Training method of signal denoising model, signal denoising method and device

Abstract

The application relates to the technical field of signal processing, and provides a training method of a signal denoising model, a signal denoising method and equipment, which can improve the denoising effect of the model. The application performs step-by-step denoising training on a multi-stage network based on training signal data corresponding to the multi-stage network, to obtain a signal denoising model; if the current stage reached by the step-by-step denoising training is a non-first stage, a loss value of the current stage network is obtained according to the training signal data corresponding to the current stage network; after gradient descent is separately performed on the current stage network based on the loss value, gradient descent is jointly performed on the current multi-stage network that has been separately subjected to gradient descent based on the loss value.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and in particular to a training method and apparatus for a signal denoising model, a signal denoising method and apparatus, a computer device, a storage medium, and a computer program product. Background Technology

[0002] A signal is a physical quantity that carries information. Most signals carry some noise, which is considered interference. Noise negatively impacts signal processing, reducing signal quality and intelligibility.

[0003] Noise reduction can decrease the noise in a signal, improve the signal-to-noise ratio, and prevent the signal from being overwhelmed by excessive noise. Traditional techniques can perform noise reduction using trained models; however, the noise reduction effect of these models is sometimes poor. Summary of the Invention

[0004] Therefore, it is necessary to provide a training method and apparatus for a signal denoising model, a signal denoising method and apparatus, a computer device, a storage medium, and a computer program product to address the above-mentioned technical problems.

[0005] This application provides a training method for a signal denoising model, the method comprising:

[0006] Based on the training signal data corresponding to the multi-level network, the multi-level network is trained to perform noise reduction step by step to obtain the signal noise reduction model.

[0007] If the current level trained by stepwise noise reduction is not the first level, the loss value of the current level network is obtained based on the training signal data corresponding to the current level network.

[0008] After performing gradient descent on the current level network individually based on the loss value, gradient descent is then jointly performed on the current multi-level networks that have already undergone individual gradient descent based on the loss value.

[0009] In one embodiment, the training signal data corresponding to the current level network is supplemented with the signal from the noise that was removed by the previous level network of the current level network.

[0010] In one embodiment, performing gradient descent on the current-level network separately based on the loss value includes:

[0011] Based on the loss value, the current level network is back-calculated layer by layer to obtain the first type of gradient value;

[0012] Gradient descent is performed on the current level network based on the first type of gradient value.

[0013] In one embodiment, performing joint gradient descent on the current multi-level individually gradient descent networks based on the loss value includes:

[0014] Based on the loss value, the current multi-level individually gradient descent network is back-calculated network by network to obtain the second type of gradient value;

[0015] Based on the second type of gradient value, gradient descent is performed on the current multi-level individual gradient descent network.

[0016] In one embodiment, the method further includes:

[0017] Based on the noise reduction calibration range corresponding to each level of the network, the training signal data corresponding to each level of the network is obtained; each training signal data includes training input signal data and training target signal data, wherein the noise level carried by the training input signal data is the upper limit of the noise reduction calibration range corresponding to the corresponding level of the network, and the noise level carried by the training target signal data is the lower limit of the noise reduction calibration range corresponding to the corresponding level of the network.

[0018] Based on the training signal data corresponding to the current level network, the loss value of the current level network is obtained, including:

[0019] Input the training input signal data corresponding to the current level network into the current level network;

[0020] The loss value of the current-level network is obtained based on the difference in noise level between the output of the current-level network and the corresponding training target signal data.

[0021] In one embodiment, before inputting the training input signal data corresponding to the current-level network into the current-level network, the method further includes:

[0022] Based on the training input signal data and training target signal data corresponding to the previous level network, the noise removed by the previous level network of the current level network is obtained;

[0023] The noise removed by the previous level network of the current level network is extracted;

[0024] The extracted signal is added to the training input signal data corresponding to the current level network.

[0025] In one embodiment, a multi-level network is either a network with the same network structure or a network with different network structures.

[0026] In one embodiment, when the multi-level network consists of networks with different network structures, the structural complexity of the preceding network in two adjacent networks is higher than that of the following network, and / or, the number of parameters of the preceding network in two adjacent networks is greater than that of the following network.

[0027] In one embodiment, the signal data includes either CT data or CT images.

[0028] In one embodiment, when the signal data includes CT generation data and the CT generation data is CT multi-level data, the training input CT multi-level data is either merged into the network or input to the network independently.

[0029] When the signal data includes CT images, and the CT images are multi-level CT images, the training input CT multi-level CT images are either merged into the network or input into the network independently.

[0030] This application provides a training device for a signal denoising model, the device comprising:

[0031] The model acquisition module is used to perform step-by-step noise reduction training on the multi-level network based on the training signal data corresponding to the multi-level network, and obtain the signal noise reduction model.

[0032] The training module is used to obtain the loss value of the current level network based on the training signal data corresponding to the current level network if the current level trained by stepwise noise reduction is not the first level; and to perform gradient descent on the current level network alone based on the loss value, and then to perform gradient descent on the current multi-level networks that have been individually gradient descent based on the loss value.

[0033] This application provides a signal noise reduction method, the method comprising:

[0034] Acquire the signal to be denoised;

[0035] The signal to be denoised is input into the signal denoising model to obtain the denoised signal; the signal denoising model is trained according to the training method of the signal denoising model described in the above embodiment.

[0036] In one embodiment, the signal to be denoised is input into a signal denoising model to obtain a denoised signal, including:

[0037] Determine the noise level of the signal to be denoised;

[0038] Based on the noise level of the signal to be denoised, a target input network is determined in the signal denoising model; the noise level of the signal to be denoised is within the denoising calibration range of the target input network.

[0039] The signal to be denoised is input into the target input network, so that the forward multi-stage network in the signal denoising model, starting from the target input network, performs step-by-step denoising on the signal to be denoised, and obtains the denoised signal.

[0040] This application provides a signal noise reduction device, the device comprising:

[0041] The signal acquisition module is used to acquire the signal to be denoised;

[0042] The noise reduction module is used to input the signal to be denoised into the signal denoising model to obtain the denoised signal; the signal denoising model is trained according to the training method of the signal denoising model described in the above embodiments.

[0043] This application provides a computer device, including a memory and a processor, wherein the memory stores a computer program and the processor executes the above-described method.

[0044] This application provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor using the methods described above.

[0045] This application provides a computer program product having a computer program stored thereon, the computer program being executed by a processor using the above-described method.

