CN112258414A - Noise suppression method for security inspection CT (computed tomography) ray signals - Google Patents
Noise suppression method for security inspection CT (computed tomography) ray signals Download PDFInfo
- Publication number
- CN112258414A CN112258414A CN202011155753.1A CN202011155753A CN112258414A CN 112258414 A CN112258414 A CN 112258414A CN 202011155753 A CN202011155753 A CN 202011155753A CN 112258414 A CN112258414 A CN 112258414A
- Authority
- CN
- China
- Prior art keywords
- data
- ray signal
- security check
- network
- training
- 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
- 238000000034 method Methods 0.000 title claims abstract description 28
- 230000001629 suppression Effects 0.000 title claims abstract description 22
- 238000002591 computed tomography Methods 0.000 title claims description 147
- 238000007689 inspection Methods 0.000 title claims description 38
- 238000012549 training Methods 0.000 claims abstract description 85
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 238000012546 transfer Methods 0.000 claims abstract description 15
- 238000013528 artificial neural network Methods 0.000 claims abstract description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 238000013508 migration Methods 0.000 claims description 7
- 230000005012 migration Effects 0.000 claims description 7
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000003786 synthesis reaction Methods 0.000 claims description 6
- 230000002194 synthesizing effect Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 3
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000013499 data model Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The invention relates to a method for suppressing noise of a security check CT ray signal, which comprises the steps of obtaining a data set of the security check CT ray signal, and dividing the data set into a training data set and a test data set; constructing a variable-step-length hole convolutional layer and a deep convolutional inverse graphic network (DC-IGN network) formed by the hole convolutional layer; training the deep convolution reverse graph network by adopting the training data set to obtain the deep convolution reverse graph network containing a security check CT ray signal denoising model; inputting the test data set into the trained deep convolution reverse graph network to perform security check CT ray signal denoising model transfer training to obtain a tested deep convolution reverse graph network; and inputting the security check CT ray signal to be subjected to noise suppression into the tested deep convolution reverse graph network to obtain a denoised security check CT ray signal. The method solves the problem that the noise reduction effect of the existing neural network on the noise suppression of the completeness of the security check CT ray signal is unstable.
Description
Technical Field
The invention relates to the technical field of security check CT, in particular to a noise suppression method for a security check CT ray signal.
Background
The security check CT becomes core security check equipment in the security market due to the characteristics of strong penetration, multiple angles, high precision and the like, but the data volume generated by the security check CT is large, the finally obtained reconstructed image is obtained by carrying out large-scale calculation on the collected security check CT signals, and the definition and the accuracy of the reconstructed image are influenced by the existence of noise.
The conventional method for denoising the CT image of the security check comprises a projection domain denoising algorithm, an image reconstruction algorithm, an image domain denoising algorithm and the like, the types of noise are various and unknown, so that accurate noise modeling needs to be selected according to noise classification, the edge of the obtained denoised image is fuzzy, and the denoising algorithm is high in complexity and poor in robustness due to the fact that ray signals with incomplete angles are generated in different acquisition modes.
In recent years, with the deep learning technology being developed, in order to solve the disadvantages of the conventional image processing method, the deep learning technology is applied to the denoising of the security inspection CT image. The Convolutional Neural Network (CNN) can well remove security CT noise, but the output image edge is blurred. At present, aiming at the problem of output image blur, a feasible solution is provided for generating a countermeasure Network (GAN), but as GAN is not easy to converge, easy to crash and difficult to control, a data set and a Network setting are complex, so that the denoising effect is unstable, and aiming at an incomplete data set, a neural Network is not optimized correspondingly at present.
Disclosure of Invention
In view of the above analysis, the present invention aims to provide a noise suppression method for a safety check CT ray signal, so as to solve the problem that the noise suppression of the completeness of the safety check CT ray signal by the existing neural network has an unstable denoising effect.
