Wavelet denoising method, system, device and medium based on measured ship distance image

Abstract

The application discloses a wavelet denoising method, system, device and medium based on a measured ship distance image, and relates to the field of target identification.The method comprises the following steps: acquiring one-dimensional distance image data of a ship target; determining a wavelet base and a decomposition order; performing wavelet decomposition on the one-dimensional distance image data according to the wavelet base and the decomposition order to obtain first wavelet coefficients; determining a wavelet threshold; determining second wavelet coefficients according to the wavelet threshold and the first wavelet coefficients by using an improved threshold function; the improved threshold function is provided with a variable factor, and the variable factor has a value range of [0, 1); when the value of the variable factor is equal to 0, the improved threshold function is a soft threshold function; when the value of the variable factor tends to 1, the improved threshold function tends to a hard threshold function; performing wavelet reconstruction on the second wavelet coefficients to obtain one-dimensional distance image data after denoising, so that the ship target can be identified.The application can improve the denoising effect and improve the ship target identification rate.

Description

TECHNICAL FIELD

[0001] The application relates to the field of target identification, in particular to a wavelet denoising method, system, device and medium based on a measured ship distance image. BACKGROUND

[0002] When a measured one-dimensional distance image of a ship is used to carry out target identification research, experimental data inevitably contains sea wave clutter, rain and snow clutter, same frequency interference, machine noise and the like, so that the data signal-to-noise ratio is reduced, the target information extracted from the one-dimensional distance image is blurred, and the target identification effect is seriously affected. Therefore, before carrying out measured classification and identification research, the one-dimensional distance image data needs to be denoised. For the same group of data, when the denoising effect is poor, the signal-to-noise ratio is improved limitedly, and the identification rate cannot be effectively improved. When the denoising effect is too high, the data is too smooth, the useful signal used for identification is lost, and the identification rate is still low. SUMMARY

[0003] The application aims to provide a wavelet denoising method, system, device and medium based on a measured ship distance image, so as to improve the denoising effect and improve the ship target identification rate.

[0004] To achieve the above-mentioned purpose, the application provides the following scheme.

[0005] A wavelet denoising method based on a measured ship distance image comprises the following steps.

[0006] Obtaining one-dimensional distance image data of a ship target;

[0007] Determining a wavelet base and a decomposition order;

[0008] According to the wavelet base and the decomposition order, wavelet decomposition is performed on the one-dimensional distance image data to obtain first wavelet coefficients;

[0009] Determining a wavelet threshold;

[0010] An improved threshold function is used to determine second wavelet coefficients according to the wavelet threshold and the first wavelet coefficients;

[0011] Wavelet reconstruction is performed on the second wavelet coefficients to obtain denoised one-dimensional distance image data; the denoised one-dimensional distance image data is used for ship target identification;

[0012] The expression of the improved threshold function is as follows:

[0013]

[0014] Wherein, sgn() is a step function, d j,k is the first wavelet coefficient, wherein, c is a first wavelet coefficient, T is a wavelet threshold, e is a natural constant, and a is a variable factor, and the variable factor is in a range of [0, 1).

[0015] Optionally, the wavelet base is any one of Haar wavelet, Daubechies wavelet, Mexican Hat wavelet, Morlet wavelet and Meyer wavelet.

[0016] Optionally, the wavelet threshold is determined by any one of unbiased likelihood estimation, fixed threshold estimation and heuristic threshold estimation.

[0017] Optionally, the wavelet base is Daubechies wavelet with a vanishing momentum of 7.

[0018] Optionally, the decomposition order is 6.

[0019] A wavelet denoising system based on a measured ship distance image, comprising:

[0020] a data acquisition module configured to acquire one-dimensional distance image data of a ship target;

[0021] a first parameter determination module configured to determine a wavelet base and a decomposition order;

[0022] a wavelet decomposition module configured to perform wavelet decomposition on the one-dimensional distance image data according to the wavelet base and the decomposition order to obtain first wavelet coefficients;

[0023] a second parameter determination module configured to determine a wavelet threshold;

[0024] a wavelet coefficient filtering module configured to determine second wavelet coefficients according to the wavelet threshold and the first wavelet coefficients by using an improved threshold function;

[0025] a wavelet reconstruction module configured to perform wavelet reconstruction on the second wavelet coefficients to obtain denoised one-dimensional distance image data; and the denoised one-dimensional distance image data is used for ship target identification.

