CN114752996A - Intelligent adjustment method for monocrystalline silicon shoulder-laying stage forking and storage medium - Google Patents
Intelligent adjustment method for monocrystalline silicon shoulder-laying stage forking and storage medium Download PDFInfo
- Publication number
- CN114752996A CN114752996A CN202210284347.8A CN202210284347A CN114752996A CN 114752996 A CN114752996 A CN 114752996A CN 202210284347 A CN202210284347 A CN 202210284347A CN 114752996 A CN114752996 A CN 114752996A
- Authority
- CN
- China
- Prior art keywords
- monocrystalline silicon
- split
- image
- length
- endpoints
- 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.)
- Pending
Links
- 229910021421 monocrystalline silicon Inorganic materials 0.000 title claims abstract description 152
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000003860 storage Methods 0.000 title claims abstract description 14
- 238000001514 detection method Methods 0.000 claims abstract description 44
- 239000013078 crystal Substances 0.000 claims abstract description 38
- 230000011218 segmentation Effects 0.000 claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 15
- 238000013527 convolutional neural network Methods 0.000 claims description 33
- 238000000605 extraction Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 16
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 238000009826 distribution Methods 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 10
- 102100037651 AP-2 complex subunit sigma Human genes 0.000 claims description 3
- 101000806914 Homo sapiens AP-2 complex subunit sigma Proteins 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 206010044625 Trichorrhexis Diseases 0.000 claims 1
- 230000008569 process Effects 0.000 abstract description 13
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 abstract description 11
- 229910052710 silicon Inorganic materials 0.000 abstract description 11
- 239000010703 silicon Substances 0.000 abstract description 11
- 230000008859 change Effects 0.000 description 10
- 238000005457 optimization Methods 0.000 description 9
- 238000012549 training Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000010899 nucleation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000003466 welding Methods 0.000 description 1
Images
Classifications
-
- C—CHEMISTRY; METALLURGY
- C30—CRYSTAL GROWTH
- C30B—SINGLE-CRYSTAL GROWTH; UNIDIRECTIONAL SOLIDIFICATION OF EUTECTIC MATERIAL OR UNIDIRECTIONAL DEMIXING OF EUTECTOID MATERIAL; REFINING BY ZONE-MELTING OF MATERIAL; PRODUCTION OF A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; SINGLE CRYSTALS OR HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; AFTER-TREATMENT OF SINGLE CRYSTALS OR A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; APPARATUS THEREFOR
- C30B15/00—Single-crystal growth by pulling from a melt, e.g. Czochralski method
- C30B15/20—Controlling or regulating
- C30B15/22—Stabilisation or shape controlling of the molten zone near the pulled crystal; Controlling the section of the crystal
- C30B15/26—Stabilisation or shape controlling of the molten zone near the pulled crystal; Controlling the section of the crystal using television detectors; using photo or X-ray detectors
-
- C—CHEMISTRY; METALLURGY
- C30—CRYSTAL GROWTH
- C30B—SINGLE-CRYSTAL GROWTH; UNIDIRECTIONAL SOLIDIFICATION OF EUTECTIC MATERIAL OR UNIDIRECTIONAL DEMIXING OF EUTECTOID MATERIAL; REFINING BY ZONE-MELTING OF MATERIAL; PRODUCTION OF A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; SINGLE CRYSTALS OR HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; AFTER-TREATMENT OF SINGLE CRYSTALS OR A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; APPARATUS THEREFOR
- C30B29/00—Single crystals or homogeneous polycrystalline material with defined structure characterised by the material or by their shape
- C30B29/02—Elements
- C30B29/06—Silicon
Landscapes
- Chemical & Material Sciences (AREA)
- Engineering & Computer Science (AREA)
- Crystallography & Structural Chemistry (AREA)
- Materials Engineering (AREA)
- Metallurgy (AREA)
- Organic Chemistry (AREA)
- Crystals, And After-Treatments Of Crystals (AREA)
Abstract
The invention provides an intelligent adjusting method of crystal pulling speed in a shouldering stage of monocrystalline silicon and a storage medium, wherein the method comprises the following steps: acquiring a monocrystalline silicon image at a shouldering stage; obtaining a plurality of split endpoints in the monocrystalline silicon image based on the key point detection; carrying out segmentation processing on the monocrystalline silicon image to obtain a constraint relation among a plurality of split endpoints; obtaining a plurality of combinations of the plurality of split endpoints according to a constraint relation, wherein each combination comprises two split endpoints; generating the actual splitting length of the monocrystalline silicon according to the distance between the two splitting end points in each combination; adjusting the pull rate of the monocrystalline silicon based on the actual length of the split. The technical problem that in the prior art, workers need to observe the real-time state of the silicon single crystal rod through the single crystal furnace by naked eyes, and then control the crystal pulling speed according to subjective experience, so that the pulling speed control precision in the shouldering process is low is solved.
Description
Technical Field
The invention relates to the field of intelligent crystal pulling, in particular to an intelligent adjustment method for forking of a monocrystalline silicon shouldering stage and a storage medium.
Background
Solar energy is an important clean energy source, monocrystalline silicon is an important raw material of a solar cell panel, the market demand for high-quality monocrystalline silicon is increased in recent years, and higher requirements are put forward in the production link of the monocrystalline silicon.
The Czochralski method is the mainstream production method of the monocrystalline silicon at present, and the main steps of the Czochralski method production are welding, seeding, shouldering, shoulder rotating and diameter equalizing. The shouldering is to enable the silicon single crystal rod to grow to a target size, but if the crystal pulling speed is not accurately controlled in the shouldering process, namely the crystal pulling speed is too high or too low, the wire breakage can occur, the success rate of crystal pulling is seriously influenced, and therefore the crystal pulling speed in the shouldering process needs to be accurately controlled.
