[go: up one dir, main page]

CN112861881A - Honeycomb lung recognition method based on improved MobileNet model - Google Patents

Honeycomb lung recognition method based on improved MobileNet model Download PDF

Info

Publication number
CN112861881A
CN112861881A CN202110252132.3A CN202110252132A CN112861881A CN 112861881 A CN112861881 A CN 112861881A CN 202110252132 A CN202110252132 A CN 202110252132A CN 112861881 A CN112861881 A CN 112861881A
Authority
CN
China
Prior art keywords
image
model
lung
network model
mobilenet
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
Application number
CN202110252132.3A
Other languages
Chinese (zh)
Inventor
李钢
张玲
李晶
张海轩
李宇
李鹏博
鄂林宁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taiyuan University of Technology
Original Assignee
Taiyuan University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Taiyuan University of Technology filed Critical Taiyuan University of Technology
Priority to CN202110252132.3A priority Critical patent/CN112861881A/en
Publication of CN112861881A publication Critical patent/CN112861881A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种基于改进MobileNet模型的蜂窝肺识别方法，研究涉及图像处理技术领域；方法包括：获取蜂窝征CT影像数据集，对数据集进行数据标注和预处理；构建分类识别模型；计算模型输出与原始数据的误差损失值；根据误差损失值更新神经网络模型的参数；遍历所有训练样本，完成神经网络模型的训练学习，构建蜂窝肺CT影像识别模型，利用评价指标获得模型的识别性能；本发明是在计算机辅助的前提下精确获得蜂窝肺CT影像识别分类结果，提高分类识别准确率和模型整体性能。

Figure 202110252132

The invention discloses a cellular lung identification method based on an improved MobileNet model, and the research relates to the technical field of image processing. The method includes: acquiring a cellular sign CT image data set, performing data labeling and preprocessing on the data set; building a classification and identification model; calculating The error loss value between the model output and the original data; update the parameters of the neural network model according to the error loss value; traverse all training samples, complete the training and learning of the neural network model, build a honeycomb lung CT image recognition model, and use the evaluation index to obtain the recognition performance of the model The invention accurately obtains the recognition and classification results of the honeycomb lung CT images under the premise of computer assistance, and improves the classification and recognition accuracy and the overall performance of the model.

