CN114418990A - Continuous industrial crystallization image processing system based on deep learning - Google Patents
Continuous industrial crystallization image processing system based on deep learning Download PDFInfo
- Publication number
- CN114418990A CN114418990A CN202210056060.XA CN202210056060A CN114418990A CN 114418990 A CN114418990 A CN 114418990A CN 202210056060 A CN202210056060 A CN 202210056060A CN 114418990 A CN114418990 A CN 114418990A
- Authority
- CN
- China
- Prior art keywords
- crystal
- target
- image
- deep learning
- network
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a continuous industrial crystallization image processing system based on deep learning, which comprises a data acquisition module, a crystal identification module and a result analysis and display module, wherein the data acquisition module comprises an imaging probe and an imaging controller, the imaging probe is inserted into a solution from the top of an industrial crystallizer to a certain depth, and a high-resolution camera is arranged at the bottom of the probe; the camera monitors the crystallization process in real time in a mode of shooting pictures at a certain frequency; the crystal identification module receives the image sent by the data acquisition module and predicts the position of the crystal in the image by adopting a neural network algorithm based on deep learning; and the result analysis and display module displays the crystal picture and the corresponding statistical result in real time based on the crystal prediction result of the crystal identification module. The system is suitable for on-line monitoring and accurate analysis of the distribution condition of the crystal particle size and the shape in various continuous industrial crystallization processes, thereby providing assistance for optimization and control of the continuous crystallization operation process.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a continuous industrial crystallization image processing system based on deep learning.
Background
At present, crystallization is an important separation operation, is widely applied to purification of solid products in the industries of chemical industry, pharmacy, food processing and the like, and has great significance for online monitoring and analysis of size and shape distribution of crystals in the industrial crystallization process. For batch crystallization operation, the solids concentration therein is usually small in the more important early stages and only becomes large in the less important later stages; for industrial scale continuous crystallization operations, the solids concentration will be maintained at a high level at all times, which tends to result in substantial crystal agglomeration or overlap.
In the prior art, traditional image recognition methods are mainly adopted, and include a multi-scale detection algorithm, a model-based recognition algorithm, a multivariate statistical model, a synthetic image analysis algorithm and the like. Although these methods work well in some simple cases, their accuracy is affected by many factors, such as image quality, solid concentration, agglomeration and overlap between crystals, etc., and thus conventional image recognition means are not suitable for crystal image analysis in continuous industrial crystallization processes.
Disclosure of Invention
The invention aims to provide a continuous industrial crystallization image processing system based on deep learning, which is suitable for online monitoring and accurate analysis of the distribution condition of crystal particle sizes and shapes in various continuous industrial crystallization processes, thereby providing assistance for optimization and control of a continuous crystallization operation process.
The purpose of the invention is realized by the following technical scheme:
a continuous industrial crystallization image processing system based on deep learning comprises a data acquisition module, a crystal identification module and a result analysis and display module, wherein:
the data acquisition module comprises an imaging probe and an imaging controller, the imaging probe is inserted into the solution from the top of the industrial crystallizer to a certain depth, and a high-resolution camera is arranged at the bottom of the probe; the camera monitors the crystallization process in real time in a mode of shooting pictures at a certain frequency, and transmits the shot images to the crystal identification module; the imaging controller is used for controlling the operation mode of the imaging probe, and comprises a camera for adjusting the image acquisition frequency;
the crystal identification module receives the image sent by the data acquisition module and predicts the position of the crystal in the image by adopting a neural network algorithm based on deep learning; wherein, the neural network algorithm comprises five parts: part 1 is a feature extraction network for outputting a feature map of a smaller size; part 2 is a target area proposal network for giving information on the position of a target frame that may contain crystals; part 3 is a regional correction network for correcting the specific position of the target frame in the feature map; the 4 th part is a crystal type output network which is used for outputting the final position of the target frame through a plurality of full connection layers and giving out the crystal type in the target frame; the 5 th part is a pixel output network used for outputting specific pixels covered by the crystal in the target frame through a plurality of convolution layers;
the result analysis and display module is used for counting the distribution characteristics of the shape and the size of the crystal based on the crystal prediction result of the crystal identification module, wherein the distribution characteristics comprise equivalent diameter, circularity and aspect ratio; and simultaneously displaying the crystal picture and a corresponding statistical result in real time.