In the prior art, workers need to visually observe the diameter change rate of a single crystal silicon rod in a single crystal furnace, and manually control the crystal pulling speed according to the diameter change rate of the single crystal silicon rod, so that the pulling speed control efficiency in the shouldering process is low.
The invention is provided in view of the above.
Disclosure of Invention
The invention provides an intelligent adjusting method for crystal pulling speed in a shouldering stage of monocrystalline silicon and a storage medium, which are used for solving the technical problems that in the prior art, workers need to visually observe the diameter change rate of a monocrystalline silicon rod in a monocrystalline furnace, and the pulling speed in the shouldering process is controlled manually according to the diameter change rate of the monocrystalline silicon rod, so that the pulling speed control efficiency is low.
According to a first aspect of the invention, a method for intelligently adjusting the pulling speed of a shouldering stage of monocrystalline silicon is provided, and comprises the following steps: obtaining a monocrystalline silicon image in a shouldering stage; obtaining a plurality of split endpoints in the monocrystalline silicon image based on key point detection; segmenting the monocrystalline silicon image to obtain a constraint relation among a plurality of forked endpoints; obtaining a plurality of combinations of the plurality of split endpoints according to a constraint relation, wherein each combination comprises two split endpoints; generating the actual splitting length of the monocrystalline silicon according to the distance between the two splitting end points in each combination; and adjusting the pulling speed of the monocrystalline silicon based on the actual jag length.
Further, after acquiring the single crystal silicon image in the shouldering stage, the method comprises the following steps: extracting features of the monocrystalline silicon image and dividing the features of the monocrystalline silicon image into a first set of features and a second set of features, wherein the step of obtaining a plurality of split endpoints in the monocrystalline silicon image based on keypoint detection comprises: inputting the first set of features into a first convolutional neural network to output a plurality of split endpoints in the single crystal silicon image; the step of performing division processing on the monocrystalline silicon image includes: inputting the second set of features into a second convolutional neural network to output a plurality of split positions and deriving a constrained relationship between the plurality of split endpoints based on the plurality of split positions.
Further, before acquiring the monocrystalline silicon image in the shouldering stage, the method further comprises the following steps:
obtaining a single-channel gray image from the monocrystalline silicon original image; extracting a gradient map in the horizontal and vertical directions of the single-channel gray image; and superposing the single-channel gray image and the transverse and longitudinal gradient images to generate the monocrystalline silicon image.
Further, the keypoint labels for training the first convolutional neural network are generated by:
obtaining an image of a monocrystalline silicon sampleMarking a split end point in the monocrystalline silicon sample image; determining a square area of the key point label picture by taking each forking end point as a center, wherein the side length of the square area is delta sigma,
sigma 2, wherein for each pixel i in each square area, the pixel value e is assigned to each pixel i-β×dFilling to generate a local Gaussian distribution map with each divergent end point in the monocrystalline silicon sample image as a center, wherein beta is a coefficient,
d is the square of the distance from each pixel i to the center point of the square region.
Further, the first convolutional neural network is optimized by a first loss function as follows:
wherein the content of the first and second substances,
in order to be a key point label, 
For the value of the ith element in the keypoint label,
to label according to the key point
The generated binary label is used for carrying out label matching on the label,
is the value of the ith element in the binary label, Y is the split end point of the first convolution neural network output, YiNumerical value of ith position in monocrystalline silicon image output by first convolution neural network。
Further, the second convolutional neural network is optimized by a second loss function as follows:
wherein the content of the first and second substances,
in order to have a binary label for the end point of the fork,
is the value of the ith element in the binary label of the split endpoint, S is the constraint relation of the constraint segmentation model output of the split endpoint, SiIs the value of the ith position in the monocrystalline silicon image output by the second convolutional neural network.
Further, the step of adjusting the pull rate of the single crystal silicon based on the actual jag length comprises: determining the sub-stage of the monocrystalline silicon in the shouldering stage; obtaining a standard divergence length associated with the sub-phase; comparing the actual splitting length with the standard splitting length; under the condition that the actual splitting length is smaller than the standard splitting length, controlling and reducing the crystal pulling speed of the monocrystalline silicon, so that the actual splitting length is increased to the standard splitting length; and under the condition that the actual splitting length is greater than the standard splitting length, controlling and increasing the crystal pulling speed of the monocrystalline silicon, so that the actual splitting length is reduced to the standard splitting length.
According to a second aspect of the invention, a method for intelligently detecting the forking of the shouldering stage of monocrystalline silicon is provided, and the method comprises the following steps: obtaining a monocrystalline silicon image in a shouldering stage; inputting the monocrystalline silicon image into a convolutional neural network, wherein the convolutional neural network comprises a feature extraction trunk network, a key point detection branch network and a segmentation branch network, and the key point detection branch network and the segmentation branch network are respectively connected with the feature extraction trunk network; the feature extraction backbone network extracts features of the monocrystalline silicon image from the monocrystalline silicon image and divides the features of the monocrystalline silicon image into a first group of features and a second group of features; the feature extraction backbone network inputs a first group of features into the key point detection branch network, and the key point detection branch network outputs a plurality of split endpoints in the monocrystalline silicon image; the feature extraction backbone network inputs a second group of features into the segmentation branch network, and the segmentation branch network outputs constraint relations among a plurality of forking endpoints; obtaining a plurality of combinations of the plurality of split endpoints according to a constraint relation, wherein each combination comprises two split endpoints; generating the actual splitting length of the monocrystalline silicon according to the distance between the two splitting end points in each combination; adjusting the pull rate of the monocrystalline silicon based on the actual length of the split.
According to a third aspect of the present invention, there is provided a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, causes the method of any one of the first and second aspects to be performed.
According to a fourth aspect of the present invention, there is provided an electronic device comprising a memory and a processor, the memory having stored thereon computer instructions, wherein the computer instructions, when executed by the processor, cause the method of any one of the first and second aspects to be performed.
