A non-line-of-sight signal recognition method, system, computer device and storage medium

Abstract

The present invention discloses a non-line-of-sight signal recognition method, system, computer equipment and storage medium. The method comprises: obtaining the GNSS raw data in the receiver and performing a solution to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle; marking the satellite type according to the raw data to construct a satellite type label; counting the number of line-of-sight satellite labels at the same time to obtain the actual number of observable satellites, and according to the preset satellite cutoff altitude angle, obtaining the theoretical number of visible satellites, and using the ratio of the two as a custom feature; using the satellite signal characteristics and the custom characteristics as feature values ​​to form a feature vector; inputting the feature vector into a deep neural network for training to obtain a non-line-of-sight signal recognition result. The present invention improves the accuracy and stability of classification recognition by implicitly increasing the amount of information of the spatial factors of the recognition task environment.

Description

Technical Field

[0001] The present invention relates to the field of satellite navigation and positioning technology, and in particular to a method, system, computer equipment and storage medium for recognizing non-line-of-sight signals in an urban environment based on a deep neural network. Background Art

[0002] At present, with the rapid development of global navigation satellite systems, GNSS has gradually expanded to many professional application fields. In complex urban environments, satellite navigation signals are easily affected by obstacles such as high-rise buildings and viaducts during space propagation, which can easily cause multipath effects and non-line-of-sight signals, thus affecting the positioning accuracy and reliability of GNSS. At the same time, terminal hardware equipment and signal processing algorithms can also affect positioning accuracy.

[0003] In order to improve the accuracy and reliability of GNSS positioning, commonly used non-line-of-sight signal detection technologies include antenna design, improved receiver algorithms, sensor integration, three-dimensional urban building modeling and machine learning. However, the above methods also have certain limitations. Solutions based on antennas, receivers and signal processing will cause terminal costs to rise, the processing process is more complicated, and can only solve some problems; methods based on three-dimensional urban modeling require detailed modeling of the city, which is labor-intensive and costly, and difficult to promote and update on a large scale; and machine learning methods are currently limited to the instability of classification algorithms, which cannot fit nonlinear relationships well, and extract features from GNSS raw observations to form feature vectors based on the theoretical level, without taking into account the actual application scenario environmental factors, and the classification effect is unstable. Summary of the invention

[0004] The purpose of the present invention is to solve the multipath effect and non-line-of-sight signals of satellite navigation positioning in urban environments. In addition to selecting common features in the original observations of satellite signals as feature vectors of deep neural networks, the amount of information of spatial factors in the recognition task environment is increased, thereby enhancing the generalization ability of the non-line-of-sight recognition model, reducing overfitting, and improving the accuracy and stability of classification and recognition.

[0005] In a first aspect, an embodiment of the present invention provides a method for recognizing non-line-of-sight signals in an urban environment based on a deep neural network, the method comprising:

[0006] Acquire GNSS raw data in the receiver, and solve the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle;

[0007] Marking the satellite type according to the GNSS raw data to construct a satellite type tag, wherein the satellite type tag includes line-of-sight satellites and non-line-of-sight satellites;

[0008] Count the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtain the theoretical number of visible satellites according to the preset satellite cutoff altitude angle, and use the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature;

[0009] The satellite signal feature and the custom feature are used as feature values ​​to form a feature vector;

[0010] The feature vector is input into a deep neural network for training to obtain a non-line-of-sight signal recognition result.

[0011] Furthermore, the step of marking the satellite type according to the GNSS raw data and constructing a satellite type tag includes:

[0012] comparing the availability of each satellite in the GNSS raw data in two receivers;

[0013] Tagging the satellites with signal types according to the availability of each satellite in the two receivers;

[0014] Satellites with different availability are labeled with satellite type tags as non-line-of-sight satellites, and satellites with the same availability are labeled with satellite type tags as line-of-sight satellites.