The technical scheme provided by the invention can show that the system is suitable for on-line monitoring and accurate analysis of the distribution condition of the crystal particle size and the shape in various continuous industrial crystallization processes, thereby providing assistance for optimization and control of a continuous crystallization operation process.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are 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 the drawings without creative efforts.
FIG. 1 is a schematic structural diagram of a continuous industrial crystallization image processing system based on deep learning according to an embodiment of the present invention;
FIG. 2 is a diagram illustrating the variation of the loss value with the number of iteration steps according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating statistical results in the form of histogram of equivalent diameters according to an embodiment of the present invention;
fig. 4 is a schematic diagram illustrating comparison between the prediction performances of the deep learning neural network algorithm according to the embodiment of the present invention and the multi-scale algorithm in the prior art.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, not all embodiments, and this does not limit the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic structural diagram of a continuous industrial crystallization image processing system based on deep learning according to an embodiment of the present invention, where the system includes a data acquisition module, a crystal identification module, and a result analysis and display module, where:
the data acquisition module comprises an imaging probe and an imaging controller, the imaging probe is inserted into the solution from the top of the industrial crystallizer to a certain depth, and a high-resolution camera is arranged at the bottom of the probe; the camera monitors the crystallization process in real time in a mode of shooting pictures at a certain frequency, and transmits the shot images to the crystal identification module; the imaging controller is used for controlling the operation mode of the imaging probe, and comprises a camera for adjusting the image acquisition frequency; for example, set to take one every 5 seconds;
the crystal identification module receives the image sent by the data acquisition module and predicts the position of the crystal in the image by adopting a neural network algorithm based on deep learning; wherein, the neural network algorithm comprises five parts: part 1 is a feature extraction network for outputting a feature map of a smaller size; part 2 is a target area proposal network for giving information on the position of a target frame that may contain crystals; part 3 is a regional correction network for correcting the specific position of the target frame in the feature map; the 4 th part is a crystal type output network which is used for outputting the final position of the target frame through a plurality of full connection layers and giving out the crystal type in the target frame; the 5 th part is a pixel output network used for outputting specific pixels covered by the crystal in the target frame through a plurality of convolution layers;
in a specific implementation, in the neural network algorithm adopted by the crystal identification module,
the feature extraction network automatically extracts deep feature information of an input image through a multilayer convolutional neural network and outputs a feature map with a smaller size; the neural network can select some classical deep learning network structures, such as ResNet, VGG, GoogleNet, and the like;
the target area suggestion network firstly generates a large number of frames to be screened which cover all positions and contain different sizes and different aspect ratios based on the output characteristic diagram, and then outputs a certain number of target candidate frame positions with higher probability under the action of a plurality of neural network layers; the position of the target candidate frame is represented by a horizontal coordinate x and a vertical coordinate y of the center point of the target candidate frame, and the width w and the height h of the target candidate frame, namely (x, y, w, h); meanwhile, outputting a two-classification result of the position of the target candidate frame, namely whether the candidate frame contains the target or not, but not distinguishing the specific class of the target, wherein if the candidate frame contains the target, the classification result is 1, otherwise, the classification result is 0;
the strategy adopted by the regional correction network is to keep the floating point numerical value of the position of the output target candidate frame, and when some information of related elements of the candidate frame is needed, the information of adjacent four points in the feature map is acquired by bilinear interpolation, so that the specific position of the target frame in the feature map is corrected. The reason is that in the process of continuously transforming the image feature map size, the position of the candidate frame is necessarily changed correspondingly, however, the transformed position coordinates of the candidate frame may be floating point numbers, if the transformed position coordinates are processed to be integers approximately, the error after multiple transformations is larger and larger, and finally the precision of the recognition algorithm is reduced.