The invention provides an intelligent adjusting method for the crystal pulling speed of a shouldering stage of monocrystalline silicon and a storage medium, wherein the method comprises the following steps: obtaining a monocrystalline silicon image in a shouldering stage; obtaining a plurality of split endpoints in the monocrystalline silicon image based on the key point detection; carrying out segmentation processing on the monocrystalline silicon image to obtain a constraint relation among a plurality of split endpoints; obtaining a plurality of combinations of the plurality of split endpoints according to a constraint relation, wherein each combination comprises two split endpoints; generating the actual splitting length of the monocrystalline silicon according to the distance between the two splitting end points in each combination; adjusting the pull rate of the monocrystalline silicon based on the actual length of the split. In the prior art, workers need to visually observe the diameter change rate of a single crystal silicon rod in a single crystal furnace, and the pulling speed control efficiency in the shouldering process is low due to the fact that the pulling speed is manually controlled according to the diameter change rate of the single crystal silicon rod.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of an intelligent detection method for forking during a shouldering stage of monocrystalline silicon according to an embodiment of the invention;
FIG. 2 is a schematic illustration of a monocrystalline silicon image of an embodiment of the invention;
FIG. 3 is a schematic line view of the divergence of an embodiment of the present invention;
fig. 4 to fig. 11 are schematic diagrams of an intelligent detection method for forking in the shouldering stage of monocrystalline silicon according to an embodiment of the invention; and
FIG. 12 is a schematic diagram of a convolutional neural network of an embodiment of the invention.
Detailed Description
In order to make the above and other features and advantages of the present invention more apparent, the present invention is further described below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are for purposes of illustration only and are not intended to be limiting.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that the specific details need not be employed to practice the present invention. In other instances, well-known steps or operations are not described in detail to avoid obscuring the invention.
Example one
The invention provides an intelligent detection method for the forking of a monocrystalline silicon shouldering stage, which comprises the following steps of:
in step S11, a silicon single crystal image at the shouldering stage is acquired.
Specifically, the main executing body of the scheme can be realized by an upper computer or other hardware equipment with data processing, wherein the upper computer or other hardware equipment with data processing establishes a communication relation with the single crystal furnace, in the process of pulling the single crystal furnace, if a shouldering stage is reached, the scheme can acquire the single crystal silicon image in the shouldering stage through an industrial camera of the single crystal furnace, the single crystal silicon image in the shouldering stage is shown in fig. 2, and as can be seen from fig. 2, four ridge lines appear on the single crystal silicon rod, and a gap can be found along the downward direction of each ridge line. The schematic diagram of the ridge and the line of the split is shown in fig. 3, AB is the ridge, and BC and BD are two splits of the ridge AB.
In step S13, a plurality of split end points in the monocrystalline silicon image are obtained based on the key point detection.
Specifically, the monocrystalline silicon image can be identified through the key point detection technology, and with reference to fig. 4, eight split endpoints in the monocrystalline silicon image can be identified and obtained through the key point detection technology (feature point detection technology), it should be noted that, under normal conditions, one split can occur in each edge line of four edge lines in the shoulder-off stage, and each split image includes two split endpoints. It should be further noted that the above-mentioned keypoint detection and identification technology may be implemented by a trained first convolution neural network, and the first convolution neural network may be trained by using keypoint (split endpoint) samples after artificial labeling.
In step S15, the silicon single crystal image is divided to obtain a plurality of constraint relationships of the split end points.
Specifically, in this scheme, with reference to fig. 6, after the key point detection is performed on the single crystal silicon image, the single crystal silicon image is segmented to obtain the constraint relationships of the multiple forking endpoints, where it should be noted that the constraint relationships of the forking endpoints can be used to represent which two forking endpoints have an association relationship and thus belong to the same fork. In connection with fig. 5, if K1 and K2 have a constraint relationship, then K1 and K2 belong to two split endpoints of the same split. It should be noted here that in step S13, only eight split endpoints are identified, but it is still unknown which two points belong to the same split, so the constrained relationship of the split endpoints is obtained by step S15.
It should be noted that the foregoing segmentation processing on the monocrystalline silicon image may be implemented by using a pre-trained convolutional neural network, and the neural network may be obtained by training a connection line sample of an artificially labeled split endpoint.
Step S17, with reference to fig. 7, obtaining a plurality of combinations of the plurality of split endpoints according to the constraint relationship, where each combination includes two split endpoints.
Step S19, the actual splitting length of the monocrystalline silicon is generated according to the distance between the two splitting end points in each combination.
Specifically, two divergence end points belonging to the same combination belong to the same divergence, and it should be noted here that the scheme can calculate the comparison between the distance between the two divergence end points in each combination and the preset distance so as to verify the accuracy of the actual divergence length of the obtained monocrystalline silicon.
In step S21, the pulling rate of the silicon single crystal is adjusted based on the actual length of the slit.
Specifically, after the actual length of each fork in the monocrystalline silicon image is obtained, the pulling rate of the monocrystalline silicon can be adjusted according to the actual length of the fork, and it should be noted that in the prior art, the morphology of the monocrystalline silicon rod needs to be observed by human eyes, and then the pulling rate of the monocrystalline silicon is manually adjusted according to the diameter change rate of the monocrystalline silicon rod, so as to avoid the pulling faults such as wire breakage, etc., while the diameter change rate of the monocrystalline silicon rod is difficult to be determined accurately by human. According to the scheme, the positive correlation relation exists between the forking length of the silicon single crystal rod and the change rate of the silicon single crystal rod, therefore, the crystal pulling speed of the silicon single crystal is adjusted based on the actual forking length of the silicon single crystal rod.
