Waveform amplitude characteristic extraction method based on entropy amplitude characteristic analysis

Abstract

The invention relates to the technical field of waveform signal feature extraction, and particularly discloses a waveform amplitude feature extraction method based on entropy amplitude feature analysis. In order to solve the problem that a waveform feature extraction method in the prior art cannot give consideration to amplitude sensitivity and complexity evaluation, entropy amplitude and compression entropy difference indexes reflecting waveform amplitude characteristics are generated through sliding window segmentation, sub-waveform compression and entropy amplitude difference calculation. The method comprises the following steps: S10, dividing an input waveform signal into a plurality of monotonically changing sub-waveform segments; s20, carrying out amplitude compression on each sub-waveform segment to enable a peak-to-peak value of each sub-waveform segment to be matched with preset quantization precision, and calculating an entropy amplitude of each sub-waveform segment; s30, traversing the input waveform signal through a sliding window, accumulating entropy amplitudes of all sub-waveform segments in the window, and generating window-level amplitude features; s40, amplifying the input waveform signal to an entropy amplitude critical point, and calculating a difference value between the amplified global information entropy and the original global information entropy to obtain a compression entropy difference; wherein the window-level amplitude feature and the compression entropy difference are used as input features of the mode recognition model. The entropy amplitude sum and the compression entropy difference can serve as independent features to be input into a mode recognition model, and the problem that traditional time domain features are not sensitive to amplitude changes is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of waveform signal feature extraction and pattern recognition, and specifically to a method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis. The method generates window-level features and global compression entropy differences reflecting the amplitude distribution characteristics of the waveform through sliding window segmentation, sub-waveform compression, and entropy amplitude difference calculation, and is used for pattern recognition tasks. Background Art

[0002] Although information entropy can measure the complexity of a waveform, its calculation is limited by the quantization accuracy of the sensor. In the prior art, information entropy only reflects the distribution uniformity of signal values and ignores the influence of amplitude scaling on the entropy value. For example, waveforms with the same shape may have different entropy values due to different amplitudes, but the prior methods do not establish a quantitative relationship between amplitude and entropy value, resulting in the inability to infer amplitude characteristics from the entropy value. In addition, the existing entropy calculation methods do not consider the local amplitude changes of the waveform (such as the amplitude accumulation effect within the sliding window), making it difficult to generate fine-grained amplitude features.

[0003] Therefore, there is an urgent need for a feature extraction method that takes into account both amplitude sensitivity and complexity evaluation, and provides more sufficient features for pattern recognition tasks by quantifying local and global amplitude distribution characteristics. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis, which solves the problem in the prior art that information entropy only reflects the distribution uniformity of signal values and ignores the influence of amplitude scaling on the entropy value, resulting in insufficient utilization of information entropy features.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis; The method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis first divides the input waveform signal into multiple monotonically changing sub-waveform segments; further, compresses the amplitude of each sub-waveform segment so that its peak-to-peak value matches the predetermined quantization accuracy, and calculates the entropy amplitude of each sub-waveform segment; then traverses the input waveform signal through a sliding window, accumulates the entropy amplitudes of all sub-waveform segments within the window, and generates window-level amplitude features; finally, amplifies the input waveform signal to the entropy amplitude critical point, calculates the difference between the global information entropy after amplification and the original global information entropy, and obtains the compression entropy difference; wherein, the window-level amplitude features and the compression entropy difference are used as input features for the pattern recognition model.

[0006] The compression of the amplitude of each sub-waveform segment so that its peak-to-peak value matches the predetermined quantization accuracy and the calculation of the entropy amplitude of each sub-waveform segment is characterized by including the following steps: Step 1, fundamental wave signal modeling, taking a segment with a peak-to-peak value equal to the predetermined quantization accuracy ( is an integer multiple of the sensor measurement error, such as = 0.000001), and a linear monotonic waveform signal that satisfies the high-frequency sampling condition (sampling interval Δt ≤ 0.01 s) is defined as the fundamental wave ; Step 2: Derivation of the fundamental wave information entropy. Divide the total sampling time of the sensor into N intervals Δt. Within each range with the quantization accuracy as the signal value interval size, distribute , , ……, sampling points. The expression for the fundamental wave information entropy is: ; Step 3: Derivation of the amplified wave information entropy. Amplify the fundamental wave by A times to obtain the amplified wave . Its information entropy expression is: ; Step 4: Definition of entropy amplitude. The entropy amplitude is defined as the difference in information entropy between the amplified wave and the fundamental wave: .