[0015] Furthermore, the step of marking the satellites by signal type according to the availability of each satellite in the two receivers further comprises:

[0016] Acquire the data time of satellites with the same availability in the GNSS raw data of the two receivers;

[0017] Calculating a time span between each data time, and comparing the time span with a preset time span threshold, and correcting satellite type labels of satellites with the same availability according to the comparison result;

[0018] The satellite type labels of the satellites whose time span is greater than the time span threshold are marked as non-line-of-sight satellites, and the satellite type labels of the satellites whose time span is less than the time span threshold are marked as line-of-sight satellites.

[0019] Furthermore, the step of counting the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtaining the theoretical number of visible satellites according to a preset satellite cutoff altitude angle, and taking the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature includes:

[0020] Count the number of satellite tags in line of sight at the same time to get the actual number of observable satellites;

[0021] Comparing the satellite altitude angle with a preset satellite cutoff altitude angle, taking satellites whose satellite altitude angle is greater than the satellite cutoff altitude angle as theoretically visible satellites, and counting the number of the theoretically visible satellites to obtain a theoretically visible satellite number;

[0022] The actual number of observable satellites is divided by the theoretical number of visible satellites to obtain a custom feature.

[0023] Furthermore, before inputting the feature vector into the deep neural network for training, the method also includes performing data preprocessing on the feature vector, and the data preprocessing includes data missing filling and data normalization.

[0024] Furthermore, the deep neural network adopts two different activation functions, and the activation functions are:

[0025]

[0026] Among them, z is the feature vector input from the input layer.

[0027] and:

[0028]

[0029] Among them, a is the output value after being processed by the g(z) function.

[0030] Furthermore, the loss function used in the deep neural network training is calculated using the following formula:

[0031]

[0032] Among them, y is a binary label whose value is 0 or 1, p(y) is the probability that the output belongs to the y label, and N is the size of the output.

[0033] In a second aspect, an embodiment of the present invention provides a system for recognizing non-line-of-sight signals in an urban environment based on a deep neural network, the system comprising:

[0034] A raw data solving module is used to obtain the GNSS raw data in the receiver and solve the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle;

[0035] A satellite type label marking module, used to mark the satellite type according to the GNSS raw data and construct a satellite type label, wherein the satellite type label includes a line-of-sight satellite and a non-line-of-sight satellite;

[0036] A custom feature calculation module is used to count the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtain the theoretical number of visible satellites according to the satellite cutoff altitude angle, and use the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature;

[0037] A feature vector combination module, used to form a feature vector by taking the satellite signal feature and the custom feature as feature values;

[0038] The signal recognition module is used to input the feature vector into a deep neural network for training to obtain a non-line-of-sight signal recognition result.

[0039] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0040] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0041] The above application provides a non-line-of-sight signal recognition method, system, computer device and storage medium based on deep neural network. Through the method, in addition to selecting the commonly used features in the original observation of satellite signals as feature values, the ratio of the actual number of observable satellites to the theoretical number of visible satellites is used as a custom feature according to the availability and data time of satellites in different receivers, and the visibility of satellites is closely linked to the surrounding environment. Environmental factors are implicitly and qualitatively incorporated into the feature vector of machine learning, and a deep neural network model with strong parallelism and adaptability is used for classification and recognition, thereby overcoming the problem that the traditional feature vector only extracts features from a theoretical level without considering the environmental factors of the actual application scenario, and the classification effect is unstable. This is very meaningful for the existing field of non-line-of-sight signal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of a non-line-of-sight signal recognition method in an embodiment of the present invention;

[0043] Figure 2 yes Figure 1 Flow chart of step S20;

[0044] Figure 3 yes Figure 2 Another flow chart of step S202;

[0045] Figure 4 yes Figure 1Flow chart of step S30;

[0046] Figure 5 It is the PR curve diagram of the experimental results of the three recognition algorithms;

[0047] Figure 6 It is the ROC curve diagram of the experimental results of the three recognition algorithms;

[0048] Figure 7 is a structural schematic diagram of a non-line-of-sight signal recognition system in an embodiment of the present invention;

[0049] Figure 8 It is a diagram of the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 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 are within the scope of protection of the present invention.