Optionally, after acquiring the single crystal silicon image in the shouldering stage in step S11, the method in this embodiment may further include: step S12 of extracting features of the monocrystalline silicon image and dividing the features of the monocrystalline silicon image into a first set of features and a second set of features, wherein,
the step of obtaining a plurality of split end points in the monocrystalline silicon image based on the key point detection in step S13 may include: step S131, inputting the first group of features into a first convolution neural network to output a plurality of split endpoints in the monocrystalline silicon image;
the step of performing the division processing on the single crystal silicon image in step S15 may include: step S151, inputting the second set of features into a second convolutional neural network to output a plurality of split positions and obtain a constraint relationship between a plurality of split endpoints based on the plurality of split positions.
Specifically, with reference to fig. 8, the single crystal silicon image may be input into a feature extraction network, the feature extraction network may be Resnet18 (including 17 convolutional layers conv and 1 full connection layer fc), the feature extraction network may extract a set of features of the single crystal silicon image from the single crystal silicon image, then perform feature grouping on the set of features of the single crystal silicon image, divide the set of features into a first group of features and a second group of features, then input the first group of features into a first convolutional neural network, the first convolutional neural network is configured to perform key point identification on the first group of features, so as to output key points (split endpoints), and the second convolutional neural network is configured to perform segmentation processing on the second group of features, so as to output constraint relationships of multiple split endpoints.
It should be noted here that, the feature grouping mode firstly reduces the number of model parameters, and at the same time, the feature grouping is equivalent to adding a sparse structure in the model to play a regular role, so that the overfitting risk is reduced, and meanwhile, features that need to be paid attention to by different tasks (key point detection, line segmentation) are different, and these features that need to be differentiated can form two highly differentiated features through the feature grouping and model training optimization, which is beneficial to improving the effects of two different tasks.
It should be noted here that, in the existing deep neural network models such as multi-classification model, multi-target detection and the like, only one branch is often used to complete multi-classification task, since the tasks of key point detection and image segmentation in the scheme have great difference on the optimization target, if one branch is still used for completing the two tasks, two optimization targets can influence each other in the optimization process, and the accuracy of final measurement can be influenced, the scheme is different from the prior art, a key point detection model (namely, a first convolution neural network) and a structure constraint segmentation model (namely, a second convolution neural network) in the scheme are two independent branches in the neural network, one group of feature map input key point detection branches are used for key point detection, the other group of feature map input segmentation branches are used for structure constraint segmentation, and the mode brings great promotion to the final measurement accuracy.
Alternatively, the first convolutional neural network or the second convolutional neural network may be formed by combining a plurality of convolutional calculation layers, batch normalization layers, activation function layers, and deconvolution layers.
Optionally, before acquiring the single crystal silicon image in the shouldering stage, the method further includes:
and S08, obtaining a single-channel gray image from the monocrystalline silicon original image.
And S09, extracting a gradient map of the single-channel gray image in the horizontal and vertical directions.
And S10, performing channel superposition on the single-channel gray-scale image and the horizontal and vertical gradient images to generate a monocrystalline silicon image.
Specifically, the monocrystalline silicon original image is an unprocessed original image collected from a monocrystalline furnace industrial camera, the scheme can firstly carry out fitting circle processing on the monocrystalline silicon original image to obtain a circle center C and a diameter D, then expanding the fitted circle by taking the center C of the fitted circle as the center to obtain a circumscribed square cut crystal region of a circle with the diameter of 1.7 x D, and scaling the cropped crystal region to 256 x 256 dimensions to obtain a single-channel grayscale image I, then extracting the horizontal and vertical direction gradient map Igx and Igy for the single-channel gray image I, and calculating the weighted sum of the absolute values to obtain the gradient map Igxy, and finally, superposing the single-channel gray image I and the gradient map Igxy to form a 2 x 256 matrix, normalizing the matrix to obtain a monocrystalline silicon image, the single crystal silicon image is a two-channel picture generated by superimposing channels, and the single crystal silicon image is used as an input of the above-mentioned feature extraction network in fig. 8.
It should be noted that, in the prior art, if feature extraction is to be performed, only a gray scale map or an RGB three-channel color map is input, and unlike the prior art, the gradient map and the gray scale map are superimposed as input, and a generated image is a two-channel image, which is equivalent to a pre-processing model with a fixed parameter added in front of the model (i.e., the feature extraction network).
Alternatively, there is a scaling ratio from the size of the original single crystal silicon image to the size of 256 × 256, and in step S19, the actual splitting length may be obtained by mapping the distance between the splitting end points to the size of the original coordinate system according to the scaling ratio.
Optionally, the keypoint labels for training the first convolutional neural network are generated by:
firstly, acquiring a monocrystalline silicon sample image, marking a forking end point and a connecting line between the forking end points in the monocrystalline silicon sample image by combining with a figure 9, then determining a square area of a key point label picture by taking each forking end point as a center, wherein the side length of the square area is delta sigma, 
sigma 2, wherein for each pixel i in each square area, the pixel value e is assigned to each pixel i-β×dFilling to generate a local Gaussian distribution map with each divergent end point in the monocrystalline silicon sample image as a center, wherein beta is a coefficient,
d is the square of the distance from each pixel i to the center point of the square region.
It should be noted that, the gaussian distribution can be used to determine the distance from the central point to the surrounding points of the key point, so as to perform model training effectively.
Specifically, the present scheme may generate a keypoint label (i.e., a split end label) for training the first convolutional neural network according to the user label in fig. 9, that is, generate a local gaussian distribution map centered at each split point, where the local gaussian distribution map of a single split point is shown in fig. 10, and the local gaussian distribution maps of four split points are shown in fig. 11. It should be noted here that the label in the second convolutional neural network can be directly generated by the straight line label for the divergent end point.
