CN114636704A - Terahertz continuous wave three-dimensional tomography method based on deep learning - Google Patents
Terahertz continuous wave three-dimensional tomography method based on deep learning Download PDFInfo
- Publication number
- CN114636704A CN114636704A CN202210233130.4A CN202210233130A CN114636704A CN 114636704 A CN114636704 A CN 114636704A CN 202210233130 A CN202210233130 A CN 202210233130A CN 114636704 A CN114636704 A CN 114636704A
- Authority
- CN
- China
- Prior art keywords
- target object
- dimensional
- terahertz
- deep learning
- image
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Geometry (AREA)
- Computer Graphics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a terahertz continuous wave three-dimensional tomography method based on deep learning, wherein projection data obtained by scanning a target object by a terahertz continuous wave tomography system is directly used as input of the deep learning method based on a generation countermeasure network provided by the invention, and the method is used for reconstructing a three-dimensional internal structure image of the target object, so that the internal structure of the target object is reproduced without damage and possible defects are found. The deep learning network model provided by the invention is used for generating a countermeasure network and comprises a generator, a discriminator and a real data set. The deep learning method provided by the invention takes the convolutional neural network of the end-to-end model of the coding-decoding structure as a generator for generating a countermeasure network. The generator is divided into two parts, the coder of the front part extracts the input high-level features, and the decoder of the rear part maps the extracted high-level features to obtain a reconstructed image. Because the number of network layers is deep, sub-modules in the encoder and the decoder both adopt a residual error structure, and the optimization and training of the model are facilitated.
Description
Technical Field
The invention belongs to the technical field of terahertz imaging, and particularly relates to a terahertz continuous wave three-dimensional tomography image reconstruction method based on deep learning.
Background
Terahertz (THz) refers to electromagnetic waves in the region of frequencies from 100GHz to 10000GHz (10THz), with corresponding wavelengths from 3mm to 3 μm, between millimeter waves and infrared light. The radiation source has the characteristics of no ionizing radiation, no harm to human bodies, strong penetrability and the like, and is suitable for being used as a radiation source of an imaging system. The Tomography technology is a main three-dimensional imaging technology, and the english word "tomographics" is derived from the combination of the greek words "tomos" and "graphics", which are respectively the meanings of section and description, so that the meaning of Tomography is to detect the internal section of a sample. Since the first successful realization of terahertz wave tomography in 1997, the terahertz wave tomography technology has been rapidly developed, and a series of new three-dimensional imaging technologies are proposed in succession. According to the structure and the principle of a three-dimensional imaging system, the terahertz wave three-dimensional imaging technology is divided into a transmission mode and a reflection mode. Currently, the mature transmission Tomography technology is terahertz computer-aided Tomography (THz-CT), which can be regarded as an extension of X-ray CT on the electromagnetic band.
When image reconstruction is carried out on terahertz tomography, the algorithm of the traditional X-ray tomography is still adopted. Such as the Filtered backprojection algorithm (FBP), the joint algebraic Reconstruction method (SART), and Ordered Set Extension Maximization (OSEM). However, the traditional algorithm has the defects of long time consumption, longer terahertz wave wavelength than that of X-rays, not strict linear transmission in an object and the like, and low imaging quality.
Disclosure of Invention
In order to solve the problems in the prior art, the invention discloses a terahertz continuous wave tomography method based on deep learning.
A terahertz continuous wave three-dimensional tomography method based on deep learning comprises the following steps:
s1, a terahertz continuous wave tomography system scans and collects data of a target to obtain projection data of the target;
s2, preprocessing terahertz data;
s3, image reconstruction is carried out by adopting a deep learning method based on generation countermeasure network training, and the model comprises a generator, a discriminator and a real data set;
s4, reconstructing by using a generator to obtain a two-dimensional sectional view of the internal structure of the imaging target object;
s5, three-dimensionally overlapping two-dimensional section images of the internal structure of the target object to obtain a three-dimensional internal structure diagram;
and S6, carrying out nondestructive testing on the internal defects of the target by using the obtained three-dimensional structure diagram.