[0051] See also Figure 1 The first embodiment of the present invention proposes a method for recognizing non-line-of-sight signals in urban environments based on a deep neural network, which includes steps S10 to S50:

[0052] Step S10, acquiring GNSS raw data in the receiver, and solving the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle.

[0053] Since the recognition accuracy of signal type is closely related to the neural network model used, the selection of feature quantities will also affect the accuracy of signal type recognition. The original data of GNSS signals usually include carrier-to-noise ratio, Doppler frequency shift measurement values, etc., which are closely related to the GNSS signal type.

[0054] Among the commonly used features, it is generally believed that the larger the satellite elevation angle, the higher the probability that the satellite signal is LOS. However, in densely populated areas, satellite signals with low angles are more likely to be LOS, and satellite signals with high angles are likely to be NLOS, i.e., non-line-of-sight. Therefore, judging the signal type only by the satellite elevation angle is prone to misjudgment. The carrier-to-noise ratio and Doppler shift measurements are often used as indicators to distinguish the types of GNSS signals. Therefore, the carrier-to-noise ratio, Doppler shift measurements and satellite elevation angle are combined, and the satellite azimuth angle is used as an auxiliary feature as a characteristic value for signal identification. Among them, the satellite elevation angle and satellite azimuth angle can be calculated through the GNSS original signal and combined with the ephemeris file, while the carrier-to-noise ratio and Doppler shift measurements can be directly extracted from the receiver. The specific acquisition process can be obtained according to the conventional calculation steps of the satellite characteristic value, which will not be repeated here.

[0055] Step S20: Mark the satellite type according to the GNSS raw data to construct a satellite type tag, where the satellite type tag includes a line-of-sight satellite and a non-line-of-sight satellite.

[0056] Among them, the step of marking the satellite type according to the GNSS raw data and constructing the satellite type label is as follows: Figure 2 As shown, specifically including:

[0057] Step S201, comparing the availability of each satellite in the GNSS raw data in two receivers.

[0058] In this embodiment, GNSS raw data from multiple receivers can be selected. In actual operation, at least two receivers are selected. For example, information on GNSS signals received by NovAtel receivers and u-blox receivers can be selected. When using the NovAtel receiver, the last group of satellites received is recorded. Next, a group of satellites is received from the u-blox receiver. If a satellite appears in NovAtel, it is recorded as available in the receiver. If the satellite does not appear in u-blox, it is recorded as unavailable in the receiver. Then, the availability of each satellite in the two receivers is compared to see if it is the same.

[0059] Since different receivers receive different satellite signals, for example, NovAtel receivers exclude some satellites in harsh environments that may be affected by NLOS, while u-blox receivers provide information on all received satellite signals. Therefore, by distinguishing the satellite signals received by different receivers, the satellite type labels can be marked more accurately.

[0060] Step S202: marking the signal type of each satellite according to the availability of each satellite in the two receivers.

[0061] Step S203: Mark the satellite type tags of satellites with different availability as non-line-of-sight satellites, and mark the satellite type tags of satellites with the same availability as line-of-sight satellites.

[0062] Among them, after comparing whether the availability of each satellite in the two receivers is the same, different signal type labels are applied to the satellites for different situations. When the satellite is available in both receivers, the satellite type label of the satellite is marked as LOS; if the satellite is unavailable in one of the receivers, the satellite type label of the satellite is marked as NLOS. This method can simply and quickly label the satellite type label of the satellite.

[0063] This embodiment also provides another labeling method, which is to further refine the satellite type label of the satellite whose satellite type label is LOS in the previous method, such as Figure 3 As shown, specifically including:

[0064] Step S204: obtaining the data time of satellites with the same availability in the GNSS raw data of the two receivers.