The purpose of the invention is mainly realized by the following technical scheme:
the invention provides a noise suppression method for a security inspection CT ray signal, which comprises the following steps:
acquiring a data set of a security check CT ray signal, and dividing the data set into a training data set and a test data set; wherein the data set comprises noisy security check CT ray signal data and normal security check CT ray signal data;
constructing a variable-step-length hole convolutional layer and a depth convolutional inverse graphic network formed by the hole convolutional layer;
training the deep convolution reverse graphic network by adopting the training data set, updating network parameters through a back propagation gradient, and obtaining the deep convolution reverse graphic network containing a security check CT ray signal denoising model after training;
inputting the test data set into the trained deep convolution reverse graph network to perform security check CT ray signal denoising model transfer training to obtain a tested deep convolution reverse graph network;
and inputting the security check CT ray signal to be subjected to noise suppression into the tested deep convolution reverse graph network to obtain a denoised security check CT ray signal.
Further, the data set of the security check CT ray signal specifically includes: full-angle full-conjugate data X of CT (computed tomography) ray signals with noise security check1，X1Full-angle full-conjugate data X of epsilon A and normal security inspection CT ray signals2，X2Epsilon A and under-angle unconjugated data Y of CT ray signal with noise security check1，Y1Epsilon A and normal security inspection CT ray signal under-angle unconjugated data Y2，Y2Belonging to A and CT ray signal under-angle mixed conjugate data Z with noise security check, wherein Z belongs to A; wherein the mixed conjugate data Z includes conjugate data and non-conjugate data; a is the domain of all types of security inspection CT ray signal noise;
the training data set comprises X1，X2，Y1，Y2(ii) a The test data set is Z.
Further, the training the deep convolution inverse graph network by using the training data set specifically includes: and obtaining a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal by using the training data set through adjusting the scaling ratio of the network.
Further, the step of inputting the test data set into the trained deep convolution inverse graph network for performing the transfer training of the security check CT ray signal denoising model specifically comprises:
judging the completeness of the security check CT ray signals in the test data set: if the signal is complete data, selecting a safety inspection CT ray signal all-angle conjugate data denoising model for carrying out transfer training to directly obtain a safety inspection CT ray signal with noise suppression; if the signal is incomplete data, dividing the test data set into conjugate data and non-conjugate data according to an acquisition angle, distributing network weights according to the magnitude of an under-acquisition angle, respectively and correspondingly selecting a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal for transfer training, and performing weighted synthesis to obtain the noise-suppressed security check CT ray signal.
Further, the constructing of the variable-step-size hole convolutional layer and the depth convolution inverse graph network formed by the hole convolutional layer includes: the deep convolution reverse graph network comprises multilayer convolution and deconvolution operators and is trained by using a random gradient variational Bayes algorithm; the convolution kernel of the cavity convolution layer adopts variable step length, the size of the convolution kernel is increased without increasing the number of network parameters, and the receptive field of the whole network is enlarged, wherein the receptive field is used for measuring the context information of the convolution neural network.
Further, the deep convolution inverse graphics network specifically comprises an encoder and a generator;
the encoder comprises a plurality of convolution layers, wherein a maximum pooling layer is arranged behind each convolution layer and is used for compressing unnecessary features of the security inspection CT ray signal;
the generator comprises a plurality of corresponding convolution layers, and an inverse pooling layer is arranged behind each convolution layer and is used for generating specific characteristics to obtain a security inspection CT ray signal with noise suppression.
Further, the deep convolution inverse graph network updates network parameters by a back propagation gradient through a random gradient descent, and adopts a variational object function as follows:
y＝-log(p(x|zi)+KL(q(zi|x)||p(zi)))；
wherein y is the difference between the predicted distribution and the true distribution, x is the sample data of the training sample set, and ziFor compressed data containing hidden variables, i is the sample number of the training data set, KL (q | p) is the inverse KL divergence, and p (x | z | p)i) As likelihood probability, q (z)i| x) is the posterior probability, p (z)i) Is a prior probability.
Further, by adjusting the scaling ratio of the network, a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal are obtained by using the training data set, and the method includes the following steps:
scaling ratio of network to 4:1:2 using training data X1，X2Training is carried out to obtain a safety inspection CT ray signal full-angle conjugate data denoising model;
scaling ratio of network to 2:1.5:2 using training data Y1，Y2And training to obtain a security inspection CT ray signal under-angle non-conjugate data denoising model.
