Line spectrum extraction method and system for LOFAR spectrogram of two-dimensional image

Abstract

The invention provides a two-dimensional image LOFAR spectrogram line spectrum extraction method and system. The method comprises the following steps: S1, establishing a dynamic background model for an LOFAR spectrogram; s2, determining a high detection threshold value and a low detection threshold value according to the background model; s3, performing multi-stage target detection and fusion to obtain a preliminary detection result; s4, performing trajectory tracking growth according to historical information of the line spectrum; s5, extracting the features of the line spectrum region to carry out line spectrum description; and S6, performing line spectrum screening according to an expert weight mode to obtain a line spectrum extraction result. According to the LOFAR spectrogram line spectrum extraction method based on two-dimensional image processing, line spectrum extraction is achieved by establishing a multi-stage detection mechanism and combining a tracking judgment mode, compared with the prior art, line spectrum frequency points can be detected under the condition of a low signal-to-noise ratio, the occurrence of line spectrum breakage conditions is reduced, and the detection efficiency is improved. And meanwhile, the detection of the pseudo-wire spectrum can be more effectively inhibited by an expert weight judgment mode based on the line spectrum characteristics.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic image processing technology, specifically to a method and system for extracting line spectra from LOFAR spectra of two-dimensional images. Background Technology

[0002] Sound waves are the most important means of monitoring marine targets. The LOFAR spectrum is the power spectrum obtained by performing a short-time Fourier transform on the signal received by passive sonar. It can reflect the characteristics of the signal in both time and frequency dimensions and is widely used in the field of passive signal analysis. The line spectrum generated by moving targets such as ships and unmanned underwater vehicles has significant characteristics in the LOFAR spectrum. Therefore, extracting the line spectrum is of great significance for target detection and identification.

[0003] Current LOFAR spectrum line extraction algorithms are mainly based on intelligent networks, statistical models, and image processing. Intelligent network-based algorithms extract target line spectra by learning from data samples, but their adaptability to location and environment is weak, and their versatility is limited. Statistical model-based line spectrum extraction algorithms calculate the optimal path for the line spectra using statistical models, thus achieving line spectrum tracking. Image processing-based line spectrum extraction methods are a widely used class of algorithms. LOFAR spectra can be viewed as two-dimensional visual images with frequency and time dimensions; combining detection and tracking can achieve line spectrum extraction. Common detection methods include using filtering techniques to establish detection windows for line spectrum extraction, or using image enhancement to improve image quality, as well as edge detection or thresholding methods for line spectrum extraction.

[0004] During sonar operation, interference noise from factors such as the marine environment and platform noise enters the system simultaneously with target noise, making the received information highly complex. This results in a large amount of background noise in the LOFAR spectrum, a low signal-to-noise ratio for underwater acoustic target radiated noise, and ultimately makes line spectrum extraction difficult. During line spectrum extraction, a large number of pseudo-line spectra are often extracted simultaneously, leading to a high false alarm rate. However, pseudo-line spectra are mostly generated by outliers, and compared to line spectra, the shape and trajectory of a region over a cumulative time period are often less stable.

[0005] The patent with publication number CN111931820B discloses a method for extracting the line spectrum of the LOFAR spectrum of underwater target radiation noise based on a convolutional residual network. This patent uses a convolutional residual regression network model to train the LOFAR spectrum. However, this method relies on a large amount of precisely labeled training data. If the training data lacks data for a certain specific scenario, it will fail in practical applications and lacks versatility.

[0006] In summary, given the problems of the existing technologies, researching a method and system for extracting line spectra from LOFAR spectra of two-dimensional images has become a critical task that urgently needs to be addressed. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for extracting LOFAR spectrum lines from two-dimensional images.

[0008] A method for extracting line spectra from LOFAR spectra of two-dimensional images according to the present invention includes the following steps: Step S1: Establish a dynamic background model for the LOFAR spectrum; Step S2: Determine two detection thresholds, high and low, based on the background model; Step S3: Perform multi-level target detection and fusion to obtain preliminary detection results; Step S4: Perform trajectory tracking growth based on the historical information of the line spectrum; Step S5: Extract features from the spectral region to perform spectral description; Step S6: Perform line spectrum filtering based on the expert weighting model to obtain the line spectrum extraction results.