[0026] The expression of the improved threshold function is:

[0027]

[0028] wherein, sgn() is a step function, d j,k is a first wavelet coefficient, is a second wavelet coefficient, T is a wavelet threshold, e is a natural constant, and a is a variable factor, and the variable factor is in a range of [0, 1).

[0029] An electronic device comprises a memory for storing a computer program and a processor for running the computer program to make the electronic device execute the wavelet denoising method.

[0030] A computer readable storage medium stores a computer program which, when executed by a processor, implements the wavelet denoising method.

[0031] According to the specific embodiments of the present application, the following technical effects are disclosed.

[0032] The wavelet denoising method based on the measured ship distance image provided by the present application improves the traditional wavelet threshold function, sets a variable factor with a value range of [0, 1) on the basis of the wavelet threshold function, so that when the value of the variable factor is equal to 0, the improved threshold function is a soft threshold function, when the value of the variable factor tends to 1, the improved threshold function tends to a hard threshold function, and when the value of the variable factor is between 0 and 1, the value of the wavelet coefficient is between the coefficients calculated by the soft and hard threshold functions, which guarantees the continuity of the wavelet threshold function and eliminates the constant deviation of the wavelet coefficient, thus improving the denoising effect and further improving the ship target recognition rate. In addition, the actual application effect can be adjusted by changing the specific value of the variable factor, which enhances the flexibility of the threshold function. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Figure 1 The flowchart of the wavelet denoising method based on the measured ship distance image provided by the present application is provided.

[0035] Figure 2 The flowchart of the wavelet decomposition provided by the present application is provided.

[0036] Figure 3 The comparison chart of the soft and hard threshold functions provided by the present application is provided.

[0037] Figure 4 The comparison chart of the improved threshold function and the soft and hard threshold functions provided by the present application is provided.

[0038] Figure 5 The flowchart of the wavelet reconstruction provided by the present application is provided.

[0039] Figure 6 The original distance image feature map provided by the present application is provided.

[0040] Figure 7 The single-recognition-rate graph based on the original range image feature provided by the present application;

[0041] Figure 8 The range image feature graph after soft threshold function denoising provided by the present application;

[0042] Figure 9 The single-recognition-rate graph based on the range image feature after soft threshold function denoising provided by the present application;

[0043] Figure 10 The range image feature graph after hard threshold function denoising provided by the present application;

[0044] Figure 11 The single-recognition-rate graph based on the range image feature after hard threshold function denoising provided by the present application;

[0045] Figure 12 The range image feature graph after improved threshold function denoising provided by the present application;

[0046] Figure 13 The single-recognition-rate graph based on the range image feature after improved threshold function denoising provided by the present application;

[0047] Figure 14 The module graph of the wavelet denoising system based on the measured ship range image provided by the present application.

[0048] Symbol explanation:

[0049] Data acquisition module - 1, first parameter determination module - 2, wavelet decomposition module - 3, second parameter determination module - 4, wavelet coefficient filtering module - 5, wavelet reconstruction module - 6. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0051] The purpose of the present application is to provide a wavelet denoising method, system, device and medium based on the measured ship range image to improve the denoising effect and improve the ship target recognition rate.

[0052] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0053] Wavelet transform is a time-frequency local transform, that is, high time resolution and low frequency resolution are maintained under high frequency, and low frequency is opposite, so wavelet transform has better signal adaptive ability. The wavelet transform has greater advantages than other signal processing methods in processing non-stationary signals. By wavelet decomposition of the non-stationary signal, the wavelet coefficient representing the variance of the noise signal is obtained, the wavelet coefficient is transformed by using the threshold function, the effect of reducing signal noise is obtained, and it has certain feasibility in theoretical analysis.