Further, the complete data or the conjugate data select a full-angle conjugate data denoising model of a security check CT ray signal to perform migration training, and data are compressed by a coder to obtain ziAmplifying the signal through a generator to obtain a denoised signal, and removing noise data in the signal to inhibit the reconstructed image noise;
the non-conjugate data selects a CT ray signal under-angle non-conjugate data denoising model for carrying out migration training, and data is compressed through a coder to obtain ziAmplifying the signal through a generator to obtain a denoised signal, and removing noise data in the signal to inhibit the reconstructed image noise;
wherein z isiCompressed data obtained by scaling of the network model for the training data set.
Further, the weighted synthesis to obtain the noise-suppressed security inspection CT ray signal includes: and (3) respectively generating denoised signals of the full-angle conjugate data denoising model and the under-angle non-conjugate data denoising model of the security check CT ray signal, and synthesizing the denoised signals of the security check CT ray signal according to the under-angle proportional weight.
The technical scheme has the beneficial effects that: the invention discloses a noise suppression method for a security check CT ray signal, which carries out weighted classification denoising on signal data according to the completeness of the security check CT ray signal and the conjugate characteristics of projection data, increases the receptive fields of a DC-IGN compiler and a generator by using a cavity convolution layer, and improves the denoising performance on the premise of not increasing the number of network parameters.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flowchart of a method for suppressing noise of a security inspection CT ray signal according to an embodiment of the present invention;
FIG. 2 is a flow chart of training a deep convolutional inverse graphics network with a training data set according to an embodiment of the present invention;
FIG. 3 is a flowchart of a transfer training of a security check CT ray signal denoising model performed by a deep convolution inverse graph network after a test data set is input into the training in the embodiment of the present invention;
FIG. 4 is a diagram of a DC-IGN neural network architecture according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a hole convolution according to an embodiment of the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
One embodiment of the present invention, as shown in fig. 1, discloses a noise suppression method for a security check CT ray signal, comprising the following steps:
acquiring a data set of a security check CT ray signal, and dividing the data set into a training data set and a test data set; wherein the data set comprises noisy security check CT ray signal data and normal security check CT ray signal data;
constructing a variable-step-length hole convolutional layer and a depth convolutional inverse graphic network formed by the hole convolutional layer; that is, the improved deep convolutional inverse graph network is based on a DC-IGN network.
Training the deep convolution reverse graphic network by adopting the training data set, updating network parameters through a back propagation gradient, and obtaining the deep convolution reverse graphic network containing a security check CT ray signal denoising model after training;
inputting the test data set into the trained deep convolution reverse graph network to perform security check CT ray signal denoising model transfer training to obtain a tested deep convolution reverse graph network;
and inputting the security check CT ray signal to be subjected to noise suppression into the tested deep convolution reverse graph network to obtain a denoised security check CT ray signal.
Compared with the prior art, the technical scheme of the invention can remove noise, can keep the edge clear and stable, and can utilize the data completeness and the data conjugate characteristics of the security check CT ray signal to carry out classification weighting denoising on the ray signal. Specifically, the present invention is primarily directed to improvements in two areas:
1. the method has the advantages that the receptive field of the network is improved by utilizing the deep convolution reverse graphic network (DC-IGN network) adopting the cavity convolution on the premise of not increasing the number of network parameters, so that the characteristics extracted by the network are more stable, the image blur caused by image denoising is reduced, and the more accurate and clear security check CT image is obtained.
2. Respectively training a conjugate data model of a complete ray signal and a non-conjugate data model of an incomplete ray signal by using an improved deep convolution reverse graph network (DC-IGN network), carrying out classification decomposition on a test security check CT ray signal, carrying out migration training on the trained models to obtain a de-noised ray signal, and preserving the edge definition of a reconstructed image.
In an embodiment of the present invention, the data set of the security check CT ray signal specifically includes: full-angle full-conjugate data X of CT (computed tomography) ray signals with noise security check1，X1Full-angle full-conjugate data X of epsilon A and normal security inspection CT ray signals2，X2Epsilon A and under-angle unconjugated data Y of CT ray signal with noise security check1，Y1Epsilon A and normal security inspection CT ray signal under-angle unconjugated data Y2，Y2Belonging to A and CT ray signal under-angle mixed conjugate data Z with noise security check, wherein Z belongs to A; wherein the mixed conjugate data Z includes conjugate data and non-conjugate data; a is the domain of all types of security inspection CT ray signal noise;
the training data set comprises X1，X2，Y1，Y2(ii) a The test data set is Z.