Optionally, the first convolutional neural network is optimized by a first loss function as follows:
wherein, the first and the second end of the pipe are connected with each other,
in order to be a key point label,
for the value of the ith element in the keypoint label,
to label according to the key point 
The generated binary label is used for carrying out label matching on the label,
is the value of the ith element in the binary label, Y is the split end point of the first convolution neural network output, YiAnd outputting the numerical value of the ith position in the monocrystalline silicon image output by the first convolution neural network.
Specifically, the conventional key point detection loss function is to calculate the sum of absolute values of differences between each pixel of an output graph and each pixel of a label graph, and the optimization method has the following problems that firstly, for a key point, processing a label into a two-dimensional gaussian distribution form is a common method when generating the label, in this way, only a central point is a real key point, and a non-zero region of the gaussian distribution has a much larger area relative to the key point, and using an inter-pixel interpolation as a loss function means that pixels in a region near the key point of an output image are required to be a certain value, which is lack of basis, in the training process, the model convergence speed is slow, and the accuracy of the key point detection result is low.
The scheme is different from the prior art, and can be seen from the formula, the central point of Gaussian distribution is regarded as a unique positive sample in the loss function loss1, other areas of the Gaussian distribution are regarded as negative samples, the classification loss function is used for training a model, the numerical values of the other areas of the Gaussian distribution are used for measuring the distance from the point to the central point, the distance is used for calculating the classification loss weight of the point, the closer to the actual central point, the lower the loss weight of the point as the negative sample is, and therefore the problems that the negative samples are too many and the positive samples account for too little total loss are solved while the problem is solved.
Optionally, the second convolutional neural network is optimized by a second loss function as follows:
wherein, the first and the second end of the pipe are connected with each other,
in order to have a binary tag for the end point of the divergence,
is the value of the ith element in the binary label of the forking endpoint, S is the constraint relation output by the constraint segmentation model of the forking endpoint, SiIs the value of the ith position in the monocrystalline silicon image output by the second convolution neural network.
Specifically, the existing segmentation task usually uses a classified cross entropy loss function to calculate loss for pixel points one by one, but for a fork detection scene, structural constraint segmentation is a binary segmentation task, positive sample pixels are far less than negative sample pixels, and the positive sample loss ratio is too low, so that model optimization is difficult due to the adoption of the classified cross entropy loss function.
The scheme is different from the prior art, and as can be seen from the formula, the used loss function loss2 increases the cross-over ratio of the soft output result and the label calculation on the basis of the common cross-over loss
The cross-over value is independent of the number of positive sample pixels as a loss function, and is only dependent on the degree of overlap of the output result and the label, so the loss function loss2 can be used to counteract the serious imbalance of the number of different types of pixels.
Optionally, in step S21, the step of adjusting the pull rate of the single crystal silicon based on the actual length of the split may include:
and step S211, determining the sub-stage of the monocrystalline silicon in the shouldering stage.
Step S212, the standard divergence length associated with the sub-phase is obtained.
Step S213, comparing the actual splitting length with the standard splitting length.
Step S214, controlling and reducing the crystal pulling speed of the monocrystalline silicon under the condition that the actual splitting length is smaller than the standard splitting length, so that the actual splitting length is increased to the standard splitting length;
and S215, controlling and increasing the pulling speed of the monocrystalline silicon under the condition that the actual splitting length is greater than the standard splitting length so as to reduce the actual splitting length to the standard splitting length.
Specifically, in the shouldering process, after the thin neck of the single crystal silicon rod reaches the specified length, the temperature is reduced, the pulling speed is reduced, the thin neck gradually grows to the specified diameter, and in the process, the change of the ridge line is as follows: closing, splitting, and the size of the split is unchanged when the split is increased to a certain size. The stage from the forking to the increasing to a certain size is each sub-stage of the shouldering stage, the standard forking length can be preset in each sub-stage, and the crystal pulling effect can be optimal under the condition that the actual forking length reaches the standard forking length, so that the difference between the current forking length and a set value (namely the standard forking length) is continuously calculated in the shouldering process, if the forking length is smaller than the preset value, the pulling speed is reduced, if the forking length is larger than the preset value, the pulling speed is improved, and the control mode can be PID (proportional Integral derivative) control. Optionally, a camera fixed on the single crystal furnace shoots an image, then the fork length is calculated through a fork detection algorithm, the value is transmitted to the PLC, the difference value between the current fork value and the target fork length (namely the standard fork length) is calculated, the proportion, the differential and the integral (PID) of the difference value are calculated to output a control signal, and finally the rotating speed of a motor at the top of the furnace is adjusted to control the pulling speed.
Optionally, in the above step S214 and step S215, the present embodiment may further calculate a difference between the actual length of the slit and the standard length of the slit, and obtain an acceleration associated with the difference, and then increase or decrease the pulling speed of the monocrystalline silicon according to the acceleration, that is, in the case of a larger difference, the acceleration is correspondingly increased, so that the adjustment of the pulling speed can be rapidly realized to rapidly reach the optimal pulling speed in the case of a larger difference between the actual length of the slit and the standard length of the slit.
Example two
The application also provides an intelligent detection method for the forking of the monocrystalline silicon shouldering stage, which comprises the following steps:
in step S31, a silicon single crystal image at the shoulder-on stage is acquired.
Step S33, inputting the monocrystalline silicon image into a convolutional neural network, wherein the convolutional neural network comprises a feature extraction trunk network, a key point detection branch network and a segmentation branch network, and the key point detection branch network and the segmentation branch network are respectively connected with the feature extraction trunk network; extracting the characteristics of the monocrystalline silicon image from the monocrystalline silicon image by the characteristic extraction backbone network and dividing the characteristics of the monocrystalline silicon image into a first group of characteristics and a second group of characteristics; inputting the first group of characteristics into a key point detection branch network by a characteristic extraction backbone network, and outputting a plurality of split endpoints in the monocrystalline silicon image by the key point detection branch network; the feature extraction backbone network inputs the second set of features into the split branch network, and the split branch network outputs a constraint relationship between the plurality of split endpoints.