The crystal type output network adopts a full-connection module to be respectively connected with two full-connection layers with different functions; the types of crystals contained in the target frame after the output correction of one full-connection layer are generally distinguished by different numbers; the other full-connection layer outputs final position information of a target frame containing the crystal, specifically represented as a horizontal coordinate x 'and a vertical coordinate y' of a center point thereof, and a width w 'and a height h' thereof, namely (x ', y', w ', h');
the pixel output network outputs mask codes of all pixels in the target frame through the plurality of convolution layers based on the corrected target frame information, namely, whether each pixel belongs to a crystal is judged, a value 1 represents that the pixel belongs to the crystal, and a value 0 represents a background. In order to avoid the mutual interference among the pixels of various types and reduce the accuracy of the algorithm, the part does not give the specific type of the pixel belonging to the crystal, and the specific type needs to be further determined by combining the types of the whole target frame of the part.
In specific implementation, a neural network prediction model adopted in the crystal identification module needs to be trained, positions of all crystals in an input image are labeled firstly, an operation mode is that a series of discrete points are drawn in a polygonal mode along the edge of each crystal, in general, if the boundary line of the crystal is more tortuous or the length of the boundary line is longer, more discrete points are needed to describe the shape of the crystal, the number of points needed for labeling one crystal is about 30 on average, the number of pictures needed by the training model is determined according to the crystal condition in the picture and specific requirements, if the concentration of the crystals in the picture is larger and the overlapping degree between the crystals is higher, the number of the needed pictures is more, and more than 1000 pictures are generally suggested. In addition, the number of the pictures is generally in direct proportion to the number of crystal types to be distinguished, all crystal position information marked on each picture is stored in a json format file, and a json format data file is finally output after marking is finished;
training a neural network prediction model based on the labeled image data, continuously calculating a loss function in an iterative process, and then further updating corresponding algorithm parameter values so as to continuously optimize the algorithm, wherein the loss function expression is as follows:
L＝Lcls+εLbox+Lmask (1)
wherein L iscls，LboxAnd LmaskRespectively corresponding to classification loss, target frame position loss and pixel mask loss; the parameter epsilon is used for balancing the numerical value size relationship among different loss types;
for example, the model training process may be performed through two mainstream deep learning frameworks of tensoflow or PyTorch, a log file is generated in the training process, which includes a loss value condition of the model, and may be displayed by a tensorbard, as shown in fig. 2, which is a schematic diagram of a change condition of the loss value with an iteration step number according to an embodiment of the present invention, the loss value rapidly decreases in an initial stage of the training and then gradually decreases, the training is completed when the loss value decreases to a sufficient hour, and a trained neural network prediction model is used to predict a crystal condition in a new image.
After the crystal identification module finishes the crystal prediction, the result analysis and display module counts distribution characteristics including equivalent diameter, circularity and aspect ratio of the crystal in terms of shape and size based on the crystal prediction result of the crystal identification module; and simultaneously displaying the crystal picture and a corresponding statistical result in real time.
In addition, in the process of counting the equivalent diameter, the circularity and the aspect ratio of the crystal by the result analysis and display module:
since the shape of the crystal usually deviates from a circle, the size of the crystal can be described by the equivalent diameter, the area S of the crystal can be determined according to the number of pixel points occupied by the crystal on an image, and each pixel point corresponds to a certain actual area. Assuming that the shape of the crystal is circular, the corresponding equivalent diameter D is calculated as follows:
wherein S is the crystal area;
by calculating the equivalent diameters of all crystals in the output image, a corresponding crystal size distribution graph can be given, and fig. 3 is a schematic diagram of a statistical result in the form of a histogram of equivalent diameters according to an embodiment of the present invention.
