Characteristic band selection method and system for hyperspectral remote sensing image target detection

Abstract

The application provides a characteristic spectrum selection method and system for hyperspectral remote sensing image target detection. It relates to the field of image detection, and the method steps are as follows: acquiring hyperspectral remote sensing image data, each of which includes M spectrum bands; performing target detection on each spectrum band in each hyperspectral remote sensing image data; recording effective recognition events and calculating corresponding signal-to-clutter ratios; determining a signal-to-clutter standard value based on the signal-to-clutter ratios corresponding to the effective recognition events; ensuring that the signal-to-clutter ratios of no less than a proportion of the effective recognition events are higher than the signal-to-clutter standard value; for each spectrum band, based on the number of corresponding effective recognition events, counting the proportion of the effective recognition events with signal-to-clutter ratios higher than the signal-to-clutter standard value to retain the corresponding spectrum band; and taking all the retained spectrum bands as initial selected spectrum bands; and adjusting the boundaries of the initial selected spectrum bands in an adaptive manner to obtain final spectrum ranges.

Description

Technical Field

[0001] This invention relates to the field of image detection, and more particularly to a method and system for selecting characteristic spectral bands for target detection in hyperspectral remote sensing images. Background Technology

[0002] Because hyperspectral remote sensing images acquired by spaceborne sensors record information in hundreds of narrow and continuous spectral bands in the same scene, they have high spectral resolution and are beneficial for recording subtle differences between targets and backgrounds, which has important practical significance in ground target detection and identification.

[0003] However, hyperspectral imagery generally suffers from low spatial resolution, high wavenumber redundancy, and significant noise interference. Therefore, in target detection applications, selecting the optimal spectral bands suitable for target detection from hundreds of complex spectral bands becomes a challenge. Current mainstream hyperspectral characteristic spectral band selection methods typically rely on spectral band combination searches based on certain evaluation criteria functions to achieve dimensionality reduction. This approach only improves the characteristic spectral band selection effect by addressing spectral information redundancy, without considering the application's specific needs, such as target detection performance.

[0004] Chinese invention patent CN118229955A, entitled "A Hyperspectral Target Detection Method, System, Device, and Medium," includes: clustering a hyperspectral image to be detected to obtain spatial sub-regions; performing adaptive two-dimensional metaspectral analysis on each spatial sub-region of each single-band image to obtain a reconstructed hyperspectral image; determining the probability score of each pixel as a target pixel using a constrained energy minimization algorithm, and determining a target dictionary and a background dictionary based on the probability scores; characterizing each pixel in the reconstructed hyperspectral image using a sparse representation model based on the target dictionary and the background dictionary to obtain background characterization residuals and target characterization residuals; determining an output result based on the difference between the background characterization residuals and the target characterization residuals, and determining the detection result of the target object based on the output result. This achieves superior target detection performance, enabling more accurate and efficient target detection and identification, demonstrating excellent technical performance. However, it does not disclose a technical solution for specifically improving target detection performance based on the signal-to-clutter ratio. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention aims to provide a characteristic spectral band selection method and system for target detection in hyperspectral remote sensing images. This method solves the problem of hyperspectral spectral band information redundancy and can specifically improve the target detection effect, thereby enhancing the performance of subsequent target detection-based applications.

[0006] To achieve the above-mentioned objectives, this invention provides a method for selecting characteristic spectral bands for target detection in hyperspectral remote sensing images, comprising the following steps:

[0007] Acquire N hyperspectral remote sensing image data collected by the same satellite, each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1;

[0008] Target detection is performed on each spectral band of each hyperspectral remote sensing image data. The effectiveness of the target identification is determined based on the detection results. If the preset effective identification conditions are met, it is recorded as a valid identification event, and the corresponding signal-to-clutter ratio is calculated. The results are then statistically analyzed. One valid identification event;

[0009] based on For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified events is higher than the standard signal-to-noise ratio;

[0010] For each spectral band i, based on the corresponding number of valid identified events... The proportion of validly identified events with a statistically higher signal-to-noise ratio than the standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments;

[0011] Where i is the spectral band number; ; This is the redundancy ratio, and ;

[0012] An adaptive approach is used to adjust the initial selected spectral band boundaries to obtain the final spectral band range;

[0013] Target recognition metrics include at least one of target recognition rate and false alarm rate.