[0007] By traversing the input waveform signal through a sliding window and accumulating the entropy amplitude values of all sub-waveform segments within the window, window-level amplitude features are generated. Specifically, it includes the following steps: Step 1: The window length L and step size S are dynamically adjusted according to the pattern recognition task, and , ; Step 2: Within the window range, divide the waveform signal into multiple monotonically increasing or decreasing sub-waveform segments according to the extreme points of the waveform. Compress each sub-waveform segment to the fundamental wave (peak-to-peak value = ), and record the compression multiple ; Step 3: Calculate the sum of the entropy amplitudes of all sub-waveforms within the window: , where u is the number of sub-waveforms within the window.

[0008] Amplify the input waveform signal to the entropy amplitude critical point, calculate the difference between the global information entropy after amplification and the original global information entropy to obtain the compression entropy difference. Specifically, it includes the following steps: Step 1: The information entropy when the waveform signal is amplified to the entropy amplitude critical point is , where N is the total number of sampling points; Step 2: The original global information entropy is calculated through the following expression: ; Step 3: The compression entropy difference is calculated through the following expression: , where is the information entropy when the waveform signal is amplified to the entropy amplitude critical point, and

[0009] A method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis. It has the following beneficial effects: (1), Through the definition of fundamental wave entropy and entropy amplitude, the waveform amplitude scaling factor A is converted into the entropy amplitude , so that the eigenvalue directly quantifies the local amplitude intensity; further combined with the window-level entropy amplitude and generated by the sliding window mechanism, the amplitude cumulative effect in the waveform can be captured; (2), Based on the fundamental wave compression operation of the sensor quantization accuracy , the entropy amplitude calculation is compatible with the actual measurement error, avoiding overfitting the ideal signal model. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a schematic diagram for defining the fundamental wave and amplified wave; Figure 2 is a schematic diagram of the sliding window sliding on the global waveform signal; Figure 3 is a complete flow chart of a method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0012] In one embodiment, as Figure 3 shown, it includes the following steps: 1. Divide the input waveform signal into multiple monotonically changing sub-waveform segments, where the sampling frequency of the input waveform signal is 100 Hz; 2. Perform amplitude compression on each sub-waveform segment to make its peak-to-peak value match the predetermined quantization accuracy, and calculate the entropy amplitude of each sub-waveform segment: (1), Take a waveform signal with a peak-to-peak value of the quantization accuracy = 0.000001, the first point at the quantization threshold, linear, and monotonic, and define it as the fundamental wave , and deduce the information entropy expression of the fundamental wave based on the actual measurement situation of the sensor, and obtain the information entropy expression of the fundamental wave as: ; (2) Establish a plane coordinate system with the first point of the fundamental wave signal as the origin, and obtain an amplified wave by amplifying the fundamental wave by A times. , as Figure 1 shown, derive the information entropy expression of the amplified wave based on the sensor measurement situation, and obtain the information entropy expression of the amplified wave as: ; (3) The entropy amplitude is defined as the difference in information entropy between the amplified wave and the fundamental wave: ; 3. Traverse the input waveform signal through a sliding window. As Figure 2 shown, accumulate the entropy amplitude values of all sub-waveform segments within the window to generate window-level amplitude features: (1) According to the device sampling frequency of 100 Hz, set the size of the sliding window to 3 s and the step size to 3 s; (2) Within the window range, divide the waveform signal into multiple monotonically increasing or decreasing sub-waveform segments according to the extreme points of the waveform signal, and compress each sub-waveform segment to the fundamental wave (peak-to-peak value = ), and record the compression ratio ; (3) Calculate the sum of the entropy amplitudes of all sub-waveforms within the window: , where u is the number of sub-waveforms in the window; 4. Amplify the input waveform signal to the entropy amplitude critical point, calculate the difference between the global information entropy after amplification and the original global information entropy to obtain the compression entropy difference; (1) The information entropy when the waveform signal is amplified to the entropy amplitude critical point is , where N is 14719000; (2) The original global information entropy is calculated through the following expression:

[0013] (2) The difference between the information entropy of the waveform signal at the entropy amplitude critical point and the global entropy amplitude is defined as the compression entropy difference. Specifically, the expression of the compression entropy difference is: , where is the compression entropy difference, is the information entropy at the entropy amplitude critical point, is the global entropy amplitude.

[0014] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis, comprising the following steps: S10 divides the input waveform signal into multiple monotonically changing sub-waveform segments; S20 compresses the amplitude of each sub-waveform segment so that its peak-to-peak value matches the predetermined quantization accuracy, and calculates the entropy amplitude of each sub-waveform segment; S30 traverses the input waveform signal through a sliding window, accumulates the entropy amplitudes of all sub-waveform segments in the window, and generates a window-level amplitude feature; S40 amplifies the input waveform signal to the entropy amplitude critical point, calculates the difference between the amplified global information entropy and the original global information entropy, and obtains the compressed entropy difference; wherein the window-level amplitude feature and the compressed entropy difference are used as input features of the pattern recognition model.

2. The method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis according to claim 1, characterized in that: The S20 performs amplitude compression on each sub-waveform segment so that its peak-to-peak value matches the predetermined quantization precision, and calculates the entropy amplitude of each sub-waveform segment by the following method: compressing the sub-waveform segment to the fundamental wave, the compression multiple A is the sum of the sub-waveform segment amplitude and the quantization precision The peak-to-peak value of the fundamental wave is equal to the predetermined quantization accuracy. , and meet the high-frequency sampling condition (sampling interval Δt ≤ 0.01 seconds); divide the total sampling time of the sensor into N sampling intervals Δt, and divide the signal value into several intervals according to the quantization accuracy, and each corresponding interval is distributed with , ,……, sampling points, so the information entropy formula It can be deduced as: Information entropy of fundamental wave The expression is: , where N is the total number of sampling points; the information entropy of the amplified wave The expression is: , where N is the total number of sampling points; The entropy amplitude of the sub-waveform segment is ,in is the compression factor of the ith sub-waveform.

3. The method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis according to claim 1, characterized in that: The entropy amplitude critical point refers to the point where the waveform signal amplitude is amplified to The state after the magnification is multiplied is as follows: 1) When the magnification is lower than the critical value, there are at least two sampling points with the same signal value; 2) When the magnification reaches or exceeds the critical value, the signal values ​​of all sampling points are unique; 3) At this critical magnification, the information entropy of the signal reaches a maximum value; 4) When the magnification continues to increase, the information entropy will remain stable and no longer increase.

4. The method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis according to claim 1, characterized in that: The difference between the amplified global information entropy and the original global information entropy is calculated to obtain the compressed entropy difference by the following steps: the information entropy when the waveform signal is amplified to the critical point of the entropy amplitude for , where N is the total number of sampling points; the original global information entropy Calculated by the following expression: ; Compression entropy difference Calculated by the following expression: ,in is the information entropy when the waveform signal is amplified to the critical point of entropy amplitude, is the original global information entropy.

5. The method for extracting waveform amplitude characteristics based on entropy amplitude feature analysis according to claim 1, characterized in that: The step S40 amplifies the input waveform signal to the entropy amplitude critical point determined by the following steps: To achieve the minimum resolution, gradually increase the magnification A of the waveform signal until it meets the following conditions: ,in It is the minimum amplitude difference between adjacent sampling points of the original waveform signal.

Images