Figure 202110252132

Description

Honeycomb lung recognition method based on improved MobileNet model
Technical Field
The invention belongs to the technical field of image processing, and relates to a honeycomb lung recognition method based on an improved MobileNet model.
Background
Cellulite is a major histopathological feature of idiopathic pulmonary fibrosis, and is the development of pulmonary alterations characterized by cellulite in a variety of advanced lung diseases. In recent years, the incidence and prevalence of interstitial lung diseases are increasing year by year due to environmental deterioration, and the diseases have high fatality rate and poor clinical prognosis, and the 5-year survival rate is less than 30%. When the honeycomb lung disease is diagnosed, a relevant honeycomb characteristic CT image can be obtained through biopsy, and medical personnel can clearly observe the distribution range or the slight change of the honeycomb lung through the CT image.
The traditional medical image recognition method is to use manual extraction of features, and most of the common methods are feature extraction based on textures, shapes and colors, and the features are classified by using classifiers such as a support vector machine and a random forest. However, the problems of low resolution, obvious noise, irregular focus shape and the like exist in the honeycomb lung CT image, so that the traditional identification method has insufficient extraction characteristics and low efficiency, the generalization capability and robustness are poor, and the identification and classification accuracy of the image is influenced.
In recent years, with the development of scientific technology, deep learning technology plays an important role in image classification due to its strong feature extraction capability. Compared with manual identification, the deep learning technology has strong objective ability and high identification speed, can process a large amount of image data in a short time, can help doctors to make better judgment, and reduces the burden of the doctors. However, the existing deep learning technology still has the problem of low classification and identification accuracy for the honeycomb lung image data set.
Disclosure of Invention
The invention overcomes the defects of the prior art, provides a honeycomb lung identification method based on an improved MobileNet model, and identifies and classifies a honeycomb lung CT image data set through multi-scale feature fusion and an improved depth separable convolution module.
In order to achieve the above object, the present invention is achieved by the following technical solutions.
A honeycomb lung recognition method based on an improved MobileNet model comprises the following steps:
a) CT images of normal people and patients in different age groups are collected and obtained, data sets are generated by the CT images and the normal people and the patients, and data labeling and preprocessing are carried out on the honeycomb lung CT images in the data sets.
b) And performing data expansion on the preprocessed cellular lung CT image data set, and dividing the data expansion into a training set and a verification set according to a preset proportion.
c) Constructing a network model based on improved MobileNet, and obtaining the output of the neural network model through a training process; the improved network model of the MobileNet automatically extracts the feature information in the cellular lung CT image by using the cavity convolution with different expansion rates, enlarges the receptive field of feature extraction on the basis of not losing the feature information, sends the feature information of different layers into a feature extraction module for channel splicing to obtain a feature fusion vector, then realizes the fusion of the feature information by splicing a plurality of kinds of feature information obtained after convolution operation through the channel, and finally keeps the feature information of each channel by using a Sigmid linear activation function.
d) Updating parameters of the network model according to loss errors between predicted values and real values of the identified and classified network model; the loss error is obtained by using a cross entropy loss function, and the calculation formula of the loss error is as follows:
Figure BDA0002966490670000021
wherein J (theta) is ginsengPartial derivatives of the number θ; y is(i)For the ith sample x(i)The label of (1); m is the number of samples; h isθ(. is the probability of the sample prediction being correct.
e) And testing the verification set by adopting the network model of the MobileNet after the parameters are updated, and obtaining the overall performance of the network model through evaluation indexes.
f) And inputting the CT image to be predicted into the network model of the MobileNet after the parameters are updated to obtain a prediction recognition result.
Preferably, in step b, the preprocessed honeycomb lung CT image is subjected to data expansion and then to normalization processing.
Preferably, the data expansion is to process the preprocessed data in one or any combination of inversion, translation, cutting and scaling.
Preferably, the preprocessing is used for image processing through mean normalization and image denoising methods, and meanwhile, the data set is manually classified for distinguishing a normal image from a lesion image.
Preferably, the CT images subjected to preprocessing and data expansion are summarized to construct a honeycomb lung CT image data set; and meanwhile, dividing the data set D into k mutually exclusive subsets with similar sizes by using a cross-validation method, then using a union set of (k-1) subsets as a training set, and using the rest subsets as a test set.
Preferably, in the step d, parameters of the network model are updated by using an Adam algorithm according to the loss error.
Compared with the prior art, the invention has the beneficial effects that.
The invention constructs the automatic identification classification model by adopting the method of multi-scale feature fusion and improved depth separable convolution module, automatically classifies the honeycomb lung CT image, and improves the classification identification accuracy and the overall performance of the model.
Drawings
In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the following drawings are taken for illustration:
fig. 1 is a flowchart of a method for automatically classifying a cellular lung CT image dataset based on an improved MobileNet network model according to an embodiment of the present invention.
Fig. 2 is a system diagram illustrating a method for automatically classifying a cellular lung CT image dataset based on an improved MobileNet network model according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects to be solved by the present invention more clearly apparent, the present invention is further described in detail with reference to the embodiments and the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The technical solution of the present invention is described in detail below with reference to the embodiments and the drawings, but the scope of protection is not limited thereto.
A honeycomb lung recognition method based on an improved MobileNet model specifically comprises the following steps:
step S1: CT images of normal people and patients in different age groups are collected and obtained, data sets are generated by the CT images and the normal people and the patients, and data labeling and preprocessing are carried out on the honeycomb lung CT images in the data sets.
Step S2: and performing data expansion on the preprocessed cellular lung CT image data set, and dividing the data expansion into a training set and a verification set according to a preset proportion.
Step S3: and constructing a classification recognition model based on a MobileNet network to train the honeycomb lung CT image training set.
