CN116595457A - E2LSH eye cornea disease classification method based on residual error network - Google Patents
E2LSH eye cornea disease classification method based on residual error network Download PDFInfo
- Publication number
- CN116595457A CN116595457A CN202310759632.5A CN202310759632A CN116595457A CN 116595457 A CN116595457 A CN 116595457A CN 202310759632 A CN202310759632 A CN 202310759632A CN 116595457 A CN116595457 A CN 116595457A
- Authority
- CN
- China
- Prior art keywords
- hash
- e2lsh
- data
- residual
- classification method
- 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
- 238000000034 method Methods 0.000 title claims abstract description 24
- 208000021921 corneal disease Diseases 0.000 title claims abstract description 19
- 238000004364 calculation method Methods 0.000 claims abstract description 11
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 28
- 239000013598 vector Substances 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000011176 pooling Methods 0.000 claims description 8
- 201000002287 Keratoconus Diseases 0.000 description 6
- 238000012549 training Methods 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 208000028006 Corneal injury Diseases 0.000 description 1
- 208000022873 Ocular disease Diseases 0.000 description 1
- 208000032023 Signs and Symptoms Diseases 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 201000009310 astigmatism Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 210000004087 cornea Anatomy 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 208000030533 eye disease Diseases 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- 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
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Public Health (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Primary Health Care (AREA)
- Epidemiology (AREA)
- Pathology (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The application provides a residual network-based E2LSH eye cornea disease classification method, which comprises the following steps: s1, loading pre-trained weights and parameters into a convolution layer of a new model; s2, deleting a traditional full connection layer in an original residual error network ResNet, designing a new full connection module, reducing the depth of the residual error network and increasing the channel number of feature mapping; s3, dropout is introduced between convolution layers, so that overfitting is effectively prevented; and S4, extracting the output characteristics of the last group of residual blocks as characteristic descriptors. The application removes unnecessary modules from the conventional ResNet architecture, so that the model is more compact, and also adopts variance calculation to construct the dynamic hash index, so that proper barrel width parameters can be selected, the size of the barrel width can be dynamically adjusted, the data in each hash barrel is relatively uniform, and the performance of the model is further improved. The average recognition rate and recall rate of the improved Resnet-v model are kept at a higher level, and the improved Resnet-v model has higher accuracy and stronger robustness.
Description
Technical Field
The application relates to a classification algorithm, in particular to an E2LSH (least squares) eye cornea disease classification method based on a residual error network, and belongs to the technical field of eye cornea disease classification methods.
Background
With age, individuals may suffer from a variety of ocular disorders, keratoconus being a primary corneal distortion characterized by a localized corneal conical bulge expansion and highly irregular myopic astigmatism. Autosomal recessive or dominant inheritance, puberty found, binocular morbidity. The pathology is characterized by fracture of the anterior elastic layer, deformation of the cornea cells, fracture of the posterior elastic layer, etc., which are complications caused by the fracture of the posterior elastic layer, and can be caused by rubbing eyes. In the past, diagnosis of keratoconus mainly depends on routine examination such as slit lamps, and clinically typical slit lamps show Fleischer rings, corneal scars and the like. If these typical clinical symptoms and signs appear, the diagnosis is easier, but for the earlier keratoconus (subclinical phase: asymptomatic, better corrected vision, negative clinical examination), the diagnosis is very difficult. Although many methods have been proposed, identifying keratoconus images remains a challenging problem.
In recent years, with the progress and development of technology, computer technology has been incorporated into people's daily life, convolutional neural networks have been used as branches of artificial intelligence, and because of their ability to extract image features, they have been applied to various fields such as image retrieval, face recognition, traffic scene recognition in traffic fields, and image classification in medical directions, where application of deep learning technology to assist doctors in diagnosis of eye diseases has become a popular research direction. The existing ResNet residual network adopts a Relu function as an activation function, and integrates local information with category differentiation in a convolution layer or a pooling layer. The fully connected output layer of the last layer was logically regressively classified using a linear classifier softmax. Although the degradation problem and gradient disappearance problem of the traditional convolutional neural network along with the continuous deepening of the network layer number are changed, the problems of more parameters, high complexity, weak expression capability and the like exist in the convolutional stage, the calculation amount is large, the function is not suitable for large gradient input in the Relu function training process, after the parameters are updated, the neurons of the ReLU can not have an activating function, the gradient is always zero, the extracted characteristics can not be well distinguished by softmax classification, the model overfitting can be caused, and the inter-class distance is even larger than the inter-class distance. On the other hand, in the E2LSH algorithm, a table of Zhang Haxi is used in order to increase the recall rate. However, the algorithm does not consider the distribution characteristic of the data, because the hash function is randomly generated based on p-steady state distribution, so that the data in the data set is not uniformly distributed in general, the data in the hash bucket of each hash table can be influenced by the original data distribution, so that the data set in part of the hash buckets is caused, and the other part of the hash buckets only has a few data, and therefore, an E2LSH eye cornea disease classification method based on a residual network is provided.
