Visible light sight distance signal identification method based on machine learning and related device

Abstract

The invention provides a visible light sight distance signal identification method based on machine learning and a related device. Receiving an optical signal carrying a synchronous training symbol; deriving a received signal strength from the received optical signal; the method comprises the following steps: performing synchronization operation on received optical signals to obtain a timing metric value, intercepting through a sliding window according to a peak value of the timing metric value to obtain a timing metric peak value sequence, and extracting characteristic quantities of the timing metric peak value sequence, including a peak value, a mean value, a standard deviation, kurtosis, skewness and a median; fusing the received signal strength and the characteristic quantity of the timing metric peak sequence to obtain a fused row vector; performing normalization processing on the fused row vector to obtain a mixed feature vector; inputting the mixed feature vector into a trained recognition model, and performing visible light sight distance signal recognition according to the mixed feature vector to obtain a recognition result; the average accuracy of signal recognition is improved, and the problem that the recognition success rate of a sight distance signal at the edge in a complex optical environment through a single RSS feature is low is effectively solved.

Description

Technical Field

[0001] This application relates to the field of communications, and more particularly to a visible light line-of-sight signal recognition method and related apparatus based on machine learning. Background Technology

[0002] Visible light communication (VLC) is a technology that uses variations in the intensity of visible light to transmit information. Visible light positioning utilizes the visible light spectrum to achieve high-bandwidth, secure, and interference-resistant signal transmission, making it particularly suitable for indoor environments. However, signal propagation in indoor VLC systems is often affected by obstacles, leading to a significant reduction in positioning accuracy under non-line-of-sight (NLOS) conditions. The complexity of indoor propagation, including multipath fading, shadows, and random obstacles, places high demands on reliable line-of-sight / non-line-of-sight identification. Therefore, accurately distinguishing between line-of-sight (LOS) and NLOS signals is a crucial prerequisite for optimizing VLC system performance. Currently, the main approach is line-of-sight signal identification based on received signal strength (RSS) characteristics, which has low requirements for the operating equipment.

[0003] The process of the line-of-sight signal identification method based on received signal strength characteristics is as follows: sample the RSS time series of LOS and NLOS signals received by the photodetector, extract simple statistical features to train a machine learning model; use the trained model to determine whether the received signal is a LOS signal or an NLOS signal.

[0004] Existing LOS / NLOS identification techniques based on received signal strength have significant limitations. The fundamental problem lies in the fact that RSS, as a scalar quantity reflecting only instantaneous power, cannot capture the structural changes of the optical signal in the time domain, resulting in insufficient discriminative power when distinguishing between direct signals of similar intensity and signals with multiple reflections. These methods typically rely heavily on long-term, high-sampling historical RSS data to extract statistical features, which not only introduces significant data overhead and processing latency, making it difficult to meet the needs of real-time applications, but also exhibits poor robustness due to its excessive sensitivity to dynamic environmental changes. Furthermore, RSS measurements are highly susceptible to inherent fluctuations such as equipment nonlinearity, noise, and ambient light, further reducing the signal-to-noise ratio and reliability of the classification. Summary of the Invention

[0005] The following is an overview of the topics described in detail in this article.

[0006] The purpose of this application is to at least partially solve one of the technical problems existing in the related technologies. The embodiments of this application provide a visible light line-of-sight signal recognition method and related apparatus based on machine learning, which can improve the accuracy of signal recognition.

[0007] An embodiment of this application provides a visible light line-of-sight signal recognition method based on machine learning, comprising: Receive an optical signal to be identified, the optical signal to be identified carrying a synchronous training symbol with parity symmetry; The received signal strength is obtained from the received optical signal. The received optical signal is synchronized using a timing measurement method to obtain a timing measurement value. The timing measurement peak sequence is obtained by using a sliding window with the peak value of the timing measurement value as the center. The feature quantities of the timing measurement peak sequence are extracted, including peak value, mean, standard deviation, kurtosis, skewness and median. A fused row vector is obtained by fusing the received signal strength and the feature values ​​of the timing metric peak sequence; The fused row vector is normalized to obtain the mixed feature vector; The mixed feature vector is input into the trained recognition model, and the visible light line-of-sight signal is identified based on the mixed feature vector to obtain the recognition result. The recognition result indicates whether the light signal to be identified is a line-of-sight signal or a non-line-of-sight signal. The trained recognition model is based on a gradient boosting decision tree.