Step S4: in order to improve the identification accuracy of the model and accelerate the model training speed, the constructed network model is improved to improve the identification accuracy of the network model.
Step S5: and updating parameters of the neural network model according to the loss error between the predicted value and the true value of the recognition classification network model, so that the model recognition accuracy is improved.
Step S6: and training the marked test set by using the classification recognition model, and acquiring the recognition accuracy and the overall performance of the model according to the evaluation index.
Step S7: and inputting the prediction picture into the recognition classification network model to obtain a prediction result.
In step S1, a honeycomb CT image data set is obtained, which includes a normal lung CT image and a honeycomb lung lesion CT image, and because the original honeycomb lung CT image has characteristics of high noise, low contrast, and a shape change of a segmented target, an image enhancement method is required to be used to preprocess the original image, and methods such as mean normalization and image denoising are used for image processing, and in addition, the data set needs to be manually classified to distinguish a normal image from a lesion image.
In step S2, because the data volume of the cellular lung CT image is limited, the data set needs to be expanded on the basis, and here, a means of data expansion is used to obtain a rotation map and a mirror image of the original CT image by performing methods such as inversion, translation, shearing, or scaling on the image in the original data set to effectively expand the data volume, so as to provide data security for training the classification recognition model, and then the CT images subjected to preprocessing and data expansion are summarized to construct the cellular lung CT image data set. And simultaneously, dividing the data set D into k mutually exclusive subsets with similar sizes by using a cross-validation method, wherein each subset is required to keep the consistency of data distribution as much as possible. Then the union of (k-1) subsets is used as the training set, and the rest subsets are used as the test set. This results in k training/test sets, allowing k training and tests to be performed, and ultimately returning the mean of the k test results. Specifically, 90% of the samples in the honeycomb lung CT image data set formed in step S2 are randomly selected as a training set, and the remaining 10% of the samples are used as a test set, so as to perform a classification test.
In step S3, a classification recognition model based on MobileNet is constructed, where the core idea of the MobileNet network is that a depth separable convolution module in the network model decomposes standard convolution into a depth convolution and a point-by-point convolution. The deep convolution is a filtering stage of the deep separable convolution, and each channel is subjected to convolution operation corresponding to a convolution kernel; the point-by-point convolution is a combined stage of deep separable convolution, a plurality of characteristic diagram information are integrated to be output in series, and the combination of the two stages realizes the separation of a channel and a space, so that the parameter quantity required by model training is reduced, the model training speed is accelerated, more characteristic information is transmitted in a network, and the identification and classification accuracy of the model is improved. The depth separable convolution has certain difference with the standard convolution, the depth convolution and the point-by-point convolution in structure, the depth separable convolution reduces a large amount of calculation and improves the classification performance of the system, wherein the standard convolution calculation process is shown as the formula (1):
DK×DK×M×N×DF×DF (1)
the depth separable convolution expression is shown in equation (2):
DK×DK×M×DF+M×N×DF×DF (2)
d in the above two formulasKIs the size of the convolution kernel, M is the number of input channels, N is the number of convolution kernels, DFFor the input, the ratio expression of the calculated amount between the two is shown as formula (3):
Figure BDA0002966490670000051
as can be seen from the above formula that the ratio is less than 1, the computation amount required for the deep separable convolution is less than that of the standard convolution, and thus the resource consumption of the computer can be reduced to some extent.
In step S4, because the obtained honeycomb lung CT images have different sizes, the different sizes of the receptive fields are obtained through convolution kernels with different sizes, wherein the feature located at the lower layer has higher resolution and contains more detailed information, but too few convolution layers are passed through, which results in too much irrelevant information and too much noise in the obtained image; high-level features have stronger semantic information, but the resolution is too low, and the perception capability of details is poorer, so that the network loses global or local information by extracting different feature information from different receptive fields, and the classification of a model to a target is hindered. By using the cavity convolution with different expansion rates, the feature information in the cellular lung CT image is automatically extracted, the receptive field of feature extraction is expanded on the basis of not losing the feature information, and the feature information of different layers is sent to a feature extraction module for channel splicing to obtain a feature fusion vector. And (3) carrying out channel splicing on the 4 kinds of feature information obtained after convolution, wherein the multi-scale feature fusion and calculation process is shown as the formula (4):
F＝[F1,F2,F3,F4] (4)
wherein F is the fused feature vector, FiEach of (i ═ 1,2,3, and 4) is a feature vector of 4 different levels obtained by hole convolution.
Since deep convolution in a MobileNet network has no ability to change channels, its extracted features are single-channel, and the ReLU activation function may cause loss of information when a convolutional layer with a small number of channels performs output operations. Therefore, in order to ensure the recognition accuracy of model training, the characteristic information of each channel is preserved by using a Sigmid linear activation function instead of the ReLU activation function.
In step S5, a loss value is obtained by using a cross entropy loss function, and the calculation is as shown in formula (5):
Figure BDA0002966490670000061
wherein J (θ) is the partial derivative of the parameter θ; y is(i)For the ith sample x(i)The label of (1); m is the number of samples; h isθ(. is the probability of the sample prediction being correct.
According to the loss value, parameters of the network model are updated by using an Adam algorithm, wherein the Adam algorithm is an algorithm based on adaptive low-order moment estimation and can perform first-order gradient optimization on a random objective function. The Adam algorithm is easy to implement, and has extremely high computational efficiency and low memory requirement. The diagonal scaling of the Adam algorithm gradient is invariant and therefore well suited to solving problems with large scale data or parameters.
In step S6, the evaluation index for training the deep learning model includes: accuracy (accuracy), sensitivity (sensitivity), specificity (specificity) and F1 value (F1-score) as shown by the formula (6):
Figure BDA0002966490670000062
wherein TP, FP, FN and TN indicate the number of true positive, false negative and true negative, respectively.
In step S7, the prediction picture is inputted into the improved classification model according to claim 1 to obtain the prediction result, so as to assist the physician to make a diagnosis.
While the invention has been described in further detail with reference to specific preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (6)

