阅读说明：本技术 一种基于混沌智能图像识别的弱信号检测方法及装置 (Weak signal detection method and device based on chaotic intelligent image recognition ) 是由 杨伏洲 沈梦琳 何思瑾 梁世远 彭晶 于 2020-01-10 设计创作，主要内容包括：本发明公开一种基于混沌智能图像识别的弱信号检测方法及装置,所述方法包括：构建Duffing振子信号检测系统,获取不同状态的Duffing振子相位图；调节Duffing振子信号检测系统的系数,使系统处于混沌状态到大尺度周期状态的临界状态,输入待测信号后,得到待测信号相空间状态图；对所述不同状态的Duffing振子相位图做预处理,得到预处理好的图像样本；提取所述图像样本的HOG特征和GLCM特征,组成图像样本的特征向量；通过所述图像样本的特征向量训练各SVM分类器,得到训练好的SVM模型；提取待测信号相空间状态图的HOG和GLCM特征并组成待测信号相空间状态图的特征向量；通过训练好的SVM模型进行弱信号检测。本发明对在强噪声环境下弱信号检测有更高的噪声容限且准确率高。(The invention discloses a weak signal detection method and a device based on chaotic intelligent image recognition, wherein the method comprises the following steps: constructing a Duffing oscillator signal detection system, and acquiring Duffing oscillator phase diagrams in different states; adjusting the coefficient of a Duffing oscillator signal detection system to enable the system to be in a critical state from a chaotic state to a large-scale periodic state, and inputting a signal to be detected to obtain a phase space state diagram of the signal to be detected; preprocessing the Duffing oscillator phase diagrams in different states to obtain preprocessed image samples; extracting HOG characteristics and GLCM characteristics of the image sample to form a characteristic vector of the image sample; training each SVM classifier through the feature vector of the image sample to obtain a trained SVM model; extracting HOG and GLCM characteristics of a signal phase space state diagram to be detected and forming a characteristic vector of the signal phase space state diagram to be detected; and weak signal detection is carried out through the trained SVM model. The invention has higher noise tolerance and high accuracy for weak signal detection in a strong noise environment.)

1. A weak signal detection method based on chaotic intelligent image recognition is characterized by comprising the following steps:

s1, constructing a Duffing oscillator signal detection system, and acquiring Duffing oscillator phase diagrams in different states;

s2, adjusting the coefficient of the Duffing oscillator signal detection system to enable the system to be in a critical state from a chaotic state to a large-scale periodic state, and inputting a signal to be detected to obtain a phase space state diagram of the signal to be detected;

s3, preprocessing the Duffing oscillator phase diagrams in different states, removing coordinate frames, and respectively labeling the Duffing oscillator phase diagrams in different states according to different categories to obtain preprocessed image samples;

s4, extracting HOG features and GLCM features of the image sample, and forming feature vectors of the image sample by the extracted HOG features and GLCM features;

s5, combining a plurality of SVM classifiers for multi-classification, and training each SVM classifier through the feature vector of the image sample to obtain a trained SVM model;

s6, extracting HOG and GLCM characteristics of the phase space state diagram of the signal to be detected in the same way as the step S4 and forming a characteristic vector of the phase space state diagram of the signal to be detected;

and S7, inputting the feature vector of the phase-space state diagram of the signal to be detected into the trained SVM model to obtain a weak signal detection result.

2. The method for detecting weak signals based on chaotic intelligent image recognition according to claim 1, wherein in step S1, the Duffing oscillator detection signal model is:

where k is the damping coefficient and ω is the signal frequency, - ω x + ω x³For the nonlinear restoring force, f is the system internal driving force, fcos (ω t) is the driving force amplitude, and s (t) + n (t) is the input signal.

3. The method for detecting weak signals based on chaotic intelligent image recognition according to claim 1, wherein in the step S2, the preprocessed image samples are respectively labeled as "homoclinic", "bifurcation", "chaos", and "large period".