[0008] According to certain embodiments of this application, training a recognition model includes: Receive training optical signals, the training optical signals carrying synchronous training symbols with parity symmetry, the training optical signals being optical signals under line-of-sight conditions or optical signals under non-line-of-sight conditions; Extract the sample received signal intensity from the training optical signal; The training optical signal is synchronized using a timing measurement method to obtain sample timing measurement values. The peak value of the sample timing measurement is used as the center to extract the sample timing measurement peak sequence through a sliding window. The sample timing measurement peak sequence is statistically analyzed under line-of-sight and non-line-of-sight conditions to extract the feature quantities of the sample timing measurement peak sequence. The feature quantities of the sample timing measurement peak sequence include peak value, mean, standard deviation, kurtosis, skewness, and median. The sample fusion row vector is obtained by fusing the feature values ​​of the received signal intensity of the sample and the timing measurement peak sequence of the sample. The sample fusion row vector is set with a classification label, which indicates whether the training optical signal is an optical signal under line-of-sight conditions or an optical signal under non-line-of-sight conditions. The sample fusion row vector is normalized to obtain the sample fusion feature vector; The sample mixed feature vector is input into a gradient boosting decision tree for training to obtain a trained recognition model.

[0009] According to certain embodiments of this application, when the training light signal is a light signal under line-of-sight conditions, the classification label is set to 0; when the training light signal is a light signal under non-line-of-sight conditions, the classification label is set to 1.

[0010] 4. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, wherein the timing measurement is expressed as: ;in, , In the formula, d represents time. Let d be the timing metric at time d, N be the size of the sliding window, and r be the baseband signal.

[0011] According to certain embodiments of this application, the time corresponding to the peak value of the timing metric is represented as follows: In the formula, The time corresponding to the peak value of the timing metric. For timed measurement values, S This is the search interval for the sliding window; The vector representation of the timing metric peak sequence is as follows: In the formula, The vector is used to measure the peak sequence at regular intervals, and the sliding window size is 2T+1, where T is a positive integer.

[0012] According to certain embodiments of this application, the peak value in the characteristic quantity of the timing-measured peak sequence is represented as follows: In the formula, To measure the characteristic quantities of the peak sequence at regular intervals, For timed measurement values, This refers to the time corresponding to the peak value of the timing measurement.

[0013] According to certain embodiments of this application, the mean value in the characteristic quantities of a timed peak sequence is represented as follows: The standard deviation of the characteristic quantities of a time-measured peak sequence is expressed as: In the formula, The mean of the characteristic quantities in the time-dependent peak sequence is denoted by d, where T is a positive integer and d is the time interval. The time corresponding to the peak value of the timing metric. For timed measurement values, The standard deviation is used to measure the characteristic quantity of a peak sequence at regular intervals.

[0014] According to certain embodiments of this application, the kurtosis in the characteristic quantities of a timed peak sequence is represented as follows: The skewness in the characteristic quantities of a time-measured peak sequence is expressed as: In the formula, To measure the kurtosis, a characteristic property of a peak sequence, at regular intervals. This is used to measure the skewness of the characteristic quantities in a peak sequence at regular intervals.

[0015] According to certain embodiments of this application, the median of the characteristic quantities in a timing-measured peak sequence is represented as follows: ; It represents the (T+1)th element in the order of elements rearranged in ascending or descending order in the timing measurement peak sequence, where T is a positive integer.

[0016] According to certain embodiments of this application, the fused row vector is normalized using the following formula to obtain the mixed feature vector: ;in, To merge row vectors, For mixed feature vectors, To merge the mean of the row vectors, This is to fuse the variance of the row vectors.