1.一种基于改进MobileNet模型的蜂窝肺识别方法，其特征在于，包括以下步骤：1. a honeycomb lung identification method based on improving MobileNet model, is characterized in that, comprises the following steps: a)收集得到不同年龄段有关正常人和病人的CT图像，利用二者生成数据集，对数据集中的蜂窝肺CT影像进行数据标注和预处理；a) Collect CT images of normal people and patients in different age groups, use them to generate a dataset, and perform data labeling and preprocessing on the honeycomb lung CT images in the dataset; b)将预处理的蜂窝肺CT影像数据集进行数据扩充，并按照预设比例划分为训练集和验证集；b) Data expansion is performed on the preprocessed honeycomb lung CT image dataset, and divided into a training set and a validation set according to a preset ratio; c)构建基于改进的MobileNet的网络模型，通过训练过程得到神经网络模型的输出；所述改进的MobileNet的网络模型是通过运用不同扩张率的空洞卷积自动提取蜂窝肺CT影像中的特征信息，在不损失特征信息的基础上扩大特征提取的感受野，并且将不同层次的特征信息送入特征提取模块进行通道拼接，获得特征融合向量，之后将通过卷积操作后获得的多种特征信息通过通道拼接实现特征信息的融合，最后通过使用Sigmid线性激活函数使各通道的特征信息得以保留；c) construct the network model based on the improved MobileNet, obtain the output of the neural network model through the training process; the network model of the described improved MobileNet is to automatically extract the feature information in the cellular lung CT image by using the hole convolution of different expansion rates, Expand the receptive field of feature extraction without losing feature information, and send feature information of different levels into the feature extraction module for channel splicing to obtain feature fusion vectors, and then pass the various feature information obtained after the convolution operation through the Channel splicing realizes the fusion of feature information, and finally the feature information of each channel is preserved by using the sigmid linear activation function; d)根据识别分类网络模型的预测值与真实值之间的损失误差更新所述网络模型的参数，所述的损失误差是通过使用交叉熵损失函数获取，所述损失误差的计算式为：d) Update the parameters of the network model according to the loss error between the predicted value and the real value of the identification classification network model, the loss error is obtained by using the cross entropy loss function, and the calculation formula of the loss error is:
Figure FDA0002966490660000011
Figure FDA0002966490660000011
 其中J(θ)为参数θ的偏导数；y(i)为第i个样本x(i)的标签；m为样本数量；hθ(*)为样本预测正确的概率；where J(θ) is the partial derivative of the parameter θ; y (i) is the label of the i-th sample x(i); m is the number of samples; hθ(*) is the correct probability of sample prediction; e)采用参数更新后的MobileNet的网络模型对验证集进行测试，通过评价指标获得所述网络模型的整体性能；e) adopt the network model of MobileNet after parameter updating to test the verification set, and obtain the overall performance of the network model through the evaluation index; f)将待预测的CT影像输入参数更新后的MobileNet的网络模型中获得预测识别结果。f) Obtain the prediction and recognition result from the MobileNet network model after updating the CT image input parameters to be predicted. 2.根据权利要求1所述的一种基于改进MobileNet模型的蜂窝肺识别方法，其特征在于，步骤b中，将预处理的蜂窝肺CT影像进行数据扩充后，再进行标准化处理。2. A kind of honeycomb lung identification method based on improved MobileNet model according to claim 1, is characterized in that, in step b, after carrying out data expansion on the preprocessed honeycomb lung CT image, standardization processing is carried out again. 3.根据权利要求1或2所述的一种基于改进MobileNet模型的蜂窝肺识别方法，其特征在于，所述数据扩充是对预处理后的数据采用反转、平移、剪切、缩放中的一种或任意组合方式进行处理。3. a kind of honeycomb lung identification method based on improved MobileNet model according to claim 1 and 2, is characterized in that, described data expansion is to adopt inversion, translation, shearing, scaling to the data after preprocessing one or any combination. 4.根据权利要求1所述的一种基于改进MobileNet模型的蜂窝肺识别方法，其特征在于，所述预处理是通过均值归一化和图像去噪方法用于图像处理，同时对数据集进行手动分类用于区分正常图像和病灶图像。4. a kind of honeycomb lung identification method based on improved MobileNet model according to claim 1, is characterized in that, described preprocessing is to be used for image processing by mean value normalization and image denoising method, simultaneously to data set. Manual classification was used to distinguish normal images from lesion images. 5.根据权利要求1所述的一种基于改进MobileNet模型的蜂窝肺识别方法，其特征在于，对经过预处理和数据扩充的CT影像进行汇总，构建蜂窝肺CT影像数据集；同时使用交叉验证法将数据集D划分为k个大小相似的互斥子集，然后用(k-1)个子集的并集作为训练集，余下的子集作为测试集。5. a kind of honeycomb lung identification method based on improved MobileNet model according to claim 1, is characterized in that, the CT image through preprocessing and data expansion is collected, builds honeycomb lung CT image data set; Use cross-validation simultaneously The method divides the dataset D into k mutually exclusive subsets of similar size, and then uses the union of (k-1) subsets as the training set and the remaining subsets as the test set. 6.根据权利要求1所述的一种基于改进MobileNet模型的蜂窝肺识别方法，其特征在于，所述步骤d中，根据损失误差，使用Adam算法对网络模型的参数进行更新。6. A kind of cellular lung identification method based on improved MobileNet model according to claim 1, is characterized in that, in described step d, according to loss error, use Adam algorithm to update the parameter of network model.
CN202110252132.3A 2021-03-08 2021-03-08 Honeycomb lung recognition method based on improved MobileNet model Pending CN112861881A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110252132.3A CN112861881A (en) 2021-03-08 2021-03-08 Honeycomb lung recognition method based on improved MobileNet model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110252132.3A CN112861881A (en) 2021-03-08 2021-03-08 Honeycomb lung recognition method based on improved MobileNet model