[0009] Preferably, a dynamic background model is established for the LOFAR spectrum, including: analyzing the LOFAR spectrum and assuming... Within time 1, the LOFAR spectral image matrix is: ,in, The row number represents the number of pixels in the time dimension. is the column number, representing the number of pixels in the frequency dimension; for the LOFAR spectral image... row data Establish There are several background models, among which... , , The maximum number of levels in the background model; in the th... Calculate the rectangular window region at each frequency pixel. mean amplitude and standard deviation ,

[0010]

[0011] in, Indicates the first Okay, number The amplitude value corresponding to the column pixel position; the average amplitude calculated from each frequency pixel. Composition of multi-level dynamic background models .

[0012] Preferably, based on the background model, two detection thresholds, high and low, are determined, including: According to the average amplitude and the standard deviation In the At different levels of dynamic background, calculate the low detection threshold vector respectively. and high detection threshold vector ,in, , The calculation formulas are as follows:

[0013]

[0014] in, Indicates the first The first calculation on a dynamic background Low detection threshold for each pixel Indicates the first The first calculation on a dynamic background A high detection threshold for each pixel. and The sum is the weighting coefficient, which is a constant value, and .

[0015] Preferably, multi-level target detection is performed and fused to obtain preliminary detection results, including: the first level of target detection on the LOFAR spectral image. row data Perform binarization processing, In the Under dynamic conditions, with the low detection threshold vector and the high detection threshold vector The comparison is performed: pixels with values ​​greater than the detection threshold are marked "1", and pixels with values ​​less than the detection threshold are marked "0", thus obtaining the binary vector detection result. and ,in, , Based on the binary vector detection results and Multi-level detection results are fused, and the multi-level detection values ​​at each frequency pixel are statistically analyzed to obtain preliminary detection values. , According to the preliminary detection values Obtain a high-threshold detection vector containing binary detection results of "1" and "0". and low threshold detection vector , , ; for the high threshold detection vector Perform region connectivity processing, grouping pixels with consecutive detection values ​​of "1" into the same connected region, resulting in... There are connected regions, where the scale of each region is the number of pixels with a value of "1", and the corresponding values ​​within the region are... The location of the maximum amplitude value is the frequency detection point of the line spectrum trajectory in each region.

[0016] Preferably, based on the binary vector detection result and Multi-level detection results are fused, and the multi-level detection values ​​at each frequency pixel are statistically analyzed to obtain preliminary detection values. This includes: the preliminary detection value The Middle Preliminary detection value at each frequency pixel The calculation formula is as follows:

[0017] in, Indicates the test results The Middle The value at each frequency pixel. Indicates the test results The Middle The value at each frequency pixel.

[0018] Preferably, based on the preliminary detection value Obtain a high-threshold detection vector containing binary detection results of "1" and "0". and low threshold detection vector This includes: points with a value of "1" in the high-threshold detection vector represent frequency pixels that are detected using the high threshold at each level of the dynamic model. The Middle Detection value at each frequency pixel The calculation formula is as follows:

[0019] In the low-threshold detection vector, points with a value of "1" represent frequency pixels that are detected using the low threshold at each level of the dynamic model. The Middle Detection value at each frequency pixel The calculation formula is as follows:

[0020] in, The preliminary detection value The Middle Preliminary detection values ​​at each frequency pixel.

[0021] Preferably, trajectory tracking growth is performed based on historical information of the line spectrum, including: based on the preliminary detection results of the first... Row data detection results and the first Linear trajectory tracking and growth are performed on the line data line trajectory tracking results, where, for the first... In the row of data, in the first row... The frequency pixel position of each line trajectory point in the row data is centered, and the bandwidth is pixels. Track within the search area; prioritize detection within the search area. ,like If a detection point exists, the trajectory grows; if the trajectory is lost, the counter is reset to zero. If no detection point is found, then... Search within this search area, if If a detection point exists, the trajectory grows, but the trajectory loss counter increments by 1; if There are no detection points in the middle, and If there are no detection points in the path, the trajectory will not grow, and the trajectory loss counter will increment by 1; when the trajectory loss counter reaches the preset upper limit... When the trajectory is terminated, the process is cancelled; for trajectories not matched by any trajectory... The detection point is initialized as a newly added line trajectory point.