[0017] The above scheme has at least the following beneficial effects: It obtains the received signal strength from the optical signal to be identified; performs synchronization operations on the synchronization symbol to obtain a timing metric value; extracts the timing metric peak sequence through a sliding window based on the peak value of the timing metric value; extracts the features of the timing metric peak sequence, including peak value, mean, standard deviation, kurtosis, skewness, and median; fuses the received signal strength and the features of the timing metric peak sequence to obtain a fused row vector; normalizes the fused row vector to obtain a mixed feature vector; inputs the mixed feature vector into a trained recognition model; and performs visible light line-of-sight signal recognition based on the mixed feature vector to obtain the recognition result. This improves the average accuracy of signal recognition and effectively solves the problem of low success rate of single RSS features in recognizing line-of-sight signals at the edges in complex optical environments. Attached Figure Description

[0018] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0019] Figure 1 This is a schematic diagram of a visible light communication model; Figure 2 This is a flowchart illustrating the steps of a machine learning-based visible light line-of-sight signal recognition method. Figure 3 This is a diagram illustrating the training steps of the recognition model; Figure 4 This is a flowchart for generating training data. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0022] Visible light communication (VLC) is a technology that uses variations in the intensity of visible light to transmit information. Visible light positioning (VLP) utilizes the visible light spectrum to achieve high-bandwidth, secure, and interference-resistant signal transmission, making it particularly suitable for indoor environments. However, signal propagation in indoor VLP systems is often affected by obstacles, leading to a significant reduction in positioning accuracy under non-line-of-sight (NLOS) conditions. The complexity of indoor propagation, including multipath fading, shadows, and random obstacles, places high demands on reliable line-of-sight / non-line-of-sight identification. Therefore, accurately distinguishing between line-of-sight (LOS) and NLOS signals is a crucial prerequisite for optimizing VLP system performance.

[0023] Currently, the main method used is line-of-sight signal identification based on Received Signal Strength (RSS) characteristics, which has low requirements for the working equipment.

[0024] Reference Figure 1 The process of the line-of-sight signal identification method based on the visible light communication model and the characteristics of the received signal intensity is as follows: sample the RSS time series of the LOS and NLOS signals received by the photodetector, extract their simple statistical features for training the machine learning model; use the trained model to determine whether the received signal is a LOS signal or an NLOS signal.

[0025] The total DC gain from LED to PD can be obtained by adding the values ​​of the LOS path and the NLOS path.

[0026] The channel gain of direct light can be expressed as: ;in, Denotes the Lambert order. For PD detection area, The straight-line distance from the LED to the PD. For the irradiation angle of the LED, It is the incident angle of the PD. and These are the condenser gain and the optical filter gain, respectively.

[0027] The channel gain of the reflection path is: ;in, It is the distance from the LED to the reflection point. It is the distance from the reflection point to the PD. It is the reflection coefficient. It is the area of ​​the reflection point. It is the irradiance angle at the reflection point. It is the angle of incidence at the point of reflection. It is the irradiation angle of the PD. It is the angle of incidence of PD.

[0028] The number of channel paths is The discrete-time channel impulse response is expressed as: ;in, It is the sampling interval. It is the arrival time of the path.

[0029] Assuming the LED emits at If the location signal is the optical power, then the optical power received by the LED is... for: .

[0030] In indoor environments, random movement of individuals or objects can obstruct visible light signals. For the receiver, this manifests primarily as the disappearance of the LOS component, which dominates the received signal, leaving only the NLOS component. When the LOS signal is blocked, the first term of the channel gain in the optical power received by the LED disappears. Therefore, under signal blocking conditions, the received optical power of the NLOS component can be expressed as: .

[0031] This algorithm for identifying LOS / NLOS based on RSS features extracts statistical features from the time series of RSS data to obtain the basic features of RSS. Training using a random forest model Used for LOS / NLOS identification.

[0032] Regarding mean and variance, the mean reflects the average energy level of the sequence, while the standard deviation measures volatility. NLOS propagation introduces multipath effects and volatility, typically lowering the mean and increasing the standard deviation. and .

[0033] Ichthyophthias bias measures the dispersion of data, describing the range of fluctuation in the core region centered on the median. Because NLOS (Not in Oriented System) makes the data distribution in the middle region more dispersed, chthyophthias bias increases. Q1 is the first quartile (lower quartile), which is the value at the 25th percentile of the RSS sequence; Q3 is the third quartile (upper quartile), which is the value at the 75th percentile of the RSS sequence.