Publications (1)

Publication Number Publication Date
CN112861881A true CN112861881A (en) 2021-05-28

Family

ID=75993482

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110252132.3A Pending CN112861881A (en) 2021-03-08 2021-03-08 Honeycomb lung recognition method based on improved MobileNet model

Country Status (1)

Country Link
CN (1) CN112861881A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115205637A (en) * 2022-09-19 2022-10-18 山东世纪矿山机电有限公司 Intelligent identification method for mine car materials
CN115810129A (en) * 2022-12-22 2023-03-17 东莞理工学院 An Object Classification Method Based on Lightweight Network
CN117275681A (en) * 2023-11-23 2023-12-22 太原理工大学 Honeycomb lung disease cycle detection and evaluation method and device based on Transformer parallel cross-fusion model

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180137406A1 (en) * 2016-11-15 2018-05-17 Google Inc. Efficient Convolutional Neural Networks and Techniques to Reduce Associated Computational Costs
CN109345507A (en) * 2018-08-24 2019-02-15 河海大学 A kind of dam image crack detection method based on transfer learning
US20190205758A1 (en) * 2016-12-30 2019-07-04 Konica Minolta Laboratory U.S.A., Inc. Gland segmentation with deeply-supervised multi-level deconvolution networks
CN110717907A (en) * 2019-10-06 2020-01-21 浙江大学 An intelligent detection method for hand tumors based on deep learning
CN111598868A (en) * 2020-05-14 2020-08-28 上海深至信息科技有限公司 Lung ultrasonic image identification method and system
US20200364870A1 (en) * 2019-05-14 2020-11-19 University-Industry Cooperation Group Of Kyung Hee University Image segmentation method and apparatus, and computer program thereof