[0046] The solution provided in this application involves training multi-level networks using training signal data to perform step-by-step denoising training, resulting in a signal denoising model. If the current level trained through step-by-step denoising is not the first level, the loss value of the current level network is obtained based on the training signal data corresponding to that level. Gradient descent is then performed on the current level network individually based on the loss value, followed by joint gradient descent on the current multi-level networks that have already undergone individual gradient descent. This application employs a step-by-step denoising approach, reducing noise at each level. During training, not only are individual gradient descent performed on single-level networks, but joint gradient descent is also performed on multiple levels, which improves the denoising capability and enhances the denoising effect of the signal denoising model. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating the training method of a signal denoising model in one embodiment;

[0049] Figure 2(a) is a schematic diagram of the series connection of a four-level network in one embodiment;

[0050] Figure 2(b) is a schematic diagram of the architecture for first-level noise reduction training in one embodiment;

[0051] Figure 2(c) is a schematic diagram of the architecture for the second-level noise reduction training in one embodiment;

[0052] Figure 2(d) is a schematic diagram of the architecture for training the third level of noise reduction in one embodiment;

[0053] Figure 2(e) is a schematic diagram of the architecture for training the fourth level of noise reduction in one embodiment;

[0054] Figure 3(a) is a schematic diagram of the architecture for first-level noise reduction training with CT multi-level data in one embodiment;

[0055] Figure 3(b) is a schematic diagram of the architecture for second-level noise reduction training with CT multi-level data in one embodiment;

[0056] Figure 4 This is a flowchart illustrating a signal noise reduction method in one embodiment;

[0057] Figure 5 This is a flowchart illustrating the inference phase in one embodiment;

[0058] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] The signal denoising model training method and signal denoising method provided in this application can be executed by computer equipment. The signal denoising model training method may include... Figure 1 The steps are shown.

[0061] Step S101: Based on the training signal data corresponding to the multi-level network, perform step-by-step noise reduction training on the multi-level network to obtain the signal noise reduction model.

[0062] When constructing a signal denoising model, the number of networks required for the model can be determined based on actual needs. The number of networks can be set to be greater than or equal to 2.

[0063] Multiple networks can be connected in series. Accordingly, there is a hierarchical relationship between the networks. The closer the network is to the model input, the lower its level. The network closest to the model input can be called the first-level network, and the network closest to the model output can be called the last-level network. Let the number of networks be an integer n. Taking n=4 as an example, as shown in Figure 2(a), in a four-level network, the first-level network is closest to the model input, and the fourth-level network is closest to the model output.

[0064] A corresponding noise reduction range is calibrated for each network level; that is, a corresponding noise reduction calibration range is set for each network level. The closer the network is to the model input (i.e., the smaller the network level), the larger the noise reduction calibration range; the closer the network is to the output (i.e., the larger the network level), the smaller the noise reduction calibration range.

[0065] According to the order of network level from large to small, the noise reduction calibration range corresponding to the multi-level network decreases in turn, showing a decreasing relationship; moreover, the noise reduction calibration ranges corresponding to the multi-level network can be independent of each other, and the noise reduction calibration ranges corresponding to the multi-level network can collectively form a larger range.

[0066] Noise reduction calibration ranges can be defined by either absolute noise magnitude or relative noise magnitude. Absolute noise magnitude is the actual physical measure of noise, a direct measurement of the noise's own size. Relative noise magnitude is a comparative value of noise relative to a specific noise level, and can be expressed as a percentage.

[0067] Let the number of networks be n, and we'll take n=4 as an example. If the noise reduction calibration range is divided by the absolute noise level, then the noise reduction calibration range for the first-level network can be 400 to 300, the second-level network can be 300 to 200, the third-level network can be 200 to 100, and the fourth-level network can be 100 to 0. The noise reduction calibration ranges for these four levels of networks decrease sequentially from the network level to the minimum. The noise reduction calibration ranges can be independent of each other and can collectively form a larger range of 400 to 0.

[0068] If the noise reduction calibration range is defined by relative noise levels, then the noise reduction calibration range for the first-level network can be 100% to 75%, for the second-level network it can be 75% to 50%, for the third-level network it can be 50% to 25%, and for the fourth-level network it can be 25% to 0. The noise reduction calibration ranges for these four levels decrease sequentially from highest to lowest network level. These noise reduction calibration ranges can be independent of each other and can collectively form a larger range of 100% to 0.

[0069] The noise reduction calibration ranges for multi-level networks can have the same or different spans. The span of the noise reduction calibration range for a single-level network is calculated by subtracting the lower limit of the noise reduction calibration range for that level from its upper limit.

[0070] When the span of the noise reduction calibration range corresponding to a multi-level network is the same, the noise reduction calibration range corresponding to the multi-level network decreases proportionally. When the span of the noise reduction calibration range corresponding to a multi-level network is different, the noise reduction calibration range corresponding to the multi-level network can decrease according to a logarithmic function.

[0071] When the noise reduction calibration range is defined by the absolute noise level, in some scenarios, the noise reduction calibration range for the first-level network is 400 to 300, for the second-level network it is 300 to 200, for the third-level network it is 200 to 100, and for the fourth-level network it is 100 to 0. In this scenario, the noise reduction calibration ranges for multiple levels of networks all have the same span, which is 100.

[0072] When the noise reduction calibration range is defined by the absolute noise level, in some scenarios, the noise reduction calibration range for the first-level network is 400 to 350, for the second-level network it is 350 to 250, for the third-level network it is 250 to 100, and for the fourth-level network it is 100 to 0. In this scenario, the span of the noise reduction calibration range for the multiple levels of networks is different: the span for the first-level network is 50, for the second-level network it is 100, for the third-level network it is 150, and for the fourth-level network it is 100.

[0073] When the noise reduction calibration range is defined by relative noise levels, in some scenarios, the noise reduction calibration range for the first-level network is 100% to 75%, for the second-level network it's 75% to 50%, for the third-level network it's 50% to 25%, and for the fourth-level network it's 25% to 0. In this scenario, the noise reduction calibration ranges for all multiple network levels have the same span, 25%.

[0074] When the noise reduction calibration range is defined by relative noise levels, in some scenarios, the noise reduction calibration range corresponding to the first-level network is 100% to 85%, the second-level network is 85% to 60%, the third-level network is 60% to 25%, and the fourth-level network is 25% to 0%. In this scenario, the span of the noise reduction calibration range corresponding to multiple levels of networks is different: the span of the noise reduction calibration range corresponding to the first-level network is 15%, the span of the second-level network is 25%, the span of the third-level network is 35%, and the span of the fourth-level network is 25%.

[0075] The noise reduction calibration range corresponding to each network level is used to calibrate how each network level can reduce the noise carried by the signal from a certain level to another level. The noise level before reduction corresponds to the upper limit of the noise reduction calibration range, and the noise levels before and after reduction correspond to the lower limit of the noise reduction calibration range.

[0076] Therefore, after determining the noise reduction calibration range corresponding to each level of the network, the noise level carried by the training signal data input to each level of the network and the noise level to be reduced can be determined based on the noise reduction calibration range of each level of the network itself. Based on the noise level carried by the training signal data input to each level of the network, training signal data with corresponding noise level as input (which can be called training input signal data) is constructed. Based on the noise level to be reduced, training signal data with corresponding noise level as target (which can be called training target signal data) is constructed. The training input signal data and the training target signal data form a training signal data.

[0077] It is understandable that each level of the network can correspond to at least one training signal data.