[0054] At present, there are many studies on wavelet threshold denoising. Cui Hua of Xi'an University of Electronic Science and Technology puts forward an improved wavelet threshold function denoising, which meets the continuity of the function and is adjustable in parameters and flexible in use. However, the threshold function does not have asymptoticity, so the deviation of the denoised signal from the original signal is large. Yang Zhengyi of Chongqing University of Software proposes a threshold function based on modulus square difference, which has continuity and deviation, but the parameter is not adjustable, so the flexibility of use cannot be guaranteed. Sun Hongxing of Liaoning University of Science and Technology combines the methods of the first two literatures to obtain a new threshold function, but it contains two adjustable parameters, and the value range of one of the parameters is too large, so that the parameter estimation is difficult, and the practicability is poor.

[0055] The present application refers to several wavelet denoising methods in the prior art, improves the wavelet threshold function, guarantees the continuity of the wavelet threshold function, eliminates the constant deviation of the wavelet coefficient, and increases the variable factor with a value range of [0, 1), so that the threshold function is more flexible to use and has better practicability. Combined with the measured data, good denoising and recognition effects are finally obtained.

[0056] As shown in Figure 1 The present application provides a wavelet denoising method based on measured ship distance image, which comprises:

[0057] Step S1: acquiring one-dimensional distance image data of a ship target.

[0058] Step S2: determining a wavelet base and a decomposition order.

[0059] Step S3: wavelet decomposing the one-dimensional distance image data according to the wavelet base and the decomposition order to obtain first wavelet coefficients.

[0060] Step S4: determining a wavelet threshold.

[0061] Step S5: determining second wavelet coefficients according to the wavelet threshold and the first wavelet coefficients by using an improved threshold function; the improved threshold function is provided with a variable factor, and the variable factor has a value range of [0, 1); when the value of the variable factor is equal to 0, the improved threshold function is a soft threshold function; and when the value of the variable factor tends to 1, the improved threshold function tends to a hard threshold function.

[0062] Step S6: performing wavelet reconstruction on the second wavelet coefficients to obtain one-dimensional range image data after noise reduction; and the one-dimensional range image data after noise reduction is used for identifying a ship target.

[0063] The wavelet noise reduction method based on the measured ship range image provided by the present application is described from the following aspects: 1) basic principle of wavelet noise reduction, including wavelet transform, wavelet decomposition and reconstruction; 2) implementation of wavelet threshold noise reduction, including wavelet threshold noise reduction principle and flow, determination of wavelet basis and wavelet threshold, selection of threshold function, and basic principle of improved wavelet threshold noise reduction; and 3) identification verification of the improved method by using measured data.

[0064] In the present application, the specific steps of reducing noise of the obtained one-dimensional range image data of a ship target are as follows:

[0065] Step 1: selecting an appropriate wavelet basis and decomposition order J.

[0066] In wavelet decomposition, the wavelet basis is not unique. Therefore, how to select a suitable wavelet basis is very important. In practice, the following three general principles should be considered when selecting a wavelet basis:

[0067] (1) self-similarity principle: if the signal has a certain similarity with the selected wavelet, the energy after wavelet transform is more concentrated, so as to reduce the calculation amount.

[0068] (2) discriminant function: by using specific parameters under different problems, a discriminant function for the related problem can be obtained, and each wavelet basis is brought into the discriminant function, so as to obtain a set of optimal criteria.

[0069] (3) support length: in most applications, the support length of the selected wavelet should be between 5-9, if the support length of the selected wavelet is too long, the boundary problem will be caused, and if the support length of the selected wavelet is too short, it is not conducive to the energy concentration of several signals.

[0070] The commonly used wavelet bases include Haar wavelet, Daubechies (dbN) wavelet, Mexican Hat (mexh) wavelet, Morlet wavelet, Meyer wavelet and the like. Preferably, the wavelet used in the present application is db7 wavelet, i.e. Daubechies wavelet with the value of vanishing moment being 7; the value of the decomposition scale J of the wavelet is 6.

[0071] Step 2: completing the wavelet decomposition of the signal (i.e. one-dimensional range image data) to obtain a group of wavelet coefficients d j,k .