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180137406A1 (en) * 2016-11-15 2018-05-17 Google Inc. Efficient Convolutional Neural Networks and Techniques to Reduce Associated Computational Costs
US20190205758A1 (en) * 2016-12-30 2019-07-04 Konica Minolta Laboratory U.S.A., Inc. Gland segmentation with deeply-supervised multi-level deconvolution networks
CN109345507A (en) * 2018-08-24 2019-02-15 河海大学 A kind of dam image crack detection method based on transfer learning
US20200364870A1 (en) * 2019-05-14 2020-11-19 University-Industry Cooperation Group Of Kyung Hee University Image segmentation method and apparatus, and computer program thereof
CN110717907A (en) * 2019-10-06 2020-01-21 浙江大学 An intelligent detection method for hand tumors based on deep learning
CN111598868A (en) * 2020-05-14 2020-08-28 上海深至信息科技有限公司 Lung ultrasonic image identification method and system

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
LI GANG等: "Recognition of honeycomb lung in CT images based on improved MobileNet model", 《MEDICAL PHYSICS》 *
LIANG-CHIEH CHEN等: "Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation", 《ARXIV》 *
刘倩倩 等: "计算机辅助肺腺癌病理图像分割识别关键技术研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 *
郭璠 等: "语义特征图引导的青光眼筛查方法", 《计算机辅助设计与图形学学报》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115205637A (en) * 2022-09-19 2022-10-18 山东世纪矿山机电有限公司 Intelligent identification method for mine car materials
CN115810129A (en) * 2022-12-22 2023-03-17 东莞理工学院 An Object Classification Method Based on Lightweight Network
CN117275681A (en) * 2023-11-23 2023-12-22 太原理工大学 Honeycomb lung disease cycle detection and evaluation method and device based on Transformer parallel cross-fusion model
CN117275681B (en) * 2023-11-23 2024-02-09 太原理工大学 Honeycomb lung disease cycle detection and evaluation method and device based on Transformer parallel cross-fusion model

Similar Documents

Publication Publication Date Title
CN110503630B (en) A method for classification, localization and prediction of cerebral hemorrhage based on 3D deep learning model
CN112529042B (en) Medical image classification method based on dual-attention multi-example deep learning
CN112446891B (en) Medical image segmentation method based on U-Net network brain glioma
CN107506761B (en) Brain image segmentation method and system based on saliency learning convolutional neural network
CN112101451A (en) Breast cancer histopathology type classification method based on generation of confrontation network screening image blocks
CN112365464A (en) GAN-based medical image lesion area weak supervision positioning method
CN111009321A (en) Application of a machine learning classification model in the auxiliary diagnosis of autism in adolescents
WO2015069824A2 (en) Diagnostic system and method for biological tissue analysis
CN113269230B (en) A multi-pneumonia CT classification method and device based on time series high-dimensional feature extraction
Naresh et al. Early detection of lung cancer using neural network techniques
CN113592769B (en) Abnormal image detection and model training method, device, equipment and medium
CN112465905A (en) Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning
CN112581450B (en) Pollen detection method based on dilated convolutional pyramid and multi-scale pyramid
CN112861881A (en) Honeycomb lung recognition method based on improved MobileNet model
CN113782184A (en) An auxiliary stroke assessment system based on facial key point and feature pre-learning
CN115564756A (en) Medical image focus positioning display method and system
CN111462082A (en) Focus picture recognition device, method and equipment and readable storage medium
CN108805181B (en) Image classification device and method based on multi-classification model
CN119131018B (en) Pathological image analysis method and equipment based on self-supervision learning and multi-task learning
CN114692748B (en) Method for identifying ground glass lung nodules
CN112633404A (en) DenseNet-based CT image classification method and device for COVID-19 patient
CN117894469A (en) A method and system for constructing a recurrence prediction model for pituitary adenoma based on full-slice pathological images
Francis et al. Diagnostic of cystic fibrosis in lung computer tomographic images using image annotation and improved PSPNet modelling
Asyhar et al. Implementation LSTM Algorithm for Cervical Cancer using Colposcopy Data
CN119360173A (en) Object detection method based on improved feature fusion

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20210528

RJ01 Rejection of invention patent application after publication