4. The method for detecting weak signals based on chaotic intelligent image recognition according to claim 1, wherein the step S4 specifically comprises:

s41, dividing the image into a plurality of cell units, constructing a gradient direction histogram for each cell unit, combining the cell units into large blocks, normalizing the gradient histograms in the blocks and collecting HOG characteristics;

s42, solving a gray level co-occurrence matrix in a preset neighborhood for each pixel on the image, solving statistics of the corresponding texture image by the gray level co-occurrence matrix, wherein the statistics comprise energy ASM, entropy ENT, contrast CON and correlation IDE, and forming GLCM (global level of similarity) characteristics of the image by the statistics;

and S43, combining the extracted HOG and GLCM characteristics into a characteristic vector of the image sample.

5. The weak signal detection method based on chaotic intelligent image recognition according to claim 4, wherein the step S42 specifically comprises:

counting a pixel (x + D) with a distance D and a gray level J from a pixel position (x, y) with a gray level i in an image_x，y+D_y) The number of simultaneous occurrences p (i, j, d, θ), the mathematical expression is:

p(i,j,d,θ)＝[(x,y),(x+D_x,y+D_x)|f(x,y)＝i,f(x+D_x,y+D_y)＝j]

wherein: x, y is 0,1,2, and N-1 is the pixel coordinates in the image; i, j ═ 0,1, 2.., L-1 is the gray level; d_x，D_yIs the position offset; d is the step length for generating the gray level co-occurrence matrix; generating a direction theta, namely taking four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, so as to generate gray level co-occurrence matrixes in different directions; carrying out normalization processing on the gray level co-occurrence matrix, and calculating statistics of corresponding texture images, wherein the statistics comprises energy ASM, entropy ENT, contrast CON and correlation IDE:

and forming the statistics into GLCM characteristics of the image.

6. A weak signal detection device based on chaotic intelligent image recognition is characterized by comprising:

a vibrator system construction module: the method is used for constructing a Duffing oscillator signal detection system and acquiring Duffing oscillator phase diagrams in different states;

the oscillator system adjusting module: the system is used for adjusting the coefficient of the Duffing oscillator signal detection system, enabling the system to be in a critical state from a chaotic state to a large-scale periodic state, and obtaining a phase space state diagram of a signal to be detected after the signal to be detected is input;

a sample image preprocessing module: the image preprocessing module is used for preprocessing the Duffing oscillator phase diagrams in different states, removing coordinate frames, and respectively labeling the Duffing oscillator phase diagrams in different states according to different categories to obtain preprocessed image samples;

an image feature extraction module: the HOG characteristic and the GLCM characteristic of the image sample are extracted, and the extracted HOG characteristic and the GLCM characteristic form a characteristic vector of the image sample;

an SVM model training module: the SVM classifier is used for combining a plurality of SVM classifiers for multi-classification, and each SVM classifier is trained through the feature vector of the image sample to obtain a trained SVM model;

weak signal detection module: the HOG and GLCM characteristics used for extracting the phase space state diagram of the signal to be detected form a characteristic vector of the phase space state diagram of the signal to be detected; and inputting the characteristic vector of the phase-space state diagram of the signal to be detected into a trained SVM model to obtain a weak signal detection result.

7. The weak signal detection method based on chaotic intelligent image recognition of claim 6, wherein in the oscillator system construction module, the Duffing oscillator signal detection system is as follows:

where k is the damping coefficient and ω is the signal frequency, - ω x + ω x³For the nonlinear restoring force, f is the system internal driving force, fcos (ω t) is the driving force amplitude, and s (t) + n (t) is the input signal.

8. The method for detecting the weak signal based on the chaotic intelligent image recognition is characterized in that in the sample image preprocessing module, the preprocessed image samples are respectively marked as four categories of homoclinic, bifurcation, chaos and large period.

9. The weak signal detection method based on chaotic intelligent image recognition according to claim 6, wherein the image feature extraction module specifically comprises:

an HOG feature extraction unit: dividing the image into a plurality of cell units, constructing a gradient direction histogram for each cell unit, combining the cell units into large blocks, normalizing the gradient histogram in the blocks and collecting HOG characteristics;

GLCM characteristic extraction unit: solving a gray level co-occurrence matrix in a preset neighborhood for each pixel on the image, solving statistics of the corresponding texture image by the gray level co-occurrence matrix, wherein the statistics comprise energy ASM, entropy ENT, contrast CON and correlation IDE, and forming GLCM characteristics of the image by the statistics;

a feature combination unit: and forming the extracted HOG and GLCM characteristics into a characteristic vector of the image sample.

