing is performed on all inputs, for example, normalization is performed on the inputs to an image with 64 pixels by 64 pixels, so that errors generated in the output result of the convolutional neural network due to inconsistency of the input images can be avoided.
Step 202: and receiving the defect type marked by each epitaxial defect image.
In practical application, in order to adjust parameters of the convolutional neural network when the convolutional neural network outputs incorrectly, the defect types of each manually-calibrated epitaxial defect image are correspondingly received (for example, different numbers are used for respectively representing different defect types) while the epitaxial defect image is acquired, so that the parameters of the convolutional neural network are adjusted according to the manually-calibrated defect types, and the defect types output by the convolutional neural network are the same as the received defect types.
Step 203: and training parameters of the first convolution neural network by adopting the plurality of electrode defect images and the defect types calibrated by the electrode defect images.
Because the human cognition on the outside world is from local to global, the spatial relation of the image is that local pixels are closer, and the correlation of the pixels at a longer distance is weaker. Therefore, each neuron of the neural network does not need to sense the global image, only needs to sense the local image, and then synthesizes the local information at a higher layer to obtain the global information.
The convolutional neural network realizes sensing of local parts by adopting a convolutional layer (adaptive convolutional layer) usually arranged at a position close to a network input end, and integrates local information by adopting a full connection layer usually arranged at a position close to a network output end.
In the specific implementation, the convolution layer performs convolution on the image by adopting at least one convolution kernel, the characteristics of each local sensing area in the image are extracted, and the image characteristics extracted by different types of convolution kernels are different. The full connection layer is used for establishing connection between each neuron of the previous layer and all neurons of the next layer. All the inputs and any one output of the full connection layer satisfy the following formula:
h＝f(∑iWi*xi+b)；
where h is the output of the full link layer, b is the offset value, xiFor each input of the fully connected layer, WiAre each of the fully-connected layersInput to output weight, i represents any one of the inputs of the fully-connected layer, ΣiRepresenting the summation of all inputs to the fully-connected layer, and f () representing a functional relationship, typically a sigmoid function or a tanh function.
The parameters of the convolutional neural network in this embodiment may include the convolutional kernels used by the convolutional layers, and the weights and bias values used by the fully-connected layers. In addition, in addition to convolutional layers and fully-connected layers, convolutional neural networks may also include pooling layers (posing layers) typically placed at the output of convolutional layers to reduce the dimensionality of image features. And the pooling layer carries out aggregation statistics on the characteristics of different positions. If the convolutional neural network further includes a pooling layer, the parameters of the convolutional neural network may further include units partitioned by the pooling layer.
In practical application, the initial values of the parameters of the convolutional neural network can be randomly set, then each defect image is sequentially input into the convolutional neural network, the defect type output by the convolutional neural network is obtained after each defect image is input into the convolutional neural network, the defect type is compared with the calibrated defect type, and when the defect type is different from the calibrated defect type, the parameters of the convolutional neural network are adjusted so that the defect type and the defect type are the same. That is, this step 203 may include:
and when the defect type obtained by the epitaxial defect image is different from the calibrated defect type, reversely propagating and adjusting the parameters of the first convolution neural network until the defect type obtained by the epitaxial defect images is the same as the calibrated defect type.
For example, firstly, inputting the 1 st epitaxial defect image into the first convolution neural network to obtain the defect type of the 1 st epitaxial defect image, and if the defect type obtained by the 1 st epitaxial defect image is the same as the calibrated defect type, keeping the parameters of the first convolution neural network unchanged; and if the defect type obtained by the 1 st epitaxial defect image is different from the calibrated defect type, adjusting the parameters of the first convolution neural network according to a set sequence until the defect type obtained by the 1 st epitaxial defect image is the same as the calibrated defect type. Inputting the 2 nd epitaxial defect image into the first convolution neural network to obtain the defect type of the 2 nd epitaxial defect image, and if the defect type obtained by the 2 nd epitaxial defect image is the same as the calibrated defect type, keeping the parameters of the first convolution neural network unchanged; if the defect type obtained by the 2 nd epitaxial defect image is different from the calibrated defect type, adjusting the parameters of the first convolution neural network according to a set sequence until the defect type obtained by the 2 nd epitaxial defect image is same as the calibrated defect type … …, and finally inputting the last epitaxial defect image into the first convolution neural network to obtain the defect type of the last epitaxial defect image; and if the defect type obtained by the last epitaxial defect image is different from the calibrated defect type, adjusting the parameters of the first convolution neural network according to a set sequence until the defect type obtained by the last epitaxial defect image is the same as the calibrated defect type. And repeating the circulation until the parameters of the first convolutional neural network are kept unchanged in one circulation, and finishing the training of the first convolutional neural network.