It should be noted that the data acquisition mode of the security check CT ray signal is conjugate acquisition, full conjugate data refers to a method of rapidly changing the position of the detector during scanning, acquiring 180 ° and 360 ° scan data respectively, and reconstructing an image using two sets of data, which can improve the longitudinal resolution of the scan image, and under-angle non-conjugate data refers to incomplete angle data generated due to incomplete data acquisition or artificial mode setting.
Conjugate sampling is 2 samples satisfying the conditions of (α, β) and (- γ, β + π +2 γ), forming conjugate sampling, where γ represents the detector angle and β represents the projection angle. Under the spiral acquisition, because of the uniform motion of the detector, 2 ray paths parallel to each other at different Z-axis positions are represented. When the scanning angle is 360 degrees, the acquired data is fully conjugate data; when the scan angle is greater than the minimum scan angle and less than 360 °, data generated by an angle through which 2 rays parallel to each other pass is conjugate data, and data generated by an angle through which only one ray passes is non-conjugate data.
Carrying out data normalization on the full conjugate data with noise and without noise of the security check CT ray signal to form two-dimensional matrix data; and carrying out data normalization on the non-conjugate data with noise and without noise of the security check CT ray signal to form two-dimensional matrix data.
In a specific embodiment of the present invention, as shown in fig. 2, the training of the deep convolution inverse graphics network by using the training data set specifically includes: and obtaining a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal by using the training data set through adjusting the scaling ratio of the network.
Specifically, the scaling ratio of the network is adjusted to 4:1:2, using dataset X1，X2Training is carried out to obtain a safety inspection CT ray signal full-angle conjugate data denoising model; adjust the scaling ratio of the network to 2:1.5:2, using data set Y1，Y2And training to obtain a security inspection CT ray signal under-angle non-conjugate data denoising model.
In a specific embodiment of the present invention, as shown in fig. 3, the step of inputting the test data set into the trained deep convolution inverse graph network for performing the transfer training of the security check CT ray signal denoising model specifically includes:
judging the completeness of the security check CT ray signals in the test data set: if the signal is complete data, selecting a safety inspection CT ray signal all-angle conjugate data denoising model for carrying out transfer training to directly obtain a safety inspection CT ray signal with noise suppression; if the signal is incomplete data, dividing the test data set into conjugate data and non-conjugate data according to an acquisition angle, distributing network weights according to the magnitude of an under-acquisition angle, respectively and correspondingly selecting a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal for transfer training, and performing weighted synthesis to obtain the noise-suppressed security check CT ray signal.
In an embodiment of the present invention, the constructing the variable-step-size hole convolutional layer and the deep convolutional inverse graphics network formed by the hole convolutional layer includes: the deep convolution reverse graph network comprises multilayer convolution and deconvolution operators and is trained by using a random gradient variational Bayes algorithm; the convolution kernel of the cavity convolution layer adopts variable step length, the size of the convolution kernel is increased without increasing the number of network parameters, and the receptive field of the whole network is enlarged, wherein the receptive field is used for measuring the context information of the convolution neural network.
Specifically, a Deep Convolution Inverse Graphics Network (DC-IGN) consists of multiple layers of Convolution and deconvolution operators and is trained using the Stochastic Gradient Variational Bayes (SGVB) algorithm. The DC-IGN is an automatic encoder, the left side is similar to a Deep Convolutional Network (DCN), the right side is similar to a Deconvolution Network (DN), the DCN is used for compressing the image to reduce unnecessary features, and the DN is used for obtaining features and generating features similar to the features. After the security check CT ray signal is input into the DC-IGN, clear edge information which needs to be reserved can be generated, interference noise which is added in the acquisition is eliminated, and therefore the quality of the reconstructed security check CT image is improved, and the structure of the DC-IGN neural network is shown in figure 4.