Technical Field

The invention belongs to the field of signal detection while drilling, and relates to two disciplines of signal detection and image processing. Mainly relates to a weak signal detection method based on intelligent identification of chaotic images.

Background

The acoustic signal transmission process is influenced by environmental factors in the logging-while-drilling work, and noises while drilling can be generated by noises generated when a drill rod moves, collision of the drill rod and a well wall, impact of a drill bit on a stratum and the like. These noises cause serious disturbances in the detection of the sound waves during their transmission. Weak signal detection based on chaos theory is a hot spot of signal detection in recent years. Under the background of strong noise, the detection of weak signals is realized by utilizing the initial value sensitivity of the chaotic system and the strong immunity to the noise. And fixing the damping coefficient k and adjusting the system driving force f to obtain a threshold Fd of a critical chaotic state in the critical state of the Duffing vibrator system phase space. The signal to be measured is added to the vibrator system so that the driving force becomes time dependent, i.e. f (t). And observing the phase space change of the system, when a weak signal exists, the system can be transited from a critical state to a large-scale periodic state, otherwise, the system can be always in the critical or chaotic state and cannot enter the large-scale periodic state. Based on a weak signal detection mode, various methods exist in the signal detection while drilling technology: correlation detection, wavelet analysis, digital phase-sensitive demodulation techniques, and the like.

The correlation detection method is based on the extraction of weak signals under the background of strong noise based on the correlation principle, and can detect the signals submerged in the noise by performing correlation operation on the target signal and the reference signal by utilizing the characteristics of the irrelevance of the signals and the noise, the noise and the complete correlation of the target signal and the provided reference signal. Because the reference signal is determined, the amplitude and the phase of the signal to be detected can be easily obtained, and therefore the detection of the weak signal is realized.

Wavelet analysis is localized time domain analysis that characterizes a signal by a combination of time and frequency domains, weak signals and noise transformed into the wavelet transform domain by a multi-scale wavelet. The wavelet system of the signal is extracted as much as possible under each scale to remove the wavelet coefficient of the noise, and the weak signal is recovered from the noise-reduced coefficient through the wavelet middle, so that the effect of removing the noise is achieved, and the detection of the weak signal is realized.

The digital phase-sensitive demodulation technology is a widely applied weak signal detection technology, and realizes the detection of the amplitude and the phase of a weak signal by utilizing high-speed floating point digital signal processing and a high-speed field programmable gate array chip to respectively process the real part and the imaginary part of an acquired signal. However, the above techniques are all studied for existing filters that detect signals linearly and smoothly and require high quality factors, and noise margin, weak signal detection accuracy, and the like are all to be improved.

Disclosure of Invention

Based on the defects, the invention provides a weak signal detection method and device based on chaotic intelligent image recognition, which extracts an image characteristic value and judges the state of a Duffing oscillator phase diagram through image recognition by analyzing the system driving force f critical range of each state of the Duffing oscillator phase diagram, thereby achieving the purpose of detecting a weak signal under a strong noise background.

The invention provides a weak signal detection method based on chaotic intelligent image recognition, which comprises the following steps:

s1, constructing a Duffing oscillator signal detection system, and acquiring Duffing oscillator phase diagrams in different states;

s2, adjusting the coefficient of the Duffing oscillator signal detection system to enable the system to be in a critical state from a chaotic state to a large-scale periodic state, and inputting a signal to be detected to obtain a phase space state diagram of the signal to be detected;

s3, preprocessing the Duffing oscillator phase diagrams in different states, removing coordinate frames, and respectively labeling the Duffing oscillator phase diagrams in different states according to different categories to obtain preprocessed image samples;

s4, extracting HOG features and GLCM features of the image sample, and forming feature vectors of the image sample by the extracted HOG features and GLCM features;

s5, combining a plurality of SVM classifiers for multi-classification, and training each SVM classifier through the feature vector of the image sample to obtain a trained SVM model;

s6, extracting HOG and GLCM characteristics of the phase space state diagram of the signal to be detected in the same way as the step S4 to form a characteristic vector of the phase space state diagram of the signal to be detected;

and S7, inputting the feature vector of the phase-space state diagram of the signal to be detected into the trained SVM model to obtain a weak signal detection result.