The data preprocessing in step S2 includes the following steps:
s21, adopting an R _ L filter, wherein the starting point of the R _ L filter is that the actual two-dimensional image has an upper frequency limit, and the R _ L filter is represented as follows:removing high-frequency components of the acquired data;
s22, converting each two-dimensional array data after filtering into a two-dimensional sine gray-scale image;
the deep learning method in the step S3 adopts a generated confrontation network model, and the generated confrontation network is composed of three parts, namely a generator (G), a discriminator (D) and a real data set. Through continuous game between the generators and the discriminators, the ideal generators are trained. And then the generator is used independently for directly reconstructing data obtained after the terahertz tomography system scans the imaging target to obtain a three-dimensional internal structure diagram of the imaging target.
The convolutional neural network using the end-to-end model of the coding-decoding structure based on the generator that generates the deep learning model of the countermeasure network in step S3. The model is divided into two parts, wherein the first half part is a coding module which is responsible for the downsampling of an input image and simultaneously continuously extracts image characteristics to obtain higher image characteristics; the second half of the model is a decoding part and is responsible for up-sampling the coded image, mapping is carried out on the features extracted by the coding module, and finally a reconstructed image is obtained.
The generator operation of the deep learning method based on generation of the countermeasure network in the step S3 includes:
s31, inputting a sinogram acquired by a system into a convolution layer, converting an input image into a larger square characteristic diagram, playing a certain role in filtering, simultaneously increasing the number of channels of the characteristic diagram, and reserving enough image information;
s32, inputting the image into a coder in the front half part of the network, and further extracting the characteristics of the image, wherein the coder consists of a down-sampling module and a bottleneck module;
and S33, each pixel of the feature map after passing through the whole encoder has a global receptive field.
S34, decoding and up-sampling the feature map, wherein the decoder comprises a bottleneck module and an up-sampling module and finally outputs a reconstructed target section internal structure diagram;
because the network structure of the generator of the whole generation confrontation network is very deep, and the network structures of the encoder and the decoder are very deep, the phenomena of gradient disappearance, degradation and the like can easily occur during training. Therefore, the sub-modules in the encoder and the decoder both adopt a residual error structure, and the optimization during network training is facilitated.
The reconstructed image from the generator is evaluated by the following three criteria:
1. the root Mean Square Error (MSE) is used for describing the difference of the precision between the reconstructed image and the corresponding actual sectional image, the smaller the MSE is, the more accurate the image reconstruction result is, otherwise, the worse is, and the formula is as follows:
2. the peak signal-to-noise ratio is an image evaluation index for evaluating the noise level of an image based on pixel errors. The formula is as follows:
3. the index evaluates the similarity of two images from three aspects of brightness, contrast and structure, and the larger the numerical value is, the more similar the images are. The formula is as follows:
in the formula, mupredAnd muGTMean, σ, of reconstructed picture and actual cross-sectional view, respectivelypredAnd σGTStandard deviation, σ, of the reconstructed picture and the actual cross-sectional view, respectivelycovFor the covariance of the reconstructed picture and the actual cross-sectional view, c1And c2The constant is not zero and is used for avoiding the condition that the denominator is zero in the calculation process.
The nondestructive testing method of the step S6 is based on the deep learning image reconstruction method based on the generation countermeasure network proposed by the invention. An imaging target is placed on a system turntable, data scanned by a system for three-dimensional point-by-point scanning of the target are acquired, after three-dimensional internal structure imaging is carried out, defects of the target are analyzed through an obtained internal three-dimensional structure diagram, and terahertz nondestructive testing is achieved.
Drawings
The drawings are for purposes of illustration and description only and are not intended to limit the scope of the present disclosure.