The neurons in the image coding generated by the encoder can represent specific features (such as object edges, illumination and the like), and only the specific features (edges, illumination and the like) can be generated in a new image according to a trained model under the condition of a given input image, and non-specific features (such as noise, artifact and the like) are ignored. The image definition is influenced by image edge features, reconstruction of edge pixels depends on corresponding signal context information, a receptive field can be used for measuring the context information of a convolutional neural network, the cavity convolution can increase the size of a convolution kernel without increasing the number of network parameters, the receptive field of the whole network is enlarged, the stability of the feature extraction of the convolutional neural network is improved, and a cavity convolution schematic diagram is shown in fig. 5, and the cavity convolution kernel with the step length of 1, the step length of 2 and the step length of 3 is sequentially arranged from left to right.
In a specific embodiment of the present invention, the deep convolution inverse graphics network specifically includes an encoder and a generator;
the encoder comprises a plurality of convolution layers, wherein a maximum pooling layer is arranged behind each convolution layer and is used for compressing unnecessary features of the security inspection CT ray signal;
the generator comprises a plurality of corresponding convolution layers, and an inverse pooling layer is arranged behind each convolution layer and is used for generating specific characteristics to obtain a security inspection CT ray signal with noise suppression.
Specifically, the DC-IGN has an encoder consisting of several convolutional layers, each followed by a max pooling layer to reduce data dimensionality and prevent overfitting, and a generator with several convolutional layers, each followed by an inverse pooling process (upsampling uses the nearest domain mode).
In a specific embodiment of the present invention, the deep convolution inverse graph network performs back propagation gradient update network parameters through random gradient descent, and adopts a variational object function as follows:
y＝-log(p(x|zi)+KL(q(zi|x)||p(zi)))；
wherein y is the difference between the predicted distribution and the true distribution, x is the sample data of the training sample set, and ziFor compressed data containing hidden variables, i is the sample number of the training data set, KL (q | p) is the inverse KL divergence, and p (x | z | p)i) As likelihood probability, q (z)i| x) is the posterior probability, p (z)i) Is a prior probability.
Specifically, during training, signal data (X or Y) enters the encoder through the input layer, producing an a posteriori approximation q (z)i| x), wherein ziConsisting of sharp features in the image (such as baggage, light, texture or shape). To change the learning parameters in DC-IGN, the gradients are propagated back using random gradient descent using the variational objective function:
y＝-log(p(x|zi)+KL(q(zi|x)||p(zi)))；
for each ziThe DC-IGN learning disentanglement function may be forced by multi-sequence graph processing of a set of inactive and active features (e.g., baggage rotation, texture changes in a certain direction, etc.).
In a specific embodiment of the present invention, a full-angle conjugate data denoising model of a security check CT ray signal and a less-angle non-conjugate data denoising model of a security check CT ray signal are obtained by adjusting a scaling ratio of a network and using the training data set, and the method includes:
scaling ratio of network to 4:1:2 using training data X1，X2Training is carried out to obtain a safety inspection CT ray signal full-angle conjugate data denoising model;
scaling ratio of network to 2:1.5:2 using training data Y1，Y2And training to obtain a security inspection CT ray signal under-angle non-conjugate data denoising model.
In a specific embodiment of the present invention, the complete data or the conjugate data is subjected to a migration training by selecting a full-angle conjugate data denoising model of a security inspection CT ray signal, and data is compressed by an encoder to obtain ziAmplifying the signal through a generator to obtain a denoised signal, and removing noise data in the signal to inhibit the reconstructed image noise;
the non-conjugate data selects a CT ray signal under-angle non-conjugate data denoising model for carrying out migration training, and data is compressed through a coder to obtain ziAmplifying the signal through a generator to obtain a denoised signal, and removing noise data in the signal to inhibit the reconstructed image noise;
wherein z isiCompressed data obtained by scaling of the network model for the training data set.
In an embodiment of the present invention, the weighted synthesis to obtain the noise-suppressed security inspection CT ray signal includes: and (3) respectively generating denoised signals of the full-angle conjugate data denoising model and the under-angle non-conjugate data denoising model of the security check CT ray signal, and synthesizing the denoised signals of the security check CT ray signal according to the under-angle proportional weight.