Preferably, in step S1, the Duffing oscillator signal detection system is:

where k is the damping coefficient and ω is the signal frequency, - ω x + ω x³For the nonlinear restoring force, f is the system internal driving force, fcos (ω t) is the driving force amplitude, and s (t) + n (t) is the input signal.

Preferably, in step S2, the preprocessed image samples are labeled as "homoclinic", "bifurcation", "chaos", and "large period", respectively.

Preferably, the step S4 specifically includes:

s41, dividing the image into a plurality of cell units, constructing a gradient direction histogram for each cell unit, combining the cell units into large blocks, normalizing the gradient histograms in the blocks and collecting HOG characteristics;

s42, solving a gray level co-occurrence matrix in a preset neighborhood for each pixel on the image, solving statistics of the corresponding texture image by the gray level co-occurrence matrix, wherein the statistics comprise energy ASM, entropy ENT, contrast CON and correlation IDE, and forming GLCM (global level of similarity) characteristics of the image by the statistics; (ii) a

And S43, combining the extracted HOG and GLCM characteristics into a characteristic vector of the image sample.

Preferably, the step S42 specifically includes:

counting a pixel (x + D) with a distance D and a gray level J from a pixel position (x, y) with a gray level i in an image_x，y+D_y) The number of simultaneous occurrences p (i, j, d, θ), the mathematical expression is:

p(i,j,d,θ)＝[(x,y),(x+D_x,y+D_x)|f(x,y)＝i,f(x+D_x,y+D_y)＝j]

wherein: x, y is 0,1,2, and N-1 is the pixel coordinates in the image; i, j ═ 0,1, 2.., L-1 is the gray level; d_x，D_yIs the position offset; d is the step length for generating the gray level co-occurrence matrix; generating a direction theta, namely taking four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, so as to generate gray level co-occurrence matrixes in different directions; carrying out normalization processing on the gray level co-occurrence matrix, and calculating statistics of corresponding texture images, wherein the statistics comprises energy ASM, entropy ENT, contrast CON and correlation IDE:

and forming the statistics into GLCM characteristics of the image.

In a second aspect of the present invention, a weak signal detection apparatus based on chaotic intelligent image recognition is provided, the apparatus comprising:

a vibrator system construction module: the method is used for constructing a Duffing oscillator signal detection system and acquiring Duffing oscillator phase diagrams in different states;

the oscillator system adjusting module: the system is used for adjusting the coefficient of the Duffing oscillator signal detection system, enabling the system to be in a critical state from a chaotic state to a large-scale periodic state, and obtaining a phase space state diagram of a signal to be detected after the signal to be detected is input;

a sample image preprocessing module: the image preprocessing module is used for preprocessing the Duffing oscillator phase diagrams in different states, removing coordinate frames, and respectively labeling the Duffing oscillator phase diagrams in different states according to different categories to obtain preprocessed image samples;

an image feature extraction module: the HOG characteristic and the GLCM characteristic of the image sample are extracted, and the extracted HOG characteristic and the GLCM characteristic form a characteristic vector of the image sample;

an SVM model training module: the SVM classifier is used for combining a plurality of SVM classifiers for multi-classification, and each SVM classifier is trained through the feature vector of the image sample to obtain a trained SVM model;

weak signal detection module: the HOG and GLCM characteristics used for extracting the phase space state diagram of the signal to be detected form a characteristic vector of the phase space state diagram of the signal to be detected; and inputting the characteristic vector of the phase-space state diagram of the signal to be detected into the trained SVM model to obtain a weak signal detection result.

Preferably, in the oscillator system building module, the Duffing oscillator signal detection system is:

where k is the damping coefficient and ω is the signal frequency, - ω x + ω x³For the nonlinear restoring force, f is the system internal driving force, fcos (ω t) is the driving force amplitude, and s (t) + n (t) is the input signal.

Preferably, in the sample image preprocessing module, the preprocessed image samples are respectively labeled as "homoclinic", "bifurcation", "chaos", and "large period".

Preferably, the image feature extraction module specifically includes:

an HOG feature extraction unit: dividing the image into a plurality of cell units, constructing a gradient direction histogram for each cell unit, combining the cell units into large blocks, normalizing the gradient histogram in the blocks and collecting HOG characteristics;

GLCM characteristic extraction unit: solving a gray level co-occurrence matrix in a preset neighborhood for each pixel on the image, solving statistics of the corresponding texture image by the gray level co-occurrence matrix, wherein the statistics comprise energy ASM, entropy ENT, contrast CON and correlation IDE, and forming GLCM characteristics of the image by the statistics;

a feature combination unit: and forming the extracted HOG and GLCM characteristics into a characteristic vector of the image sample.