[0072] Specifically, the wavelet decomposition of the signal is performed by using a wavelet transform method, which includes continuous wavelet transform and discrete wavelet transform. The principles of the continuous wavelet transform and the discrete wavelet transform are described below.

[0073] (1) Continuous wavelet transform

[0074] The function space of the wavelet transform is generally L 2 (R). L 2 (R) refers to the function space of square-integrable functions on the real number field R, i.e.

[0075] ψ(t)∈L 2 (R)

[0076] wherein t is time. If ψ(t)∈L 2 (R), ψ(t) is called an energy-limited signal. If ψ(t)∈L 2 (R), the Fourier transform ψ(ω) of ψ(t) satisfies the admissible condition:

[0077]

[0078] wherein ω is the frequency of the signal, and ψ(t) is called a basic wavelet or mother wavelet. The stretching and translation of the mother wavelet can generate a group of wavelet sequences wherein a, b ∈ R, and a ≠ 0. a is called a stretching factor, and b is called a translation factor.

[0079] The continuous wavelet transform of a function f(t) with respect to a basic wavelet ψ is as follows:

[0080]

[0081] wherein W ψ is the continuous wavelet transform of a function f with respect to a basic wavelet ψ, ψ is the basic wavelet, f is the function to be subjected to the continuous wavelet transform, and represents the continuous wavelet transform of The inverse wavelet transform of a function f(t) with respect to a basis wavelet ψ is given by:

[0082]

[0083] where x is the time domain signal, C is the wavelet constant. The inverse wavelet transform can restore the signal generated by the continuous wavelet transform to the original signal. ψ

[0084] (2) Discrete wavelet transform

[0085] The scaling factor and the translation factor of the continuous wavelet transform are continuous, so the continuous integral needs to be calculated, which is inconvenient to use. In solving practical problems, the discrete wavelet transform (DWT) is usually used. The scaling factor a and the translation factor b of the continuous wavelet transform are discretized, which is the discrete wavelet transform. Usually take m, n∈Z, where a0 is a real number greater than 0, b0 is a real number greater than 0, m is the power, and n is the coefficient, that is, a set of wavelet sequences can be obtained:

[0086]

[0087] At this time, the wavelet function is discrete, and the corresponding DWT is as follows:

[0088]

[0089] In particular, take a0=2, b0=1, and a dyadic wavelet can be obtained:

[0090]

[0091] In the discrete wavelet transform (DWT), according to the theory of orthogonal multi-resolution analysis, the linear function space {V j : j∈Z} can be decomposed into the direct sum of V j-1 and the projection of V j in W j-1 , that is, j is the resolution. And V j-1 ⊥W j-1 , j∈Z. Then for each signal x(t) represented in V j , it can be represented by the basis functions φ j-1,k (t) and ψ j-1,k (t) in V j-1 and W j-1 , that is: ​

[0092]

[0093] Where k is the ordinal number of the digitized signal, which is an ordinal number representing time, and k∈Z,φ j-1,k (t) represents the scaling function, ψ j-1,k (t) represents the wavelet function; c j-1,k For subspace V j-1 The scaling factor, d j-1,k For subspace V j-1 Wavelet coefficients on.

[0094] The decomposition of the above equation can be achieved using a certain filter bank. When the wavelet and the scale are orthogonal in space, the scale coefficient c can be calculated using the inner product formula. j-1,k With wavelet coefficients d j-1,k :

[0095]

[0096]

[0097] Where n∈Z. h(n-2k) and g(n-2k) can be considered as a low-pass filter and a high-pass filter, respectively, then... Figure 2 As shown, the principle of wavelet decomposition is as follows: the signal is passed through a pair of high-pass and low-pass filters to obtain a set of wavelet coefficients and scaling coefficients. The scaling coefficients are then passed through high-pass and low-pass filters to obtain the next set of wavelet coefficients and scaling coefficients. This process is continued until the specified number of decomposition levels is reached, thus realizing the wavelet decomposition of the signal.