Fig. 1 is a schematic diagram of a terahertz continuous wave tomography method based on deep learning.
FIG. 2 is a block diagram of a deep learning model based on generation of a countermeasure network and its sub-modules, (a) is a block diagram of a convolutional neural network based on generation of a countermeasure network proposed by the present invention; (b) - (d) is a block diagram of the submodule in the generator for generating the countermeasure network.
Detailed Description
This embodiment is a possible implementation based on Python, and the present invention will be explained and explained in more detail with reference to the accompanying drawings and embodiments.
1. Fig. 1 is a basic block diagram of an embodiment of a terahertz continuous wave tomography method based on deep learning according to the present invention. The system comprises a data acquisition part, an image reconstruction part based on deep learning and nondestructive internal detection of an object after three-dimensional imaging.
2. FIGS. 2(a) - (d) are diagrams of deep learning models based on the structure of the generated countermeasure network and the sub-modules of the generator in the present invention. The structure diagram of the generation countermeasure network is shown in fig. 2(a), and the generation countermeasure network is composed of three parts, namely a generator, a discriminator and a real data set. The generator is used as a main part and is used for directly reconstructing data obtained after the terahertz tomography system scans an imaging target to obtain a three-dimensional internal structure diagram of the imaging target. The function of the discriminator is to train the generator in combination with the real data set, and when the discriminator cannot identify whether the generated image of the generator is a real image, the generator is successfully trained. The discriminator is a neural network structure and a convolutional neural network, the two latter layers are full connection layers, and finally, the two classification outputs judge whether the image generated by the generator is real. Output 1 represents true and output 0 represents not true. If the output 1 is all-out, it represents that the discrimination ability of the discriminator is too poor, and if the output 0 is all-out, it represents that the discrimination ability of the discriminator is too strong. The game process of the generator and the discriminators is realized, and the final discriminators cannot identify whether the images generated by the generator are real or not through continuous training, so that the generator is in the optimal state. And then the trained generator is used independently, and the data acquired by the tomography system is directly used for reconstructing an internal structure diagram of the target.
3. Where the generator is a convolutional neural network of an end-to-end model of the coding-decoding structure. The model is divided into two parts, wherein the first half part is a coding module which is responsible for the downsampling of an input image and simultaneously continuously extracts image characteristics to obtain higher image characteristics; the second half of the model is a decoding part and is responsible for up-sampling the coded image, mapping is carried out on the features extracted by the coding module, and finally a reconstructed image is obtained. The input sinogram enters a convolution layer, an input image is converted into a larger square characteristic diagram, a certain filtering effect is achieved, the channel number of the characteristic diagram is increased to 64, sufficient image information is reserved, and the design of the subsequent network size is facilitated. The images are then input into the first half of the network encoder, which consists of 4 downsampling modules (B2) and 8 bottleneck modules (B1), to further extract the features of the images. Each pixel of the feature map after passing through the whole encoder has a global receptive field. And then decoding and upsampling the feature map, wherein the feature map comprises 7 bottleneck modules (B1) and 5 upsampling modules (B3), and finally outputting a reconstructed target cross-section internal structure map.
Fig. 2(b) - (d) are diagrams of sub-modules in the network, which include a bottleneck residual module, a downsampling residual module, and an upsampling residual module. As can be seen from fig. 2(a), the whole network structure is very deep, both the encoder and the decoder have 36 layers of convolutional neural networks, and there are 72 layers, and the whole network structure is very deep, and the phenomena of gradient disappearance, degradation and the like easily occur during training. Therefore, the sub-modules in the network all adopt a residual error structure, and the optimization during network training is facilitated.
The activation function in each sub-module is chosen to be Leaky ReLU, whose expression is:where a is a fixed parameter in the interval (1, + ∞), which is added with a hyper-parameter α in the interval compared to the ReLU functionThe phenomenon of neuron death is avoided.