The invention provides a weak signal detection method and device based on chaotic intelligent image recognition, which have the beneficial effects that:

1) the invention has higher noise tolerance to weak signal detection in strong noise environment due to the strong immunity of the chaotic system to the initial value sensitivity and the noise;

2) the SVM model has sparsity, namely a good classification effect can be obtained by a small amount of samples;

3) compared with the prior art, the method is simpler;

4) the traditional detection system is mainly based on a linear theory, and the basic characteristics presented by the chaos theory are nonlinearity, unbalance and nonuniformity and higher accuracy.

Drawings

In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the technical description of the present invention will be briefly introduced below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive labor.

FIG. 1 is a schematic flow chart of a weak signal detection method based on chaotic intelligent image recognition according to the present invention;

FIG. 2 is a diagram of x-y phase of Duffing oscillator critical state without added signal;

fig. 3 shows that the amplitude of the added signal is a 1 × 10 in the critical state^-5V, ω ═ 2. pi. 1000rad/s, large periodic regime x-y phase of the systemA bitmap;

fig. 4 shows that the amplitude of the added signal is a 1 × 10 in the critical state^-5V, ω ═ 2 × pi 1000rad/s, noise amplitude 1 × 10^-6Large period state x-y phase diagram of-10 dBW in dB, snr;

FIG. 5 is a diagram of the x-y phase of a chaotic state with an added noise amplitude of 0.001dB in a critical state;

FIG. 6 is a time domain and frequency spectrum diagram of randomly acquired 1s single frequency signals;

FIG. 7 is a time domain diagram of a single frequency signal s (t) and noise n (t) added to Duffing oscillators;

fig. 8 is a diagram of x-y phase of Duffing's oscillator in a large period state after adding the signal and noise of fig. 7 in a critical state.

Detailed Description

In order to make the objects, features and advantages of the present invention more obvious and understandable, 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, and it is obvious that the embodiments described below are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The idea of the invention is that: compared with the traditional weak signal detection methods such as correlation detection, wavelet analysis and the like, the chaos theory is combined with intelligent image recognition, and due to the characteristics of the sensitivity of the chaotic oscillator to weak signals and strong immunity to noise, the state of the phase diagram changes differently under the critical state of the Duffing oscillator. Through image intelligent identification, the purpose of detecting weak signals under the background of strong noise is achieved. Therefore, the intelligent chaotic image identification method is used for detection, the noise tolerance can be improved, and the method is more accurate.

Referring to fig. 1, the present invention provides a weak signal detection method based on chaotic intelligent image recognition, which includes the following steps:

s1, constructing a Duffing oscillator signal detection system, and acquiring Duffing oscillator phase diagrams in different states;

the Duffing oscillator detection signal model is as follows:

where k is the damping coefficient and ω is the signal frequency, - ω x + ω x³For the nonlinear restoring force, f is the system internal driving force, fcos (ω t) is the driving force amplitude, and s (t) + n (t) is the input signal.

S2, adjusting the coefficient of the Duffing oscillator signal detection system to enable the system to be in a critical state from a chaotic state to a large-scale periodic state, and inputting a signal to be detected to obtain a phase space state diagram of the signal to be detected;

and taking proper k and omega, and simultaneously taking f as a chaotic critical threshold value of the driving force in the system, so that the system is in a critical state from a chaotic state to a large-scale periodic state, and inputting a signal to be measured to obtain a phase space state diagram of the system at the moment.

S3, preprocessing the Duffing oscillator phase diagrams in different states, removing coordinate frames, and respectively labeling the Duffing oscillator phase diagrams in different states according to different categories to obtain preprocessed image samples;

putting different types of phase space state diagrams as much as possible under different catalogs according to different categories, respectively marking the phase space state diagrams as four types of homoclinic, bifurcation, chaos and large period to obtain preprocessed image samples, and using the preprocessed image samples as training samples of the SVM model.