In summary, the invention discloses a noise suppression method for a security check CT ray signal, comprising the following steps: acquiring a data set of a security check CT ray signal, and dividing the data set into a training data set and a test data set; wherein the data set comprises noisy security check CT ray signal data and normal security check CT ray signal data; constructing a variable-step-length hole convolutional layer and a depth convolutional inverse graphic network formed by the hole convolutional layer; training the deep convolution reverse graphic network by adopting the training data set, updating network parameters through a back propagation gradient, and obtaining the deep convolution reverse graphic network containing a security check CT ray signal denoising model after training; inputting the test data set into the trained deep convolution reverse graph network to perform security check CT ray signal denoising model transfer training to obtain a tested deep convolution reverse graph network; and inputting the security check CT ray signal to be subjected to noise suppression into the tested deep convolution reverse graph network to obtain a denoised security check CT ray signal. Compared with the traditional image processing method, the technical scheme of the invention does not need to design an accurate noise model, can achieve the aim of suppressing various noises only by expanding training data, and simultaneously utilizes the conjugate characteristic of a CT ray signal for security inspection to classify complete and incomplete CT ray signals, uses different networks to train classified data, and performs weighted synthesis on the generated data according to proportion to finally obtain the denoised signal data. Namely, the embodiment of the invention can denoise the under-angle security check CT ray signal, expands the denoising application range of the security check CT ray signal, constructs the convolutional layer by using the cavity convolution, can enlarge the receptive field of a neural network, enables the signal characteristics extracted by the convolutional layer to be more stable and accurate, and achieves the purpose of finally improving the quality of the reconstructed image.
Those skilled in the art will appreciate that all or part of the processes for implementing the methods in the above embodiments may be implemented by a computer program, which is stored in a computer-readable storage medium, to instruct associated hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (10)
1. A noise suppression method for a security check CT ray signal is characterized by comprising the following steps:
acquiring a data set of a security check CT ray signal, and dividing the data set into a training data set and a test data set; wherein the data set comprises noisy security check CT ray signal data and normal security check CT ray signal data;
constructing a variable-step-length hole convolutional layer and a depth convolutional inverse graphic network formed by the hole convolutional layer;
training the deep convolution reverse graphic network by adopting the training data set, updating network parameters through a back propagation gradient, and obtaining the deep convolution reverse graphic network containing a security check CT ray signal denoising model after training;
inputting the test data set into the trained deep convolution reverse graph network to perform security check CT ray signal denoising model transfer training to obtain a tested deep convolution reverse graph network;
and inputting the security check CT ray signal to be subjected to noise suppression into the tested deep convolution reverse graph network to obtain a denoised security check CT ray signal.
2. The method according to claim 1, characterized in that the data set of security check CT-ray signals comprises in particular: full-angle full-conjugate data X of CT (computed tomography) ray signals with noise security check1，X1Full-angle full-conjugate data X of epsilon A and normal security inspection CT ray signals2，X2Epsilon A and under-angle unconjugated data Y of CT ray signal with noise security check1，Y1Epsilon A and normal security inspection CT ray signal under-angle unconjugated data Y2，Y2Belonging to A and CT ray signal under-angle mixed conjugate data Z with noise security check, wherein Z belongs to A; wherein the mixed conjugate data Z includes conjugate data and non-conjugate data; a is the domain of all types of security inspection CT ray signal noise;
the training dataSet includes X1，X2，Y1，Y2(ii) a The test data set is Z.
3. The method of claim 2, wherein the training the deep convolutional inverse graphics network with the training data set specifically comprises: and obtaining a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal by using the training data set through adjusting the scaling ratio of the network.
4. The method according to claim 1 or 2, wherein the inputting of the test data set into the trained deep convolution inverse graph network for the security check CT ray signal denoising model transfer training specifically comprises:
judging the completeness of the security check CT ray signals in the test data set: if the signal is complete data, selecting a safety inspection CT ray signal all-angle conjugate data denoising model for carrying out transfer training to directly obtain a safety inspection CT ray signal with noise suppression; if the signal is incomplete data, dividing the test data set into conjugate data and non-conjugate data according to an acquisition angle, distributing network weights according to the magnitude of an under-acquisition angle, respectively and correspondingly selecting a full-angle conjugate data denoising model of the security check CT ray signal and an under-angle non-conjugate data denoising model of the security check CT ray signal for transfer training, and performing weighted synthesis to obtain the noise-suppressed security check CT ray signal.