Disclosure of Invention
In view of the above, the present application provides a method for classifying E2LSH ocular cornea diseases based on a residual network, so as to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial choice.
The technical scheme of the embodiment of the application is realized as follows: a method for classifying E2LSH eye cornea diseases based on a residual error network comprises the following steps:
s1, loading pre-trained weights and parameters into a convolution layer of a new model;
s2, deleting a traditional full connection layer in an original residual error network ResNet, designing a new full connection module, reducing the depth of the residual error network and increasing the channel number of feature mapping;
s3, dropout is introduced between convolution layers, so that overfitting is effectively prevented;
s4, extracting output features of the last group of residual blocks as feature descriptors, and replacing a combination of a ReLU layer and a softmax classifier in a standard residual function by combining an improved E2LSH local sensitive hash algorithm;
s5, calculating a corresponding variance matrix of the extracted feature descriptors to obtain feature vectors, constructing a dynamic hash index, selecting proper barrel width parameters to dynamically adjust the barrel width, improving the balance degree of data distribution in each hash barrel, and calculating and obtaining probability of each category by a classifier according to the vectors;
and S6, connecting an average pooling layer to obtain classified output.
Further preferred is: in the S1, the model input image is 224×224×3, and the input is set to be three layers of convolution kernels of 3×3, so that a feature map with a feeling of 5 can be obtained, and the number of parameters can be effectively reduced by using a small convolution kernel, so that training and testing become more effective.
Further preferred is: in the step S1, the output channel size of the first and second 3×3 convolution layers in the three-layer convolution kernel is 32, the stride size is 2, and the output channel size of the third convolution layer is 64, so that the calculation cost of the classification network is greatly reduced under the condition of ensuring the consistency with the previous output trunk information, and the calculation amount of the network model is reduced.
Further preferred is: in the step S2, the full connection modules are divided into 5 modules in total, and each Resnet module has five groups of convolutions.
Further preferred is: in said S3 Dropout layers with a P value of 0.5 are introduced between the convolution layers.
Further preferred is: in the step S4, since a large amount of computation is required by the full connection layer, the full connection layer is modified, the output features of the last group of residual blocks are extracted as feature descriptors, the combination of the ReLU layer and the softmax classifier in the standard residual functions is replaced by combining the improved E2LSH local sensitive hash algorithm, the spatial resolution of the feature map is reduced by using the pooling layer and downsampling in the residual network, so that a lot of details are lost, thereby affecting the accuracy of the model in classifying the graphics, in order to alleviate the gradient disappearance problem in the residual, the output features of the last group of residual blocks are extracted as feature descriptors, the combination of the ReLU layer and the softmax classifier in the standard residual functions is replaced by combining the improved E2LSH local sensitive hash algorithm, the corresponding variance matrix is calculated by combining the extracted feature descriptors to obtain feature vectors, a dynamic hash index is constructed, a proper bucket width parameter is selected, the size of the bucket width is adjusted dynamically, the degree of data distribution in each hash bucket is improved, and each class probability is calculated by the classifier according to the vector.
Further preferred is: in the step S5, when calculating the variance of each data, the ratio of the data to the set is defined as:
when calculating the variance of each data, the average eigenvalue is set as follows:
。
further preferred is: in S5, the intra-class variance is noted as:
the optimal variance value is defined as the corresponding one when the variance is maximumValues, namely:
and sorting eigenvalues of the variance matrix, wherein the corresponding eigenvectors are that the eigenvectors reflect the data density.
Further preferred is: in S5, the hash function in the E2LSH is randomly generated based on the p-steady state distribution, and the data in the hash bucket changes with the data distribution, resulting in a situation that the data in one part of the hash bucket is concentrated, and the other part of the hash bucket has only a few data, so the improved hash function is as follows:
wherein The dynamic hash index is constructed by feature vectors obtained through calculation in the corresponding variance matrix, and proper barrel width parameters are selected so as to dynamically adjust the size of the barrel width and improve the balance degree of data distribution in each hash barrel.