[0014] According to one technical solution of the present invention, the process of determining the noise standard value includes:

[0015] Establish a spectrum-signal-noise ratio coordinate plot;

[0016] The band-signal-noise ratio plot is a two-dimensional rectangular coordinate system, with the horizontal axis representing the band number and the vertical axis representing the signal-noise ratio.

[0017] Based on the spectral band and signal-to-noise ratio corresponding to each valid identification event, plot the representation points representing the valid identification event at the corresponding coordinates in the spectral band-signal-to-noise ratio coordinate graph;

[0018] Define a signal-to-noise ratio (SNR) standard line parallel to the x-axis; such that the proportion of characterization points above the SNR standard line among all characterization points is . ;

[0019] Obtain the ordinate value of the signal-to-noise ratio standard line as the basis for determining the signal-to-noise ratio standard value.

[0020] According to one technical solution of the present invention, an adaptive method is used to adjust the initial selected spectral band boundary to obtain the final spectral band range, including:

[0021] Target detection was performed on all hyperspectral remote sensing image data based on the initially selected spectral band range, and the baseline target detection rate and baseline false alarm rate were obtained respectively.

[0022] The boundaries of the initially selected spectral band range are adjusted respectively, and after each adjustment, target detection is performed on all hyperspectral remote sensing image data based on the adjusted spectral band range to obtain the current target detection rate and the current false alarm rate.

[0023] Adjust to expand or shrink the spectral range by a set spectral step size;

[0024] If the current target detection rate is less than the baseline target detection rate, or the current false alarm rate is greater than the baseline false alarm rate, the target recognition index is determined to have deteriorated relative to the initial spectral range, and the adjustment is terminated.

[0025] The most recent spectral range before the target recognition index deteriorated will be used as the final selected spectral range.

[0026] According to one technical solution of the present invention, the calculation process of the signal-to-noise ratio includes:

[0027] Process hyperspectral remote sensing image data into two-dimensional remote sensing images;

[0028] Based on the labeled target region, the average value of the target pixels is calculated. ;

[0029] The average value of background pixels is calculated based on the background region. and background pixel standard deviation ;

[0030] The background area is the region in the two-dimensional remote sensing image that excludes the target area;

[0031] The signal-to-noise ratio is calculated using the following formula. :

[0032] .

[0033] According to one technical solution of the present invention, the target region is a first side with a length of The length of the second side is A rectangular area;

[0034] The first side length of the background area is The length of the second side is ;

[0035] in, =3~5;

[0036] The target area is located in the center of the background area.

[0037] The present invention also provides a characteristic spectral band selection system for target detection in hyperspectral remote sensing images, comprising:

[0038] The signal-to-noise ratio standard value extraction module is used to acquire N hyperspectral remote sensing image data collected by the same satellite. Each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1.

[0039] Furthermore, it is used to perform target detection on each spectral band in each hyperspectral remote sensing image data; based on the target detection results, it determines whether the identification is effective. If the preset effective identification conditions are met, it is recorded as an effective identification event, and the corresponding signal-to-clutter ratio is calculated, and then statistically obtained. One valid identification event;

[0040] And, for use based on For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified events is higher than the standard signal-to-noise ratio;

[0041] The characteristic spectral band sensitivity analysis module is used to analyze the corresponding number of valid identified events for each spectral band i. The proportion of validly identified events with a statistically higher signal-to-noise ratio than the standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments;

[0042] Where i is the spectral band number; ; This is the redundancy ratio, and ;

[0043] And, it is used to adaptively adjust the initial selected spectral band boundaries to obtain the final spectral band range;

[0044] Target recognition metrics include at least one of target recognition rate and false alarm rate.

[0045] According to one technical solution of the present invention, a signal-to-noise ratio standard value extraction module includes:

[0046] The coordinate graph generation module is used to create spectral band-signal-noise ratio coordinate graphs;

[0047] The band-signal-noise ratio plot is a two-dimensional rectangular coordinate system, with the horizontal axis representing the band number and the vertical axis representing the signal-noise ratio.

[0048] The characterization point marking module is used to draw characterization points representing valid identification events at the corresponding coordinates in the spectral band-signal-noise ratio coordinate graph based on the spectral band and signal-to-noise ratio corresponding to each valid identification event.

[0049] The signal-to-noise ratio (SNR) standard value determination module is used to determine a SNR standard line parallel to the x-axis; such that the proportion of characterization points above the SNR standard line among all characterization points is 1 / 2. ;

[0050] Obtain the ordinate value of the signal-to-noise ratio standard line as the basis for determining the signal-to-noise ratio standard value.