Further, backpropagating adjusts the parameters of the convolutional neural network, which may include:
and adjusting parameters of the convolutional neural network by adopting a gradient descent method.
In practical applications, the parameters of the convolutional neural network may be gradually adjusted (gradually increased or gradually decreased), for example, the parameters of the convolutional neural network may be continuously increased or decreased until the obtained defect type is the same as the calibrated defect type.
It should be noted that, steps 201 to 203 are optional steps, and the training of the parameters of the first convolutional neural network can be implemented through steps 201 to 203.
Step 204: a plurality of electrode defect images are acquired.
Illustratively, this step 204 may include:
acquiring images of a plurality of chips;
processing images of a plurality of chips by adopting an edge detection model, and dividing the images of each chip into images of an extension part and images of an electrode part;
and sequentially comparing the images of the electrode parts divided by the images of the chips with the images of the non-defective electrodes, and taking the parts of the images of the electrode parts, which are different from the images of the non-defective electrodes, as the images of the electrode defects.
In practical applications, an image of the chip may also be acquired using an AOI device.
In addition, as mentioned above, in order to ensure the training effect of the parameters of the convolutional neural network (e.g., the output of the convolutional neural network is accurate), a large number of defect images need to be acquired for training. For example, the number of electrode defect images for each defect type is above 1000, so that the output of the second convolutional neural network is accurate.
Optionally, after step 204, the method may further include:
and carrying out normalization processing on the electrode defect image to obtain an electrode defect image with a preset specification.
The predetermined specification may include a set size or format, among others.
In practical applications, in order to ensure the accuracy of the output result of the convolutional neural network, normalization processing is performed on all inputs, for example, normalization is performed on the inputs to an image with 64 pixels by 64 pixels, so that errors generated in the output result of the convolutional neural network due to inconsistency of the input images can be avoided.
Step 205: and receiving the defect type marked by each electrode defect image.
Alternatively, step 205 may be similar to step 202 and will not be described in detail herein.
Step 206: and training the second convolutional neural network by adopting the plurality of electrode defect images and the defect types calibrated by the electrode defect images.
Optionally, this step 206 may be similar to step 203, i.e. this step 206 may comprise:
and when the defect type obtained from the electrode defect image is different from the calibrated defect type, the parameters of the second convolutional neural network are reversely propagated and adjusted until the defect type obtained from the plurality of electrode defect images is the same as the calibrated defect type.
It should be noted that steps 204-206 are optional steps, and the training of the parameters of the second convolutional neural network can be implemented through steps 204-206. In addition, the steps 201 to 203 and the steps 204 to 206 are not performed in sequence, the steps 201 to 203 may be performed first, and then the steps 204 to 206 may be performed, the steps 204 to 206 may be performed first, and then the steps 201 to 203 may be performed, or the steps 201 to 203 and the steps 204 to 206 may be performed simultaneously.
Step 207: and acquiring an image of the chip to be detected.
Illustratively, this step 207 may be the same as step 101 and will not be described in detail here.
Step 208: and processing the image of the chip to be detected by adopting an edge detection model, and dividing the image of the chip to be detected into an image of the extension part and an image of the electrode part.
Illustratively, this step 208 may be the same as step 102 and will not be described in detail herein.