It should be noted here that the split branch network may output a plurality of split positions first and obtain a constraint relationship between a plurality of split endpoints based on the plurality of split positions.
And step S35, obtaining a plurality of combinations of the plurality of split endpoints according to the constraint relationship, wherein each combination comprises two split endpoints.
Step S37, actual slit lengths of the single crystal silicon are generated according to the distance between the two slit end points in each combination.
In step S39, the pulling rate of the silicon single crystal is adjusted based on the actual length of the slit.
Specifically, the structure of the convolutional neural network is shown in fig. 12, in the present scheme, a monocrystalline silicon image may be input into the convolutional neural network, a feature extraction trunk network in the convolutional neural network may be Resnet18 (including 17 convolutional layers conv and 1 full connection layer fc), the feature extraction trunk network divides a set of features of the monocrystalline silicon image into a first group of features and a second group of features, then the feature extraction trunk network inputs the first group of features into a key point detection branch network, the key point detection branch network is used for performing key point identification on the first group of features, so as to output key points (split endpoints), the feature extraction trunk network inputs the second group of features into a segmentation branch network, and the segmentation branch network is used for performing segmentation processing on the second group of features, so as to generate a constraint relationship of the split endpoints.
It should be noted here that, in the existing deep neural network models such as a multi-class model and a multi-target detection model, a multi-class task is often completed by using only one branch, but the key point detection and image segmentation tasks in the scheme have great difference in the optimization targets, if the two tasks are still completed by using one branch, the two optimization targets will affect each other in the optimization process, and the final measurement precision will be affected. It should be noted that the technical effects of steps S11 to S21 in the first embodiment are the same as those of the second embodiment.
It will be understood that the specific features, operations, and details described herein above with respect to the method of the present invention may be similarly applied to the apparatus and system of the present invention, or vice versa. Further, each step of the method of the invention described above may be performed by a respective component or unit of the device or system of the invention.
It should be understood that the various modules/units of the apparatus of the present invention may be implemented in whole or in part by software, hardware, firmware, or a combination thereof. The modules/units may be embedded in the processor of the computer device in the form of hardware or firmware or independent from the processor, or may be stored in the memory of the computer device in the form of software for being called by the processor to execute the operations of the modules/units. Each of the modules/units may be implemented as a separate component or module, or two or more modules/units may be implemented as a single component or module.
In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored thereon computer instructions executable by the processor, the computer instructions, when executed by the processor, instructing the processor to perform the steps of the method of embodiments one, two of the present invention. The computer device may broadly be a server, a terminal, or any other electronic device having the necessary computing and/or processing capabilities. In one embodiment, the computer device may include a processor, memory, a network interface, a communication interface, etc., connected by a system bus. The processor of the computer device may be used to provide the necessary computing, processing and/or control capabilities. The memory of the computer device may include non-volatile storage media and internal memory. An operating system, a computer program, and the like may be stored in or on the non-volatile storage medium. The internal memory may provide an environment for the operating system and the computer programs in the non-volatile storage medium to run. The network interface and the communication interface of the computer device may be used to connect and communicate with an external device through a network. Which when executed by a processor performs the steps of the method of the invention.
The invention may be implemented as a computer-readable storage medium, having stored thereon a computer program, which, when executed by a processor, causes the steps of the method of embodiment one of the invention to be performed. In one embodiment, the computer program is distributed across a plurality of computer devices or processors coupled by a network such that the computer program is stored, accessed, and executed by one or more computer devices or processors in a distributed fashion. A single method step/operation, or two or more method steps/operations, may be performed by a single computer device or processor, or by two or more computer devices or processors. One or more method steps/operations may be performed by one or more computer devices or processors, and one or more other method steps/operations may be performed by one or more other computer devices or processors. One or more computer devices or processors may perform a single method step/operation, or perform two or more method steps/operations.
It will be appreciated by those of ordinary skill in the art that the method steps of the present invention may be directed to associated hardware, such as a computer device or processor, for performing the steps of the present invention by a computer program, which may be stored in a non-transitory computer readable storage medium, which when executed causes the steps of the present invention to be performed. Any reference herein to memory, storage, databases, or other media may include non-volatile and/or volatile memory, as appropriate. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (prom), electrically programmable ROM (eprom), electrically erasable programmable ROM (eeprom), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, hard disk, solid state disk, and the like. Examples of volatile memory include Random Access Memory (RAM), external cache memory, and the like.
The respective technical features described above may be arbitrarily combined. Although not all possible combinations of features are described, any combination of features should be considered to be covered by the present specification as long as there is no contradiction between such combinations.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for intelligently adjusting the crystal pulling speed of a monocrystalline silicon shouldering stage is characterized by comprising the following steps:
obtaining a monocrystalline silicon image in a shouldering stage;
obtaining a plurality of split endpoints in the monocrystalline silicon image based on the key point detection;
carrying out segmentation processing on the monocrystalline silicon image to obtain a constraint relation among a plurality of split endpoints;
Obtaining a plurality of combinations of the plurality of split endpoints according to a constraint relation, wherein each combination comprises two split endpoints;
generating the actual splitting length of the monocrystalline silicon according to the distance between the two splitting end points in each combination;
adjusting the pull rate of the monocrystalline silicon based on the actual length of the split.
2. The method of claim 1, wherein after acquiring the monocrystalline silicon image at the shouldering stage, the method comprises: extracting features of the monocrystalline silicon image and classifying the features of the monocrystalline silicon image into a first set of features and a second set of features, wherein,
the step of obtaining a plurality of split endpoints in the monocrystalline silicon image based on the keypoint detection comprises: inputting the first set of features into a first convolutional neural network to output a plurality of split endpoints in the single crystal silicon image;
the step of performing division processing on the monocrystalline silicon image includes: inputting the second set of features into a second convolutional neural network to output a plurality of split positions and deriving a constrained relationship between the plurality of split endpoints based on the plurality of split positions.