The circularity of a crystal can be used to characterize the shape of the crystal, with a circularity value of 1.0 for an ideal circular crystal, a convex crystal typically having a circularity value of less than 1.0, and a concave crystal having a circularity of greater than 1.0. The perimeter L of the crystal boundary line can be calculated by determining the number of the pixel points occupied by the crystal boundary line. The circularity e of each crystal is calculated according to the following formula:
wherein L is the perimeter of the crystal;
in order to characterize the shape characteristics of the crystal in different directions, the aspect ratio of the crystal can be further analyzed, the width W and the height H of the crystal are respectively defined as the lengths of the crystal in the horizontal direction and the vertical direction, in order to eliminate the influence of different orientations of the crystal, the calculation mode of the aspect ratio AR is determined according to the length ratio of the width of the crystal to the height of the crystal, and when the length ratio is greater than 1.0, the calculation mode of the aspect ratio AR is as follows:
if the length ratio is less than 1.0, the aspect ratio AR is calculated as follows:
wherein W is the width of the crystal; h is the height of the crystal;
thus, regardless of the particular orientation of the crystal, its aspect ratio AR will be greater than 1.0.
It is noted that those skilled in the art will recognize that embodiments of the present invention are not described in detail herein.
The effect brought by the deep learning-based neural network algorithm adopted by the crystal identification module is explained in detail below, and fig. 4 is a schematic diagram showing comparison between the prediction performances of the deep learning neural network algorithm and the multi-scale algorithm in the prior art, wherein the diagram shows comparison between the prediction performances of 180 image test sets, including two aspects of accuracy and recall rate, and the accuracy rate is defined as the proportion of the real crystal number in all crystals identified by the algorithm; recall is defined as the proportion of the number of successfully identified portions of all crystals contained in the picture by the algorithm.
The crystallization process image was acquired online by the data acquisition module, and the total number of crystals in the test set was 12,820, with an average of 71 crystals per picture. As can be seen from fig. 4: the accuracy rate of the crystal identification module adopting the neural network algorithm based on deep learning is 72.7% to 100%, and the average rate is 87.8%; whereas the average accuracy of the prior art multi-scale algorithm is only 25.2%, ranging between 2.3% and 58.8%. On the other hand, the recall rate of the multi-scale algorithm is between 1.7% and 37%, with an average value of 15.2%; the average recall rate of the application reaches 90.7%, and the minimum value and the maximum value are 75% and 100% respectively.
Therefore, the accuracy rate and the recall rate of the crystal identification module adopting the neural network algorithm based on deep learning are far superior to those of the multi-scale algorithm in the prior art. In addition, the accuracy and recall of multi-scale algorithms generally decrease as the number of crystals in the image increases, mainly because it is not easy to identify clusters or overlapping crystals that may be present when the concentration of crystals is high, which is not the case in the present application.
Therefore, the continuous industrial crystallization image processing system based on deep learning is suitable for crystal identification of online images under continuous crystallization operation and has high precision.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims. The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Claims (4)
1. The continuous industrial crystallization image processing system based on deep learning is characterized by comprising a data acquisition module, a crystal identification module and a result analysis and display module, wherein:
the data acquisition module comprises an imaging probe and an imaging controller, the imaging probe is inserted into the solution from the top of the industrial crystallizer to a certain depth, and a high-resolution camera is arranged at the bottom of the probe; the camera monitors the crystallization process in real time in a mode of shooting pictures at a certain frequency, and transmits the shot images to the crystal identification module; the imaging controller is used for controlling the operation mode of the imaging probe, and comprises a camera for adjusting the image acquisition frequency;
the crystal identification module receives the image sent by the data acquisition module and predicts the position of the crystal in the image by adopting a neural network algorithm based on deep learning; wherein, the neural network algorithm comprises five parts: part 1 is a feature extraction network for outputting a feature map of a smaller size; part 2 is a target area proposal network for giving information on the position of a target frame that may contain crystals; part 3 is a regional correction network for correcting the specific position of the target frame in the feature map; the 4 th part is a crystal type output network which is used for outputting the final position of the target frame through a plurality of full connection layers and giving out the crystal type in the target frame; the 5 th part is a pixel output network used for outputting specific pixels covered by the crystal in the target frame through a plurality of convolution layers;
the result analysis and display module is used for counting the distribution characteristics of the shape and the size of the crystal based on the crystal prediction result of the crystal identification module, wherein the distribution characteristics comprise equivalent diameter, circularity and aspect ratio; and simultaneously displaying the crystal picture and a corresponding statistical result in real time.