[0098] Step 3: Select an appropriate wavelet threshold T. Wavelet coefficients below this threshold are considered noise coefficients, and those above this threshold are considered signal coefficients. The threshold is estimated using a fixed threshold.

[0099] D.L. Donoho and I.M. Johnstone first used wavelet thresholding for noise reduction and developed soft and hard thresholding methods. Signal energy is mainly distributed in some wavelet coefficients, while noise energy is spread throughout the entire wavelet threshold. Since signals are usually continuous in space, the effective wavelet coefficients are relatively large in the wavelet domain; however, noise is discontinuous in space, so after wavelet decomposition, the noise still exhibits strong randomness, resulting in smaller wavelet coefficients. Based on these reasons, by setting a wavelet threshold, wavelet coefficients below this threshold are set to zero. The remaining wavelet coefficients are then transformed using a threshold function to obtain a new set of wavelet coefficients. Wavelet reconstruction using these new coefficients yields a reconstructed signal with a signal-to-noise ratio that is improved compared to the original wavelet denoising.

[0100] The determination of threshold is an important step in wavelet threshold denoising process, one of the reference standards is to make the difference between d j,k and as small as possible, so as not to deviate too much from the original signal. The common threshold selection methods are: unbiased likelihood estimation, fixed threshold estimation and heuristic threshold estimation.

[0101] (1) Unbiased likelihood estimation (Rigrsure)

[0102] The square of the wavelet coefficient d l is arranged in order, and the square of the wavelet coefficient d is obtained. l represents the length of the wavelet coefficient vector. The risk vector is denoted as p = 1, 2,..., l, and the minimum risk point is l min , then the Rigrsure threshold is defined as follows, where σ represents the mean square error of the noise signal.

[0103]

[0104] (2) Fixed threshold estimation (Sqtwolog)

[0105] The threshold calculation method is:

[0106]

[0107] l represents the length of the wavelet coefficient vector, and σ represents the mean square error of the noise signal. Since the exact variance of the noise cannot be known in specific application and data denoising, the median estimator can be used to obtain the noise variance, that is, σ = median(d j,k ) / 0.6745.

[0108] (3) Heuristic threshold estimation

[0109] Let represent the first variable, represent the second variable, then the calculation method of heuristic threshold estimation is:

[0110]

[0111] Step 4: An improved threshold function with variable parameters is proposed, which substitutes the wavelet coefficient d j,k (i.e. the first wavelet coefficient, specifically the first wavelet coefficient on the subspace V j ) into it to obtain the new wavelet coefficient d (i.e. the second wavelet coefficient, specifically the second wavelet coefficient on the subspace V j ).

[0112] The determination process of the improved threshold function is as follows:

[0113] In wavelet threshold denoising, how to select the threshold function is one of the key problems to determine the final denoising effect. The commonly used wavelet threshold selection functions are hard threshold function and soft threshold function. The soft threshold processing is to compare the signal with the threshold, and the part below the threshold is zero, and the part above or equal to the threshold is contracted to the threshold; for the hard threshold processing, the data absolute value greater than or equal to the threshold is unchanged, and the data absolute value less than the threshold is zero. The expressions of the soft threshold function and the hard threshold function are as follows:

[0114] The expression of the soft threshold function is:

[0115]

[0116] The expression of the hard threshold function is:

[0117]

[0118] Suppose T = 10, then the comparison chart of the hard threshold function and the soft threshold function is as shown in Figure 3 . The horizontal coordinate of the wavelet coefficient is the first wavelet coefficient, and the vertical coordinate of the new wavelet coefficient is the second wavelet coefficient.

[0119] The hard threshold function and the soft threshold function are widely used in actual denoising. Although the two methods can ensure that the noise is suppressed to a certain extent, due to the defects of the two methods, the denoising effect still has a lot of room for improvement.

[0120] The soft threshold function sets the coefficient less than the threshold T to zero, and the coefficient greater than the threshold T is subtracted from the threshold. This will cause a constant deviation between the wavelet coefficient after the threshold function processing and the original coefficient. When the wavelet coefficient after the transformation is used for signal reconstruction, the target will be distorted and the useful signal will be lost, which will directly affect the subsequent target identification.