[0034] The number of outliers measures the number of extreme values ​​in a sequence. Deep fading and anomalous reflections in NLOS environments introduce more extremely strong or weak signals, leading to a significant increase in the number of outliers. .in This is an indicator function that takes the value 1 when the condition is true and 0 otherwise. The lower bound; The upper boundary.

[0035] Existing LOS / NLOS identification techniques based on received signal strength have significant limitations. The fundamental problem lies in the fact that RSS, as a scalar quantity reflecting only instantaneous power, cannot capture the structural changes of the optical signal in the time domain, resulting in insufficient discriminative power when distinguishing between direct signals of similar intensity and signals with multiple reflections. These methods typically rely heavily on long-term, high-sampling historical RSS data to extract statistical features, which not only introduces significant data overhead and processing latency, making it difficult to meet the needs of real-time applications, but also exhibits poor robustness due to its excessive sensitivity to dynamic environmental changes. Furthermore, RSS measurements are highly susceptible to inherent fluctuations such as equipment nonlinearity, noise, and ambient light, further reducing the signal-to-noise ratio and reliability of the classification.

[0036] To address the above problems, embodiments of this application provide a visible light line-of-sight signal recognition method and related apparatus based on machine learning.

[0037] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0038] Reference Figure 2 Visible light line-of-sight signal recognition methods based on machine learning include: Step S210: Receive the optical signal to be identified; wherein the optical signal to be identified carries a synchronization training symbol with parity-even symmetry; Step S220: Obtain the received signal strength from the received optical signal; Step S230: The received optical signal is synchronized using a timing measurement method to obtain a timing measurement value. The timing measurement peak sequence is obtained by using a sliding window with the peak value of the timing measurement value as the center. The feature quantities of the timing measurement peak sequence are extracted. The feature quantities of the timing measurement peak sequence include peak value, mean, standard deviation, kurtosis, skewness and median. Step S240: The characteristic quantities of the received signal strength and the timing metric peak sequence are fused to obtain the fused row vector; Step S250: Normalize the fused row vectors to obtain the mixed feature vectors; Step S260: Input the mixed feature vector into the trained recognition model, and perform visible light line-of-sight signal recognition based on the mixed feature vector to obtain the recognition result; wherein, the recognition result indicates whether the light signal to be recognized is a line-of-sight signal or a non-line-of-sight signal, and the trained recognition model is based on a gradient boosting decision tree.

[0039] Reference Figure 3 Training the recognition model includes the following steps: Step S110: Receive training optical signal; wherein the training optical signal carries a synchronous training symbol with parity symmetry, and the training optical signal is an optical signal under line-of-sight conditions or an optical signal under non-line-of-sight conditions. Step S120: Extract the sample received signal intensity from the training optical signal; Step S130: The training optical signal is synchronized using a timing measurement method to obtain sample timing measurement values. The peak value of the sample timing measurement is used as the center to extract the sample timing measurement peak sequence through a sliding window. The sample timing measurement peak sequence is statistically analyzed under line-of-sight and non-line-of-sight conditions to extract the feature quantities of the sample timing measurement peak sequence. The feature quantities of the sample timing measurement peak sequence include peak value, mean, standard deviation, kurtosis, skewness, and median. Step S140: The sample fusion row vector is obtained by fusing the feature values ​​of the received signal intensity of the sample and the timing measurement peak sequence of the sample; wherein, the sample fusion row vector is provided with a classification label, and the classification label indicates that the training optical signal is an optical signal under line-of-sight conditions or an optical signal under non-line-of-sight conditions. Step S150: Normalize the sample fusion row vector to obtain the sample fusion feature vector; Step S160: Input the sample mixed feature vector into the gradient boosting decision tree for training to obtain the trained recognition model.

[0040] The receiver receives training optical signals from each LED under line-of-sight (LOS) and non-line-of-sight (NLOS) conditions, i.e., optical signals under LOS and non-line-of-sight conditions. In the simulation environment, there are four LEDs, therefore there will be four training optical signals at one location.