S4, extracting HOG features and GLCM features of the image sample, and forming feature vectors of the image sample by the extracted HOG features and GLCM features;

further, the step S4 specifically includes the following sub-steps:

s41, dividing the image into a plurality of cell units, constructing a gradient direction histogram for each cell unit, combining the cell units into a large block (block), normalizing the gradient histogram in the block and collecting HOG characteristics;

and S42, calculating a gray level co-occurrence matrix in a preset neighborhood for each pixel on the image, and calculating the statistic of the corresponding texture image according to the gray level co-occurrence matrix, wherein the statistic comprises energy ASM, entropy ENT, contrast CON and correlation IDE. The method specifically comprises the following steps:

counting a pixel (x + D) with a distance D and a gray level J from a pixel position (x, y) with a gray level i in an image_x，y+D_y) The number of simultaneous occurrences p (i, j, d, θ), the mathematical expression is:

p(i,j,d,θ)＝[(x,y),(x+D_x,y+D_x)|f(x,y)＝i,f(x+D_x,y+D_y)＝j]

wherein: x, y is 0,1,2, …, M-1 is the pixel coordinate in the image, M is the pixel number; i, j is 0,1,2, …, L-1 is the gray scale level, L is the number of gray scale levels; d_x，D_yIs the position offset; d is the step length for generating the gray level co-occurrence matrix; generating a direction theta, namely taking four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, so as to generate gray level co-occurrence matrixes in different directions; carrying out normalization processing on the gray level co-occurrence matrix, and calculating statistics of corresponding texture images, wherein the statistics comprises energy ASM, entropy ENT, contrast CON and correlation IDE:

and forming the statistics into GLCM characteristics of the image.

And S43, combining the extracted HOG and GLCM characteristics into a characteristic vector of the image sample. Specifically, the extracted HOG features and GLCM features are combined into a feature vector for image classification.

S5, combining a plurality of SVM classifiers for multi-classification, and training each SVM classifier through the feature vector of the image sample to obtain a trained SVM model;

the multi-class classification problem is solved by combining a plurality of SVM classifiers: (x)_k,y_k),k＝1,2,…,N；x_k∈Rⁿ；y_k∈ (-1,1) is a sample set, a classification surface wx + b is constructed for the sample set to be 0, the distance between two classified samples is maximized, wherein w and x are n-dimensional vectors, g (x) wx + b is a general form of a linear discriminant function, w and b are normalized by enlarging and reducing the same times, the sample closest to the classification surface satisfies | g (x) | ≧ 1, at this time, the two types of samples satisfy | g (x) | ≧ 1, and the classification interval of the two types of samples is | g (x) | ≧ 1The optimal plane solving problem is converted into an optimization solving problem:

wherein, a_kIs a constrained lagrange multiplier. For the following steps:

the partial derivation in (a) results in a dual form:

the optimal solution of the simultaneous optimization problem must satisfy the following condition a_ky_k(w^Tx_k+b)-1＝0，B can be obtained.

Due to the fact thatThe result can be found as a large part_k＝0，a_kThe samples corresponding to not equal to 0 are called support vectors, and the classification plane is determined by the support vectors.

The support vector machine implements nonlinear classification by mapping the input vectors to a high-dimensional feature space through some kind of pre-selected nonlinear mapping (kernel function), and constructing an optimal classification hyperplane in the space. The classification plane of the transformed space isSimilar to the linear case, the optimization equation is:

wherein the content of the first and second substances,is the transformed spatial inner product. Constructing a kernel function K (x)_k,x_l) To make it equal toSubstitution solutionA total of 4 x (4-1) to 12 planes were constructed from the 4 types of samples.

S6, extracting HOG and GLCM characteristics of the phase space state diagram of the signal to be detected in the same way as the step S4 to form a characteristic vector of the phase space state diagram of the signal to be detected;

and S7, inputting the feature vector of the phase-space state diagram of the signal to be detected into the trained SVM model to obtain a weak signal detection result.

Specifically, a signal to be detected is input in a Duffing oscillator critical state, the phase space change of the system is observed, when a weak signal exists, the system can transit from the critical state to a large-scale periodic state, otherwise, the system can be in the critical or chaotic state all the time and cannot enter the large-scale periodic state. Weak signals are detected by identifying the system phase space state diagram.