Further preferred is: in the step S5, when the data is hashed and mapped to a hash bucket, selection is requiredThe hash functions form a complete data hash map, and the method is as follows:
wherein ,representing the dimension of the dimension reduced data, +.>Representing functions in the hash function cluster. Original data pass function->Mapping down to->Dimension, the result is->Layering is carried out by adopting a mode of combining an array and a linked list, indexes are established by the hash through a bucket layering mode, and an average pooling layer is connected to obtain classified output.
By adopting the technical scheme, the embodiment of the application has the following advantages:
the application removes unnecessary modules from the conventional ResNet architecture, so that the model is more compact, and also adopts variance calculation to construct the dynamic hash index, so that proper barrel width parameters can be selected, the size of the barrel width can be dynamically adjusted, the data in each hash barrel is relatively uniform, and the performance of the model is further improved. The average recognition rate and recall rate of the improved Resnet-v model are kept at a higher level, the improved Resnet-v model has higher accuracy and stronger robustness, the time for training the model is shorter, and correct classification can be made when the keratoconus feature is detected in the eye picture.
The foregoing summary is for the purpose of the specification only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present application will become apparent by reference to the drawings and the following detailed description.
Drawings
In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings that are necessary for the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the application and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of an algorithm of the present application.
Detailed Description
Hereinafter, only certain exemplary embodiments are briefly described. As will be recognized by those of skill in the pertinent art, the described embodiments may be modified in various different ways without departing from the spirit or scope of the present application. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive.
Embodiments of the present application will be described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the embodiment of the application provides a method for classifying E2LSH eye cornea diseases based on a residual network, which comprises the following steps:
s1, loading pre-trained weights and parameters into a convolution layer of a new model;
s2, deleting a traditional full connection layer in an original residual error network ResNet, designing a new full connection module, reducing the depth of the residual error network and increasing the channel number of feature mapping;
s3, dropout is introduced between convolution layers, so that overfitting is effectively prevented;
s4, extracting output features of the last group of residual blocks as feature descriptors, and replacing a combination of a ReLU layer and a softmax classifier in a standard residual function by combining an improved E2LSH local sensitive hash algorithm;
s5, calculating a corresponding variance matrix of the extracted feature descriptors to obtain feature vectors, constructing a dynamic hash index, selecting proper barrel width parameters to dynamically adjust the barrel width, improving the balance degree of data distribution in each hash barrel, and calculating and obtaining probability of each category by a classifier according to the vectors;
and S6, connecting an average pooling layer to obtain classified output.
In this embodiment, specific: in S1, the model input image is 224×224×3, and the input is set to be three layers of convolution kernels of 3×3, so that a feature map with a feeling of 5 can be obtained, and the number of parameters can be effectively reduced by using a small convolution kernel, so that training and testing become more effective.
In this embodiment, specific: in S1, the output channel sizes of the first and second 3×3 convolution layers in the three-layer convolution kernel are 32, the stride size is 2, and the output channel size of the third convolution layer is 64, so that the calculation cost of the classification network is greatly reduced under the condition of ensuring the consistency with the previous output trunk information, and the calculation amount of the network model is reduced.
In this embodiment, specific: in S2, the fully connected modules are divided into 5 modules in total, and each Resnet module has five sets of convolutions.
In this embodiment, specific: in S3 Dropout layers with a P value of 0.5 are introduced between the convolution layers.
In this embodiment, specific: in S4, since the full-connection layer requires a large amount of computation, the full-connection layer is modified, the output features of the last group of residual blocks are extracted as feature descriptors, and the combination of the ReLU layer and the softmax classifier in the standard residual function is replaced by combining the improved E2LSH locality sensitive hashing algorithm.
In this embodiment, specific: in S5, when calculating the variance of each data, the ratio of the data to the set is defined as:
when calculating the variance of each data, the average eigenvalue is set as follows:
。
in this embodiment, specific: in S5, the intra-class variance is noted as:
the optimal variance value is defined as the corresponding one when the variance is maximumValues, namely:
and sorting eigenvalues of the variance matrix, wherein the corresponding eigenvectors are that the eigenvectors reflect the data density.
In this embodiment, specific: in S5, the hash function in E2LSH is randomly generated based on p-steady state distribution, and the data in the hash bucket changes with the data distribution, resulting in the situation that part of the hash bucket has data concentrated, and another part of the hash bucket has only a few data, so the improved hash function is:
wherein The dynamic hash index is constructed by feature vectors obtained through calculation in the corresponding variance matrix, and proper barrel width parameters are selected so as to dynamically adjust the size of the barrel width and improve the balance degree of data distribution in each hash barrel.