[0041] The simulation environment was a 4m × 4m indoor area. This 4m × 4m indoor area was discretized into a grid with a spatial resolution of 10cm × 10cm, providing sufficient granularity for evaluating positioning performance. At each grid point, under LOS and NLOS conditions, ten instantaneous received signal strength (RSS) measurements and corresponding timing metric peak sequence (TMPS) signals were collected from four LEDs.

[0042] At the LED transmitter, synchronous training symbols with parity symmetry are generated by applying Hermite symmetry and inverse fast Fourier transform.

[0043] The input data consists of Shapiro-Rudin sequences. By applying Hermitian symmetry and inverse fast Fourier transform (IFFT), time-domain synchronization training symbols of length N are generated, whose elements satisfy the following parity symmetry: .

[0044] By adding a cyclic prefix to the synchronous training symbols, a real-valued DC-biased optical orthogonal frequency division multiplexing (DCO-OFDM) signal is obtained. In visible light communication (VLC) systems, the nonlinear characteristics of light-emitting diodes (LEDs) are a key factor affecting the performance of DCO-OFDM. To mitigate this effect, the time-domain signal needs to be clipped to limit the symbol amplitude within the linear operating range of the LED, thus obtaining a clipped signal. After digital-to-analog (D / A) conversion, a DC bias is added. To generate LED driving signals, represented as: The LED is driven by an LED driving signal to emit a training light signal carrying synchronous training symbols with parity-even symmetry.

[0045] At the receiving end, training optical signals from each LED are received under LOS and NLOS conditions.

[0046] The received signal intensity of samples is extracted from the training optical signal. Specifically, an RSS feature-based recognition algorithm extracts the basic features of the RSS time series by statistically analyzing the RSS data. .

[0047] The baseband signal after analog-to-digital (A / D) conversion can be represented as: ;in This represents the convolution operation. It is the channel impulse response, and It is additive white Gaussian noise (AWGN), which includes thermal noise and shot noise.

[0048] The timing metric value is obtained by synchronizing the synchronization symbol using a timing metric method. In timing synchronization, a sliding window extracts a segmented sequence of length N from the received signal, which is used to perform correlation operations at sampling time d. The timing metric based on parity symmetry is expressed as: .

[0049] in, , In the formula, d represents time. d is the timing metric at time d, N is the size of the sliding window, and r is the baseband signal.

[0050] The timing metric peak sequence is obtained by using a sliding window to capture the peak value of the timing metric. Specifically, the timing metric value at time d can be calculated and expressed as follows: Then, the time corresponding to the detected timing metric peak is defined as: Where S represents the search interval of the sliding window. The center of the window with length 2T+1 (choosing T=70) is located at... The vector definition for timing the peak sequence is: In the formula, The vector is used to measure the peak sequence at regular intervals, and the sliding window size is 2T+1, where T is a positive integer.

[0051] Reference Figure 4 The timing measurement peak sequence was statistically analyzed under both line-of-sight and non-line-of-sight conditions to extract its characteristic quantities. Specifically, through statistical analysis of a large number of TMPS samples under LOS and NLOS conditions, the following discriminant features of TMPS were obtained, including peak value, mean, standard deviation, kurtosis, skewness, and median as basic features.

[0052] For peak ( The value (), represents the maximum intensity in the TMPS sequence. Compared to LOS conditions, the attenuation of reflection and diffraction significantly reduces the peak value. The peak value is expressed as: In the formula, To measure the characteristic quantities of the peak sequence at regular intervals, For timed measurement values, This refers to the time corresponding to the peak value of the timing measurement.

[0053] For mean and standard deviation ( , The mean reflects the average energy level of the sequence, while the standard deviation measures volatility. NLOS propagation introduces multipath effects and volatility, typically lowering the mean and increasing the standard deviation. The mean is expressed as: Standard deviation is expressed as: In the formula, The mean of the characteristic quantities in the time-dependent peak sequence is denoted by d, where T is a positive integer and d is the time interval. The time corresponding to the peak value of the timing metric. For timed measurement values, The standard deviation is used to measure the characteristic quantity of a peak sequence at regular intervals.