5. The method of claim 1, wherein constructing the variable-step-size hole convolutional layer and the deep convolutional inverse graphics network of the hole convolutional layers comprises: the deep convolution reverse graph network comprises multilayer convolution and deconvolution operators and is trained by using a random gradient variational Bayes algorithm; the convolution kernel of the cavity convolution layer adopts variable step length, the size of the convolution kernel is increased without increasing the number of network parameters, and the receptive field of the whole network is enlarged, wherein the receptive field is used for measuring the context information of the convolution neural network.
6. The method according to claim 5, characterized in that said deep convolutional inverse graphics network comprises in particular an encoder and a generator;
the encoder comprises a plurality of convolution layers, wherein a maximum pooling layer is arranged behind each convolution layer and is used for compressing unnecessary features of the security inspection CT ray signal;
the generator comprises a plurality of corresponding convolution layers, and an inverse pooling layer is arranged behind each convolution layer and is used for generating specific characteristics to obtain a security inspection CT ray signal with noise suppression.
7. The method according to claim 1 or 5, wherein the deep convolution inverse graph network updates network parameters by backpropagating gradient through stochastic gradient descent using a variational object function as follows:
y＝-log(p(x|zi)+KL(q(zi|x)||p(zi)))；
wherein y is the difference between the predicted distribution and the true distribution, x is the sample data of the training sample set, and ziFor compressed data containing hidden variables, i is the sample number of the training data set, KL (q | p) is the inverse KL divergence, and p (x | z | p)i) As likelihood probability, q (z)i| x) is the posterior probability, p (z)i) Is a prior probability.
8. The method of claim 3, wherein the obtaining a full-angle conjugate data denoising model of the security CT ray signal and a less-angle non-conjugate data denoising model of the security CT ray signal by using the training data set by adjusting the scaling ratio of the network comprises:
scaling ratio of network to 4:1:2 using training data X1，X2Training is carried out to obtain a safety inspection CT ray signal full-angle conjugate data denoising model;
scaling ratio of network to 2:1.5:2 using training data Y1，Y2And training to obtain a security inspection CT ray signal under-angle non-conjugate data denoising model.
9. The method of claim 4, wherein the complete data or the conjugate data is subjected to a migration training by selecting a full-angle conjugate data denoising model of the security check CT ray signal, and z is obtained by compressing data through an encoderiAmplifying the signal through a generator to obtain a denoised signal, and removing noise data in the signal to inhibit the reconstructed image noise;
the non-conjugate data selects a CT ray signal under-angle non-conjugate data denoising model for carrying out migration training, and data is compressed through a coder to obtain ziAmplifying the signal through a generator to obtain a denoised signal, and removing noise data in the signal to inhibit the reconstructed image noise;
wherein z isiCompressed data obtained by scaling of the network model for the training data set.