[0065] Step S205, calculating the time span between each data time, and comparing the time span with a preset time span threshold, and correcting the satellite type labels of satellites with the same availability according to the comparison result.

[0066] Step S206: Mark the satellite type tags of the satellites whose time span is greater than the time span threshold as non-line-of-sight satellites, and mark the satellite type tags of the satellites whose time span is less than the time span threshold as line-of-sight satellites.

[0067] After receiving a group of satellites from the receiver, in addition to recording the availability of each satellite, the data time of the satellites with the same availability in the receiver is also recorded separately, and the time span is calculated based on the two data times of the satellite. For example, the data time of a satellite in the NovAtel receiver and the u-blox receiver is recorded, and the difference between the two data times is used as the time span since the last update of NovAtel.

[0068] The satellite's time span is compared with the preset time span threshold. If the time span is greater than the time span threshold, it means that the interval between the satellite's appearance is too long, and the satellite type label of the satellite is changed from LOS to NLOS; if the time span is less than the time span threshold, the satellite type label of the satellite is retained as LOS. The advantage of this is that the LOS satellite type label is further corrected by the length of the time span, so that the satellite label classification is more accurate, and no additional complex calculations are required.

[0069] Step S30, counting the number of the line-of-sight satellite tags at the same time, obtaining the actual number of observable satellites, obtaining the theoretical number of visible satellites according to the satellite cutoff altitude angle, and taking the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature.

[0070] After the satellite label is marked, a custom feature value is obtained by further calculating the satellite label and feature value, such as Figure 4 As shown, the specific steps include:

[0071] Step S301, counting the number of satellite tags within the same time range to obtain the actual number of observable satellites.

[0072] Step S302, comparing the satellite altitude angle with a preset satellite cutoff altitude angle, taking the satellites whose satellite altitude angle is greater than the satellite cutoff altitude angle as theoretically visible satellites, and counting the number of the theoretically visible satellites to obtain the theoretically visible satellite number.

[0073] Step S303: Divide the actual number of observable satellites by the theoretical number of visible satellites to obtain a custom feature.

[0074] Although the Doppler shift measurement values, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle in the GNSS raw data were selected as eigenvalues ​​in the above steps, the selection of these eigenvalues ​​did not take environmental factors into account too much. Therefore, we customized a eigenvalue to incorporate the amount of environmental information into the eigenvalue of signal recognition.

[0075] Among them, after the satellite type labels are marked, the number of satellite type labels is classified and counted. Since we will obtain satellite data at different times, here, the number of LOS satellite type labels at the same time is taken as the actual number of observable satellites. The advantage of this is to ensure the accuracy of the data used.

[0076] In GPS measurement, a satellite cutoff elevation angle is usually set to shield the influence of obstructions and multipath effects. Satellites below this angle in the visual airspace are not tracked. Selecting an appropriate satellite elevation angle can effectively improve the influence caused by factors such as multipath effects, which is very beneficial for purifying raw data. Therefore, we preset a satellite cutoff elevation angle in advance, and compare the satellite elevation angle calculated based on the raw data in the receiver combined with the precise ephemeris with the satellite cutoff elevation angle preset according to the actual situation, and regard the satellites with satellite elevation angles greater than the satellite cutoff elevation angle as theoretically visible satellites.

[0077] In fact, the actual number of observable satellites and the theoretical number of visible satellites we obtain are calculated based on the satellite data received by the receiver at the same time. Therefore, the ratio of the actual number of observable satellites divided by the theoretical number of visible satellites at the same time can be used as a custom feature value. Since the satellite type label and the satellite cutoff elevation angle reflect the impact of environmental factors on satellite signals, this custom feature value is introduced to qualitatively consider the impact of environmental information on the recognition task.