4. Based on the deep learning image reconstruction method provided by the invention, the process of the embodiment of the nondestructive detection of the internal defect of the target object is as follows:
s1, building a terahertz tomography system by adopting a terahertz continuous wave transmission imaging mode, wherein the terahertz tomography system comprises a terahertz wave emission source, a terahertz wave collector, a rotating platform, a lens and imaging algorithm software of the method;
s2, placing a target object to be subjected to nondestructive inspection on a rotary table of a terahertz tomography system, transmitting terahertz continuous waves by a terahertz wave emission source and focusing the terahertz continuous waves at the position of the rotary table through a lens, transmitting the target object by the terahertz continuous waves, and scanning and acquiring continuous wave data of the target object by the terahertz tomography system;
s3, preprocessing terahertz continuous wave data;
s4, reconstructing to obtain a two-dimensional cross-sectional diagram of the internal structure of the target by adopting the terahertz image reconstruction method based on the generation countermeasure network model;
s5, three-dimensionally overlapping the two-dimensional cross-sectional drawing of the internal structure of the target object to obtain a three-dimensional internal structure drawing of the target object;
and S6, analyzing the internal three-dimensional structure diagram of the target object to realize the detection of the internal defects of the target.
Claims (5)
1. A terahertz continuous wave three-dimensional tomography method based on deep learning is characterized in that a depth learning method based on a generation countermeasure network is used for carrying out inversion imaging on terahertz continuous wave data acquired by a terahertz tomography system, an internal structure diagram of a target object is obtained through reconstruction, and nondestructive detection on an internal defect structure of the target object is achieved; the method mainly comprises the following steps:
s1, scanning and continuous wave data acquisition are carried out on a target object by a terahertz tomography system;
s2, preprocessing terahertz continuous wave data;
s3, image reconstruction is carried out by adopting a deep learning method based on a generated confrontation network model, and the model comprises a generator, a discriminator and a real data set;
s4, reconstructing to obtain a two-dimensional cross-sectional view of the internal structure of the target object;
and S5, three-dimensionally overlapping the two-dimensional cross-sectional drawing of the internal structure of the target object to obtain a three-dimensional internal structure drawing of the target object.
And S6, analyzing the internal three-dimensional structure diagram of the target object to realize the detection of the internal defects of the target.
2. The terahertz continuous wave three-dimensional tomography method based on deep learning as claimed in claim 1, characterized in that the generator for generating the countermeasure network structure model is used as a main part of the deep learning, and is used for directly reconstructing data obtained after the terahertz tomography system scans a target object to obtain a three-dimensional internal structure diagram of the target object; the function of the discriminator is to combine with the real data set for training the generator, and when the discriminator can not identify whether the generated image of the generator is a real image, the training of the generator is successful; the discriminator is a neural network structure, a convolutional neural network and a full connection layer are combined, and finally, two classification outputs are used for judging whether the image generated by the generator is real or not. And then the trained generator is used independently, and the data acquired by the tomography system is directly used for reconstructing an internal structure diagram of the target.
3. The terahertz continuous wave three-dimensional tomography method based on deep learning as claimed in claim 1, characterized in that the generator for generating the countermeasure network structure model adopts an end-to-end model convolutional neural network of a coding-decoding structure. The model is divided into two parts, wherein the first half part is a coding module which is responsible for the downsampling of an input image and simultaneously continuously extracts image characteristics to obtain higher image characteristics; the second half of the model is a decoding part and is responsible for up-sampling the coded image, mapping is carried out on the features extracted by the coding module, and finally a reconstructed image is obtained.