[0121] The hard threshold function sets the coefficient less than the threshold T to zero, and the coefficient greater than the threshold T remains unchanged. Although there is no constant deviation between the wavelet coefficient after the threshold function processing and the original coefficient, the threshold function at the threshold T is discontinuous, which causes the denoised signal to oscillate and no longer have smoothness, and makes the one-dimensional range image signal after reconstruction appear false target scattering point peaks, affecting the subsequent target recognition effect.

[0122] Therefore, the threshold function should have continuity and no discontinuity at the threshold T; at the same time, the wavelet coefficient after the threshold processing should be as close as possible to the original coefficient to reduce the constant deviation between them; and the advantages of the traditional threshold function should be retained as much as possible. Based on the above considerations, combined with a large amount of experimental research on measured data, an improved threshold function with variable parameters is proposed as follows:

[0123]

[0124] where sgn() is a step function, e is a natural constant, and a is a variable factor with a range of [0, 1). The variable factor a can enhance the flexibility of the threshold function, and the actual application effect can be adjusted by changing the value of a. It can be seen that, since the variable factor a ∈ [0, 1), the value of is between the coefficients calculated by the soft and hard threshold functions. When a is 0, the improved threshold function is restored to the soft threshold function, and when a -> 1, the improved threshold function tends to the hard threshold function.

[0125] Next, the continuity of the improved threshold function is verified.

[0126] When d j,k → T + (where T + represents the right limit of the threshold T, that is, approaching T from the right side of the threshold T), there is:

[0127]

[0128] When d j,k → T - (where T - represents the left limit of the threshold T, that is, approaching T from the left side of the threshold T), there is:

[0129]

[0130] From the above formula, it can be seen that The improved threshold function is continuous at the threshold T, which reduces the signal oscillation existing in the hard threshold function.

[0131] Next, the asymptotic property of the improved threshold function is verified:

[0132] When d j,k > 0, there is:

[0133]

[0134] Similarly, when d j,k < 0, there is:

[0135]

[0136] From the above, it can be obtained that:

[0137]

[0138] Therefore, when d j,k → ∞, the wavelet coefficient The coefficients gradually approach the original values, eliminating the constant deviation between them. Therefore, the reconstructed signal is not lost, and the noise reduction effect is further improved. A comparison of the improved threshold function image with the soft and hard threshold function images is shown below. Figure 4 As shown, by Figure 4 It can be seen that the image of the improved threshold function at threshold T is continuous, and it increases with d. j,k →∞, Gradually approaching d j,k There is no constant deviation between the two.

[0139] Step 5: Apply new wavelet coefficients Reconstruction is performed to obtain the desired signal, i.e., the denoised one-dimensional range image data. Specifically, the wavelet reconstruction algorithm can be viewed as the inverse process of the decomposition algorithm. The flowchart of wavelet reconstruction is as follows: Figure 5 As shown.

[0140] The improved wavelet threshold denoising method provided by this invention is verified using actual measurement data.

[0141] Using 200 sets of multi-angle high-resolution range image data (50 sets for each of four different types of ship targets, totaling 200 sets), the improved wavelet threshold denoising method provided in this invention was verified. The wavelet used in the experiment was the db7 wavelet, with the variable factor α set to 0.8, the wavelet decomposition scale J set to 6, and a fixed threshold estimation method used. Simultaneously, SNR was used as the performance quantification criterion for the denoising algorithm.

[0142]

[0143] In the above formula, s(n) is the original signal. It is a signal that has undergone noise reduction processing.

[0144] The experiment is divided into four groups, respectively, with the original distance image, soft threshold denoising distance image, hard threshold denoising distance image, improved wavelet threshold denoising distance image as four groups of recognition features, with support vector machine (Support Vector Machine, SVM) as the classifier, to carry out target classification and recognition, and calculate the recognition rate of four groups of data. Support vector machine is a supervised learning method, which can be widely used in statistical classification and regression analysis. Because it does not involve probability measure and law of large numbers, etc., it also simplifies the usual classification and regression problems, so support vector machine can be better used for target recognition under the condition of small sample. In this embodiment, the number of samples is small, the sample dimension is small, and it belongs to small sample target recognition, so the use of support vector machine can obtain better recognition result. The kernel function of support vector machine is selected as radial basis function (Radial Basis Function, RBF). Radial basis function is a typical shift invariant kernel, which is one of the most commonly used kernel functions, and has the characteristics of being able to realize nonlinear mapping and low complexity. The penalty factor of support vector machine and the parameter selection of radial basis function are selected by cross validation method.