10. The method of claim 4 or 9, wherein the weighted synthesizing of the noise suppressed security check CT ray signals comprises: and (3) respectively generating denoised signals of the full-angle conjugate data denoising model and the under-angle non-conjugate data denoising model of the security check CT ray signal, and synthesizing the denoised signals of the security check CT ray signal according to the under-angle proportional weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011155753.1A CN112258414B (en) | 2020-10-26 | 2020-10-26 | Noise suppression method for security inspection CT ray signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011155753.1A CN112258414B (en) | 2020-10-26 | 2020-10-26 | Noise suppression method for security inspection CT ray signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112258414A true CN112258414A (en) | 2021-01-22 |
| CN112258414B CN112258414B (en) | 2023-11-10 |
Family
ID=74262046
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011155753.1A Active CN112258414B (en) | 2020-10-26 | 2020-10-26 | Noise suppression method for security inspection CT ray signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112258414B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180240219A1 (en) * | 2017-02-22 | 2018-08-23 | Siemens Healthcare Gmbh | Denoising medical images by learning sparse image representations with a deep unfolding approach |
| CN109064521A (en) * | 2018-07-25 | 2018-12-21 | 南京邮电大学 | A kind of CBCT using deep learning removes pseudo- image method |
| CN109584324A (en) * | 2018-10-24 | 2019-04-05 | 南昌大学 | A kind of positron e mission computed tomography (PET) method for reconstructing based on autocoder network |
| CN110473150A (en) * | 2019-06-24 | 2019-11-19 | 浙江工业大学之江学院 | CNN medicine CT image denoising method based on multi-feature extraction |
-
2020
- 2020-10-26 CN CN202011155753.1A patent/CN112258414B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180240219A1 (en) * | 2017-02-22 | 2018-08-23 | Siemens Healthcare Gmbh | Denoising medical images by learning sparse image representations with a deep unfolding approach |
| CN109064521A (en) * | 2018-07-25 | 2018-12-21 | 南京邮电大学 | A kind of CBCT using deep learning removes pseudo- image method |
| CN109584324A (en) * | 2018-10-24 | 2019-04-05 | 南昌大学 | A kind of positron e mission computed tomography (PET) method for reconstructing based on autocoder network |
| CN110473150A (en) * | 2019-06-24 | 2019-11-19 | 浙江工业大学之江学院 | CNN medicine CT image denoising method based on multi-feature extraction |
Non-Patent Citations (1)
| Title |
|---|
| 上官宏: "低剂量X线CT统计迭代重建方法研究", 中国博士学位论文全文数据库 医药卫生科技辑, no. 08, pages 1 - 133 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112258414B (en) | 2023-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110211045B (en) | Super-resolution face image reconstruction method based on SRGAN network | |
| CN108460726B (en) | Magnetic resonance image super-resolution reconstruction method based on enhanced recursive residual network | |
| CN106940816B (en) | CT image pulmonary nodule detection system based on 3D full convolution neural network | |
| CN107633486B (en) | Structural magnetic resonance image denoising method based on three-dimensional full-convolution neural network | |
| CN109214989B (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
| CN109035142B (en) | Satellite image super-resolution method combining countermeasure network with aerial image prior | |
| CN111951344B (en) | Magnetic resonance image reconstruction method based on cascade parallel convolution network | |
| CN110728729A (en) | Unsupervised CT projection domain data recovery method based on attention mechanism | |
| CN112767417B (en) | Multi-modal image segmentation method based on cascaded U-Net network | |
| CN102231204A (en) | Sequence image self-adaptive regular super resolution reconstruction method | |
| He et al. | Remote sensing image super-resolution using deep–shallow cascaded convolutional neural networks | |
| Paul et al. | Modified convolutional neural network with pseudo-CNN for removing nonlinear noise in digital images | |
| CN113538616A (en) | Magnetic resonance image reconstruction method combining PUGAN and improved U-net | |
| CN111353935A (en) | Magnetic resonance imaging optimization method and device based on deep learning | |
| CN113808234A (en) | Rapid magnetic particle imaging reconstruction method based on undersampling | |
| Li et al. | Speckle noise removal based on structural convolutional neural networks with feature fusion for medical image | |
| CN116051382A (en) | Data enhancement method based on deep reinforcement learning generation type antagonistic neural network and super-resolution reconstruction | |
| Wang et al. | An adaptive learning image denoising algorithm based on eigenvalue extraction and the GAN model | |
| Zuluaga et al. | Blind microscopy image denoising with a deep residual and multiscale encoder/decoder network | |
| CN106023116B (en) | A kind of compressed sensing image rebuilding method and device based on block Weighted Constraint | |
| Soltanmohammadi et al. | A comparative analysis of super-resolution techniques for enhancing micro-CT images of carbonate rocks | |
| Yang et al. | GAN-based sample expansion for underwater acoustic signal | |
| Annavarapu et al. | An adaptive watershed segmentation based medical image denoising using deep convolutional neural networks | |
| El-Shafai et al. | Hybrid Single Image Super-Resolution Algorithm for Medical Images. | |
| CN117726540A (en) | Image denoising method for enhanced gate control converter |
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 |