Step 209: and comparing the image of the epitaxial part with the image of the defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, inputting the image of the epitaxial defect part into a first convolution neural network to obtain the defect type of the epitaxial defect part, and training the parameters of the first convolution neural network by adopting a plurality of epitaxial defect images for calibrating the defect type.
Optionally, inputting the image of the epitaxial defect portion into the first convolutional neural network to obtain the defect type of the epitaxial defect portion, which may include:
normalizing the images of the epitaxial defect parts to obtain epitaxial defect images with preset specifications;
and inputting the epitaxial defect image with the preset specification into the first convolution neural network to obtain the defect type of the epitaxial defect part.
Corresponding to step 201, the image is normalized and then input into the convolutional neural network, so that errors generated in the output result of the convolutional neural network due to inconsistency of the input images can be avoided.
Step 210: and comparing the image of the electrode part with the image of the non-defective electrode, intercepting the image of the defective electrode part from the image of the electrode part, inputting the image of the defective electrode part into a second convolutional neural network to obtain the defect type of the defective electrode part, and training the parameters of the second convolutional neural network by adopting a plurality of electrode defect images with calibrated defect types.
Optionally, inputting the image of the defective electrode portion into a second convolutional neural network to obtain a defect type of the defective electrode portion, which may include:
normalizing the image of the electrode defect part to obtain an electrode defect image with a preset specification;
and inputting the electrode defect image with the preset specification into a second convolution neural network to obtain the defect type of the electrode defect part.
Corresponding to step 204, the image is normalized and then input into the convolutional neural network, so that errors generated in the output result of the convolutional neural network due to inconsistency of the input images can be avoided.
Step 211: and counting the number, the size and the defect type of the images of the epitaxial defect parts intercepted in the images of the epitaxial parts, and the number, the size and the defect type of the images of the electrode defect parts intercepted in the images of the electrode parts, and determining the quality grade of the chip to be detected.
In practical application, the chips to be detected can be classified into corresponding grades according to the statistical result and the requirements on the number, size, type and the like of defects in the universal product quality standard in the industry.
It should be noted that step 211 is an optional step, and the screening of the chip can be realized through step 211, so as to perform different processes on the chip. For example, chips with high quality (e.g., defect rate less than 10%) and chips with good quality (e.g., defect rate between 20% and 30%) are sorted differently, and chips with poor quality (e.g., defect rate more than 40%) are discarded.
The embodiment of the invention provides a device for detecting defects of a light-emitting diode chip, which is suitable for realizing the method for detecting the defects of the light-emitting diode chip shown in figure 1 or figure 3. Fig. 4 is a schematic structural diagram of an apparatus for detecting defects of a light emitting diode chip according to an embodiment of the present invention. Referring to fig. 4, the apparatus includes:
a chip image obtaining module 301, configured to obtain an image of a chip to be detected;
the image dividing module 302 is configured to process an image of a chip to be detected by using an edge detection model, and divide the image of the chip to be detected into an image of an extension portion and an image of an electrode portion;
the epitaxial image processing module 303 is configured to compare the image of the epitaxial portion with the image of the defect-free epitaxial wafer, intercept the image of the epitaxial defect portion from the image of the epitaxial portion, input the image of the epitaxial defect portion into a first convolution neural network, and obtain a defect type of the epitaxial defect portion, where parameters of the first convolution neural network are obtained by training a plurality of epitaxial defect images with the defect types calibrated;
and the electrode image processing module 304 is configured to compare the image of the electrode portion with the image of the non-defective electrode, intercept the image of the defective electrode portion from the image of the electrode portion, and input the image of the defective electrode portion into a second convolutional neural network to obtain a defect type of the defective electrode portion, where parameters of the second convolutional neural network are obtained by training with a plurality of electrode defect images with the defect types calibrated.
Alternatively, the epi image processing module 303 may include:
the extension image normalization submodule is used for normalizing the images of the extension defect parts to obtain extension images with preset specifications;
and the epitaxial defect type determining submodule is used for inputting an epitaxial image with a preset specification into the first convolution neural network to obtain the defect type of the epitaxial defect part.