[0078] After obtaining the training signal data corresponding to the multi-level network, the multi-level network can be trained for noise reduction step by step based on the training signal data, in ascending order of network level. First, the first-level network is trained, then the second-level network, and so on, until the last-level network is trained. After all levels of the network have been trained, it is considered that one iteration is completed. Multiple iterations can be performed according to actual needs.

[0079] The following describes the relevant training process for a single iteration.

[0080] If the current level trained through successive denoising is considered the first level, meaning we are currently performing first-level denoising training, we can obtain the loss value of the first-level network based on the training signal data corresponding to that network. Based on this loss value, we then perform gradient descent on the first-level network independently. At this point, the first-level denoising training is complete, and we can proceed to the next level of denoising training.

[0081] Step S102, wherein if the current level trained by stepwise noise reduction is not the first level, the loss value of the current level network is obtained according to the training signal data corresponding to the current level network; after performing gradient descent on the current level network alone based on the loss value, gradient descent is jointly performed on the current multi-level networks that have been individually gradient descent based on the loss value.

[0082] For example, if the current level trained through stepwise denoising is the second level, that is, if the current training is for the second level of denoising, the loss value of the second-level network can be obtained based on the training signal data corresponding to the second-level network. Based on the loss value of the second-level network, gradient descent is performed on the second-level network alone. After performing gradient descent on the second-level network alone, the current multi-level networks that have undergone individual gradient descent include the second-level network and networks at levels lower than the second level (i.e., the first-level network). Based on the loss value of the second-level network, gradient descent can be jointly performed on the first-level network and the second-level network. At this point, the second-level denoising training is completed, and the next level of denoising training can begin.

[0083] The current level trained through successive denoising is the third level. That is, we are currently training for the third level of denoising, and the loss value of the third-level network can be obtained based on the training signal data corresponding to the third-level network. Based on the loss value of the third-level network, gradient descent is performed on the third-level network alone. After performing gradient descent on the third-level network alone, the current multi-level networks that have undergone individual gradient descent include the third-level network and networks at levels lower than the third level (i.e., the first-level and second-level networks). Based on the loss value of the third-level network, gradient descent can be jointly performed on the first-level, second-level, and third-level networks. At this point, the training for the third-level denoising is complete, and the next level of denoising training can begin.

[0084] Following the method described above, perform the remaining levels of noise reduction training until the final level of noise reduction training is completed. Completing the final level of noise reduction training can be considered as completing one iteration.

[0085] In the training method of the aforementioned signal denoising model, the multi-level networks are trained stepwise for denoising based on the training signal data corresponding to the multi-level networks to obtain the signal denoising model. If the current level trained through stepwise denoising is not the first level, the loss value of the current level network is obtained based on the training signal data corresponding to the current level network. Gradient descent is then performed on the current level network individually based on the loss value, and finally, gradient descent is performed jointly on the current multi-level networks that have already undergone individual gradient descent based on the loss value. This application adopts a stepwise denoising approach, with each level network reducing some noise. During training, not only is individual gradient descent performed on single-level networks, but joint gradient descent is also performed on multi-level networks, which can improve the denoising capability of the signal denoising model and enhance the model's denoising effect.

[0086] In one embodiment, the training signal data corresponding to the current-level network is supplemented with the signal from the noise that was removed by the previous-level network.

[0087] For example, if the current level trained through successive denoising is the first level, that is, if the current training is for the first level of denoising, the loss value of the first level network can be obtained based on the training signal data corresponding to the first level network. Based on the loss value of the first level network, gradient descent is performed on the first level network alone. At this point, the first level of denoising training is completed, and the next level of denoising training can be performed.

[0088] The current level trained through successive denoising is the second level. That is, when training the second level of denoising, the signal removed by the first level network (which was present in the noise) is added to the training signal data corresponding to the second level network. The loss value of the second level network is then obtained based on the training signal data with this added signal. Gradient descent is then performed on the second level network independently based on its loss value. After performing gradient descent on the second level network alone, the current multi-level networks that have undergone individual gradient descent include both the second level network and networks at lower levels (i.e., the first level network). Gradient descent can then be jointly performed on both the first and second level networks based on the loss value of the second level network. At this point, the second level of denoising training is complete, and the next level of denoising training can begin.

[0089] The current level trained through successive denoising is the third level. That is, when training for the third level of denoising, the signal removed by the second level network (which was present in the noise) is added to the training signal data corresponding to the third level network. The loss value of the third level network is then obtained based on this added signal. Gradient descent is then performed on the third level network independently based on its loss value. After performing gradient descent on the third level network alone, the current multi-level networks that have undergone individual gradient descent include the third level network and networks at lower levels (i.e., the first and second levels). Based on the loss value of the third level network, gradient descent can then be jointly performed on the first, second, and third levels. At this point, the third level of denoising training is complete, and the next level of denoising training can begin.

[0090] Following the method described above, perform the remaining levels of noise reduction training until the final level of noise reduction training is completed. Completing the final level of noise reduction training can be considered as completing one iteration.

[0091] Specifically, when adding the signal from the noise removed by the previous stage network to the training signal data corresponding to the current stage network, it can be done by adding a predetermined proportion of the signal from the noise removed by the previous stage network to the training signal data corresponding to the current stage network. The predetermined proportion can be set according to actual needs and can be between 0 and 1. The predetermined proportion can be the same or different when training different levels of noise reduction.

[0092] In scenarios where the same proportion is used during training at different levels of noise reduction, for example, 80% of the signal from the noise removed by the first-level network is added to the training signal data corresponding to the second-level network; and 80% of the signal from the noise removed by the second-level network is added to the training signal data corresponding to the third-level network.

[0093] In scenarios where different proportions are used during training at different levels of noise reduction, for example, 90% of the signal from the noise removed by the first-level network is added to the training signal data corresponding to the second-level network; and 80% of the signal from the noise removed by the second-level network is added to the training signal data corresponding to the third-level network.

[0094] In this embodiment, during the stepwise noise reduction training process, the signal from the noise removed by the previous level network is added to the training signal data corresponding to the current level network to compensate for and restore the signal. This avoids the signal being weakened after each level network is denoised during training, allowing the signal denoising model to perform denoising while ensuring signal accuracy. In particular, when the signal denoising model processes some signals carrying large noise, it can significantly alleviate the signal weakening situation.

[0095] In one embodiment, gradient descent is performed on the current level network separately based on the loss value, including:

[0096] Based on the loss value, back-calculate layer by layer in the current level of the network to obtain the first type of gradient value; based on the first type of gradient value, perform gradient descent on the current level of the network.

[0097] Each network level can include multiple layers. Therefore, when performing gradient descent on a single-level network, the gradient values ​​can be obtained by back-calculating layer by layer. The direction of back-calculation is from the layer closest to the model output to the layer closest to the model input. The gradient values ​​obtained by back-calculation are mainly used for gradient descent alone. To distinguish them, the gradient values ​​obtained by back-calculation can be called the first type of gradient values.