[0145] The test set of recognition comes from 10 random groups of data in 50 groups of data of four types of ship targets, and the remaining 40 groups of data are used as recognition training set. The experiment is carried out for ten times, Figure 6 to Figure 13 The four groups of distance image features and the recognition rate of this group in a certain experiment are shown.

[0146] Table 1 shows the signal-to-noise ratio of 10 random groups of one-dimensional distance image data after denoising, and table 2 shows the recognition rate of each experiment and the average recognition rate.

[0147] Table 1 shows the signal-to-noise ratio of 10 random groups of one-dimensional distance image data after denoising, and table 2 shows the recognition rate of each experiment and the average recognition rate.

[0148]

[0149] Table 2 shows the classification recognition rate (%) of four groups of features.

[0150]

[0151] It can be seen from the data in Table 1 that the wavelet threshold denoising can indeed eliminate most of the noise in the distance image, and the smoothness and signal-to-noise ratio of the denoised distance image are in turn soft threshold denoising, improved wavelet threshold denoising and hard threshold denoising. It should be noted that signal denoising will cause the loss of part of the effective information, therefore, although the one-dimensional distance image is more smooth after soft threshold denoising, part of the effective information is filtered out as noise in the denoising process, resulting in signal distortion and recognition rate reduction. It can be seen from the data in Table 2 that the one-dimensional distance image data after improved wavelet threshold denoising has the highest target recognition rate. Therefore, it can be known that the improved wavelet threshold function denoising can better eliminate data noise interference and retain useful information, thereby improving the target recognition effect.

[0152] In order to perform the method corresponding to the above-mentioned embodiment to realize the corresponding functions and technical effects, a wavelet denoising system based on measured ship distance image is provided below. As shown in Figure 14 The wavelet denoising system comprises:

[0153] A data acquisition module 1 is configured to acquire one-dimensional distance image data of a ship target.

[0154] A first parameter determination module 2 is configured to determine a wavelet base and a decomposition order.

[0155] A wavelet decomposition module 3 is configured to perform wavelet decomposition on the one-dimensional distance image data according to the wavelet base and the decomposition order to obtain first wavelet coefficients.

[0156] A second parameter determination module 4 is configured to determine a wavelet threshold.

[0157] A wavelet coefficient filtering module 5 is configured to determine second wavelet coefficients according to the wavelet threshold and the first wavelet coefficients by using an improved threshold function.

[0158] A wavelet reconstruction module 6 is configured to perform wavelet reconstruction on the second wavelet coefficients to obtain denoised one-dimensional distance image data; the denoised one-dimensional distance image data is used for ship target recognition.

[0159] The expression of the improved threshold function is:

[0160]

[0161] Wherein, sgn() is a step function, d j,k is the first wavelet coefficient, is the second wavelet coefficient, T is the wavelet threshold, e is a natural constant, and a is a variable factor, and the value range of the variable factor is [0, 1).

[0162] Further, the present application also provides an electronic device, comprising a memory for storing a computer program and a processor for running the computer program to make the electronic device execute the wavelet denoising method in the above-mentioned embodiments. The electronic device can be a server.

[0163] In addition, the present application also provides a computer readable storage medium storing a computer program, which, when executed by a processor, implements the wavelet denoising method in the above-mentioned embodiments.

[0164] To sum up, aiming at the noise interference problem existing in the measured ship one-dimensional range profile data, the present application uses the wavelet denoising method to pre-process the measured data, and proposes an improved threshold function on the basis of the traditional threshold function, thereby improving the target recognition effect.