Accordingly, the electrode image processing module 304 may include:
the electrode image normalization submodule is used for normalizing the image of the electrode defect part to obtain an electrode image with a preset specification;
and the electrode defect type determining submodule is used for inputting the electrode image with the preset specification into the second convolutional neural network to obtain the defect type of the electrode defect part.
Optionally, the apparatus may further include:
the epitaxial defect image acquisition module is used for acquiring a plurality of epitaxial defect images;
the epitaxial defect type receiving module is used for receiving the defect types marked by the epitaxial defect images;
and the epitaxial defect type training module is used for training the first convolution neural network by adopting a plurality of epitaxial defect images and the defect types calibrated by the epitaxial defect images.
Further, an epitaxial defect type training module may be used to,
and when the defect type obtained by the epitaxial defect image is different from the calibrated defect type, reversely propagating and adjusting the parameters of the first convolution neural network until the defect type obtained by the epitaxial defect images is the same as the calibrated defect type.
Optionally, the apparatus may further include:
the electrode defect image acquisition module is used for acquiring a plurality of electrode defect images;
the electrode defect type receiving module is used for receiving the defect types calibrated by the defect images of the electrodes;
and the electrode defect type training module is used for training the second convolutional neural network by adopting a plurality of electrode defect images and the defect types calibrated by the electrode defect images.
Further, an electrode defect type training module may be used,
and when the defect type obtained from the electrode defect image is different from the calibrated defect type, the parameters of the second convolutional neural network are reversely propagated and adjusted until the defect type obtained from the plurality of electrode defect images is the same as the calibrated defect type.
Optionally, the apparatus may further include:
and the determining module is used for counting the number, the size and the defect type of the images of the epitaxial defect parts intercepted in the images of the epitaxial parts, and the number, the size and the defect type of the images of the electrode defect parts intercepted in the images of the electrode parts, and determining the quality grade of the chip to be detected.
It should be noted that: in the light emitting diode chip defect detecting apparatus provided in the above embodiment, when detecting a light emitting diode chip defect, only the division of the functional modules is exemplified, and in practical applications, the function distribution may be completed by different functional modules according to needs, that is, the internal structure of the apparatus is divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus for detecting defects of a light emitting diode chip and the method for detecting defects of a light emitting diode chip provided by the above embodiments belong to the same concept, and specific implementation processes thereof are described in the method embodiments and are not described herein again.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. A method for detecting defects of a light emitting diode chip is characterized by comprising the following steps:
acquiring an image of a chip to be detected;
processing the image of the chip to be detected by adopting an edge detection model, and dividing the image of the chip to be detected into an image of an extension part and an image of an electrode part;
comparing the image of the epitaxial part with the image of a defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, and inputting the image of the epitaxial defect part into a first convolution neural network to obtain the defect type of the epitaxial defect part, wherein the parameter of the first convolution neural network is obtained by training a plurality of epitaxial defect images with calibrated defect types; comparing the image of the epitaxial part with the image of a defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, selecting all pixels with different characteristic values or difference values exceeding a set range for sequentially comparing the characteristic value of each pixel in the image of the epitaxial part with the characteristic value of a pixel at the same position on the image of the defect-free epitaxial wafer, and classifying two pixels, the distance between which does not exceed a set value, in the selected pixels into the image of the same defect part;
comparing the image of the electrode part with an image of a non-defective electrode, intercepting the image of the defective electrode part from the image of the electrode part, and inputting the image of the defective electrode part into a second convolutional neural network to obtain the defect type of the defective electrode part, wherein the parameters of the second convolutional neural network are obtained by training a plurality of electrode defect images with the defect types calibrated; comparing the image of the electrode part with the image of the non-defective electrode, intercepting the image of the defective part of the electrode from the image of the electrode part, sequentially comparing the characteristic value of each pixel in the image of the electrode part with the characteristic value of the pixel at the same position on the image of the non-defective electrode, selecting all pixels with different characteristic values or difference values exceeding a set range, and classifying two pixels, the distance between which does not exceed a set value, in the selected pixels into the image of the same defective part.