Corresponding to the embodiment of the method, the invention also provides a weak signal detection device based on chaotic intelligent image recognition, which comprises the following steps:

a vibrator system construction module: the method is used for constructing a Duffing oscillator signal detection system and acquiring Duffing oscillator phase diagrams in different states; further, in the oscillator system building module, the Duffing oscillator signal detection system is:

where k is the damping coefficient and ω is the signal frequency, - ω x + ω x³For the nonlinear restoring force, f is the system internal driving force, fcos (ω t) is the driving force amplitude, and s (t) + n (t) is the input signal.

The oscillator system adjusting module: the system is used for adjusting the coefficient of the Duffing oscillator signal detection system, enabling the system to be in a critical state from a chaotic state to a large-scale periodic state, and obtaining a phase space state diagram of a signal to be detected after the signal to be detected is input;

a sample image preprocessing module: the image preprocessing module is used for preprocessing the Duffing oscillator phase diagrams in different states, removing coordinate frames, and respectively labeling the Duffing oscillator phase diagrams in different states according to different categories to obtain preprocessed image samples; further, in the sample image preprocessing module, the preprocessed image samples are respectively labeled as "homoclinic", "bifurcation", "chaos", and "large period".

An image feature extraction module: the HOG characteristic and the GLCM characteristic of the image sample are extracted, and the extracted HOG characteristic and the GLCM characteristic form a characteristic vector of the image sample; further, the image feature extraction module specifically includes:

an HOG feature extraction unit: dividing the image into a plurality of cell units, constructing a gradient direction histogram for each cell unit, combining the cell units into large blocks, normalizing the gradient histogram in the blocks and collecting HOG characteristics;

GLCM characteristic extraction unit: solving a gray level co-occurrence matrix in a preset neighborhood for each pixel on the image, solving statistics of the corresponding texture image by the gray level co-occurrence matrix, wherein the statistics comprise energy ASM, entropy ENT, contrast CON and correlation IDE, and forming GLCM characteristics of the image by the statistics; (ii) a

A feature combination unit: and forming the extracted HOG and GLCM characteristics into a characteristic vector of the image sample.

An SVM model training module: the SVM classifier is used for combining a plurality of SVM classifiers for multi-classification, and each SVM classifier is trained through the feature vector of the image sample to obtain a trained SVM model;

weak signal detection module: the method comprises the steps of extracting HOG characteristics and GLCM characteristics of a phase space state diagram of a signal to be detected to form a characteristic vector of the phase space state diagram of the signal to be detected; and inputting the characteristic vector of the phase-space state diagram of the signal to be detected into the trained SVM model to obtain a weak signal detection result.

The effect of the invention is verified below with reference to specific experimental data:

for the Duffing oscillator signal detection systemLet k be 0.5, ω be 2 × pi 1000rad/s, f be 0.8276, and the value is unchanged in the simulation process, at this time the system is in the critical state, as shown in fig. 2, it is the x-y phase diagram of the critical state of Duffing oscillator when no signal is added;

fig. 3 shows that the amplitude of the added signal is a 1 × 10 in the critical state^-5V; ω ═ 2 × pi × 1000 rad/s; a large periodic state x-y phase diagram of the system;

fig. 4 shows that the amplitude of the added signal is a 1 × 10 in the critical state^-5V, ω 2 pi 1000rad/s, noise amplitude of 1 × 10^-6dB isLarge period state x-y phase diagram of-10 dBW snr;

FIG. 5 is a diagram of the x-y phase of a chaotic state with an added noise amplitude of 0.001dB in a critical state;

the signal environment adopted by the invention is a single-frequency signal s (t) with the transmitted signal frequency of omega-2 x pi 1000rad/s in the background of a natural environment, as shown in fig. 6, fig. 6 is a time domain and frequency spectrogram of randomly collecting 1s single-frequency signals;

FIG. 7 is a time domain diagram of a single frequency signal s (t) and noise n (t) added to a Duffing oscillator signal detection system;

adding a randomly acquired single-frequency signal s (t) and noise n (t) into a determined Duffing oscillator signal detection system, as shown in fig. 8, a large-period state x-y phase diagram of the Duffing oscillator signal detection system is obtained after the signal and noise of fig. 7 are added in a critical state, and the system is in a large-period state, which proves that the chaotic system can detect weak signals in a noise background.