[0051] According to one technical solution of the present invention, the signal-to-noise ratio standard value extraction module further includes:

[0052] The adaptive adjustment module is used for:

[0053] Target detection was performed on all hyperspectral remote sensing image data based on the initially selected spectral band range, and the baseline target detection rate and baseline false alarm rate were obtained respectively.

[0054] The boundaries of the initially selected spectral band range are adjusted respectively, and after each adjustment, target detection is performed on all hyperspectral remote sensing image data based on the adjusted spectral band range to obtain the current target detection rate and the current false alarm rate.

[0055] Adjust to expand or shrink the spectral range by a set spectral step size;

[0056] If the current target detection rate is less than the baseline target detection rate, or the current false alarm rate is greater than the baseline false alarm rate, the target recognition index is determined to have deteriorated relative to the initial spectral range, and the adjustment is terminated.

[0057] The most recent spectral range before the target recognition index deteriorated will be used as the final selected spectral range.

[0058] According to one technical solution of the present invention, the signal-to-noise ratio standard value extraction module further includes:

[0059] The signal-to-noise ratio calculation module is used for:

[0060] Process hyperspectral remote sensing image data into two-dimensional remote sensing images;

[0061] Based on the labeled target region, the average value of the target pixels is calculated. ;

[0062] The average value of background pixels is calculated based on the background region. and background pixel standard deviation ;

[0063] The background area is the region in the two-dimensional remote sensing image that excludes the target area;

[0064] The signal-to-noise ratio is calculated using the following formula. :

[0065] .

[0066] According to one technical solution of the present invention, the target region is a first side with a length of The length of the second side is A rectangular area;

[0067] The first side length of the background area is The length of the second side is ;

[0068] in, =3~5;

[0069] The target area is located in the center of the background area.

[0070] Compared with the prior art, the present invention has the following advantages:

[0071] The present invention provides a characteristic spectral band selection method and system for target detection in hyperspectral remote sensing images, which has the following beneficial effects:

[0072] By combining the signal-to-noise ratio (SNR) of a single spectral band with the target detection results, the lowest SNR value is selected when almost all (around 95%) targets are detectable, and this is taken as the "standard value" of the SNR for target detection. Based on this, the initial spectral bands for characteristic spectral band selection are used. Then, by adjusting the spectral band boundaries, sensitivity analysis of characteristic spectral band selection is performed to determine the final characteristic spectral band selection results suitable for target detection.

[0073] The ability to selectively combine target detection results with characteristic spectral bands not only solves the problem of information redundancy in hyperspectral bands, but also improves target detection results in a targeted manner, thereby enhancing the performance of subsequent target detection-based applications.

[0074] Compared with the existing detection results, the addition of characteristic spectral bands has improved the detection results of different spectral ranges to varying degrees, with an average improvement of 6%, and the detection accuracy meets the project requirements. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0076] Figure 1 A flowchart illustrating a characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to an embodiment of the present invention;

[0077] Figure 2 The diagram illustrates a spectral band-signal-noise ratio coordinate graph of a characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to an embodiment of the present invention.

[0078] Figure 3 The diagram illustrates the principle of signal-to-noise ratio calculation in a characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to an embodiment of the present invention. Detailed Implementation

[0079] The description of the embodiments in this specification should be taken in conjunction with the accompanying drawings, which should form part of the complete specification. In the drawings, the shape or thickness of the embodiments may be exaggerated and may be indicated in a simplified or convenient manner. Furthermore, parts of the various structures in the drawings will be described separately; it is worth noting that elements not shown in the figures or not described in words are in a form known to those skilled in the art.

[0080] The descriptions of the embodiments herein, including any references to directions and orientations, are for ease of description only and should not be construed as limiting the scope of the invention. The following description of preferred embodiments involves combinations of features, which may exist independently or in combination; the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims. Specific Implementation Method 1

[0082] This embodiment provides a characteristic spectral band selection method for target detection in hyperspectral remote sensing images, comprising the following steps:

[0083] Acquire N hyperspectral remote sensing image data collected by the same satellite, each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1;

[0084] Target detection is performed on each spectral band of each hyperspectral remote sensing image data. The effectiveness of the target identification is determined based on the detection results. If the preset effective identification conditions are met, it is recorded as a valid identification event, and the corresponding signal-to-clutter ratio is calculated. The results are then statistically analyzed. One valid identification event;

[0085] based on For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified events is higher than the standard signal-to-noise ratio;

[0086] For each spectral band i, based on the corresponding number of valid identified events... The proportion of validly identified events with a statistically higher signal-to-noise ratio than the standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments;

[0087] Where i is the spectral band number; ; This is the redundancy ratio, and ;

[0088] An adaptive approach is used to adjust the initial selected spectral band boundaries to obtain the final spectral band range;

[0089] Target recognition metrics include at least one of target recognition rate and false alarm rate.