[0022] Preferably, the feature extraction of the line spectrum region for line spectrum description includes: analyzing the connected region information of the line spectrum trajectory based on the tracked line spectrum trajectory, and extracting the scale, amplitude, and curvature features of the line spectrum. The scale feature includes the average and standard deviation of the scale of the line spectrum region, the amplitude feature includes the average and standard deviation of the amplitude of the line spectrum region, and the curvature feature includes the average and standard deviation of the curvature of the line spectrum trajectory points.

[0023] Preferably, the line spectrum is screened according to an expert weighting model to obtain the line spectrum extraction results, including: establishing an expert weighting equation, calculating the probability of each line spectrum, performing line spectrum screening, and removing pseudo-line spectra that do not meet the conditions. The expression of the expert weighting equation is:

[0024] in, Represents the probability value of the line spectrum. This represents the total dimension of the features. Indicates the first dimensional features, and Indicates the first Weighted values ​​and eigenvalues ​​of the dimensional features; based on conditional thresholds Perform line spectrum analysis, if If, then the musical line is preserved, if If so, the musical line is removed.

[0025] This invention also provides a system for extracting line spectra from LOFAR spectra of two-dimensional images. This system can be implemented by executing the steps of the method for extracting line spectra from LOFAR spectra of two-dimensional images. That is, those skilled in the art can understand the method for extracting line spectra from LOFAR spectra of two-dimensional images as a preferred embodiment of the system. The system includes: Module M1 establishes a dynamic background model for the LOFAR spectrum; Module M2 determines two detection thresholds, high and low, based on the background model; Module M3 performs multi-level target detection and fusion to obtain preliminary detection results; Module M4 performs trajectory tracking growth based on historical information of the line spectrum; Module M5 extracts features from the spectral region for spectral description; Module M6 performs line spectrum filtering based on the expert weighting model to obtain the line spectrum extraction results.

[0026] Compared with the prior art, the present invention has the following beneficial effects: The LOFAR spectrum line spectrum extraction method based on two-dimensional image processing disclosed in the present invention achieves line spectrum extraction by establishing a multi-level detection mechanism and combining it with a tracking discrimination method. Compared with the prior art, it can detect line spectrum frequency points under low signal-to-noise ratio conditions, reduce the occurrence of line spectrum breakage, and the expert weight discrimination method based on line spectrum features can more effectively suppress the detection of pseudo line spectra. Attached Figure Description

[0027] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A flowchart illustrating a method for extracting line spectra from LOFAR spectra of two-dimensional images, provided in an embodiment of the present invention; Figure 2 The underwater target LOFAR spectrum and line spectrum extraction trajectory map provided in the embodiments of the present invention; Figure 3 This is a two-level dynamic background model diagram based on LOFAR spectrum provided in an embodiment of the present invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0029] Figure 1 A flowchart of a method for extracting line spectra from LOFAR spectra of two-dimensional images provided in an embodiment of the present invention is shown below. Figure 1 As shown, it includes the following steps: Step 1: Establish a dynamic background model for the LOFAR spectrum.

[0030] Specifically, the LOFAR spectrum is analyzed, and the following is assumed: Within a given time period, the LOFAR spectral image matrix ,in The row number represents the number of pixels in the time dimension. Let be the column number, representing the number of pixels in the frequency dimension, for the . The moment is the first time on the image. row data Establish a multi-level background model and use Implemented using rectangular windows of different sizes, in the first... Calculate the rectangular window region at each frequency pixel. mean amplitude and standard deviation The expressions are as follows:

[0031]

[0032] in, Indicates the first Okay, number The amplitude value corresponding to the column pixel position, if Then take ,like Then take The average amplitude calculated from each frequency pixel. Composition of multi-level dynamic background models ,in , .

[0033] Step 2: Determine the detection threshold.