[0165] Firstly, the wavelet base function and the decomposition order are determined, secondly, the wavelet decomposition of the signal is completed to obtain a group of wavelet coefficients, then the wavelet coefficients are processed by selecting the wavelet threshold and the improved threshold function to obtain a group of new wavelet coefficients, and finally the denoised signal is obtained through the wavelet reconstruction algorithm. Among them, the present application proposes a parameter-variable improved threshold function, and the effectiveness of the improved wavelet threshold denoising method is verified in combination with the measured data, and the following conclusions are obtained:

[0166] 1. The improved threshold function has continuity. The function is continuous at the threshold T, which ensures that the denoised signal is smooth and reduces the oscillation of the signal caused by denoising.

[0167] 2. The improved threshold function has asymptoticity. The wavelet coefficient of the threshold function after transformation gradually approaches the original coefficient, eliminating the constant deviation between them, and reducing the loss of the reconstructed signal.

[0168] 3. The improved threshold function is parameter-variable. The variable factor a∈[0,1), The value of the variable factor a∈[0,1),

[0169] 4. It can be verified through experiments that, compared with the traditional threshold denoising method, the improved wavelet threshold denoising method can better reduce noise while retaining more effective information, thereby improving the target recognition effect.

[0170] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts are described in the method part.

[0171] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.

Claims

1. A wavelet denoising method based on measured ship distance image, characterized in that, The method comprises the following steps: acquiring one-dimensional range image data of a ship target; determining a wavelet base and a decomposition order; performing wavelet decomposition on the one-dimensional range image data according to the wavelet base and the decomposition order to obtain first wavelet coefficients; determining a wavelet threshold value; determining second wavelet coefficients according to the wavelet threshold value and the first wavelet coefficients by using an improved threshold function; performing wavelet reconstruction on the second wavelet coefficients to obtain one-dimensional range image data after noise reduction; and the one-dimensional range image data after noise reduction is used for ship target identification. An expression of the improved threshold function is as follows: wherein sgn() is a step function, d j,k is a first wavelet coefficient, is a second wavelet coefficient, T is a wavelet threshold, e is a natural constant, and a is a variable factor, and the variable factor has a value range of [0, 1).

2. The wavelet denoising method based on measured ship distance image according to claim 1, characterized in that, The wavelet base is any one of a Haar wavelet, a Daubechies wavelet, a Mexican Hat wavelet, a Morlet wavelet and a Meyer wavelet.

3. The wavelet denoising method based on measured ship distance image according to claim 1, characterized in that, The wavelet threshold value is determined by any one of unbiased likelihood estimation, fixed threshold estimation and heuristic threshold estimation.

4. The wavelet denoising method based on measured ship distance image according to claim 2, characterized in that, The wavelet base is a Daubechies wavelet with a vanishing momentum of 7.

5. The wavelet denoising method based on measured ship distance image according to claim 1, characterized in that, The decomposition order is 6.

6. A wavelet denoising system based on measured ship distance image, characterized in that, The method comprises the following steps: a data acquisition module is configured to acquire one-dimensional range image data of a ship target; a first parameter determination module is configured to determine a wavelet base and a decomposition order; a wavelet decomposition module is configured to perform wavelet decomposition on the one-dimensional range image data according to the wavelet base and the decomposition order to obtain first wavelet coefficients; a second parameter determination module is configured to determine a wavelet threshold value; a wavelet coefficient filtering module is configured to determine second wavelet coefficients according to the wavelet threshold value and the first wavelet coefficients by using an improved threshold function; a wavelet reconstruction module is configured to perform wavelet reconstruction on the second wavelet coefficients to obtain one-dimensional range image data after noise reduction; and the one-dimensional range image data after noise reduction is used for ship target identification. An expression of the improved threshold function is as follows: wherein sgn() is a step function, d j,k is a first wavelet coefficient, is a second wavelet coefficient, T is a wavelet threshold, e is a natural constant, and a is a variable factor, and the variable factor has a value range of [0, 1).

7. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the wavelet denoising method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed by the processor to implement the wavelet denoising method according to any one of claims 1 to 5.

Images