[0090] In this embodiment, such as Figure 1 As shown, the specific steps of the method are as follows:

[0091] S1. Acquire N hyperspectral remote sensing image data collected by the same satellite. Each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1.

[0092] Specifically, sufficient hyperspectral remote sensing image data is acquired, and a hyperspectral remote sensing image dataset containing the target is constructed through screening and preprocessing.

[0093] Among them, the hyperspectral remote sensing image data covers complex backgrounds (residential areas, farmland, mountains, etc.) and single backgrounds (clouds, sea, etc.), as well as different time phases (different solar altitude angles).

[0094] Background data for various targets should be balanced (the number of targets with complex backgrounds should be balanced with those with simple backgrounds). Hyperspectral remote sensing image data should be preprocessed to ensure consistent data levels (e.g., all data should be radiance data, grayscale data, or reflectance / emissivity data). Preprocessing methods such as data resampling should be used to ensure consistent spectral and spatial resolution of the sample data. Preprocessing methods such as projection transformation should be used to ensure consistent projection systems for the sample data, guaranteeing that target locations are on the same reference.

[0095] Then, use annotation tools (such as Envi software) to annotate the targets in the hyperspectral remote sensing image data, using polygon, rectangle and other annotation forms to annotate the target outline as standardly as possible, obtain the target pixel coordinates, mark the location of the target, and construct the target sample dataset.

[0096] S2. Target detection is performed on each spectral band in each hyperspectral remote sensing image data. Based on the target detection results, it is determined whether the identification is effective. If the preset effective identification conditions are met, it is recorded as a valid identification event, and the corresponding signal-to-noise ratio is calculated, and then statistically obtained. One valid identification event;

[0097] based on For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified events is higher than the standard signal-to-noise ratio;

[0098] Specifically, for each spectral band in the sample dataset, the signal-to-clutter ratio (SCR) of the target to the background is calculated; that is, each spectral band in the manually labeled target sample dataset is treated as a single image, and for all images of the same spectral band, the SCR of the target to the background is calculated, while simultaneously using an anomaly detection algorithm for all spectral bands ( ) to perform target detection.

[0099] For each spectral band, target detection is performed using the classic RX anomaly detection algorithm;

[0100]

[0101]

[0102]

[0103] in The spectrum of the pixel to be detected. This is the average value of the background window. The background window covariance matrix, It is the inverse of the covariance matrix. This represents the total number of pixels in the image.

[0104] Among them, meeting the preset valid identification conditions means meeting the preset anomaly detection threshold.

[0105] Target detectable signal-to-noise ratio calculation;

[0106] By combining the signal-to-noise ratio (SNR) and target detection performance of a single spectral band, and summarizing all continuous spectral bands, it was found that all spectral bands achieved a 95% (proportion) accuracy. ) Target detectable At least a proportion of valid identified events The minimum signal-to-noise ratio (SNR) value is used as the standard SNR value for the target to be detectable (effective identification event).

[0107] S3. For each spectral segment i, based on the corresponding number of valid identification events... The proportion of validly identified events with a statistically higher signal-to-noise ratio than the standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments;

[0108] Where i is the spectral band number; ; This is the redundancy ratio, and ;

[0109] Specifically, based on the detectable "signal-to-noise ratio standard value", characteristic spectral bands are selected from the hyperspectral remote sensing image, and the signal-to-noise ratio is calculated. At that time, the range of spectral values ​​is used as the starting spectral range for the sensitivity analysis of the characteristic spectral range, where, It is the standard value of signal-to-noise ratio, and i is the wavenumber (the number of the spectral band).

[0110] S4. The initial selected spectral band boundaries are adjusted using an adaptive method to obtain the final spectral band range;

[0111] Target recognition metrics include at least one of target recognition rate and false alarm rate.