3. The method of claim 1, wherein prior to acquiring the single crystal silicon image in the shouldering stage, the method further comprises:
Obtaining a single-channel gray image from the monocrystalline silicon original image;
extracting a gradient map in the horizontal and vertical directions of the single-channel gray image;
and superposing the single-channel gray image and the transverse and longitudinal gradient images to generate the monocrystalline silicon image.
4. The method of claim 2, wherein the keypoint labels used to train the first convolutional neural network are generated by:
acquiring a monocrystalline silicon sample image, wherein a fork end point is marked in the monocrystalline silicon sample image;
determining key points centered on each of the divergent end pointsA square region of the label picture, the square region having a side length of delta sigma,
sigma 2, wherein each pixel i in each square region is assigned a pixel value e-β×dFilling to generate a local Gaussian distribution map with each of the split ends in the single crystal silicon sample image as a center, wherein beta is a coefficient,
d is the square of the distance of each pixel i to the center point of the square region.
5. The method of claim 2, wherein the first convolutional neural network is optimized by a first loss function as follows:
wherein the content of the first and second substances,
in order to be a key point label,
the value of the ith element in the keypoint label, 
Is based on key point label
The generated binary label is used for carrying out label matching on the label,
is the value of the ith element in the binary label, and Y is the split end of the first convolution neural network outputPoint, YiAnd outputting the numerical value of the ith position in the monocrystalline silicon image output by the first convolution neural network.
6. The method of claim 2, wherein the second convolutional neural network is optimized by a second loss function as follows:
wherein, the first and the second end of the pipe are connected with each other,
in order to have a binary tag for the end point of the divergence,
is the value of the ith element in the binary label of the forking endpoint, S is the constraint relation output by the constraint segmentation model of the forking endpoint, SiIs the value of the ith position in the monocrystalline silicon image output by the second convolution neural network.
7. The method of claim 1, wherein the step of adjusting the pull rate of the single crystal silicon based on the actual jag length comprises:
determining the sub-stage of the monocrystalline silicon in the shouldering stage;
obtaining a standard divergence length associated with the sub-phase;
comparing the actual splitting length with the standard splitting length;
under the condition that the actual splitting length is smaller than the standard splitting length, controlling and reducing the crystal pulling speed of the monocrystalline silicon, so that the actual splitting length is increased to the standard splitting length;
And under the condition that the actual forking length is greater than the standard forking length, controlling and increasing the crystal pulling speed of the monocrystalline silicon, so that the actual forking length is reduced to the standard forking length.
8. An intelligent detection method for the split of a monocrystalline silicon shouldering stage is characterized by comprising the following steps:
obtaining a monocrystalline silicon image in a shouldering stage;
inputting the monocrystalline silicon image into a convolutional neural network, wherein the convolutional neural network comprises a feature extraction trunk network, a key point detection branch network and a segmentation branch network, and the key point detection branch network and the segmentation branch network are respectively connected with the feature extraction trunk network; the feature extraction backbone network extracts features of the monocrystalline silicon image from the monocrystalline silicon image and divides the features of the monocrystalline silicon image into a first group of features and a second group of features; the feature extraction backbone network inputs the first group of features into the key point detection branch network, and the key point detection branch network outputs a plurality of forking endpoints in the monocrystalline silicon image; the feature extraction backbone network inputs a second group of features into the segmentation branch network, and the segmentation branch network outputs constraint relations among a plurality of forking endpoints;
Obtaining a plurality of combinations of the plurality of split endpoints according to a constraint relation, wherein each combination comprises two split endpoints;
generating the actual splitting length of the monocrystalline silicon according to the distance between the two splitting end points in each combination;
and adjusting the pulling speed of the monocrystalline silicon based on the actual jag length.
9. A non-transitory computer readable storage medium having stored thereon a computer program, characterized in that the computer program, when executed by a processor, causes the method of any of claims 1 to 8 to be performed.
10. An electronic device comprising a memory and a processor, the memory having stored thereon computer instructions, wherein the computer instructions, when executed by the processor, cause the method of any of claims 1-8 to be performed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210284347.8A CN114752996A (en) | 2022-03-22 | 2022-03-22 | Intelligent adjustment method for monocrystalline silicon shoulder-laying stage forking and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210284347.8A CN114752996A (en) | 2022-03-22 | 2022-03-22 | Intelligent adjustment method for monocrystalline silicon shoulder-laying stage forking and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114752996A true CN114752996A (en) | 2022-07-15 |
Family
ID=82327087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210284347.8A Pending CN114752996A (en) | 2022-03-22 | 2022-03-22 | Intelligent adjustment method for monocrystalline silicon shoulder-laying stage forking and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114752996A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115984276A (en) * | 2023-03-20 | 2023-04-18 | 内蒙古晶环电子材料有限公司 | Shoulder-laying defect real-time detection method and device, computer equipment and storage medium |