[0078] Step S40: Using the satellite signal feature and the user-defined feature as feature values ​​to form a feature vector.

[0079] The Doppler shift measurement value, carrier-to-noise ratio, satellite altitude angle, satellite azimuth angle and custom features obtained above are used as five one-dimensional eigenvalues ​​to form a feature vector. These one-dimensional eigenvalues ​​may have data missing or inconsistent lengths, so it is necessary to perform data preprocessing on the feature vector, such as data missing filling and data normalization, so that it can form a complete five-dimensional feature vector. The feature vector contains parameters closely related to the GNSS signal type, as well as the influence of environmental factors, so that the recognition results of the deep neural network trained for non-line-of-sight signal recognition using the feature vector are more accurate. It should be understood that the data preprocessing of the feature vector not only includes data missing filling and data normalization, but also other processing methods according to actual conditions, which will not be repeated here.

[0080] Step S50: input the feature vector into a deep neural network for training to obtain a non-line-of-sight signal recognition result.

[0081] In addition to constructing the feature vector, it is also necessary to build an NLOS recognition model for identifying non-line-of-sight signals. Since the deep neural network has good adaptability and robustness and strong comprehensive information capabilities, we choose the deep neural network as the NLOS recognition model. In this embodiment, the deep neural network uses two activation functions, which are the ReLU function and the sigmoid function.

[0082] Among them, the ReLU function is:

[0083]

[0084] z is the feature vector input from the input layer.

[0085] The sigmoid function is:

[0086]

[0087] a is the output value after being processed by the g(z) function.

[0088] The loss function of the deep neural network uses the binary cross entropy loss function:

[0089]

[0090] y is a binary label whose value is 0 or 1, p(y) is the probability that the output belongs to the y label, and N is the size of the output.

[0091] The above-mentioned feature vector is input into the constructed deep neural network for training to obtain a trained deep neural network, and the test set is input into the deep neural network for signal recognition to obtain the recognition result of the non-line-of-sight signal.

[0092] In order to further verify that this embodiment has better stability and accuracy than other recognition algorithms, Figure 5 and Figure 6 , the recognition results of data collected from different regions using this embodiment, the SVM algorithm and the DT algorithm are described in detail.

[0093] Commonly used indicators for measuring the effectiveness of NLOS signal recognition models include accuracy, precision, recall, F1 score, PR curve and ROC curve, etc. In this comparative experiment, we selected PR curve and ROC curve as performance indicators of experimental results. For data collected in different regions, the training models of the above three algorithms are used to test the adaptability to different actual environments, and the five-fold cross-validation algorithm is used to evaluate the model performance. The specific calculation process can follow the steps described in this embodiment, as well as the conventional steps of the other two measurement methods, which will not be repeated here.

[0094] By using different recognition algorithms to identify the data, we can get the following Figure 5 The PR curve shown in the figure and Figure 6 The ROC curve is shown in Figure 2. Figure 5It can be seen that the PR curve of the deep neural network algorithm DNN of this embodiment is above the curves of the other two algorithms, and better covers the other two curves, indicating that the performance of the DNN algorithm is better than the other two recognition algorithms, and the overall jitter of the DNN algorithm is smaller, and the prediction effect of the recognition algorithm of this embodiment is less affected by the change of threshold.

[0095] exist Figure 6 It can be seen that the curve of the deep neural network algorithm DNN is also above the curves of the other two algorithms, and its area under the ROC curve is the largest, that is, the AUC is the largest. Therefore, the classifier of the deep neural network algorithm of this application is better than that of the other two algorithms.

[0096] In summary, in the calculation of signal recognition for the data set, the PR curve and ROC curve of the SVM algorithm and the DT algorithm are both below the curve of the algorithm in this embodiment. Therefore, compared with the other two methods, the non-line-of-sight signal recognition method provided in this embodiment has better stability and more accurate recognition results.