[0054] For kurtosis ( The kurtosis of a probability distribution is measured. Excess kurtosis based on a normal distribution (kurtosis of 0) is used. The LOS condition produces a concentrated distribution with high kurtosis, while the NLOS condition disperses energy, producing a flatter distribution and lower kurtosis values. Kurtosis is expressed as: .

[0055] For skewness ( The quantification of distribution asymmetry. Under LOS, TMPS are concentrated and symmetrical with low skewness. Under NLOS, multiple reflections lead to an asymmetric, dispersed distribution with higher skewness. Skewness is expressed as: .

[0056] For the median ( The median is a robust metric that is not very sensitive to outliers such as impulse noise. It provides complementary information to the mean and variance, especially in noisy environments. The median is expressed as: .in This indicates the order of elements after rearranging them in ascending or descending order in TMPS. It represents the (T+1)th element in the order of elements rearranged in ascending or descending order in the timing metric peak sequence.

[0057] For example, a total of V=128000 original samples were obtained, and each sample was represented as a fused row vector formed by concatenating RSS and TMPS features. The fused row vector contained... , , , , , , .

[0058] The merged row vectors are normalized to obtain the mixed feature vector. Since the original features extracted by RSS and TMPS have different units and numerical ranges, feature normalization is necessary to ensure balanced contributions during training. Z-score normalization is used to map each feature to a distribution with a mean of 1 and a variance of 0. The mixed feature vector is expressed as: .in Represents the j-th raw feature of RSS and TMPS (j=1,2,…,7, corresponding to instantaneous RSS value, peak value, mean, standard deviation, kurtosis, skewness and median). Includes , , , , , , ,and and It is the mean and standard deviation of the sample over all V training samples.

[0059] For each sample v, the mixed feature vector is used as the basic feature and concatenated with its corresponding classification label to form a training sample of length 7+1. Specifically, the first 7 elements correspond to the normalized features derived from RSS and TMPS, and the last element encodes the sample's classification label. If the sample was obtained under LOS conditions, its classification label is set to 0; if the sample was obtained under NLOS conditions, its classification label is set to 1. This generates a dataset containing 8 columns and 128,000 rows, where the first 7 columns represent the sample's RSS and TMPS features, and the 8th column represents the corresponding label. Finally, this training set is fed into the GBDT model for training to obtain the trained recognition model.

[0060] The classification model is implemented using Gradient Boosting Decision Tree (GBDT). This model directly receives temporal feature vectors (such as TMPS and instantaneous RSS) extracted from the original signal. It iteratively trains multiple weak decision trees, gradually correcting the prediction errors of previous rounds through gradient boosting, thereby constructing a powerful ensemble prediction model to complete the classification task. The overall structure is suitable for handling high-dimensional features and can effectively capture complex nonlinear relationships between features.

[0061] The trained recognition model is used to perform online recognition of the optical signal to be recognized, and to determine whether the optical signal is under LOS conditions or NLOS conditions.

[0062] Specifically, the optical signal to be identified is received; wherein the optical signal to be identified carries a synchronization symbol with parity symmetry; The received signal strength is obtained from the received optical signal. The received optical signal is synchronized using a timing measurement method to obtain timing measurement values. The timing measurement peak sequence is obtained by using a sliding window with the peak value of the timing measurement value as the center, and the feature quantities of the timing measurement peak sequence are extracted. The feature quantities of the timing measurement peak sequence include peak value, mean, standard deviation, kurtosis, skewness and median. The fused row vector is obtained by fusing the characteristics of the received signal strength and the timing metric peak sequence. The fused row vectors are normalized to obtain the mixed feature vectors; The mixed feature vector is input into the trained recognition model, and the recognition result is obtained by recognizing the visible light line-of-sight signal based on the mixed feature vector; wherein, the recognition result indicates whether the light signal to be recognized is a line-of-sight signal or a non-line-of-sight signal, and the trained recognition model is based on a gradient boosting decision tree.

[0063] Understandably, the online recognition process is essentially the same as the training process. The difference lies in the fact that the received light signal to be recognized is uncertain—whether it's a line-of-sight signal or a non-line-of-sight signal—and the type of light signal to be recognized needs to be identified by the recognition model. The mixed feature vector input to the recognition model does not have a classification label.