| CN116934727A (en) * | 2023-07-28 | 2023-10-24 | 保定景欣电气有限公司 | Seed crystal welding control method and device in crystal pulling process and electronic equipment |
| CN117350984A (en) * | 2023-10-23 | 2024-01-05 | 保定景欣电气有限公司 | Method and device for detecting shoulder-opening and fork-opening of monocrystalline silicon |
| CN117604618A (en) * | 2023-11-24 | 2024-02-27 | 保定景欣电气有限公司 | Method and device for determining liquid mouth distance in crystal pulling process |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109112625A (en) * | 2018-09-28 | 2019-01-01 | 宁晋晶兴电子材料有限公司 | A kind of monocrystalline silicon speed change shouldering technique |
| CN109183141A (en) * | 2018-10-29 | 2019-01-11 | 上海新昇半导体科技有限公司 | A kind of crystal growth control method, device, system and computer storage medium |
| CN109234795A (en) * | 2018-10-29 | 2019-01-18 | 上海新昇半导体科技有限公司 | A kind of crystal growth control method, device, system and computer storage medium |
| CN111690980A (en) * | 2019-03-11 | 2020-09-22 | 上海新昇半导体科技有限公司 | Crystal growth control method, device and system for shouldering process and computer storage medium |
| CN111893563A (en) * | 2020-08-25 | 2020-11-06 | 连城凯克斯科技有限公司 | Single crystal furnace capable of automatically adjusting shouldering process parameters and control method |
| CN113463185A (en) * | 2021-07-02 | 2021-10-01 | 无锡松瓷机电有限公司 | Single crystal growth control method, device, equipment and computer storage medium |
| CN113789568A (en) * | 2021-09-18 | 2021-12-14 | 无锡唯因特数据技术有限公司 | Single crystal growth control method, apparatus and storage medium |
-
2022
- 2022-03-22 CN CN202210284347.8A patent/CN114752996A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109112625A (en) * | 2018-09-28 | 2019-01-01 | 宁晋晶兴电子材料有限公司 | A kind of monocrystalline silicon speed change shouldering technique |
| CN109183141A (en) * | 2018-10-29 | 2019-01-11 | 上海新昇半导体科技有限公司 | A kind of crystal growth control method, device, system and computer storage medium |
| CN109234795A (en) * | 2018-10-29 | 2019-01-18 | 上海新昇半导体科技有限公司 | A kind of crystal growth control method, device, system and computer storage medium |
| CN111690980A (en) * | 2019-03-11 | 2020-09-22 | 上海新昇半导体科技有限公司 | Crystal growth control method, device and system for shouldering process and computer storage medium |
| CN111893563A (en) * | 2020-08-25 | 2020-11-06 | 连城凯克斯科技有限公司 | Single crystal furnace capable of automatically adjusting shouldering process parameters and control method |
| CN113463185A (en) * | 2021-07-02 | 2021-10-01 | 无锡松瓷机电有限公司 | Single crystal growth control method, device, equipment and computer storage medium |
| CN113789568A (en) * | 2021-09-18 | 2021-12-14 | 无锡唯因特数据技术有限公司 | Single crystal growth control method, apparatus and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 马博渊;刘楚妮;高鸣飞;班晓娟;黄海友;王浩;薛维华;: "基于深度学习和区域感知的多晶体显微组织图像分割方法", 中国体视学与图像分析, no. 02, 25 June 2020 (2020-06-25), pages 32 - 39 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115984276A (en) * | 2023-03-20 | 2023-04-18 | 内蒙古晶环电子材料有限公司 | Shoulder-laying defect real-time detection method and device, computer equipment and storage medium |
| CN115984276B (en) * | 2023-03-20 | 2023-05-16 | 内蒙古晶环电子材料有限公司 | Shoulder defect real-time detection method, device, computer equipment and storage medium |
| CN116934727A (en) * | 2023-07-28 | 2023-10-24 | 保定景欣电气有限公司 | Seed crystal welding control method and device in crystal pulling process and electronic equipment |
| CN116934727B (en) * | 2023-07-28 | 2024-03-08 | 保定景欣电气有限公司 | Seed crystal welding control method and device in crystal pulling process and electronic equipment |
| CN117350984A (en) * | 2023-10-23 | 2024-01-05 | 保定景欣电气有限公司 | Method and device for detecting shoulder-opening and fork-opening of monocrystalline silicon |
| CN117604618A (en) * | 2023-11-24 | 2024-02-27 | 保定景欣电气有限公司 | Method and device for determining liquid mouth distance in crystal pulling process |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114752996A (en) | Intelligent adjustment method for monocrystalline silicon shoulder-laying stage forking and storage medium | |
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| WO2022160771A1 (en) | Method for classifying hyperspectral images on basis of adaptive multi-scale feature extraction model | |
| Dias et al. | Multispecies fruit flower detection using a refined semantic segmentation network | |
| CN111091109B (en) | Method, system and equipment for predicting age and gender based on face image | |
| CN111652892A (en) | Remote sensing image building vector extraction and optimization method based on deep learning | |
| CN110751644B (en) | Road surface crack detection method | |
| CN113205063A (en) | Visual identification and positioning method for defects of power transmission conductor | |
| CN112541508A (en) | Fruit segmentation and recognition method and system and fruit picking robot | |
| CN111461006B (en) | Optical remote sensing image tower position detection method based on deep migration learning | |
| CN111160407A (en) | Deep learning target detection method and system | |
| CN111161244B (en) | Industrial product surface defect detection method based on FCN + FC-WXGboost | |
| CN111882620A (en) | Road drivable area segmentation method based on multi-scale information | |
| CN109829507B (en) | Aerial high-voltage transmission line environment detection method | |
| CN112749675A (en) | Potato disease identification method based on convolutional neural network | |
| CN111695640A (en) | Foundation cloud picture recognition model training method and foundation cloud picture recognition method | |
| CN113408505B (en) | Chromosome polarity identification method and system based on deep learning | |
| CN110727817B (en) | Three-dimensional model retrieval method based on t-CNN, terminal equipment and storage medium | |
| CN115861686A (en) | Litchi key growth period identification and detection method and system based on edge deep learning | |
| CN113408573B (en) | Method and device for automatically classifying and classifying tile color numbers based on machine learning | |
| CN108288273B (en) | Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network | |
| CN116206208B (en) | Forestry plant diseases and insect pests rapid analysis system based on artificial intelligence | |
| CN106951888B (en) | Relative coordinate constraint method and positioning method of human face characteristic point | |
| CN116542962A (en) | Improved Yolov5m model-based photovoltaic cell defect detection method | |
| Nawawi et al. | Comprehensive pineapple segmentation techniques with intelligent convolutional 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 |