[0097] A non-line-of-sight signal recognition method provided by an embodiment of the present invention cannot fit nonlinear relationships well and has unstable classification effects compared to conventional intelligent algorithms that extract characteristic values ​​from original GNSS observations at a theoretical level for classification and recognition. By increasing the amount of information of spatial factors in the recognition task environment, the generalization ability of the non-line-of-sight recognition model is enhanced, overfitting is reduced, and the accuracy and stability of classification and recognition are improved.

[0098] See also Figure 7 Based on the same inventive concept, a second embodiment of the present invention provides a non-line-of-sight signal recognition system, comprising:

[0099] The raw data solving module 10 is used to obtain the GNSS raw data in the receiver and solve the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle;

[0100] A satellite type label marking module 20, used to mark the satellite type according to the GNSS raw data and construct a satellite type label, wherein the satellite type label includes a line-of-sight satellite and a non-line-of-sight satellite;

[0101] A custom feature calculation module 30 is used to count the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtain the theoretical number of visible satellites according to a preset satellite cutoff altitude angle, and use the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature;

[0102] A feature vector combination module 40, configured to use the satellite signal feature and the user-defined feature as feature values ​​to form a feature vector;

[0103] The signal recognition module 50 is used to input the feature vector into a deep neural network for training to obtain a non-line-of-sight signal recognition result.

[0104] The technical features and technical effects of the non-line-of-sight signal recognition system proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention, and are not described in detail here. Each module in the above-mentioned non-line-of-sight signal recognition system can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0105] See also Figure 8 , an internal structure diagram of a computer device in an embodiment, the computer device can specifically be a terminal or a server. The computer device includes a processor, a memory, a network interface, a display and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a non-line-of-sight signal recognition method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0106] It can be understood by those skilled in the art that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the present application scheme, and does not constitute a limitation on the computer device to which the present application scheme is applied. The specific computing device may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0107] In addition, an embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0108] In addition, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0109] In summary, the embodiments of the present invention provide a non-line-of-sight signal recognition method, system, computer device and storage medium. The method obtains GNSS raw data in a receiver and solves the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement values, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle; the satellite type is marked according to the GNSS raw data to construct a satellite type label, wherein the satellite type label includes line-of-sight satellites and non-line-of-sight satellites; the number of line-of-sight satellite labels at the same time is counted to obtain the actual number of observable satellites, and the theoretical number of visible satellites is obtained according to a preset satellite cutoff altitude angle, and the ratio of the actual number of observable satellites to the theoretical number of visible satellites is used as a custom feature; the satellite signal characteristics and the custom features are used as feature values ​​to form a feature vector; the feature vector is input into a deep neural network for training to obtain a non-line-of-sight signal recognition result. In addition to selecting the commonly used signal recognition parameters in the GNSS raw data as feature vectors, this method also introduces custom eigenvalues, closely links the satellite visibility with the surrounding environment, and invisibly and qualitatively incorporates environmental factors into the feature vectors of machine learning. It considers the regional differences in the collected data sets, increases the amount of information about environmental spatial factors in the recognition task, links the recognition and classification tasks with environmental factors, forms a correct and comprehensive feature vector, enhances the generalization ability of the GNSS NLOS recognition model, reduces overfitting, enhances the understanding of eigenvalues, reduces the difficulty of building the model, and outputs higher quality results.

[0110] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not 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 specification.

[0111] The above-described embodiments only express several preferred implementations of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be based on the protection scope of the claims.