[0034] Specifically, based on step 1, according to the mean amplitude and standard deviation In the Calculate the low detection threshold vector under dynamic background. , High threshold vector , ; The expressions are as follows:

[0035]

[0036] in, Indicates the first The first calculation on a dynamic background Low detection threshold for each pixel Indicates the first The first calculation on a dynamic background A high detection threshold for each pixel. and The sum is the weighting coefficient, which is a constant value, and .

[0037] Step 3: Perform multi-level detection.

[0038] Step 3.1: For each level of dynamic background, use high and low detection thresholds to perform detection, and then perform detection on the image from Step 1. row data Perform binarization processing, In the Under dynamic background, compared with the low detection threshold vector calculated in step 2. and high detection threshold vector The comparison is performed: pixels with values ​​greater than the detection threshold are marked "1", and pixels with values ​​less than the detection threshold are marked "0", thus obtaining the binary vector detection result. and , , .

[0039] Step 3.2: Perform multi-level detection result fusion. By statistically analyzing the multi-level detection values ​​at each frequency pixel, a preliminary detection value is formed. , , of which Preliminary detection value at each frequency pixel The calculation formula is as follows:

[0040] in, Indicates the test results The Middle The value at each frequency pixel. Indicates the test results The Middle The value at each frequency pixel.

[0041] Step 3.3: Based on the detection values ​​from Step 3.2, form a high-threshold detection vector containing binary detection results of "1" and "0". and low threshold detection vector , , High threshold detection vector Points with a median value of "1" represent the frequency pixels that were detected using a high threshold at each level of the dynamic model. The Middle Detection value at each frequency pixel The calculation formula is as follows:

[0042] Low threshold detection vector Points with a median value of "1" represent the frequency pixels that were detected using a low threshold at each level of the dynamic model. The Middle Detection value at each frequency pixel The calculation formula is as follows:

[0043] Step 3.4, the detection results from step 3.3. Perform region connectivity processing, grouping pixels with consecutive detection values ​​of "1" into the same connected region, resulting in... Each region is a connected region, and the scale of each region is the number of pixels with a value of "1". The corresponding values ​​within each region are... The location of the maximum amplitude value is the frequency detection point of the line spectrum trajectory in each region.

[0044] Step 4: Perform line trajectory tracking growth.

[0045] The line spectrum trajectory is categorized into three states: newly created trajectory, established trajectory in growth, and trajectory to be cancelled. To reduce the occurrence of line spectrum trajectory breakage, the line spectrum trajectory is discussed in different cases, based on step 3.4. The detection results in the row data and the first Line trajectory tracking and growth are performed on the line trajectory tracking results of the data lines, and the line trajectory tracking is performed on the first line. In the row of data, in the first row... The frequency pixel position of each line trajectory point in the row data is centered, and the bandwidth is pixels. Within the search area, if If a detection point exists, then that detection point is a trajectory tracking growth point; if If no detection point is found, then... Search within this search area, if If a detection point exists, it becomes a trajectory tracking growth point, but the trajectory loss count is incremented by 1; if and If no detection points are found in the middle, the trajectory loss count is 1. When the number of track loss is not less than When, the spectral trajectory is cancelled; when If no detection point is found in the search, the detection point is the newly added line trajectory point.

[0046] Step 5: Extract line spectrum features.

[0047] Specifically, based on the line trajectory obtained in step 4, the information of the connected region where the line trajectory is located is analyzed, and the scale, amplitude, and curvature features of the line are extracted. The scale features include the average and standard deviation of the scale of the line region, the amplitude features include the average and standard deviation of the amplitude of the line region, and the curvature features include the average and standard deviation of the curvature of the line trajectory points.

[0048] Step 6: Perform line spectrum screening.

[0049] An expert weight equation is established to calculate the probability of each line spectrum, and line spectrum screening is performed to remove pseudo-line spectra that do not meet the conditions. Its expression is:

[0050] in, Represents the probability value of the line spectrum. This represents the total dimension of the features. Indicates the first dimensional features, and Indicates the first Weighted values ​​and eigenvalues ​​of the dimensional features; based on conditional thresholds Perform line spectrum analysis. Then the musical line is preserved. If so, the musical line is removed.