[0112] Specifically, by expanding or shrinking the boundary of the initially selected spectral band, the spectral band with the best detection effect is selected in an adaptive manner, thereby determining the characteristic spectral band selection result suitable for target detection.

[0113] The final spectral range obtained in this embodiment can be applied to target detection in hyperspectral remote sensing images of the same satellite without any changes to the satellite's remote sensing hardware. After acquiring other hyperspectral remote sensing images from the same satellite, the final determined spectral range can be used directly for target detection. Specific Implementation Method Two

[0115] This embodiment is a further explanation of embodiment one. In this embodiment, the process of determining the noise standard value includes:

[0116] Establish a spectrum-signal-noise ratio coordinate plot;

[0117] The band-signal-noise ratio plot is a two-dimensional rectangular coordinate system, with the horizontal axis representing the band number and the vertical axis representing the signal-noise ratio.

[0118] Based on the spectral band and signal-to-noise ratio corresponding to each valid identification event, plot the representation points representing the valid identification event at the corresponding coordinates in the spectral band-signal-to-noise ratio coordinate graph;

[0119] Define a signal-to-noise ratio (SNR) standard line parallel to the x-axis; such that the proportion of characterization points above the SNR standard line among all characterization points is . ;

[0120] Obtain the ordinate value of the signal-to-noise ratio standard line as the basis for determining the signal-to-noise ratio standard value.

[0121] In this embodiment, such as Figure 2 As shown, to facilitate statistical analysis, the signal-to-noise ratio (SNR) of the detectable spectral bands of the target is plotted as a spectral band-SNR coordinate graph. Based on statistical or observational methods, the minimum SNR value under 95% target detectability is obtained as the "standard SNR value" for measuring target detectability. Specific Implementation Method 3

[0123] This embodiment is a further explanation of embodiment one or two. In this embodiment, an adaptive approach is used to adjust the initial selected spectral band boundary to obtain the final spectral band range, including:

[0124] Target detection was performed on all hyperspectral remote sensing image data based on the initially selected spectral band range, and the baseline target detection rate and baseline false alarm rate were obtained respectively.

[0125] The boundaries of the initially selected spectral band range are adjusted respectively, and after each adjustment, target detection is performed on all hyperspectral remote sensing image data based on the adjusted spectral band range to obtain the current target detection rate and the current false alarm rate.

[0126] Adjust to expand or shrink the spectral range by a set spectral step size;

[0127] If the current target detection rate is less than the baseline target detection rate, or the current false alarm rate is greater than the baseline false alarm rate, the target recognition index is determined to have deteriorated relative to the initial spectral range, and the adjustment is terminated.

[0128] The most recent spectral range before the target recognition index deteriorated will be used as the final selected spectral range.

[0129] In this embodiment, the characteristic spectral band sensitivity analysis is performed by expanding or shrinking the boundary of the initially selected spectral band (e.g., performing target detection to obtain the detection rate and false alarm rate). The local optimum principle of "stopping spectral band amplification when the detection rate decreases or the false alarm rate increases" is adopted to select the spectral band with the best detection effect, thereby determining the characteristic spectral band selection result suitable for target detection.

[0130] This involves a combination problem of selecting characteristic spectral bands. If the existing spectral band selection method is used, it requires traversing a total of [number missing] times. There are several methods for selecting characteristic spectral bands, but the number of hyperspectral bands is very high, resulting in extremely low algorithm efficiency. According to the characteristic spectral band sensitivity analysis method of this patent, spectral band selection only needs to start from the initial spectral band range and then select bands by scaling the boundaries, greatly improving the efficiency of spectral band selection. Specific Implementation Method Four

[0132] This embodiment is a further explanation of embodiment three. In this embodiment, the calculation process of the signal-to-noise ratio includes:

[0133] Process hyperspectral remote sensing image data into two-dimensional remote sensing images;

[0134] Based on the labeled target region, the average value of the target pixels is calculated. ;

[0135] The average value of background pixels is calculated based on the background region. and background pixel standard deviation ;

[0136] The background area is the region in the two-dimensional remote sensing image that excludes the target area;

[0137] The signal-to-noise ratio is calculated using the following formula. :

[0138] .

[0139] In this embodiment, based on Figure 3 The target region and background region are compared, and the SNR (Signal-to-Noise Ratio) is calculated. SNR represents the separability of the target and background in that frequency band. The differences between the target and background are considered, enabling the distinction between "significant differences" and "differences caused by noise."