2. The method of claim 1, wherein inputting the image of the electrode defect portion into a second convolutional neural network to obtain a defect type of the electrode defect portion comprises:
normalizing the image of the electrode defect part to obtain an electrode image with a preset specification;
and inputting the electrode image with the preset specification into the second convolutional neural network to obtain the defect type of the defective part of the electrode.
3. The method according to claim 1 or 2, characterized in that the method further comprises:
acquiring a plurality of electrode defect images;
receiving the defect type calibrated by each electrode defect image;
and training the second convolutional neural network by adopting a plurality of electrode defect images and the defect types calibrated by the electrode defect images.
4. The method of claim 3, wherein said training said second convolutional neural network with a plurality of said electrode defect images and defect types specified for each of said electrode defect images comprises:
and when the defect type obtained by the electrode defect image is different from the calibrated defect type, the parameters of the second convolutional neural network are adjusted in a back propagation mode until the defect type obtained by the electrode defect image is the same as the calibrated defect type.
5. The method according to claim 1 or 2, characterized in that the method further comprises:
and counting the number, the size and the defect type of the images of the epitaxial defect parts intercepted in the images of the epitaxial parts, and the number, the size and the defect type of the images of the electrode defect parts intercepted in the images of the electrode parts, and determining the quality grade of the chip to be detected.
6. An apparatus for detecting defects of a light emitting diode chip, the apparatus comprising:
the chip image acquisition module is used for acquiring an image of a chip to be detected;
the image dividing module is used for processing the image of the chip to be detected by adopting an edge detection model and dividing the image of the chip to be detected into an image of an extension part and an image of an electrode part;
the epitaxial image processing module is used for comparing the image of the epitaxial part with the image of a defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, inputting the image of the epitaxial defect part into a first convolution neural network to obtain the defect type of the epitaxial defect part, and training the parameters of the first convolution neural network by adopting a plurality of epitaxial defect images with calibrated defect types; comparing the image of the epitaxial part with the image of a defect-free epitaxial wafer, intercepting the image of the epitaxial defect part from the image of the epitaxial part, selecting all pixels with different characteristic values or difference values exceeding a set range for sequentially comparing the characteristic value of each pixel in the image of the epitaxial part with the characteristic value of a pixel at the same position on the image of the defect-free epitaxial wafer, and classifying two pixels, the distance between which does not exceed a set value, in the selected pixels into the image of the same defect part;
the electrode image processing module is used for comparing the image of the electrode part with the image of a non-defective electrode, intercepting the image of the defective electrode part from the image of the electrode part, inputting the image of the defective electrode part into a second convolutional neural network to obtain the defect type of the defective electrode part, and the parameters of the second convolutional neural network are obtained by adopting a plurality of electrode defect images with calibrated defect types for training; comparing the image of the electrode part with the image of the non-defective electrode, intercepting the image of the defective part of the electrode from the image of the electrode part, sequentially comparing the characteristic value of each pixel in the image of the electrode part with the characteristic value of the pixel at the same position on the image of the non-defective electrode, selecting all pixels with different characteristic values or difference values exceeding a set range, and classifying two pixels, the distance between which does not exceed a set value, in the selected pixels into the image of the same defective part.
7. The apparatus of claim 6, wherein the electrode image processing module comprises:
the electrode image normalization submodule is used for normalizing the image of the electrode defect part to obtain an electrode image with a preset specification;
and the electrode defect type determining submodule is used for inputting the electrode image with the preset specification into the second convolutional neural network to obtain the defect type of the electrode defect part.
8. The apparatus of claim 6 or 7, further comprising:
the electrode defect image acquisition module is used for acquiring a plurality of electrode defect images;
the electrode defect type receiving module is used for receiving the defect type calibrated by each electrode defect image;
and the electrode defect type training module is used for training the second convolutional neural network by adopting a plurality of electrode defect images and the defect types calibrated by the electrode defect images.