4. The terahertz continuous wave three-dimensional tomography method based on deep learning as claimed in claim 1, characterized in that the data processing procedure of step 3 is as follows:
s31, inputting an array obtained after the terahertz tomography system is scanned into a convolutional layer, converting an input image into a larger square characteristic diagram, playing a certain filtering role, and simultaneously reserving enough image information;
s32, inputting the image into a first half encoder of a network, and further extracting the characteristics of the image;
s33, each pixel of the feature map after passing through the whole encoder has a global receptive field;
s34, decoding and up-sampling the characteristic graph, and finally outputting a reconstructed target section internal structure graph;
s35, all sub-modules in the network adopt a residual error structure, so that the optimization during network training is facilitated, and the structure is used as a generator for generating the countermeasure network.
5. The terahertz continuous wave three-dimensional tomography method based on deep learning as claimed in claim 1, characterized in that the target object to be subjected to nondestructive inspection is placed in the terahertz tomography system, the terahertz tomography system is adopted to perform three-dimensional point-by-point scanning and data acquisition on the target object, and the three-dimensional internal structure diagram of the target object is reconstructed by using the deep learning method based on generation of the countermeasure network, so as to realize nondestructive inspection on the internal defects of the target object.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210233130.4A CN114636704B (en) | 2022-03-10 | 2022-03-10 | Terahertz continuous wave three-dimensional tomography method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210233130.4A CN114636704B (en) | 2022-03-10 | 2022-03-10 | Terahertz continuous wave three-dimensional tomography method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114636704A true CN114636704A (en) | 2022-06-17 |
CN114636704B CN114636704B (en) | 2023-05-26 |
Family
ID=81946872
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210233130.4A Active CN114636704B (en) | 2022-03-10 | 2022-03-10 | Terahertz continuous wave three-dimensional tomography method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114636704B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104883629A (en) * | 2014-06-19 | 2015-09-02 | 电子科技大学 | Apparatus and method for generating spectral amplitude code label by using adjustable optical filter |
US20160063755A1 (en) * | 2014-08-29 | 2016-03-03 | Wal-Mart Stores, Inc. | Simultaneous item scanning in a pos system |
JP2016057259A (en) * | 2014-09-12 | 2016-04-21 | 日本電信電話株式会社 | Tomographic imaging apparatus |
US20160377557A1 (en) * | 2014-03-12 | 2016-12-29 | National University Corporation Kobe University | Scattering tomography method and scattering tomography device |
CN110163179A (en) * | 2019-05-29 | 2019-08-23 | 浙江啄云智能科技有限公司 | A kind of contraband detecting recognition methods, system, equipment and its storage medium based on deep learning |
CN111311574A (en) * | 2020-02-14 | 2020-06-19 | 山西大学 | Terahertz lesion detection method and system based on artificial intelligence |
CN111652949A (en) * | 2020-04-17 | 2020-09-11 | 北京理工大学 | Terahertz focusing tomography image reconstruction method based on AdaFM-Net |
CN111784573A (en) * | 2020-05-21 | 2020-10-16 | 昆明理工大学 | Passive terahertz image super-resolution reconstruction method based on transfer learning |
CN112130221A (en) * | 2019-06-24 | 2020-12-25 | 清华大学 | Human body safety inspection apparatus and human body safety inspection method |
US20220034824A1 (en) * | 2018-09-26 | 2022-02-03 | Northwestern University | Space-time scattering network for inverse design and tomography |
-
2022
- 2022-03-10 CN CN202210233130.4A patent/CN114636704B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160377557A1 (en) * | 2014-03-12 | 2016-12-29 | National University Corporation Kobe University | Scattering tomography method and scattering tomography device |
CN104883629A (en) * | 2014-06-19 | 2015-09-02 | 电子科技大学 | Apparatus and method for generating spectral amplitude code label by using adjustable optical filter |