[0098] Let's take the current first-level noise reduction training as an example. Based on the training signal data corresponding to the first-level network, the loss value of the first-level network is obtained. Based on the loss value of the first-level network, layer-by-layer backward calculation is performed to obtain the first type of gradient values. Based on the first type of gradient values, gradient descent is performed on the first-level network to update the parameters within the first-level network.

[0099] In this embodiment, the loss value of the current level network is used to perform back-calculation layer by layer to obtain the first type of gradient value. Based on the first type of gradient value, gradient descent is performed on the current level network so that the parameters in the current level network can be updated in the direction of reducing the loss function value.

[0100] In one embodiment, performing joint gradient descent on the current multi-level individually gradient descent networks based on the loss value includes:

[0101] Based on the loss value, the current multi-level individually gradient descent network is back-calculated network by network to obtain the second type of gradient value; based on the second type of gradient value, the current multi-level individually gradient descent network is subjected to gradient descent.

[0102] In joint gradient descent, the current individual gradient descent network has two or more levels. The gradient values ​​are obtained by backcomputing the current multi-level individual gradient descent network network by network; the direction of backcomputing is from the network closest to the model output to the network closest to the model input. The gradient values ​​obtained by backcomputing the network by network are mainly used for joint gradient descent. To distinguish them, the gradient values ​​obtained by backcomputing the network by network can be called second-type gradient values.

[0103] Taking the current second-level noise reduction training as an example, the current multi-level gradient descent network includes a first-level network and a second-level network. Based on the loss value of the second-level network, reverse calculation is performed on the first-level network and the second-level network to obtain the second type of gradient value; based on the second type of gradient value, gradient descent is performed on the first-level network and the second-level network.

[0104] Taking the current third-level noise reduction training as an example, the current multi-level gradient descent network includes a first-level network, a second-level network, and a third-level network. Based on the loss value of the third-level network, backward computation is performed on the first-level network, the second-level network, and the third-level network to obtain the second type of gradient value; based on the second type of gradient value, gradient descent is performed on the first-level network, the second-level network, and the third-level network.

[0105] This embodiment performs layer-by-layer backward computation on the current multi-level gradient descent network based on the loss value of the current level network to obtain the second type of gradient value. Based on the second type of gradient value, gradient descent is performed on the current multi-level gradient descent network so that the parameters within the current multi-level gradient descent network can be updated in the direction of reducing the loss function value. By using a joint gradient descent approach, the optimal solution can be converged more quickly.

[0106] In one embodiment, the training method for the signal denoising model provided in this application further includes:

[0107] Based on the noise reduction calibration range corresponding to each network level, the training signal data corresponding to each network level is obtained. Each training signal data includes training input signal data and training target signal data. The noise level carried by the training input signal data is the upper limit of the noise reduction calibration range corresponding to the corresponding network level, and the noise level carried by the training target signal data is the lower limit of the noise reduction calibration range corresponding to the corresponding network level.

[0108] Based on the training signal data corresponding to the current level network, the loss value of the current level network is obtained, including:

[0109] Input the training input signal data corresponding to the current level network into the current level network; based on the difference in noise level between the output result of the current level network and the training target signal data corresponding to the current level network, obtain the loss value of the current level network.

[0110] The noise reduction calibration range corresponding to each network level is used to calibrate how each network level can reduce the noise carried by the signal from a certain level to another level. The noise level before reduction corresponds to the upper limit of the noise reduction calibration range, and the noise levels before and after reduction correspond to the lower limit of the noise reduction calibration range.

[0111] Therefore, after determining the noise reduction calibration range corresponding to each level of the network, the noise level carried by the training signal data input to each level of the network can be determined based on the noise reduction calibration range of each level of the network itself, thereby constructing training input signal data with the corresponding noise level; the noise level to be reduced to is determined based on the noise reduction calibration range of each level of the network itself, thereby constructing training target signal data with the corresponding noise level; the training input signal data and the training target signal data form a training signal data.

[0112] In scenarios where the noise reduction calibration range is divided by the absolute noise level, we will take a primary network with a noise reduction calibration range of 400 to 300 as an example. The upper limit of the noise reduction calibration range for the primary network is 400, which can construct training input signal data with a noise level of 400. The lower limit of the noise reduction calibration range for the primary network is 300, which can construct training target signal data with a noise level of 300.

[0113] In some scenarios, a preset noise-free signal can be obtained. For any level of network, noise with a magnitude equal to the upper limit of the noise reduction calibration range corresponding to that level of network is added to the noise-free signal to obtain the training input signal data corresponding to that level of network. Noise with a magnitude equal to the lower limit of the noise reduction calibration range corresponding to that level of network is added to the noise-free signal to obtain the training target signal data corresponding to that level of network.

[0114] This paper takes a primary network with a noise reduction calibration range of 400 to 300 as an example. After obtaining a noise-free signal, noise of magnitude 400 can be simulated and added to the noise-free signal to obtain training input signal data with an absolute noise magnitude of 400. Alternatively, noise of magnitude 300 can be simulated and added to the noise-free signal to obtain training target signal data with an absolute noise magnitude of 300.

[0115] In other scenarios, a preset noisy signal can be obtained and the noise level of the noisy signal can be determined; for any level of network, the relative magnitude between the noise level of the noisy signal and the upper limit of the noise reduction calibration range can be determined, as well as the relative magnitude between the noise level of the noisy signal and the lower limit of the noise reduction calibration range can be determined.

[0116] If the noise level of the noise signal is greater than the upper limit of the noise reduction calibration range and the noise level of the noise signal is greater than the lower limit of the noise reduction calibration range, then the noise level of the noise signal can be reduced to the upper limit of the noise reduction calibration range through a preset noise reduction algorithm, thereby obtaining the training input signal data corresponding to this level of network; and the noise level of the noise signal can be reduced to the lower limit of the noise reduction calibration range through a noise reduction algorithm, thereby obtaining the training target signal data corresponding to this level of network.

[0117] If the noise level of the noisy signal is less than the upper limit of the noise reduction calibration range and greater than the lower limit of the noise reduction calibration range, then noise can be added to the noisy signal to increase its noise level to the upper limit of the noise reduction calibration range, thereby obtaining the training input signal data corresponding to this level of the network; through the noise reduction algorithm, the noise level of the noisy signal is reduced to the lower limit of the noise reduction calibration range, thereby obtaining the training target signal data corresponding to this level of the network.

[0118] If the noise level of the noisy signal is less than the upper limit of the noise reduction calibration range and less than the lower limit of the noise reduction calibration range, then noise can be added to the noisy signal to increase its noise level to the upper limit of the noise reduction calibration range, thereby obtaining the training input signal data corresponding to that level of the network; conversely, adding noise to the noisy signal to increase its noise level to the lower limit of the noise reduction calibration range will also yield the training target signal data corresponding to that level of the network.

[0119] If the noise level of the noise signal is equal to the upper limit of the noise reduction calibration range, then the noisy signal can be directly used as the training input signal data for that level of the network.