2. The continuous industrial crystal image processing system based on deep learning of claim 1, wherein, in the neural network algorithm adopted by the crystal identification module,
the feature extraction network automatically extracts deep feature information of an input image through a multilayer convolutional neural network and outputs a feature map with a smaller size;
the target area suggestion network firstly generates a large number of frames to be screened which cover all positions and contain different sizes and different aspect ratios based on the output characteristic diagram, and then outputs a certain number of target candidate frame positions with higher probability under the action of a plurality of neural network layers; the position of the target candidate frame is represented by a horizontal coordinate x and a vertical coordinate y of the center point of the target candidate frame, and the width w and the height h of the target candidate frame, namely (x, y, w, h); meanwhile, outputting a two-classification result of the position of the target candidate frame, namely whether the candidate frame contains the target or not, but not distinguishing the specific class of the target, wherein if the candidate frame contains the target, the classification result is 1, otherwise, the classification result is 0;
the strategy adopted by the regional correction network is to keep the floating point numerical value of the position of the output target candidate frame, and when some information of related elements of the candidate frame is needed, the information of the adjacent four points in the feature map is acquired by bilinear interpolation, so that the specific position of the target frame in the feature map is corrected;
the crystal type output network adopts a full-connection module to be respectively connected with two full-connection layers with different functions; one of the full connection layers outputs the type of the crystal contained in the corrected target frame; the other full-connection layer outputs final position information of a target frame containing the crystal, specifically represented as a horizontal coordinate x 'and a vertical coordinate y' of a center point thereof, and a width w 'and a height h' thereof, namely (x ', y', w ', h');
the pixel output network outputs mask codes of all pixels in the target frame through the plurality of convolution layers based on the corrected target frame information, namely, whether each pixel belongs to a crystal is judged, a value 1 represents that the pixel belongs to the crystal, and a value 0 represents a background.
3. The deep learning-based continuous industrial crystalline image processing system according to claim 1, wherein the neural network prediction model adopted in the crystal recognition module is further trained, positions of all crystals in the input image are labeled first, an operation mode is that a series of discrete points are drawn in a polygonal form along an edge of each crystal, and a json-format data file is finally output;
training a neural network prediction model based on the labeled image data, continuously calculating a loss function in an iterative process, and then further updating corresponding algorithm parameter values so as to continuously optimize the algorithm, wherein the loss function expression is as follows:
L＝Lcls+εLbox+Lmask (1)
wherein L iscls,LboxAnd LmaskRespectively corresponding to classification loss, target frame position loss and pixel mask loss; the parameter epsilon is used for balancing the numerical value size relationship among different loss types;
the loss value is rapidly reduced in the initial training stage and then gradually reduced, the training is completed when the loss value is reduced to be small enough, and the trained neural network prediction model is used for predicting the crystal situation in a new image.