In this embodiment, specific: in S5, when the data is hashed and mapped to a certain hash bucket, selection is requiredThe hash functions form a complete data hash map, and the method is as follows:
wherein ,representing the dimension of the dimension reduced data, +.>Representing functions in the hash function cluster. Original data pass function->Mapping down to->Dimension, the result is->Layering is carried out by adopting a mode of combining an array and a linked list, indexes are established by the hash through a bucket layering mode, and an average pooling layer is connected to obtain classified output.
The application works when in work: on the basis of feature extraction ResNet-50, a new residual network model is designed, pre-trained weights and parameters are loaded into a convolution layer of the new model, an improved E2LSH local sensitive hash algorithm is combined, a corresponding variance matrix is calculated to obtain a feature vector, a dynamic hash index is constructed, a proper barrel width parameter is selected to realize dynamic adjustment of the barrel width, and finally the probability of each category is obtained through calculation of a classifier according to the vector. Through the operation result, the loss can be effectively reduced, and the accuracy rate on the verification set can reach about 97% after 50 epoch training. Therefore, the algorithm can effectively increase the accuracy of the classification result, avoid overfitting, and train the model in a short time, and can make correct classification when the keratoconus feature in the eye picture is detected.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that various changes and substitutions are possible within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. The E2LSH eye cornea disease classification method based on the residual error network is characterized by comprising the following steps of:
s1, loading pre-trained weights and parameters into a convolution layer of a new model;
s2, deleting a traditional full connection layer in an original residual error network ResNet, designing a new full connection module, reducing the depth of the residual error network and increasing the channel number of feature mapping;
s3, dropout is introduced between convolution layers, so that overfitting is effectively prevented;
s4, extracting output features of the last group of residual blocks as feature descriptors, and replacing a combination of a ReLU layer and a softmax classifier in a standard residual function by combining an improved E2LSH local sensitive hash algorithm;
s5, calculating a corresponding variance matrix of the extracted feature descriptors to obtain feature vectors, constructing a dynamic hash index, selecting proper barrel width parameters to dynamically adjust the barrel width, improving the balance degree of data distribution in each hash barrel, and calculating and obtaining probability of each category by a classifier according to the vectors;
and S6, connecting an average pooling layer to obtain classified output.
2. The residual network-based E2LSH ocular corneal disease classification method of claim 1, wherein: in S1, the model input image is 224×224×3, and the input is set to a three-layer 3×3 convolution kernel, so that a feature map having a feeling of 5 can be obtained.
3. The residual network-based E2LSH ocular corneal disease classification method of claim 2, wherein: in S1, the output channel sizes of the first and second 3×3 convolution layers in the three-layer convolution kernel are 32, the stride size is 2, and the output channel size of the third convolution layer is 64.
4. The residual network-based E2LSH ocular corneal disease classification method of claim 1, wherein: in the step S2, the full connection modules are divided into 5 modules in total, and each Resnet module has five groups of convolutions.
5. The residual network-based E2LSH ocular corneal disease classification method of claim 1, wherein: in said S3 Dropout layers with a P value of 0.5 are introduced between the convolution layers.
6. The residual network-based E2LSH ocular corneal disease classification method of claim 1, wherein: in the step S4, since the full connection layer needs a large amount of computation, the full connection layer is modified, the output features of the last group of residual blocks are extracted as feature descriptors, and the combination of the ReLU layer and the softmax classifier in the standard residual function is replaced by combining the improved E2LSH local sensitive hash algorithm.
7. The residual network-based E2LSH ocular corneal disease classification method of claim 6, wherein: in the step S5, when calculating the variance of each data, the ratio of the data to the set is defined as:
when calculating the variance of each data, the average eigenvalue is set as follows:
。
8. the residual network-based E2LSH ocular corneal disease classification method of claim 7, wherein: in S5, the intra-class variance is noted as:
the optimal variance value is defined as the corresponding one when the variance is maximumValues, namely:
and sorting eigenvalues of the variance matrix, wherein the corresponding eigenvectors are that the eigenvectors reflect the data density.