[0120] If the noise level of the noisy signal is equal to the lower limit of the noise reduction calibration range, then the noisy signal can be directly used as the training target signal data for that level of the network.

[0121] The preset noise reduction algorithm may include, but is not limited to, low-pass filtering, iterative noise reduction, deep learning noise reduction, and other algorithms.

[0122] This paper takes a primary network with a noise reduction calibration range of 400 to 300 as an example. Following the previously described method, we can obtain training input signal data with an absolute noise level of 400 for the primary network, and training target signal data with an absolute noise level of 300 for the primary network. Then, we can input the training input signal data with an absolute noise level of 400 into the primary network to obtain its output. Based on the difference in noise level between the primary network's output and the training target signal data with an absolute noise level of 300, we can obtain the loss value of the primary network. Subsequently, we can perform individual gradient descent and joint gradient descent based on the loss value of the primary network.

[0123] This paper takes a second-level network with a noise reduction calibration range of 300 to 200 as an example. Following the previously described method, we can obtain training input signal data with an absolute noise level of 300 for the second-level network, and training target signal data with an absolute noise level of 200 for the second-level network. Then, we can input the training input signal data with an absolute noise level of 300 into the second-level network to obtain its output. Based on the difference in noise level between the output of the second-level network and the training target signal data with an absolute noise level of 200, we can obtain the loss value of the second-level network. Subsequently, we can perform individual gradient descent and joint gradient descent based on the loss value of the second-level network.

[0124] In scenarios where the noise reduction calibration range is defined by relative noise magnitude, we will take a primary network with a noise reduction calibration range of 100% to 75% as an example. In this scenario, we can first calibrate a noise magnitude, which is called the calibration noise value. For example, we will use a calibration noise value of 500.

[0125] Since the upper limit of the noise reduction calibration range corresponding to the primary network is 100%, the product of the calibration noise value and this upper limit is 500, which can be used to construct training input signal data with a noise level of 500. Since the lower limit of the noise reduction calibration range corresponding to the primary network is 75%, the product of the calibration noise value and this lower limit is 375, which can be used to construct training target signal data with a noise level of 375.

[0126] In some scenarios, a preset noise-free signal can be obtained. For any level of network, the product of the upper limit of the noise reduction calibration range corresponding to that level of network and the calibration noise value can be obtained (denoted as the first multiplication result), and the product of the lower limit of the noise reduction calibration range corresponding to that level of network and the calibration noise value can be obtained (denoted as the second multiplication result). Adding noise of magnitude equal to the first multiplication result to the noise-free signal yields the training input signal data corresponding to that level of network. Adding noise of magnitude equal to the second multiplication result to the noise-free signal yields the training target signal data corresponding to that level of network.

[0127] Taking a primary network with a noise reduction calibration range of 100% to 75% and a calibration noise value of 500 as an example, the product of the upper limit of the primary network's noise reduction calibration range and the calibration noise value is 500. The product of the lower limit of the primary network's noise reduction calibration range and the calibration noise value is 375. After obtaining a noise-free signal, noise of magnitude 500 can be simulated and added to the noise-free signal to obtain training input signal data with a noise magnitude of 500. Alternatively, noise of magnitude 375 can be simulated and added to the noise-free signal to obtain training target signal data with a noise magnitude of 375.

[0128] In other scenarios, a preset noisy signal can be obtained and the noise level of the noisy signal can be determined; for any level of network, the relative magnitude relationship between the noise level of the noisy signal and the first multiplication result can be determined, and the relative magnitude relationship between the noise level of the noisy signal and the second multiplication result can be determined.

[0129] If the noise of the noisy signal is greater than the first multiplication result and the noise of the noisy signal is greater than the second multiplication result, then the noise level of the noisy signal can be reduced to the value corresponding to the first multiplication result through a preset denoising algorithm, thereby obtaining the training input signal data corresponding to this level of the network; and the noise level of the noisy signal can be reduced to the value corresponding to the second multiplication result through a denoising algorithm, thereby obtaining the training target signal data corresponding to this level of the network.

[0130] If the noise of the noisy signal is less than the first multiplication result and the noise of the noisy signal is greater than the second multiplication result, then noise can be added to the noisy signal so that the noise level of the noisy signal increases to the value corresponding to the first multiplication result, thereby obtaining the training input signal data corresponding to this level of the network; through a noise reduction algorithm, the noise level of the noisy signal is reduced to the value corresponding to the second multiplication result, thereby obtaining the training target signal data corresponding to this level of the network.

[0131] If the noise level of the noisy signal is less than the first multiplication result and the noise level of the noisy signal is less than the second multiplication result, then noise can be added to the noisy signal to increase its noise level to the value corresponding to the first multiplication result, thereby obtaining the training input signal data corresponding to this level of the network; similarly, noise can be added to the noisy signal to increase its noise level to the value corresponding to the second multiplication result, thereby obtaining the training target signal data corresponding to this level of the network.

[0132] If the noise of the noisy signal is equal to the result of the first multiplication, then the noisy signal can be directly used as the training input signal data for that level of the network.

[0133] If the noise of the noisy signal is equal to the result of the second multiplication, then the noisy signal can be directly used as the training target signal data for that level of the network.

[0134] The preset noise reduction algorithm may include, but is not limited to, low-pass filtering, iterative noise reduction, deep learning noise reduction, and other algorithms.

[0135] This section uses a primary network with a noise reduction calibration range of 100% to 75% as an example. Following the previously described method, we can obtain training input signal data with 100% relative noise and training target signal data with 75% relative noise for the primary network. Then, we can input the training input signal data with 100% relative noise into the primary network to obtain its output. Based on the difference in noise level between the primary network's output and the training target signal data with 75% relative noise, we can obtain the loss value of the primary network. Subsequently, we can perform individual gradient descent and joint gradient descent based on the loss value of the primary network.

[0136] Let's take a second-level network with a noise reduction calibration range of 75% to 50% as an example. Following the previously described method, we can obtain training input signal data with 75% relative noise and training target signal data with 50% relative noise for the second-level network. Then, we can input the training input signal data with 75% relative noise into the second-level network to obtain its output. Based on the difference in noise level between the second-level network's output and the training target signal data with 50% relative noise, we can obtain the loss value of the second-level network. Subsequently, we can perform individual gradient descent and joint gradient descent based on the loss value of the second-level network.

[0137] In one embodiment, before inputting the training input signal data corresponding to the current-level network into the current-level network, the training method of the signal denoising model provided in this application further includes:

[0138] Based on the training input signal data and training target signal data corresponding to the previous level network, the noise removed by the previous level network of the current level network is obtained; the noise removed by the previous level network of the current level network is extracted; and the extracted signal is added to the training input signal data corresponding to the current level network.

[0139] Taking the current second-level noise reduction training, where the noise reduction calibration range is divided by the absolute noise level, as an example, if the noise reduction calibration range of the second-level network is 300 to 200, then the training input signal data corresponding to the second-level network with an absolute noise level of 300, and the training target signal data corresponding to the second-level network with an absolute noise level of 200, can be obtained. Accordingly, the network preceding the second-level network is the first-level network.

[0140] The training input signal data and training target signal data corresponding to the primary network can be subtracted to obtain the difference signal data. The difference signal data is not significantly different from the noise actually removed by the primary network; therefore, it can be considered as the noise removed by the primary network. Next, a signal compensation algorithm can be used to extract the signal from the noise removed by the primary network, extracting the signals containing valid information from the noise. The signal extraction method can be:

[0141] Method 1: Multiply the noise removed by the first-level network by a set coefficient to obtain the signal containing valid information from the noise removed by the first-level network.

[0142] Method 2: Perform low-frequency domain extraction processing on the noise removed by the first-level network to obtain the signal that contains valid information from the noise removed by the first-level network.

[0143] Method 3 involves inputting the noise removed by the primary network into a signal extraction network using deep learning to obtain the signal containing valid information from the noise removed by the primary network.

[0144] In the training input signal data corresponding to the second-level network, which carries an absolute noise level of 300, a predetermined proportion of the signal removed by the first-level network is first added. Then, the training input signal data is input into the second-level network to obtain its output. Based on the difference in noise level between the second-level network's output and the corresponding training target signal data carrying an absolute noise level of 200, the loss value of the second-level network can be obtained. Subsequently, individual gradient descent and joint gradient descent can be performed based on the loss value of the second-level network.

[0145] In this embodiment, between the training of two adjacent network levels, the signal from the noise removed by the previous network is added to the training input signal data of the current network, thereby ensuring that the signal is not lost too much during the stepwise noise reduction training process, so as to achieve the purpose of noise reduction under better signal-to-noise ratio conditions.

[0146] In one embodiment, a multi-level network is either a network with the same network structure or a network with different network structures.

[0147] Multilevel networks can have the same network structure, or they can use different network structures based on requirements.

[0148] In one embodiment, when the multi-level network consists of networks with different network structures, the structural complexity of the preceding network in two adjacent networks is higher than that of the following network, and / or, the number of parameters of the preceding network in two adjacent networks is greater than that of the following network.

[0149] The introduction will take two adjacent network levels as the first-level network and the second-level network as examples.

[0150] The first-level network is located closer to the model input than the second-level network. The first-level network faces more noise and requires a higher signal-to-noise ratio (SNR) preservation capability. Therefore, the first-level network can have higher structural complexity and / or a larger number of parameters. Conversely, the second-level network faces less noise, has lower requirements for SNR preservation capability, and requires higher signal reconstruction capability. Therefore, the second-level network can have lower structural complexity and / or a smaller number of parameters.

[0151] In one embodiment, the signal data includes either CT data or CT images.

[0152] CT stands for Computed Tomography. In signals such as CT images, the structures within the CT images are considered valid information. Therefore, the signal extracted from the noise removed by the previous stage of the network can include the structural information from the CT images.

[0153] In one embodiment, when the signal data includes CT generation data, and the CT generation data is CT multi-level data, the training input CT multi-level data is either merged into the network or independently input into the network.

[0154] When the signal data includes CT images, and the CT images are multi-level CT images, the training input CT multi-level CT images are either merged into the network or input into the network independently.

[0155] When the signal data includes CT generation data, and the CT generation data is CT multi-level data, the training input signal data can be called training input CT multi-level data. The training input CT multi-level data can be merged into the network. The merging methods include, but are not limited to: (1) merging the training input CT multi-level data in the channel direction, and (2) merging the training input CT multi-level data in independent dimensions, such as merging in the z-direction. The training input CT multi-level data can also be independently input into the network.

[0156] The training input CT multi-level data is either merged into the network or input independently into the network, and then the loss value of the input CT multi-level data is trained together.

[0157] When the signal data includes CT images, and the CT images are CT multi-level images, the training input signal data can be called training input CT multi-level images. The training input CT multi-level images can be merged into the network. The merging methods include, but are not limited to: (1) merging the training input CT multi-level images along the channel direction, and (2) merging the training input CT multi-level images along independent dimensions, such as merging along the z-direction. The training input CT multi-level images can also be independently input into the network.

[0158] The training input CT multi-level images are either merged into the network or input independently into the network, and then used together to train the loss values ​​of the input CT multi-level images.

[0159] To better understand the training method of the above signal denoising model, an application example is described in detail below. In this application example, the signal denoising model includes a four-level network, and the denoising calibration range is divided according to the absolute noise level. Let the denoising calibration range corresponding to the first-level network be 400 to 300, the denoising calibration range corresponding to the second-level network be 300 to 200, the denoising calibration range corresponding to the third-level network be 200 to 100, and the denoising calibration range corresponding to the fourth-level network be 100 to 0.

[0160] Based on the noise reduction calibration range corresponding to each network level, the training signal data corresponding to each network level is obtained. Each training signal data includes training input signal data and training target signal data. The noise level carried by the training input signal data is the upper limit of the noise reduction calibration range corresponding to the corresponding network level, and the noise level carried by the training target signal data is the lower limit of the noise reduction calibration range corresponding to the corresponding network level.

[0161] The training input signal data for the first-level network is denoted as signal-input(400), and the training target signal data for the first-level network is denoted as signal-target(300). The training input signal data for the second-level network is denoted as signal-input(300), and the training target signal data for the second-level network is denoted as signal-target(200). The training input signal data for the third-level network is denoted as signal-input(200), and the training target signal data for the third-level network is denoted as signal-target(100). The training input signal data for the fourth-level network is denoted as signal-input(100), and the training target signal data for the fourth-level network is denoted as signal-target(0).

[0162] During the first-level noise reduction training, as shown in Figure 2(b), the training input signal data signal-input(400) corresponding to the first-level network can be input into the first-level network. The difference in noise magnitude between the output of the first-level network and the training target signal data signal-target(300) corresponding to the first-level network is calculated using the loss function, yielding the loss value loss-1 of the first-level network. Based on the loss value loss-1 of the first-level network, backpropagation and gradient descent are performed separately on the first-level network to complete the first-level noise reduction training.

[0163] After completing the first-level noise reduction training, the second-level noise reduction training is performed. As shown in Figure 2(c), the signal from the noise removed by the first-level network can be added to the training input signal data signal-input(300) of the second-level network, and then the training input signal data signal-input(300) is input into the second-level network. The difference in noise magnitude between the output of the second-level network and the training target signal data signal-target(200) of the second-level network is calculated using the loss function, and the loss value loss-2 of the second-level network is obtained. Based on the loss value loss-2 of the second-level network, backpropagation and gradient descent are performed on the second-level network separately. After backpropagation and gradient descent are performed separately, backpropagation and gradient descent are then performed jointly on the first-level network and the second-level network based on the loss value loss-2 of the second-level network to complete the second-level noise reduction training.

[0164] After completing the second-level noise reduction training, the third-level noise reduction training is performed. As shown in Figure 2(d), the signal removed by the second-level network can be added to the training input signal data signal-input(200) of the third-level network, and then the training input signal data signal-input(200) is input into the third-level network. The difference in noise magnitude between the output of the third-level network and the training target signal data signal-target(100) of the third-level network is calculated using the loss function, and the loss value loss-3 of the third-level network is obtained. Based on the loss value loss-3 of the third-level network, backpropagation and gradient descent are performed on the third-level network alone. After backpropagation and gradient descent are performed alone, backpropagation and gradient descent are then performed on the first-level network, the second-level network, and the third-level network together based on the loss value loss-3 of the third-level network to complete the third-level noise reduction training.

[0165] After completing the third-level noise reduction training, the fourth-level noise reduction training is performed. As shown in Figure 2(e), the signal removed by the third-level network can be added to the training input signal data signal-input(100) of the fourth-level network, and then the training input signal data signal-input(100) is input into the fourth-level network. The difference in noise magnitude between the output of the fourth-level network and the training target signal data signal-target(0) of the fourth-level network is calculated using the loss function, and the loss value loss-4 of the fourth-level network is obtained. Based on the loss value loss-4 of the fourth-level network, backpropagation and gradient descent are performed on the fourth-level network alone. After backpropagation and gradient descent are performed alone, backpropagation and gradient descent are then performed on the first-level network, the second-level network, the third-level network, and the fourth-level network together based on the loss value loss-4 of the fourth-level network to complete the fourth-level noise reduction training.

[0166] Completing the fourth level of noise reduction training can be considered as completing one iteration. Two or more iterations can be performed based on the actual situation. After all iterations are complete, the signal noise reduction model is obtained.

[0167] When the signal data is CT multi-level data, if there are two levels, the architecture diagram of the first-level noise reduction training is shown in Figure 3(a), and the architecture diagram of the second-level noise reduction training is shown in Figure 3(b). The architecture diagrams of the third-level noise reduction training and the fourth-level noise reduction training are not shown.

[0168] The signal denoising method provided in this application belongs to the inference stage. This signal denoising method can be executed by a computer device, and the signal denoising method may include... Figure 4 The steps are shown.

[0169] Step S401: Obtain the signal to be denoised.

[0170] If the signal during the training phase is CT raw data, then the CT raw data to be denoised can be obtained during the inference phase. If the signal during the training phase is a CT image, then the CT image to be denoised can be obtained during the inference phase.

[0171] Step S402: Input the signal to be denoised into the signal denoising model to obtain the denoised signal.

[0172] The signal denoising model was trained using the training method described in the above embodiments.

[0173] After obtaining the signal to be denoised, the signal can be input into the signal denoising model. Based on the output of the signal denoising model, the denoised signal can be obtained.

[0174] The signal denoising model used in this embodiment is obtained through a step-by-step denoising training method. During the training process, not only is individual gradient descent performed on single-level networks, but joint gradient descent is also performed on multi-level networks, which can improve the denoising capability of the signal denoising model. Furthermore, in the training signal data corresponding to the current level network, the signal removed by the previous level network is added to compensate for and restore the signal, avoiding the situation where the signal is weakened after each level network is denoised during training. This allows the signal denoising model to perform denoising while ensuring signal accuracy. As a result, the denoised signal has higher signal accuracy and retains more signal.

[0175] In one embodiment, the signal to be denoised is input into a signal denoising model to obtain the denoised signal, including:

[0176] Determine the noise level of the signal to be denoised; based on the noise level of the signal to be denoised, determine the target input network in the signal denoising model; ensure that the noise level of the signal to be denoised is within the denoising calibration range of the target input network; input the signal to be denoised into the target input network so that the forward multi-stage network in the signal denoising model, starting from the target input network, performs step-by-step denoising on the signal to be denoised, and obtain the denoised signal.

[0177] Forward (also known as positive direction) is the direction from the model input to the model output, and correspondingly, backward (also known as negative direction) is the direction from the model output to the model input.

[0178] The noise reduction calibration range is defined by the absolute noise level. The signal noise reduction model includes a four-level network. For example, the noise reduction calibration range corresponding to the first-level network is 400 to 300, the noise reduction calibration range corresponding to the second-level network is 300 to 200, the noise reduction calibration range corresponding to the third-level network is 200 to 100, and the noise reduction calibration range corresponding to the fourth-level network is 100 to 0.

[0179] If the noise level of the signal to be denoised is 250, which falls within the denoising calibration range of 300 to 200, then the second-stage network can be used as the target input network. The feedforward multi-stage network starting with the second-stage network includes the second-stage network, the third-stage network, and the fourth-stage network. For example... Figure 5 As shown, after the signal to be denoised is input into the second-level network, it undergoes progressive denoising processing through the second-level, third-level, and fourth-level networks. The denoised signal is obtained based on the output of the fourth-level network. The noise level of the signal to be denoised can be estimated using a noise extraction algorithm.

[0180] The noise reduction calibration range is defined by the relative noise level. The signal noise reduction model includes a four-level network. For example, the noise reduction calibration range corresponding to the first-level network is 100% to 75%, the noise reduction calibration range corresponding to the second-level network is 75% to 50%, the noise reduction calibration range corresponding to the third-level network is 50% to 25%, and the noise reduction calibration range corresponding to the fourth-level network is 25% to 0.

[0181] If the calibrated noise value is 500, and the noise level of the signal to be denoised is 300, the relative noise level of the signal to be denoised compared to the calibrated noise value is 60%. 60% falls within the denoising calibration range of 75% to 50%. Therefore, the second-level network can be used as the target input network. The feedforward multi-stage network starting with the second-level network includes the second-level network, the third-level network, and the fourth-level network. For example... Figure 5 As shown, after the signal to be denoised is input into the second-level network, it undergoes progressive denoising processing through the second-level, third-level, and fourth-level networks. The denoised signal is obtained based on the output of the fourth-level network. The noise level of the signal to be denoised can be estimated using a noise extraction algorithm.

[0182] In this embodiment, during inference, the input from which network level can be dynamically determined based on the noise level of the signal to be denoised, thereby obtaining results faster and improving denoising efficiency.

[0183] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0184] Based on the same inventive concept, this application also provides a training apparatus for a signal denoising model to implement the training method for the signal denoising model described above. The solution provided by this apparatus is similar to the implementation described in the above method. Therefore, the specific limitations of one or more training apparatus embodiments for signal denoising models provided below can be found in the limitations of the training method for the signal denoising model described above, and will not be repeated here.

[0185] This application provides a training device for a signal denoising model, comprising:

[0186] The model acquisition module is used to perform step-by-step noise reduction training on the multi-level network based on the training signal data corresponding to the multi-level network, and obtain the signal noise reduction model.

[0187] The training module is used to obtain the loss value of the current level network based on the training signal data corresponding to the current level network if the current level trained by stepwise noise reduction is not the first level; and to perform gradient descent on the current level network alone based on the loss value, and then to perform gradient descent on the current multi-level networks that have been individually gradient descent based on the loss value.

[0188] In one embodiment, the training signal data corresponding to the current level network is supplemented with the signal from the noise that was removed by the previous level network of the current level network.

[0189] In one embodiment, the training module is further configured to:

[0190] Based on the loss value, the current level network is back-calculated layer by layer to obtain the first type of gradient value; based on the first type of gradient value, gradient descent is performed on the current level network.

[0191] In one embodiment, the training module is further configured to:

[0192] Based on the loss value, the current multi-level individually gradient descent network is back-calculated network by network to obtain the second type of gradient value; based on the second type of gradient value, the current multi-level individually gradient descent network is subjected to gradient descent.

[0193] In one embodiment, the training device for the signal denoising model further includes a sample acquisition module, used for:

[0194] Based on the noise reduction calibration range corresponding to each level of the network, the training signal data corresponding to each level of the network is obtained; each training signal data includes training input signal data and training target signal data, wherein the noise level carried by the training input signal data is the upper limit of the noise reduction calibration range corresponding to the corresponding level of the network, and the noise level carried by the training target signal data is the lower limit of the noise reduction calibration range corresponding to the corresponding level of the network.

[0195] The training module is also used for:

[0196] The training input signal data corresponding to the current level network is input into the current level network; the loss value of the current level network is obtained based on the difference in noise level between the output result of the current level network and the training target signal data corresponding to the current level network.

[0197] In one embodiment, the training module is further configured to:

[0198] Based on the training input signal data and training target signal data corresponding to the previous level network, the noise removed by the previous level network of the current level network is obtained; the noise removed by the previous level network of the current level network is extracted; and the extracted signal is added to the training input signal data corresponding to the current level network.

[0199] In one embodiment, a multi-level network is either a network with the same network structure or a network with different network structures.

[0200] In one embodiment, when the multi-level network consists of networks with different network structures, the structural complexity of the preceding network in two adjacent networks is higher than that of the following network, and / or, the number of parameters of the preceding network in two adjacent networks is greater than that of the following network.

[0201] In one embodiment, the signal data includes either CT data or CT images.

[0202] In one embodiment, when the signal data includes CT raw data, and the CT raw data is CT multi-level data, the training input CT multi-level data is either merged into the network or input independently into the network; when the signal data includes CT images, and the CT images are CT multi-level images, the training input CT multi-level images are either merged into the network or input independently into the network.

[0203] Each module in the training device for the aforementioned signal denoising model can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0204] Based on the same inventive concept, this application also provides a signal denoising apparatus for implementing the signal denoising method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more signal denoising apparatus embodiments provided below can be found in the limitations of the signal denoising method described above, and will not be repeated here.

[0205] This application provides a signal noise reduction device, comprising:

[0206] The signal acquisition module is used to acquire the signal to be denoised;

[0207] The noise reduction module is used to input the signal to be denoised into the signal denoising model to obtain the denoised signal; the signal denoising model is trained according to the training method of the signal denoising model described in the above embodiments.

[0208] In one embodiment, the noise reduction module is further configured to:

[0209] The noise level of the signal to be denoised is determined; based on the noise level of the signal to be denoised, a target input network is determined in the signal denoising model; the noise level of the signal to be denoised is within the denoising calibration range of the target input network; the signal to be denoised is input into the target input network, so that the forward multi-stage network in the signal denoising model, starting from the target input network, performs step-by-step denoising processing on the signal to be denoised to obtain the denoised signal.

[0210] Each module in the aforementioned signal noise reduction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0211] In one exemplary embodiment, a computer device is provided, the internal structure of which can be as shown in the figure. Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores data related to the methods described above. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements one of the methods described in the above embodiments.

[0212] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0213] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the various method embodiments described above.

[0214] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the various method embodiments described above.

[0215] In one embodiment, a computer program product is provided having a computer program stored thereon, the computer program being executed by a processor of the steps described in the various method embodiments above.

[0216] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0217] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0218] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0219] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A training method for a signal denoising model, characterized in that, The method includes: Based on the training signal data corresponding to the multi-level network, the multi-level network is trained to perform noise reduction step by step to obtain the signal noise reduction model. If the current level trained by progressive noise reduction is not the first level, the loss value of the current level network is obtained based on the training signal data corresponding to the current level network; the training input signal data in the training signal data corresponding to the current level network is supplemented with the signal from the noise removed by the previous level network; the noise removed by the previous level network is obtained based on the training input signal data and the training target signal data corresponding to the previous level network. After performing gradient descent on the current level network individually based on the loss value, gradient descent is then jointly performed on the current multi-level networks that have already undergone individual gradient descent based on the loss value; wherein, the signal data includes either CT raw data or CT images.

2. The method according to claim 1, characterized in that, The training signal data includes the training target signal data.

3. The method according to claim 1, characterized in that, Based on the loss value, gradient descent is performed separately on the current level network, including: Based on the loss value, the current level network is back-calculated layer by layer to obtain the first type of gradient value; Gradient descent is performed on the current level network based on the first type of gradient value.

4. The method according to claim 1, characterized in that, Based on the aforementioned loss value, perform joint gradient descent on the current multi-level individually gradient descent networks, including: Based on the loss value, the current multi-level individually gradient descent network is back-calculated network by network to obtain the second type of gradient value; Based on the second type of gradient value, gradient descent is performed on the current multi-level individual gradient descent network.

5. The method according to claim 1, characterized in that, The method further includes: Based on the noise reduction calibration range corresponding to each network level, the training signal data corresponding to each network level is obtained; the noise level carried by the training input signal data included in the training signal data corresponding to each network level is the upper limit of the noise reduction calibration range corresponding to the corresponding network level; the noise level carried by the training target signal data included in the training signal data corresponding to each network level is the lower limit of the noise reduction calibration range corresponding to the corresponding network level. Based on the training signal data corresponding to the current level network, the loss value of the current level network is obtained, including: Input the training input signal data corresponding to the current level network into the current level network; The loss value of the current-level network is obtained based on the difference in noise level between the output of the current-level network and the corresponding training target signal data.

6. The method according to claim 5, characterized in that, In a multi-level network, the closer the network is to the model input, the larger the noise reduction calibration range, and the closer the network is to the model output, the smaller the noise reduction calibration range.

7. The method according to claim 5, characterized in that, The noise reduction calibration ranges corresponding to multi-level networks may have the same or different spans.

8. A signal noise reduction method, characterized in that, The method includes: Acquire the signal to be denoised; The signal to be denoised is input into the signal denoising model to obtain the denoised signal; the signal denoising model is trained by the training method of the signal denoising model according to any one of claims 1 to 7.

9. The method according to claim 8, characterized in that, The signal to be denoised is input into a signal denoising model to obtain the denoised signal, including: Determine the noise level of the signal to be denoised; Based on the noise level of the signal to be denoised, a target input network is determined in the signal denoising model; the noise level of the signal to be denoised is within the denoising calibration range of the target input network. The signal to be denoised is input into the target input network, so that the forward multi-stage network in the signal denoising model, starting from the target input network, performs step-by-step denoising on the signal to be denoised, and obtains the denoised signal.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 9.

Images