4. The continuous industrial crystalline image processing system based on deep learning of claim 1, wherein in the process of the result analysis and demonstration module counting the equivalent diameter, circularity and aspect ratio of the crystal:
assuming that the shape of the crystal is circular, the corresponding equivalent diameter is calculated as follows:
wherein S is the crystal area; the corresponding crystal size distribution diagram can be given by calculating the equivalent diameters of all crystals in the output image;
the circularity e of each crystal is calculated according to the following formula:
wherein L is the perimeter of the crystal;
the calculation mode of the aspect ratio AR is determined according to the length ratio of the width of the crystal to the height of the crystal, and when the length ratio is greater than 1.0, the calculation mode of the aspect ratio AR is as follows:
if the length ratio is less than 1.0, the aspect ratio AR is calculated as follows:
wherein W is the width of the crystal; h is the height of the crystal;
thus, regardless of the particular orientation of the crystal, its aspect ratio AR will be greater than 1.0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210056060.XA CN114418990A (en) | 2022-01-18 | 2022-01-18 | Continuous industrial crystallization image processing system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210056060.XA CN114418990A (en) | 2022-01-18 | 2022-01-18 | Continuous industrial crystallization image processing system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114418990A true CN114418990A (en) | 2022-04-29 |
Family
ID=81273770
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210056060.XA Pending CN114418990A (en) | 2022-01-18 | 2022-01-18 | Continuous industrial crystallization image processing system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114418990A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117092149A (en) * | 2023-10-19 | 2023-11-21 | 大连高佳化工有限公司 | On-line monitoring system for solvency crystallization |
-
2022
- 2022-01-18 CN CN202210056060.XA patent/CN114418990A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117092149A (en) * | 2023-10-19 | 2023-11-21 | 大连高佳化工有限公司 | On-line monitoring system for solvency crystallization |
| CN117092149B (en) * | 2023-10-19 | 2024-01-09 | 大连高佳化工有限公司 | On-line monitoring system for solvency crystallization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10930037B2 (en) | Image processing device for displaying object detected from input picture image | |
| EP3667286A1 (en) | Tire image recognition method and tire image recognition device | |
| US20190077015A1 (en) | Machine learning device, robot system, and machine learning method | |
| CN111242026B (en) | Remote sensing image target detection method based on spatial hierarchy perception module and metric learning | |
| CN111652869B (en) | Slab void identification method, system, medium and terminal based on deep learning | |
| JP6584250B2 (en) | Image classification method, classifier configuration method, and image classification apparatus | |
| CN112052876B (en) | Improved RA-CNN-based fine-grained image detection method and system | |
| CN113298053B (en) | Multi-target unmanned aerial vehicle tracking identification method and device, electronic equipment and storage medium | |
| CN113538603B (en) | Optical detection method and system based on array product and readable storage medium | |
| JP7391504B2 (en) | Information processing device, information processing method and program | |
| CN112419202B (en) | Automatic wild animal image recognition system based on big data and deep learning | |
| CN114418990A (en) | Continuous industrial crystallization image processing system based on deep learning | |
| CN113052873A (en) | Single-target tracking method for on-line self-supervision learning scene adaptation | |
| CN111627050A (en) | Training method and device for target tracking model | |
| JP6924031B2 (en) | Object detectors and their programs | |
| CN116563312B (en) | Method for dividing display image of double-screen machine | |
| CN113869211A (en) | Automatic image annotation and automatic annotation quality evaluation method and system | |
| DE102018102833A1 (en) | POSITION CONTROL SYSTEM, POSITION DETECTION DEVICE AND CONTROL PROGRAM | |
| CN115147418A (en) | Compression training method and device for defect detection model | |
| CN112749675A (en) | Potato disease identification method based on convolutional neural network | |
| CN116883763A (en) | Deep learning-based automobile part defect detection method and system | |
| JP2023123387A (en) | Defect detection method and system | |
| CN110163132B (en) | Correlation filtering tracking method based on maximum response value change rate updating strategy | |
| US20200258254A1 (en) | Method of processing object detection data | |
| CN111507414A (en) | Deep learning skin disease picture comparison and classification method, storage medium and robot |
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 |