9. The residual network-based E2LSH ocular corneal disease classification method of claim 8, wherein: in S5, the hash function in the E2LSH is randomly generated based on the p-steady state distribution, and the data in the hash bucket changes with the data distribution, resulting in a situation that the data in one part of the hash bucket is concentrated, and the other part of the hash bucket has only a few data, so the improved hash function is as follows:
wherein The dynamic hash index is constructed by feature vectors obtained through calculation in the corresponding variance matrix, and proper barrel width parameters are selected so as to dynamically adjust the size of the barrel width and improve the balance degree of data distribution in each hash barrel.
10. The residual network-based E2LSH ocular corneal disease classification method of claim 9, wherein: in the step S5, when the data is hashed and mapped to a hash bucket, selection is requiredThe hash functions form a complete data hash map, and the method is as follows:
wherein ,representing the dimension of the dimension reduced data, +.>Representing functions in the hash function cluster. Raw data passing functionMapping down to->Dimension, the result is->Layering is carried out by adopting a mode of combining an array and a linked list, indexes are established by the hash through a bucket layering mode, and an average pooling layer is connected to obtain classified output.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310759632.5A CN116595457A (en) | 2023-06-26 | 2023-06-26 | E2LSH eye cornea disease classification method based on residual error network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310759632.5A CN116595457A (en) | 2023-06-26 | 2023-06-26 | E2LSH eye cornea disease classification method based on residual error network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116595457A true CN116595457A (en) | 2023-08-15 |
Family
ID=87599303
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310759632.5A Pending CN116595457A (en) | 2023-06-26 | 2023-06-26 | E2LSH eye cornea disease classification method based on residual error network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116595457A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117390515A (en) * | 2023-11-01 | 2024-01-12 | 江苏君立华域信息安全技术股份有限公司 | Data classification method and system based on deep learning and SimHash |
-
2023
- 2023-06-26 CN CN202310759632.5A patent/CN116595457A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117390515A (en) * | 2023-11-01 | 2024-01-12 | 江苏君立华域信息安全技术股份有限公司 | Data classification method and system based on deep learning and SimHash |
CN117390515B (en) * | 2023-11-01 | 2024-04-12 | 江苏君立华域信息安全技术股份有限公司 | Data classification method and system based on deep learning and SimHash |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109376636B (en) | Capsule network-based eye fundus retina image classification method | |
de Sousa et al. | Texture based on geostatistic for glaucoma diagnosis from fundus eye image | |
CN109376767A (en) | Retina OCT image classification method based on deep learning | |
CN111340776B (en) | Method and system for identifying keratoconus based on multi-dimensional feature adaptive fusion | |
Ovreiu et al. | Deep learning & digital fundus images: Glaucoma detection using DenseNet | |
CN110070531A (en) | For detecting the model training method of eyeground picture, the detection method and device of eyeground picture | |
Araújo et al. | Glaucoma diagnosis in fundus eye images using diversity indexes | |
CN116595457A (en) | E2LSH eye cornea disease classification method based on residual error network | |
CN112580580A (en) | Pathological myopia identification method based on data enhancement and model fusion | |
CN116523840A (en) | Lung CT image detection system and method based on deep learning | |
Sorić et al. | Using convolutional neural network for chest X-ray image classification | |
CN115641309A (en) | Method and device for identifying age of eye ground color photo of residual error network model and storage medium | |
CN112233742A (en) | Medical record document classification system, equipment and storage medium based on clustering | |
Sharma et al. | Harnessing the Strength of ResNet50 to Improve the Ocular Disease Recognition | |
CN114667539A (en) | Fundus image classification device and method based on deep learning for diagnosing eye diseases | |
CN113284140B (en) | Binocular keratoconus diagnosis method based on multi-modal data | |
CN116091449A (en) | Retina OCT (optical coherence tomography) image lesion classification method based on unsupervised heterogeneous distillation framework | |
CN114882284A (en) | Color fundus image classification system and method and electronic equipment | |
Santos et al. | Generating photorealistic images of people's eyes with strabismus using Deep Convolutional Generative Adversarial Networks | |
CN111899879A (en) | Automatic eye table disease screening method and system and block chain | |
Al Sariera et al. | Automated Cataract Detection and Classification Using Random Forest Classifier in Fundus Images. | |
Manju et al. | Determination of Early Onset Glaucoma Using OCT Image | |
Zhang et al. | Resnet-V: a residual networks classification algorithm of keratoconus | |
Veena et al. | An Enhanced RNN-LSTM Model for Fundus Image Classification to Diagnose Glaucoma | |
CN116246331B (en) | Automatic keratoconus grading method, device and storage medium |
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 |