[0051] This embodiment, in conjunction with the following simulation experiments, further illustrates the technical effects of the present invention: Simulation conditions: All simulation experiments were implemented using Matlab 2015b software under the Windows 10 operating system; Simulation content and result analysis: The underwater target LOFAR spectrum and the obtained line spectrum trajectory diagram obtained by the method of this invention are shown below. Figure 2 As shown, the multi-level dynamic background model established by analyzing the LOFAR spectrum is as follows: Figure 3 As shown in the figure, the method of the present invention can extract line spectra from LOFAR spectral images.

[0052] This invention also provides a system for extracting line spectra from LOFAR spectra of two-dimensional images. This system can be implemented by executing the steps of the method for extracting line spectra from LOFAR spectra of two-dimensional images. That is, those skilled in the art can understand the method for extracting line spectra from LOFAR spectra of two-dimensional images as a preferred embodiment of the system. The system includes: Module M1 establishes a dynamic background model for the LOFAR spectrum; Module M2 determines two detection thresholds, high and low, based on the background model; Module M3 performs multi-level target detection and fusion to obtain preliminary detection results; Module M4 performs trajectory tracking growth based on historical information of the line spectrum; Module M5 extracts features from the spectral region for spectral description; Module M6 performs line spectrum filtering based on the expert weighting model to obtain the line spectrum extraction results.

[0053] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0054] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for extracting line spectra from LOFAR spectra of two-dimensional images, characterized in that, include: Step S1: Establish a dynamic background model for the LOFAR spectrum; Step S2: Determine two detection thresholds, high and low, based on the background model; Step S3: Perform multi-level target detection and fusion to obtain preliminary detection results; Step S4: Perform trajectory tracking growth based on the historical information of the line spectrum; Step S5: Extract features from the spectral region to perform spectral description; Step S6: Perform line spectrum filtering based on the expert weighting model to obtain the line spectrum extraction results.

2. The method for extracting line spectra from LOFAR spectra of two-dimensional images according to claim 1, characterized in that, The process of establishing a dynamic background model for the LOFAR spectrum includes: Analyze the LOFAR spectrum and assume... Within time 1, the LOFAR spectral image matrix is: ,in, The row number represents the number of pixels in the time dimension. The column number represents the number of pixels in the frequency dimension; On the LOFAR spectral image row data Establish There are several background models, among which... , , The maximum number of levels in the background model; In the Calculate the rectangular window region at each frequency pixel. mean amplitude and standard deviation , in, Indicates the first Okay, number The amplitude value corresponding to the column pixel position; The average amplitude calculated at each frequency pixel point Composition of multi-level dynamic background models .

3. The method for extracting line spectra from LOFAR spectra of two-dimensional images according to claim 1, characterized in that, The determination of two detection thresholds, high and low, based on the background model includes: According to the average amplitude and the standard deviation In the At different levels of dynamic background, calculate the low detection threshold vector respectively. and high detection threshold vector ,in, , The calculation formulas are as follows: in, Indicates the first The first calculation on a dynamic background Low detection threshold for each pixel Indicates the first The first calculation on a dynamic background A high detection threshold for each pixel. and The sum is the weighting coefficient, which is a constant value, and .

4. The method for extracting line spectra from LOFAR spectra of two-dimensional images according to claim 1, characterized in that, The process of performing multi-level target detection and fusion to obtain preliminary detection results includes: On the LOFAR spectral image row data Perform binarization processing, In the Under dynamic conditions, with the low detection threshold vector and the high detection threshold vector The pixels are compared: pixels with values ​​greater than the detection threshold are marked "1", and pixels with values ​​less than the detection threshold are marked "0", thus obtaining the binary vector detection result. and ,in, , ; Based on the binary vector detection results and Multi-level detection results are fused, and the multi-level detection values ​​at each frequency pixel are statistically analyzed to obtain preliminary detection values. , ; Based on the preliminary detection values Obtain a high-threshold detection vector containing binary detection results of "1" and "0". and low threshold detection vector , , ; For the high threshold detection vector Perform region connectivity processing, grouping pixels with consecutive detection values ​​of "1" into the same connected region, to obtain... There are connected regions, where the scale of each region is the number of pixels with a value of "1", and the corresponding values ​​within the region are... The location of the maximum amplitude value is the frequency detection point of the line spectrum trajectory in each region.