[0140] Furthermore, in hyperspectral analysis, the brightness of each spectral band may differ, but SNR can place them within the same comparable framework. This allows for comparison of the effects of different spectral bands. Detailed Implementation Method Five

[0142] This embodiment is a further explanation of embodiment four. In this embodiment, the target region is the first region with a side length of... The length of the second side is A rectangular area;

[0143] The first side length of the background area is The length of the second side is ;

[0144] in, =3~5;

[0145] The target area is located in the center of the background area.

[0146] In this embodiment, such as Figure 3 As shown, That is, the background area is 3 to 5 times the number of pixels occupied by the target area extrapolated from the target area (excluding the target position). Specific Implementation Method Six

[0148] This embodiment provides a characteristic spectral band selection system for target detection in hyperspectral remote sensing images, comprising:

[0149] The signal-to-noise ratio standard value extraction module is used to acquire N hyperspectral remote sensing image data collected by the same satellite. Each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1.

[0150] Furthermore, it is used to perform target detection on each spectral band in each hyperspectral remote sensing image data; based on the target detection results, it determines whether the identification is effective. If the preset effective identification conditions are met, it is recorded as an effective identification event, and the corresponding signal-to-clutter ratio is calculated, and then statistically obtained. One valid identification event;

[0151] And, for use based on For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified events is higher than the standard signal-to-noise ratio;

[0152] The characteristic spectral band sensitivity analysis module is used to analyze the corresponding number of valid identified events for each spectral band i. The proportion of validly identified events with a statistically higher signal-to-noise ratio than the standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments;

[0153] Where i is the spectral band number; ; This is the redundancy ratio, and ;

[0154] And, it is used to adaptively adjust the initial selected spectral band boundaries to obtain the final spectral band range;

[0155] Target recognition metrics include at least one of target recognition rate and false alarm rate. Detailed Implementation Method Seven

[0157] This embodiment is a further explanation of embodiment six. In this embodiment, the signal-to-noise ratio standard value extraction module includes:

[0158] The coordinate graph generation module is used to create spectral band-signal-noise ratio coordinate graphs;

[0159] The band-signal-noise ratio plot is a two-dimensional rectangular coordinate system, with the horizontal axis representing the band number and the vertical axis representing the signal-noise ratio.

[0160] The characterization point marking module is used to draw characterization points representing valid identification events at the corresponding coordinates in the spectral band-signal-noise ratio coordinate graph based on the spectral band and signal-to-noise ratio corresponding to each valid identification event.

[0161] The signal-to-noise ratio (SNR) standard value determination module is used to determine a SNR standard line parallel to the x-axis; such that the proportion of characterization points above the SNR standard line among all characterization points is 1 / 2. ;

[0162] Obtain the ordinate value of the signal-to-noise ratio standard line as the basis for determining the signal-to-noise ratio standard value. Detailed Implementation Method Eight

[0164] This embodiment is a further explanation of embodiment six or seven. In this embodiment, the signal-to-noise ratio standard value extraction module further includes:

[0165] The adaptive adjustment module is used for:

[0166] Target detection was performed on all hyperspectral remote sensing image data based on the initially selected spectral band range, and the baseline target detection rate and baseline false alarm rate were obtained respectively.

[0167] The boundaries of the initially selected spectral band range are adjusted respectively, and after each adjustment, target detection is performed on all hyperspectral remote sensing image data based on the adjusted spectral band range to obtain the current target detection rate and the current false alarm rate.

[0168] Adjust to expand or shrink the spectral range by a set spectral step size;

[0169] If the current target detection rate is less than the baseline target detection rate, or the current false alarm rate is greater than the baseline false alarm rate, the target recognition index is determined to have deteriorated relative to the initial spectral range, and the adjustment is terminated.

[0170] The most recent spectral range before the target recognition index deteriorated will be used as the final selected spectral range. Detailed Implementation Method Nine

[0172] This embodiment is a further explanation of embodiment eight. In this embodiment, the signal-to-noise ratio standard value extraction module further includes:

[0173] The signal-to-noise ratio calculation module is used for:

[0174] Process hyperspectral remote sensing image data into two-dimensional remote sensing images;

[0175] Based on the labeled target region, the average value of the target pixels is calculated. ;

[0176] The average value of background pixels is calculated based on the background region. and background pixel standard deviation ;

[0177] The background area is the region in the two-dimensional remote sensing image that excludes the target area;

[0178] The signal-to-noise ratio is calculated using the following formula. :

[0179] . Detailed Implementation Method Ten

[0181] This embodiment is a further explanation of embodiment nine. In this embodiment, the target region is the first region with a side length of... The length of the second side is A rectangular area;

[0182] The first side length of the background area is The length of the second side is ;

[0183] in, =3~5;

[0184] The target area is located in the center of the background area.