US20160063755A1 (en) * | 2014-08-29 | 2016-03-03 | Wal-Mart Stores, Inc. | Simultaneous item scanning in a pos system |
JP2016057259A (en) * | 2014-09-12 | 2016-04-21 | 日本電信電話株式会社 | Tomographic imaging apparatus |
US20220034824A1 (en) * | 2018-09-26 | 2022-02-03 | Northwestern University | Space-time scattering network for inverse design and tomography |
CN110163179A (en) * | 2019-05-29 | 2019-08-23 | 浙江啄云智能科技有限公司 | A kind of contraband detecting recognition methods, system, equipment and its storage medium based on deep learning |
CN112130221A (en) * | 2019-06-24 | 2020-12-25 | 清华大学 | Human body safety inspection apparatus and human body safety inspection method |
CN111311574A (en) * | 2020-02-14 | 2020-06-19 | 山西大学 | Terahertz lesion detection method and system based on artificial intelligence |
CN111652949A (en) * | 2020-04-17 | 2020-09-11 | 北京理工大学 | Terahertz focusing tomography image reconstruction method based on AdaFM-Net |
CN111784573A (en) * | 2020-05-21 | 2020-10-16 | 昆明理工大学 | Passive terahertz image super-resolution reconstruction method based on transfer learning |
Non-Patent Citations (2)
Title |
---|
XI YANG 等: "CNN with spatio-temporal information for fast suspicious object detection and recognition in THz security images", 《SIGNAL PROCESSING》 * |
赵博研 等: "改进YOLOv4算法的GFRP内部缺陷检测与识别", 《空军工程大学学报(自然科学版)》 * |
Also Published As
Publication number | Publication date |
---|---|
CN114636704B (en) | 2023-05-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110175971B (en) | Deep learning image reconstruction method for multispectral single-pixel imaging | |
CN110363215B (en) | Method for converting SAR image into optical image based on generating type countermeasure network | |
CN111062880B (en) | Underwater image real-time enhancement method based on condition generation countermeasure network | |
CN109816742B (en) | Cone beam CT geometric artifact removing method based on fully-connected convolutional neural network | |
Bhunia et al. | Improving document binarization via adversarial noise-texture augmentation | |
CN115578406B (en) | CBCT jaw bone region segmentation method and system based on context fusion mechanism | |
CN117274759A (en) | Infrared and visible light image fusion system based on distillation-fusion-semantic joint driving | |
CN114693524A (en) | Side-scan sonar image accurate matching and fast splicing method, equipment and storage medium | |
CN114266957A (en) | Hyperspectral image super-resolution restoration method based on multi-degradation mode data augmentation | |
CN112200264B (en) | High-flux imaging-free classification method and device based on scattering multiplexing | |
CN113298700B (en) | High-resolution image reconstruction method in scattering scene | |
CN112633123B (en) | Heterogeneous remote sensing image change detection method and device based on deep learning | |
CN114636704B (en) | Terahertz continuous wave three-dimensional tomography method based on deep learning | |
CN117523095A (en) | Sparse angle THz-CT image reconstruction method based on deep learning | |
CN116664587A (en) | Pseudo-color enhancement-based mixed attention UNet ultrasonic image segmentation method and device | |
CN116579959A (en) | Fusion imaging method and device for hyperspectral image | |
CN112508862B (en) | Method for enhancing magneto-optical image of crack by improving GAN | |
CN114998469A (en) | Regional infrared digital holographic method based on neural convolutional network | |
CN115690001A (en) | Method for detecting defects in steel pipe welding digital radiographic image | |
CN112001956B (en) | CNN-based image denoising method for strong laser far-field focal spot measurement by using schlieren method | |
Feng et al. | X-ray Cherenkov-luminescence tomography reconstruction with a three-component deep learning algorithm: Swin transformer, convolutional neural network, and locality module | |
CN114897785B (en) | Single-view CT reconstruction method combining global and local in defect detection | |
Nyberg | Exploring the Capabilities of Generative Adversarial Networks in Remote Sensing Applications | |
CN118229981B (en) | CT image tumor segmentation method, device and medium combining convolutional network and transducer | |
CN116563237B (en) | Deep learning-based chicken carcass defect hyperspectral image detection method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |