High-precision signal intelligent detection and identification method and device based on neural network

Abstract

The invention discloses a high-precision signal intelligent detection and identification method and device based on a neural network, and relates to the technical field of computers. The method comprises the following steps: loading a trained signal detection network, wherein the signal detection network is used for positioning and identifying a signal in a signal time-frequency image; wherein the signal detection network comprises a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network comprises a four-scale down-sampling detection head, the lightweight backbone network comprises a union attention mechanism module, and the union attention mechanism comprises a channel attention sub-module and a space attention sub-module; acquiring a to-be-detected signal, and performing time-frequency transformation on the to-be-detected signal to obtain a time-frequency image of the to-be-detected signal; and positioning and identifying the time-frequency image of the to-be-detected signal based on the signal detection network to obtain signal detection data of the to-be-detected signal. According to the invention, the accuracy of signal detection can be considered while the efficiency is ensured.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a high-precision signal intelligent detection and recognition method and device based on a neural network. Background Art

[0002] As communication technology continues to expand in depth and breadth in the electromagnetic spectrum, signal complexity continues to increase, and the electromagnetic environment is complex and changeable. Intelligent modulation recognition technology based on deep learning, combined with other signal blind processing technologies, can be used in scenarios such as confirmation of unknown signals, identification of interference signals, and spectrum management and detection, thereby ensuring the normal communication of legitimate users, detecting illegal users and eliminating their interference effects.

[0003] At present, signal detection methods based on deep learning are divided into two-stage detection and one-stage detection. The main representative of the two-stage target detection algorithm is the R-CNN algorithm. Compared with traditional target detection algorithms, R-CNN has greatly improved its accuracy, but it still has the following disadvantages: the generated candidate areas will overlap, and there will be a lot of repeated calculations when performing convolution operations on them, resulting in long model training time, high space cost, and slow detection speed. The one-stage algorithm is more efficient than the two-stage algorithm and can meet real-time requirements, but its accuracy is not as good as the two-stage algorithm. Summary of the invention

[0004] The present invention provides a high-precision signal intelligent detection and recognition method and device based on a neural network to solve the problem that the existing signal detection method based on deep learning cannot take into account both efficiency and accuracy.

[0005] According to one aspect of the present invention, a high-precision signal intelligent detection and recognition method based on a neural network is provided, comprising:

[0006] Loading a trained signal detection network, the signal detection network is used to locate and identify the signal in the signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network includes a four-scale downsampling detection head, the lightweight backbone network includes a union attention mechanism module, and the union attention mechanism includes a channel attention submodule and a spatial attention submodule;

[0007] Acquire a signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain a time-frequency image of the signal to be detected;

[0008] The time-frequency image of the signal to be detected is located and identified based on the signal detection network to obtain signal detection data of the signal to be detected.

[0009] According to another aspect of the present invention, there is provided a high-precision signal intelligent detection and recognition device based on a neural network, comprising:

[0010] A signal detection network loading module is used to load a trained signal detection network, wherein the signal detection network is used to locate and identify signals in a signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network, and a prediction network, wherein the lightweight backbone network includes a four-scale downsampling detection head, and the lightweight backbone network includes a union attention mechanism module, wherein the union attention mechanism includes a channel attention submodule and a spatial attention submodule;

[0011] The signal to be detected conversion module is used to obtain the signal to be detected, and perform time-frequency conversion on the signal to be detected to obtain a time-frequency image of the signal to be detected;

[0012] A positioning and identification module is used to locate and identify the time-frequency image of the signal to be detected based on the signal detection network to obtain signal detection data of the signal to be detected.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the high-precision signal intelligent detection and recognition method based on neural network described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the high-precision signal intelligent detection and recognition method based on a neural network as described in any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention is to load the trained signal detection network, and the signal detection network is used to locate and identify the signal in the signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network includes a four-scale downsampling detection head, the lightweight backbone network includes a union attention mechanism module, and the union attention mechanism includes a channel attention submodule and a spatial attention submodule; obtain the signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain the time-frequency image of the signal to be detected; locate and identify the time-frequency image of the signal to be detected based on the signal detection network, and obtain the signal detection data of the signal to be detected. The signal detection network is constructed using a lightweight backbone network to reduce the dimension of the backbone network, and the lightweight backbone network uses a four-scale downsampling detection head, so that the network can learn better multi-level feature information of the learning target, increase the network's detection ability of detail features, and perform fusion detection at four scales, while ensuring efficiency and taking into account the accuracy of signal detection.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 This is a flowchart of a high-precision signal intelligent detection and recognition method based on a neural network provided in the first embodiment of the present invention;

[0022] Figure 2 is a schematic diagram of the structure of a signal detection network provided in Embodiment 1 of the present invention;

[0023] Figure 3 is a structural diagram of the union attention mechanism module provided in the first embodiment of the present invention;

[0024] Figure 4 is a schematic diagram of a first sub-region and a second sub-region obtained by dividing the filter window up and down according to the first embodiment of the present invention;

[0025] Figure 5 This is a flow chart of a high-precision signal intelligent detection and recognition method based on a neural network provided in Embodiment 2 of the present invention;

[0026] Figure 6 It is a structural schematic diagram of a high-precision signal intelligent detection and identification device based on a neural network provided in Embodiment 3 of the present invention;

[0027] Figure 7 It is a structural schematic diagram of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first sub-region", "second sub-region", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Embodiment 1

[0031] Figure 1 This is a flowchart of a high-precision signal intelligent detection and identification method based on a neural network provided in the first embodiment of the present invention. This embodiment can be applied to the identification and positioning of signals in scenarios such as confirmation of unknown signals, identification of interference signals, and spectrum management and detection. The method can be executed by a high-precision signal intelligent detection and identification device based on a neural network. The high-precision signal intelligent detection and identification device based on a neural network can be implemented in the form of hardware and / or software. The high-precision signal intelligent detection and identification device based on a neural network can be configured in electronic devices such as computers and servers. Figure 1 As shown, the method includes:

[0032] S110. Loading a trained signal detection network, wherein the signal detection network is used to locate and identify signals in a signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network, and a prediction network, wherein the lightweight backbone network includes a four-scale downsampling detection head, and wherein the lightweight backbone network includes a union attention mechanism module, wherein the union attention mechanism includes a channel attention submodule and a spatial attention submodule.

[0033] In an embodiment of the present invention, Figure 2 FIG. 1 is a schematic diagram of a signal detection network structure provided by Embodiment 1 of the present invention. Figure 2 As shown, the backbone network is used to extract image feature data of the signal time-frequency image. The first convolutional layer performs a 2-fold downsampling operation on the input image. Downsampling is performed five times in the backbone network, and finally the 32-fold downsampled feature map is input into the feature fusion network. The feature fusion network is used to fuse image features. Specifically, the feature fusion network includes two feature fusion paths: upsampling fusion and downsampling fusion. The prediction network is used to output prediction results. Specifically, the prediction network includes OUT1, OUT2, OUT3 and OUT4 prediction layers, and each prediction layer outputs the classification and positioning results of the corresponding scale target respectively. Among them, the lightweight backbone network can be a MobilenetV3 network. When sampling on the MobilenetV3 network, feature extraction is performed on multiple anchor points on different feature maps, and the bottom and high-level links in the network increase the fine-grained features of the prediction layer. Exemplarily, the structure of the lightweight backbone network in the present invention is shown in Table 1 below:

[0034] Table 1

[0035]

[0036] It should be noted that the lightweight backbone network includes a union attention mechanism module, which is used to filter the information in the task to find key information related to the target. Figure 3 Schematic diagram of the structure of the union attention mechanism module provided in the first embodiment of the present invention. Figure 3 As shown, the union attention mechanism module includes a channel attention submodule and a spatial attention submodule. The channel attention submodule is connected in parallel with the spatial attention submodule, and the input feature map F∈R of the lightweight backbone network is input. C×H×W At the same time, it passes through the channel attention submodule and the spatial attention submodule to obtain the channel attention vector Mc∈R C×1×1 and the spatial attention vector M s ∈R 1×H×W , for the channel attention vector Mc∈R C×1×1 , spatial attention vector M s ∈R 1×H×WAnd the input feature map F∈R C×H×W Perform product operation to get the output feature map

[0037] Among them, the channel attention submodule uses global maximum pooling and global average pooling to compress the feature map in the spatial direction to obtain two C×1×1 feature maps. Then the two feature maps are shared and input into a two-layer fully connected network, where the first layer reduces the dimension of the features, and the second layer fuses the two channel attention vectors, and inputs the activation function to obtain the channel attention vector Mc. The spatial attention submodule performs global maximum pooling and global average pooling on the features along the channel direction to obtain a 1×H×W feature map, which is concatenated to obtain 2×H×W features, and then a convolution operation is performed to obtain 1×H×W features, and the spatial attention vector Ms is obtained after the activation function.

[0038] Based on the above embodiments, optionally, the training method of the signal detection network includes: obtaining a signal time-frequency image set; training the lightweight backbone network based on the signal time-frequency image; determining an initial signal detection network based on the trained lightweight backbone network and target detection network; obtaining a signal time-frequency standard data set, the signal time-frequency standard data set including a signal time-frequency image and annotation information corresponding to the signal time-frequency image; training the initial signal detection network based on the signal time-frequency standard data set to obtain a trained signal detection network.

[0039] Among them, the signal time-frequency image set is composed of the time-frequency image of the modulated signal. Specifically, the signal time-frequency image set includes the signal time-frequency image and the signal system corresponding to the signal time-frequency image. The signal system includes but is not limited to binary frequency shift keying (Frequency Shift Keying, 2FSK), Gaussian frequency shift keying (Gaussian Frequency-Shift Keying, GFSK), binary phase shift keying (Binary Phase Shift Keying, BPSK), quadrature phase shift keying (QuadraturePhaseShift Keying, QPSK), frequency modulation (Frequency Modulation, FM) and other modulation signals. In an embodiment of the present invention, the lightweight backbone network is pre-trained, and the pre-trained lightweight backbone network is used to replace the backbone network in the target detection network to obtain an initial signal detection network. Further, a signal time-frequency standard data set is obtained, and the signal time-frequency standard data set includes the signal time-frequency image and the annotation information corresponding to the signal time-frequency image, and the annotation information includes the signal system and bounding box corresponding to the signal time-frequency image. The initial signal detection network is trained based on the signal time-frequency standard data set to obtain a trained signal detection network.

[0040] Based on the above embodiment, optionally, the loss function of the signal detection network includes a bounding box regression loss function, a category loss function and a loss function for whether a target exists at an anchor point, and the loss function for whether a target exists at an anchor point includes a loss function for whether a target exists at the anchor point and a loss function for whether a target does not exist at the anchor point; wherein, the bounding box regression loss function is used to measure the difference between the predicted bounding box and the true bounding box, the category loss function is used to measure the difference between the predicted category and the true category of the signal, and the loss function for whether a target exists at the anchor point is used to perform a binary classification prediction for the anchor point position to determine whether there is a target signal at the anchor point.

[0041] Specifically, the loss function of the signal detection network is:

[0042] loss = λ coord lbox+λ class lcls+λ noobj lnoobj+λ obj lobj;

[0043] Among them, coord is the proportion of the bounding box regression loss function, lbox represents the bounding box regression loss function, λ class is the proportion of the category loss function, lcls represents the category loss function, λ noobj is the proportion of the loss function when there is no target at the anchor point, lnoobj represents the loss function when there is no target at the anchor point, and λ obj is the proportion of the loss function of the target at the anchor point, and lobj represents the loss function of the target at the anchor point.

[0044] The bounding box regression loss function is as follows:

[0045]

[0046] Among them, S 2 Indicates the output feature map size. When the network input size is a 416×416 feature map, the output feature map sizes of the four prediction layers OUT1, OUT2, OUT3, and OUT4 are 13×13, 26×26, 54×54, and 104×104, respectively; B indicates the number of scale boxes of the bounding box; Indicates that the bounding box at anchor point i, j has an object, and its value is 1, otherwise it is 0; w i and h i Respectively represent the width and height of the target, x i and i Respectively represent the target center point coordinates, and Represent the predicted coordinates of the target center point, and Represent the width and height of the target prediction respectively.

[0047] The category loss function is as follows:

[0048]

[0049] Among them, p i (c) is 0 or 1, It is the probability of category prediction, and its value is between 0 and 1.

[0050] The loss function for whether there is a target at the anchor point is as follows:

[0051] The loss function for the target at the anchor point is as follows:

[0052]

[0053] The loss function when there is no target at the anchor point is as follows:

[0054]

[0055] in, Indicates that there is no target in the bounding box at anchor point i, j, and its value is 1, otherwise it is 0; c i Whether there is a target at the anchor point, the value is 0 or 1, It is the predicted probability of whether there is a target at the anchor point, and its value is between 0 and 1.

[0056] Based on the above embodiment, optionally, the acquiring of the signal time-frequency image set includes: acquiring a signal training set, the signal training set including signal data and a signal system corresponding to the signal data; performing a time-frequency transform on the signal data to obtain a signal time-frequency image; and generating the signal time-frequency image set based on the signal time-frequency image and the signal system corresponding to the signal time-frequency image.

[0057] Among them, the signal training set is used to train a lightweight backbone network, such as a MobilenetV3 network. Specifically, the signal training set includes signal data and a signal system corresponding to the signal data, which is used to train the lightweight backbone network's signal recognition ability. In an embodiment of the present invention, a signal training set is obtained, and a signal time-frequency image is obtained by performing a time-frequency transformation on the signal data; a signal time-frequency image set is generated based on the signal time-frequency image and the signal system corresponding to the signal time-frequency image. It can be understood that since the lightweight backbone network cannot directly recognize the signal data, it is necessary to perform a time-frequency transformation on the signal data and convert the signal data into a signal time-frequency image.

[0058] For example, take STFT time-frequency transform as an example:

[0059] The characteristics that signal data of different modes can reflect are often quite different. Although the time domain waveform can reflect certain signal characteristics, the time-frequency distribution diagram shows the changes of the signal's frequency distribution over time more intuitively, and has the advantages of both time and frequency domain representation. The time-frequency diagram expression of the signal obtained by performing a short-time Fourier transform (STFT) on the training signal is as follows:

[0060]

[0061] Where N is the signal sequence length, h is the sampling function, y is the original signal data, k and τ are the time domain scale and frequency domain scale respectively.

[0062] In some embodiments, optionally, after obtaining the signal time-frequency image, the method further includes: for each pixel point in the signal time-frequency image, determining a filter window based on the pixel point, and dividing the filter window into a first sub-region and a second sub-region; for any sub-region, determining the regional variance of the sub-region based on the pixel value at each coordinate of the sub-region, the number of pixel columns, the number of pixel rows and the mean of all pixels in the sub-region; determining a target region in the first sub-region and the second sub-region based on the regional variance, and taking the grayscale median in the target region as the grayscale value of the pixel point.

[0063] The noise existing in the time domain of the signal data is manifested as Gaussian noise with pixel grayscale fluctuation transformation in the signal time-frequency image. In an embodiment of the present invention, for each pixel point in the signal time-frequency image, a rectangular area with a preset range around the pixel point is selected as a filter window, and the filter window is divided into a first sub-area and a second sub-area. The first sub-area and the second sub-area can be obtained by dividing the filter window up and down, or by dividing the filter window left and right. There is no limitation here. For example, the up and down division is taken as an example. Figure 4 It is a schematic diagram of a first sub-region and a second sub-region obtained by dividing the filter window up and down provided in Embodiment 1 of the present invention.

[0064] Further, for any sub-region, the regional variance of the sub-region is determined based on the pixel value at each coordinate of the sub-region, the number of pixel columns, the number of pixel rows and the mean value of all pixels in the sub-region; the regional variance of the first sub-region is compared with the regional variance of the second sub-region, the sub-region with the smaller variance is taken as the target sub-region, the grayscale values ​​of all pixels in the target sub-region are arranged in ascending or descending order, and the middle value is taken as the grayscale value of the pixel. The embodiment of the present invention traverses all pixels in the signal time-frequency image and performs the above operation on each pixel to achieve noise reduction processing on the signal time-frequency image.

[0065] Specifically, the regional variance calculation formula is as follows:

[0066]

[0067] Among them, θ k (k∈{1,2}) represents two regions, f(i,j) is the pixel value at coordinate (i,j), M is the number of pixel columns, N is the number of pixel rows, MN is the number of pixels in the current region, and m k is the mean value of all pixels in the direction region.

[0068] S120: Acquire a signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain a time-frequency image of the signal to be detected.

[0069] Wherein, the signal to be detected refers to the signal data to be detected. In an embodiment of the present invention, the signal to be detected is obtained, and the signal to be detected is subjected to time-frequency transformation to obtain a time-frequency image of the signal to be detected. Wherein, the time-frequency image of the signal to be detected is the time-frequency image after the time-frequency transformation of the signal to be detected. Exemplarily, taking STFT time-frequency transformation as an example, the time-frequency diagram expression of the signal obtained by performing short-time Fourier transform (STFT) on the signal to be detected is as follows:

[0070]

[0071] Where N is the signal sequence length, h is the sampling function, y is the original signal data, k and τ are the time domain scale and frequency domain scale respectively.

[0072] S130: Positioning and identifying the time-frequency image of the signal to be detected based on the signal detection network to obtain signal detection data of the signal to be detected.

[0073] In an embodiment of the present invention, the time-frequency image of the signal to be detected is input into the signal detection network for positioning and identification, and the signal detection data of the signal to be detected is obtained. The signal detection data includes the position and signal system of the signal to be detected, and the position of the signal to be detected is displayed in the form of a boundary box.

[0074] The technical solution of this embodiment is to load the trained signal detection network, and the signal detection network is used to locate and identify the signal in the signal time-frequency image; wherein, the signal detection network includes a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network includes a four-scale downsampling detection head, the lightweight backbone network includes a union attention mechanism module, and the union attention mechanism includes a channel attention submodule and a spatial attention submodule; obtain the signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain the time-frequency image of the signal to be detected; locate and identify the time-frequency image of the signal to be detected based on the signal detection network, and obtain the signal detection data of the signal to be detected. The signal detection network is constructed using a lightweight backbone network to reduce the dimension of the backbone network, and the lightweight backbone network uses a four-scale downsampling detection head, so that the network can learn better multi-level feature information of the learning target, increase the network's detection ability of detail features, and perform fusion detection at four scales, while ensuring efficiency and taking into account the accuracy of signal detection.

[0075] Embodiment 2

[0076] Figure 5 It is a flowchart of a high-precision signal intelligent detection and identification method based on a neural network provided in the second embodiment of the present invention. Based on the above embodiments, this embodiment adds "for each pixel point in the signal time-frequency image to be detected or the signal time-frequency image, determine the filter window based on the pixel point, and divide the filter window into a first sub-region and a second sub-region; for any sub-region, determine the regional variance of the sub-region based on the pixel value at each coordinate of the sub-region, the number of pixel columns, the number of pixel rows and the mean value of all pixels in the sub-region; determine the target region in the first sub-region and the second sub-region based on the regional variance, and use the grayscale median in the target region as the grayscale value of the pixel point" after obtaining the signal time-frequency image to be detected or the signal time-frequency image. The explanation of the terms that are the same or corresponding to the above embodiments will not be repeated here.

[0077] like Figure 5 As shown, the method includes:

[0078] S210. Loading a trained signal detection network, wherein the signal detection network is used to locate and identify signals in a signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network, and a prediction network, wherein the lightweight backbone network includes a four-scale downsampling detection head, and wherein the lightweight backbone network includes a union attention mechanism module, wherein the union attention mechanism includes a channel attention submodule and a spatial attention submodule.

[0079] S220 , obtaining a signal to be detected, and performing time-frequency transformation on the signal to be detected to obtain a time-frequency image of the signal to be detected.

[0080] S230. For each pixel point in the time-frequency image of the signal to be detected, determine a filter window based on the pixel point, and divide the filter window into a first sub-region and a second sub-region; for any sub-region, determine the regional variance of the sub-region based on the pixel value at each coordinate of the sub-region, the number of pixel columns, the number of pixel rows, and the mean value of all pixels in the sub-region; determine a target region in the first sub-region and the second sub-region based on the regional variance, and use the grayscale median in the target region as the grayscale value of the pixel point.

[0081] The noise existing in the time domain of the signal data is manifested as Gaussian noise with pixel grayscale fluctuation transformation in the signal time-frequency image. In an embodiment of the present invention, for each pixel point in the time-frequency image of the signal to be detected, a rectangular area of ​​a preset range is selected around the pixel point as a filter window, and the filter window is divided into a first sub-area and a second sub-area. The first sub-area and the second sub-area can be obtained by dividing the filter window up and down, or by dividing the filter window left and right. There is no limitation here. For example, taking the up and down division as an example, Figure 4 It is a schematic diagram of a first sub-region and a second sub-region obtained by dividing the filter window up and down provided in Embodiment 1 of the present invention.

[0082] Further, for any sub-region, the regional variance of the sub-region is determined based on the pixel value at each coordinate of the sub-region, the number of pixel columns, the number of pixel rows and the mean value of all pixels in the sub-region; the regional variance of the first sub-region is compared with the regional variance of the second sub-region, the sub-region with the smaller variance is taken as the target sub-region, the grayscale values ​​of all pixels in the target sub-region are arranged in ascending or descending order, and the middle value is taken as the grayscale value of the pixel. The embodiment of the present invention traverses all pixels in the signal time-frequency image and performs the above operation on each pixel to achieve noise reduction processing on the signal time-frequency image to be detected.

[0083] Specifically, the regional variance calculation formula is as follows:

[0084]

[0085] Among them, θ k (k∈{1,2}) represents two regions, f(i,j) is the pixel value at coordinate (i,j), M is the number of pixel columns, N is the number of pixel rows, MN is the number of pixels in the current region, and mk is the mean of all pixels in the direction region.

[0086] S240: Positioning and identifying the time-frequency image of the signal to be detected based on the signal detection network to obtain signal detection data of the signal to be detected.

[0087] The technical solution of this embodiment traverses all pixel points in the time-frequency image of the signal to be detected, processes each pixel point, and realizes noise reduction processing on the signal time-frequency image, thereby improving the accuracy of signal detection.

[0088] Embodiment 3

[0089] Figure 6 Schematic diagram of the structure of a high-precision signal intelligent detection and recognition device based on a neural network provided by the third embodiment of the present invention. Figure 6 As shown, the device comprises:

[0090] A signal detection network loading module 310 is used to load a trained signal detection network, wherein the signal detection network is used to locate and identify signals in a signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network, and a prediction network, wherein the lightweight backbone network includes a four-scale downsampling detection head, and the lightweight backbone network includes a union attention mechanism module, wherein the union attention mechanism includes a channel attention submodule and a spatial attention submodule;

[0091] The signal-to-be-detected conversion module 320 is used to obtain the signal to be detected, and perform time-frequency conversion on the signal to be detected to obtain a time-frequency image of the signal to be detected;

[0092] The positioning and identification module 330 is used to locate and identify the time-frequency image of the signal to be detected based on the signal detection network to obtain signal detection data of the signal to be detected.

[0093] The technical solution of this embodiment is to load the trained signal detection network, and the signal detection network is used to locate and identify the signal in the signal time-frequency image; wherein, the signal detection network includes a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network includes a four-scale downsampling detection head, the lightweight backbone network includes a union attention mechanism module, and the union attention mechanism includes a channel attention submodule and a spatial attention submodule; obtain the signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain the time-frequency image of the signal to be detected; locate and identify the time-frequency image of the signal to be detected based on the signal detection network, and obtain the signal detection data of the signal to be detected. The signal detection network is constructed using a lightweight backbone network to reduce the dimension of the backbone network, and the lightweight backbone network uses a four-scale downsampling detection head, so that the network can learn better multi-level feature information of the learning target, increase the network's detection ability of detail features, and perform fusion detection at four scales, while ensuring efficiency and taking into account the accuracy of signal detection.

[0094] Based on the above embodiment, optionally, the device further includes a training module of a signal detection network, which is used to:

[0095] Obtain signal time-frequency image set;

[0096] Training the lightweight backbone network based on the signal time-frequency image;

[0097] Determine an initial signal detection network based on the trained lightweight backbone network and target detection network;

[0098] Acquire a signal time-frequency standard data set, wherein the signal time-frequency standard data set includes a signal time-frequency image and annotation information corresponding to the signal time-frequency image;

[0099] The initial signal detection network is trained based on the signal time-frequency standard data set to obtain a trained signal detection network.

[0100] Based on the above embodiment, optionally, the training module of the signal detection network further includes a signal time-frequency image set acquisition module, which is used to:

[0101] Acquire a signal training set, wherein the signal training set includes signal data and a signal system corresponding to the signal data;

[0102] Performing time-frequency transformation on the signal data to obtain a signal time-frequency image;

[0103] The signal time-frequency image set is generated based on the signal time-frequency image and a signal system corresponding to the signal time-frequency image.

[0104] Based on the above embodiment, optionally, the loss function of the signal detection network includes a bounding box regression loss function, a category loss function and a loss function for whether a target exists at an anchor point, and the loss function for whether a target exists at an anchor point includes a loss function for whether a target exists at the anchor point and a loss function for whether a target does not exist at the anchor point; wherein, the bounding box regression loss function is used to measure the difference between the predicted bounding box and the true bounding box, the category loss function is used to measure the difference between the predicted category and the true category of the signal, and the loss function for whether a target exists at the anchor point is used to perform a binary classification prediction for the anchor point position to determine whether there is a target signal at the anchor point.

[0105] Based on the above embodiment, optionally, the channel attention submodule is connected in parallel with the spatial attention submodule.

[0106] Based on the above embodiment, optionally, the device further includes a noise reduction module, which is used to:

[0107] For each pixel point in the time-frequency image of the signal to be detected or the time-frequency image of the signal, determine a filter window based on the pixel point, and divide the filter window into a first sub-region and a second sub-region;

[0108] For any sub-region, determining the regional variance of the sub-region based on the pixel value, the number of pixel columns, the number of pixel rows and the mean value of all pixels in the sub-region at each coordinate of the sub-region;

[0109] A target region is determined in the first sub-region and the second sub-region based on the regional variance, and a grayscale median value in the target region is used as a grayscale value of the pixel point.

[0110] Based on the above embodiment, optionally, the lightweight backbone network is a MobilenetV3 network.

[0111] The high-precision signal intelligent detection and identification device based on a neural network provided in an embodiment of the present invention can execute the high-precision signal intelligent detection and identification method based on a neural network provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0112] Embodiment 4

[0113] Figure 71 is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0114] like Figure 7 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0115] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0116] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as a high-precision signal intelligent detection and recognition method based on a neural network.

[0117] In some embodiments, a high-precision signal intelligent detection and identification method based on a neural network may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the high-precision signal intelligent detection and identification method based on a neural network described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute a high-precision signal intelligent detection and identification method based on a neural network in any other appropriate manner (e.g., by means of firmware).

[0118] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] The computer program for implementing the neural network-based high-precision signal intelligent detection and recognition method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, partially on the machine as an independent software package and partially on a remote machine, or completely on a remote machine or server.

[0120] Embodiment 5

[0121] Embodiment 5 of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute a high-precision signal intelligent detection and recognition method based on a neural network, the method comprising:

[0122] Load the trained signal detection network, which is used to locate and identify the signal in the signal time-frequency image; the signal detection network includes a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network includes a four-scale downsampling detection head, the lightweight backbone network includes a union attention mechanism module, and the union attention mechanism includes a channel attention submodule and a spatial attention submodule;

[0123] Acquire the signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain a time-frequency image of the signal to be detected;

[0124] Based on the signal detection network, the time-frequency image of the signal to be detected is located and identified to obtain the signal detection data of the signal to be detected.

[0125] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0126] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0127] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0128] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0129] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0130] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A high-precision signal intelligent detection and recognition method based on neural network, characterized in that: include: Loading a trained signal detection network, the signal detection network is used to locate and identify the signal in the signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network and a prediction network, the lightweight backbone network includes a four-scale downsampling detection head, the lightweight backbone network includes a union attention mechanism module, and the union attention mechanism includes a channel attention submodule and a spatial attention submodule; Acquire a signal to be detected, and perform time-frequency transformation on the signal to be detected to obtain a time-frequency image of the signal to be detected; The time-frequency image of the signal to be detected is located and identified based on the signal detection network to obtain signal detection data of the signal to be detected.

2. The method according to claim 1, characterized in that The signal detection network training method comprises: Obtain signal time-frequency image set; Training the lightweight backbone network based on the signal time-frequency image; Determine an initial signal detection network based on the trained lightweight backbone network and target detection network; Acquire a signal time-frequency standard data set, wherein the signal time-frequency standard data set includes a signal time-frequency image and annotation information corresponding to the signal time-frequency image; The initial signal detection network is trained based on the signal time-frequency standard data set to obtain a trained signal detection network.

3. The method according to claim 2, characterized in that The step of acquiring a signal time-frequency image set comprises: Acquire a signal training set, wherein the signal training set includes signal data and a signal system corresponding to the signal data; Performing time-frequency transformation on the signal data to obtain a signal time-frequency image; The signal time-frequency image set is generated based on the signal time-frequency image and a signal system corresponding to the signal time-frequency image.

4. The method according to claim 2, characterized in that: The loss function of the signal detection network includes a bounding box regression loss function, a category loss function, and a loss function of whether a target exists at an anchor point, and the loss function of whether a target exists at an anchor point includes a loss function of a target existing at an anchor point and a loss function of a target not existing at an anchor point; Among them, the bounding box regression loss function is used to measure the difference between the predicted bounding box and the true bounding box, the category loss function is used to measure the difference between the predicted category and the true category of the signal, and the loss function of whether there is a target at the anchor point is used to perform binary classification prediction for the anchor point position to determine whether there is a target signal at the anchor point.

5. The method according to claim 1, characterized in that The channel attention submodule is connected in parallel with the spatial attention submodule.

6. The method according to claim 3, characterized in that After obtaining the time-frequency image of the signal to be detected or the time-frequency image of the signal, the method further includes: For each pixel point in the time-frequency image of the signal to be detected or the time-frequency image of the signal, determine a filter window based on the pixel point, and divide the filter window into a first sub-region and a second sub-region; For any sub-region, determining the regional variance of the sub-region based on the pixel value, the number of pixel columns, the number of pixel rows and the mean value of all pixels in the sub-region at each coordinate of the sub-region; A target region is determined in the first sub-region and the second sub-region based on the regional variance, and a grayscale median value in the target region is used as a grayscale value of the pixel point.

7. The method according to claim 1, characterized in that The lightweight backbone network is the MobilenetV3 network.

8. A high-precision signal intelligent detection and recognition device based on a neural network, characterized in that: include: A signal detection network loading module is used to load a trained signal detection network, wherein the signal detection network is used to locate and identify signals in a signal time-frequency image; wherein the signal detection network includes a lightweight backbone network, a feature fusion network, and a prediction network, wherein the lightweight backbone network includes a four-scale downsampling detection head, and the lightweight backbone network includes a union attention mechanism module, wherein the union attention mechanism includes a channel attention submodule and a spatial attention submodule; The signal to be detected conversion module is used to obtain the signal to be detected, and perform time-frequency conversion on the signal to be detected to obtain a time-frequency image of the signal to be detected; A positioning and identification module is used to locate and identify the time-frequency image of the signal to be detected based on the signal detection network to obtain signal detection data of the signal to be detected.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the high-precision signal intelligent detection and recognition method based on a neural network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the high-precision signal intelligent detection and recognition method based on a neural network as described in any one of claims 1 to 7 when executed.

Images