9. The apparatus of claim 8, wherein the electrode defect type training module is to,
and when the defect type obtained by the electrode defect image is different from the calibrated defect type, the parameters of the second convolutional neural network are adjusted in a back propagation mode until the defect type obtained by the electrode defect image is the same as the calibrated defect type.
10. The apparatus of claim 6 or 7, further comprising:
and the determining module is used for counting the number, the size and the defect type of the images of the epitaxial defect parts intercepted in the images of the epitaxial parts, and the number, the size and the defect type of the images of the electrode defect parts intercepted in the images of the electrode parts, and determining the quality grade 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 CN109919908A (en) | 2019-06-21 |
| CN109919908B true 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) |
Families Citing this family (9)
| 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 |
| 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 |
| 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 |
| CN112304960B (en) * | 2020-12-30 | 2021-08-10 | 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 |
| CN114397306B (en) * | 2022-03-25 | 2022-07-29 | 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 |
| 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 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009236593A (en) * | 2008-03-26 | 2009-10-15 | Mitsubishi Electric Corp | Visual inspection supporting device |
| CN103278511B (en) * | 2013-05-17 | 2015-04-29 | 南京大学 | Wafer defect detection method based on multi-scale corner feature extraction |
| CN104597057B (en) * | 2015-02-02 | 2017-06-06 | 东华大学 | A kind of column Diode facets defect detecting 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 |
| 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 |
| CN108982508B (en) * | 2018-05-23 | 2020-11-24 | 江苏农林职业技术学院 | Plastic package IC chip defect detection method based on feature template matching and deep learning |
| CN109242033B (en) * | 2018-09-21 | 2021-08-20 | 长鑫存储技术有限公司 | Wafer defect mode classification method and device, storage medium and electronic equipment |
-
2019
- 2019-01-23 CN CN201910065080.1A patent/CN109919908B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109919908A (en) | 2019-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919908B (en) | Method and device for detecting defects of light-emitting diode chip | |
| CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
| JP7087397B2 (en) | Substrate defect inspection equipment, substrate defect inspection method and storage medium | |
| US20200356718A1 (en) | Implementation of deep neural networks for testing and quality control in the production of memory devices | |
| JP7146857B2 (en) | Method and system for classifying wafer defects using wafer defect images based on deep learning | |
| US20210150700A1 (en) | Defect detection device and method | |
| CN109829914A (en) | The method and apparatus of testing product defect | |
| CN107966454A (en) | A kind of end plug defect detecting device and detection method based on FPGA | |
| TWI834005B (en) | Deep learning networks for nuisance filtering | |
| CN112508857B (en) | Aluminum product surface defect detection method based on improved Cascade R-CNN | |
| KR100868884B1 (en) | Flat glass defect information system and classification method | |
| TW202226027A (en) | Deep generative models for optical or other mode selection | |
| TW202242390A (en) | Defect inspection device, defect inspection method, and manufacturing method | |
| Lin et al. | Determination of the varieties of rice kernels based on machine vision and deep learning technology | |
| Yuan-Fu et al. | Double feature extraction method for wafer map classification based on convolution neural network | |
| CN114627090A (en) | Convolutional neural network optical lens defect detection method based on attention mechanism | |
| US11967060B2 (en) | Wafer level spatial signature grouping using transfer learning | |
| CN115082449B (en) | Electronic component defect detection method | |
| CN113269251A (en) | Fruit flaw classification method and device based on machine vision and deep learning fusion, storage medium and computer equipment | |
| CN113657539A (en) | Display panel micro-defect detection method based on two-stage detection network | |
| CN111862039A (en) | Rapid visual detection method for braid | |
| Chaugule et al. | Seed technological development—a survey | |
| US12020418B2 (en) | Image processing method and system, and non-transitory computer readable medium | |
| Kavitha et al. | Component identification and defect detection of printed circuit board using artificial intelligence | |
| Jamal et al. | Fine-tuning strategy for re-classification of false call in automated optical inspection post reflow |
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 |