Claims

1. A non-line-of-sight signal recognition method based on deep neural network, characterized in that: include: Acquire GNSS raw data in the receiver, and solve the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle; Marking the satellite type according to the GNSS raw data to construct a satellite type tag, wherein the satellite type tag includes line-of-sight satellites and non-line-of-sight satellites; Count the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtain the theoretical number of visible satellites according to the preset satellite cutoff altitude angle, and use the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature; The satellite signal feature and the custom feature are used as feature values ​​to form a feature vector; Inputting the feature vector into a deep neural network for training to obtain a non-line-of-sight signal recognition result; The deep neural network uses two different activation functions, which are: Among them, z is the feature vector input from the input layer; and: Among them, a is the output value after being processed by the g(z) function; The loss function used in the deep neural network training is calculated using the following formula: Among them, y is a binary label whose value is 0 or 1, p(y) is the probability that the output belongs to the y label, and N is the size of the output.

2. The non-line-of-sight signal recognition method based on deep neural network according to claim 1 is characterized in that: The step of marking the satellite type according to the GNSS raw data and constructing the satellite type label comprises: comparing the availability of each satellite in the GNSS raw data in two receivers; Tagging the satellites with signal types according to the availability of each satellite in the two receivers; Satellites with different availability are labeled with satellite type tags as non-line-of-sight satellites, and satellites with the same availability are labeled with satellite type tags as line-of-sight satellites.

3. The non-line-of-sight signal recognition method based on deep neural network according to claim 2 is characterized in that: The step of marking the satellites with signal types according to the availability of each satellite in the two receivers further comprises: Acquire the data time of satellites with the same availability in the GNSS raw data of the two receivers; Calculating a time span between each data time, and comparing the time span with a preset time span threshold, and correcting satellite type labels of satellites with the same availability according to the comparison result; The satellite type labels of the satellites whose time span is greater than the time span threshold are marked as non-line-of-sight satellites, and the satellite type labels of the satellites whose time span is less than the time span threshold are marked as line-of-sight satellites.

4. The non-line-of-sight signal recognition method based on deep neural network according to claim 1, characterized in that: The step of counting the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtaining the theoretical number of visible satellites according to the satellite cutoff altitude angle, and taking the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature includes: Count the number of satellite tags in the same time range to get the actual number of observable satellites; Comparing the satellite altitude angle with a preset satellite cutoff altitude angle, taking satellites whose satellite altitude angle is greater than the satellite cutoff altitude angle as theoretically visible satellites, and counting the number of the theoretically visible satellites to obtain a theoretically visible satellite number; The actual number of observable satellites is divided by the theoretical number of visible satellites to obtain a custom feature.

5. The non-line-of-sight signal recognition method based on deep neural network according to claim 1, characterized in that: Before inputting the feature vector into the deep neural network for training, the method also includes performing data preprocessing on the feature vector, wherein the data preprocessing includes data missing filling and data normalization.

6. A non-line-of-sight signal recognition system based on deep neural network, characterized in that: The system comprises: A raw data solving module is used to obtain the GNSS raw data in the receiver and solve the GNSS raw data to obtain satellite signal characteristics, wherein the satellite signal characteristics include Doppler frequency shift measurement value, carrier-to-noise ratio, satellite altitude angle and satellite azimuth angle; A satellite type label marking module, used to mark the satellite type according to the GNSS raw data and construct a satellite type label, wherein the satellite type label includes a line-of-sight satellite and a non-line-of-sight satellite; A custom feature calculation module is used to count the number of the line-of-sight satellite tags at the same time to obtain the actual number of observable satellites, obtain the theoretical number of visible satellites according to a preset satellite cutoff altitude angle, and use the ratio of the actual number of observable satellites to the theoretical number of visible satellites as a custom feature; A feature vector combination module, used to form a feature vector by taking the satellite signal feature and the user-defined feature as feature values; A signal recognition module, used for inputting the feature vector into a deep neural network for training to obtain a non-line-of-sight signal recognition result; The deep neural network uses two different activation functions, which are: Among them, z is the feature vector input from the input layer; and: Among them, a is the output value after being processed by the g(z) function; The loss function used in the deep neural network training is calculated using the following formula: Among them, y is a binary label whose value is 0 or 1, p(y) is the probability that the output belongs to the y label, and N is the size of the output.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images