[0064] To evaluate model performance, test data is dynamically generated during the evaluation phase to ensure unbiasedness. During online testing, a single LED light is randomly selected and blocked to simulate a real-world scenario. For each test grid point, 100 LOS / NLOS identification trials are performed to account for random channel fluctuations and ensure statistical reliability.

[0065] Compared with RSS-based RF identification methods, this machine learning-based visible light line-of-sight signal identification method has significant improvements in cumulative distribution function (CDF) curves and positioning accuracy.

[0066] Matlab simulation parameters: The simulation scene size is 4 m × 4 m × 3 m, the coordinates of the 4 LEDs are (1 m, 1 m, 3 m), (1 m, 3 m, 3 m), (3 m, 1 m, 3 m), (3 m, 3 m, 3 m), and the height of the receiver PD from the ceiling is 1 m.

[0067] The cumulative distribution function (CDF) curves and histograms of the error rates of different methods were compared, with the histogram reflecting the dispersion of the error rates. The vertical axes on the left and right sides represent the number of error rates and their corresponding proportions, respectively. The RF recognition method based on RSS is denoted as RF (comparison method); Example 1 of the visible light LOS / NLOS recognition method based on TMPS and RSS GBDT in Example 1 is denoted as GBDT. It can be observed that the error rate distribution of RF exhibits a longer tail and larger errors, with errors below 16.0% accounting for 50% and errors below 44.2% accounting for 95%, mainly caused by errors in the corners of the room. The error rate distribution of GBDT, on the other hand, exhibits a very short tail and smaller errors, with errors below 4.5% accounting for 50% and errors below 22.3% accounting for 95%. Compared with the RF method, a significant performance leap has been achieved.

[0068] Table 1. Comparison of recognition performance of different methods

[0069] Table 1 compares the recognition performance of different methods. In this simulation scenario, the average accuracy of the GBDT method and the RF method are 93.60% and 82.05%, respectively. Compared with the RF method, the accuracy of the GBDT method is improved by 14.1%. The spatial recognition success rates (SISR) of the above methods are 78.56% and 46.89%, respectively. The SISR of the GBDT method is 67.5% higher than that of the RF method, indicating a significant improvement in the success rate of identifying the location of room edges.

[0070] In summary, this machine learning-based visible light line-of-sight signal recognition method achieves a breakthrough improvement over the traditional RSS-RF scheme: (1) the average accuracy is increased to 114.1% of the RF method; (2) the SISR is improved by more than 60%; and (3) it effectively solves the problem that the single RSS feature has a low success rate in recognizing line-of-sight signals at the edge in complex optical environments.

[0071] Embodiments of this application provide an electronic device. The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the machine learning-based visible light line-of-sight signal recognition method described above.

[0072] This electronic device can be any smart terminal, including computers.

[0073] In general, for the hardware structure of electronic devices, the processor can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., to execute relevant programs and implement the technical solutions provided in the embodiments of this application.

[0074] The memory can be implemented in the form of read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory and is called and executed by the processor.

[0075] Input / output interfaces are used to implement information input and output.

[0076] The communication interface is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0077] The bus transmits information between various components of a device, such as the processor, memory, input / output interfaces, and communication interfaces. The processor, memory, input / output interfaces, and communication interfaces communicate with each other within the device via the bus.

[0078] Embodiments of this application provide a computer storage medium. The computer storage medium stores computer-executable instructions for performing the machine learning-based visible light line-of-sight signal recognition method described above.

[0079] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0080] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0081] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0083] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed between each other may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms. Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0085] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A visible light line-of-sight signal recognition method based on machine learning, characterized in that, include: Receive an optical signal to be identified, the optical signal to be identified carrying a synchronous training symbol with parity symmetry; The received signal strength is obtained from the received optical signal. The received optical signal is synchronized using a timing measurement method to obtain a timing measurement value. The timing measurement peak sequence is obtained by using a sliding window with the peak value of the timing measurement value as the center. The feature quantities of the timing measurement peak sequence are extracted, including peak value, mean, standard deviation, kurtosis, skewness and median. A fused row vector is obtained by fusing the received signal strength and the feature values ​​of the timing metric peak sequence; The fused row vector is normalized to obtain the mixed feature vector; The mixed feature vector is input into the trained recognition model, and the visible light line-of-sight signal is identified based on the mixed feature vector to obtain the recognition result. The recognition result indicates whether the light signal to be identified is a line-of-sight signal or a non-line-of-sight signal. The trained recognition model is based on a gradient boosting decision tree.

2. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, characterized in that, Training the recognition model includes: Receive training optical signals, the training optical signals carrying synchronous training symbols with parity symmetry, the training optical signals being optical signals under line-of-sight conditions or optical signals under non-line-of-sight conditions; Extract the sample received signal intensity from the training optical signal; The training optical signal is synchronized using a timing measurement method to obtain sample timing measurement values. The peak value of the sample timing measurement is used as the center to extract the sample timing measurement peak sequence through a sliding window. The sample timing measurement peak sequence is statistically analyzed under line-of-sight and non-line-of-sight conditions to extract the feature quantities of the sample timing measurement peak sequence. The feature quantities of the sample timing measurement peak sequence include peak value, mean, standard deviation, kurtosis, skewness, and median. The sample fusion row vector is obtained by fusing the feature values ​​of the received signal intensity of the sample and the timing measurement peak sequence of the sample. The sample fusion row vector is set with a classification label, which indicates whether the training optical signal is an optical signal under line-of-sight conditions or an optical signal under non-line-of-sight conditions. The sample fusion row vector is normalized to obtain the sample fusion feature vector; The sample mixed feature vector is input into a gradient boosting decision tree for training to obtain a trained recognition model.

3. The visible light line-of-sight signal recognition method based on machine learning according to claim 2, characterized in that, When the training light signal is a light signal under line-of-sight conditions, the classification label is set to 0; when the training light signal is a light signal under non-line-of-sight conditions, the classification label is set to 1.

4. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, characterized in that, The timing metric is expressed as: ;in, , In the formula, d represents time. Let d be the timing metric at time d, N be the size of the sliding window, and r be the baseband signal.

5. The visible light line-of-sight signal recognition method based on machine learning according to claim 4, characterized in that, The time corresponding to the peak value of the timing metric is represented as follows: In the formula, The time corresponding to the peak value of the timing metric. For timed measurement values, S This is the search interval for the sliding window; The vector representation of the timing metric peak sequence is as follows: In the formula, The vector is used to measure the peak sequence at regular intervals, and the sliding window size is 2T+1, where T is a positive integer.

6. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, characterized in that, The peak value in the characteristic quantities of a timed peak sequence is represented as: In the formula, To measure the characteristic quantities of the peak sequence at regular intervals, For timed measurement values, This refers to the time corresponding to the peak value of the timing measurement.

7. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, characterized in that, The mean value among the characteristic quantities of a time-measured peak sequence is expressed as: The standard deviation of the characteristic quantities of a time-measured peak sequence is expressed as: In the formula, The mean of the characteristic quantities in the time-dependent peak sequence is denoted by d, where T is a positive integer and d is the time interval. The time corresponding to the peak value of the timing metric. For timed measurement values, The standard deviation is used to measure the characteristic quantity of a peak sequence at regular intervals.

8. The visible light line-of-sight signal recognition method based on machine learning according to claim 7, characterized in that, Kurtosis, a characteristic of a time-measured peak sequence, is represented as: The skewness in the characteristic quantities of a time-measured peak sequence is expressed as: In the formula, To measure the kurtosis, a characteristic property of a peak sequence, at regular intervals. This is used to measure the skewness of the characteristic quantities in a peak sequence at regular intervals.

9. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, characterized in that, The median of the features in a time-measured peak sequence is expressed as: ; It represents the (T+1)th element in the order of elements rearranged in ascending or descending order in the timing measurement peak sequence, where T is a positive integer.

10. The visible light line-of-sight signal recognition method based on machine learning according to claim 1, characterized in that, The fused row vector is normalized using the following formula to obtain the fused feature vector: ;in, To merge row vectors, For mixed feature vectors, To merge the mean of the row vectors, This is to fuse the variance of the row vectors.

Images