5. The method for extracting LOFAR spectrum lines from a two-dimensional image according to claim 3, characterized in that, The detection result based on the binary vector and Multi-level detection results are fused, and the multi-level detection values ​​at each frequency pixel are statistically analyzed to obtain preliminary detection values. ,include: The preliminary test values The Middle Preliminary detection value at each frequency pixel The calculation formula is as follows: in, Indicates the test results The Middle The value at each frequency pixel. Indicates the test results The Middle The value at each frequency pixel.

6. The method for extracting LOFAR spectrum lines from a two-dimensional image according to claim 3, characterized in that, Based on the preliminary detection value Obtain a high-threshold detection vector containing binary detection results of "1" and "0". and low threshold detection vector ,include: In the high-threshold detection vector, points with a value of "1" represent frequency pixels that are detected using the high threshold at each level of the dynamic model. The Middle Detection value at each frequency pixel The calculation formula is as follows: In the low-threshold detection vector, points with a value of "1" represent frequency pixels that are detected using the low threshold at each level of the dynamic model. The Middle Detection value at each frequency pixel The calculation formula is as follows: in, The preliminary detection value The Middle Preliminary detection values ​​at each frequency pixel.

7. The method for extracting line spectra from LOFAR spectra of two-dimensional images according to claim 1, characterized in that, The trajectory tracking growth based on historical information of the line spectrum includes: According to the preliminary test results, the first Row data detection results and the first Linear trajectory tracking and growth are performed on the line data line trajectory tracking results, where, for the first... In the row of data, in the first row... The frequency pixel position of each line trajectory point in the row data is centered, and the bandwidth is pixels. Track within the search area; Prioritize detection within the search area ,like If a detection point exists, the trajectory grows; if the trajectory is lost, the counter is reset to zero. like If no detection point is found, then... Search within this search area, if If a detection point exists, the trajectory grows, but the trajectory loss counter increments by 1; if There are no detection points in the middle, and If there are no detection points in the middle, the trajectory will not grow, and the trajectory loss counter will increment by 1; When the trajectory loss counter reaches the preset upper limit When the time comes, determine that the trajectory of the line spectrum has ended and cancel it; For those not matched by any trajectory The detection point is initialized as a newly added line trajectory point.

8. The method for extracting line spectra from LOFAR spectra of two-dimensional images according to claim 1, characterized in that, The extraction of features from the spectral region for spectral description includes: Based on the tracked line spectrum trajectory, analyze the information of the connected region where the line spectrum trajectory is located, and extract the scale, amplitude, and curvature features of the line spectrum. Among them, the scale feature includes the average and standard deviation of the scale of the line spectrum region, the amplitude feature includes the average and standard deviation of the amplitude of the line spectrum region, and the curvature feature includes the mean and standard deviation of the curvature of the line spectrum trajectory points.

9. The method for extracting line spectra from LOFAR spectra of two-dimensional images according to claim 1, characterized in that, The line spectrum filtering based on the expert weighting model, to obtain the line spectrum extraction results, includes: An expert weight equation is established to calculate the probability of each line spectrum, and line spectrum screening is performed to remove pseudo-line spectra that do not meet the conditions. The expression of the expert weight equation is as follows: in, Represents the probability value of the line spectrum. This represents the total dimension of the features. Indicates the first dimensional features, and Indicates the first The weighted values ​​and eigenvalues ​​of the dimensional features; Based on condition threshold Perform line spectrum analysis, if If, then the musical line is preserved, if If so, the musical line is removed.

10. A system for extracting line spectra from LOFAR spectra of two-dimensional images, characterized in that, include: Module M1 establishes a dynamic background model for the LOFAR spectrum; Module M2 determines two detection thresholds, high and low, based on the background model; Module M3 performs multi-level target detection and fusion to obtain preliminary detection results; Module M4 performs trajectory tracking growth based on historical information of the line spectrum; Module M5 extracts features from the spectral region for spectral description; Module M6 performs line spectrum filtering based on the expert weighting model to obtain the line spectrum extraction results.

Images