[0185] The present invention discloses a characteristic spectral band selection method and system for target detection in hyperspectral remote sensing images. The method comprises the following steps: acquiring hyperspectral remote sensing image data, each hyperspectral remote sensing image data including M spectral bands; performing target detection on each spectral band in each hyperspectral remote sensing image data; recording valid identification events and calculating the corresponding signal-to-noise ratio (SNR); determining a SNR standard value based on the SNR corresponding to the valid identification events; ensuring that the SNR is not lower than the specified proportion. The effective identification events have a signal-to-noise ratio (SNR) higher than the SNR standard value. For each spectral band, based on the number of corresponding effective identification events, the proportion of effective identification events with an SNR higher than the SNR standard value is counted to retain the corresponding spectral band. All retained spectral bands are used as the initial selected spectral bands. The boundaries of the initial selected spectral bands are adjusted in an adaptive manner to obtain the final spectral band range.

[0186] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0187] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0188] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0189] It should also be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0190] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for selecting characteristic spectral bands for target detection in hyperspectral remote sensing images, characterized in that, The steps are as follows: Acquire N hyperspectral remote sensing image data collected by the same satellite, each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1; Target detection is performed on each spectral band of each hyperspectral remote sensing image data. The effectiveness of the target identification is determined based on the detection results. If the preset effective identification conditions are met, it is recorded as a valid identification event, and the corresponding signal-to-clutter ratio is calculated. The results are then statistically analyzed. One valid identification event; Based on the above For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified event is higher than the aforementioned signal-to-noise standard value; For each spectral band i, based on the corresponding number of valid identified events... The proportion of valid identification events with a signal-to-noise ratio higher than the aforementioned signal-to-noise standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments; Where i is the spectral band number; ; This is the redundancy ratio, and ; The initial selected spectral band boundaries are adjusted in an adaptive manner to obtain the final spectral band range; The target recognition metrics include at least one of target recognition rate and false alarm rate.

2. The characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to claim 1, characterized in that, The process of determining the standard values ​​for information and noise includes: Establish a spectrum-signal-noise ratio coordinate plot; The spectral band-signal-noise ratio coordinate graph is a two-dimensional rectangular coordinate system, with the horizontal axis representing the spectral band number and the vertical axis representing the signal-noise ratio. Based on the spectral band and signal-to-noise ratio corresponding to each valid identification event, a representation point representing the valid identification event is plotted at the corresponding coordinates in the spectral band-signal-to-noise ratio coordinate graph; Define a signal-to-noise ratio (SNR) standard line parallel to the x-axis; such that the proportion of characterization points located above the SNR standard line among all characterization points is . ; Obtain the ordinate value of the signal-to-noise ratio standard line, and use it as the determined signal-to-noise ratio standard value.

3. The characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to claim 1 or 2, characterized in that, The initial selected spectral band boundaries are adjusted adaptively to obtain the final spectral band range, including: Target detection was performed on all hyperspectral remote sensing image data based on the initially selected spectral band range, and the baseline target detection rate and baseline false alarm rate were obtained respectively. The boundaries of the initially selected spectral band range are adjusted respectively, and after each adjustment, target detection is performed on all hyperspectral remote sensing image data based on the adjusted spectral band range to obtain the current target detection rate and the current false alarm rate. The adjustment is to expand or shrink the spectral range by a set spectral range step size; If the current target detection rate is less than the benchmark target detection rate, or the current false alarm rate is greater than the benchmark false alarm rate, then the target recognition index is determined to have deteriorated relative to the initial spectral range, and the adjustment is terminated. The most recent spectral range before the target recognition index deteriorated will be used as the final selected spectral range.

4. The characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to claim 3, characterized in that, The calculation process for the signal-to-noise ratio includes: Process hyperspectral remote sensing image data into two-dimensional remote sensing images; Based on the labeled target region, the average value of the target pixels is calculated. ; The average value of background pixels is calculated based on the background region. and background pixel standard deviation ; The background region is the region in the two-dimensional remote sensing image excluding the target region; The signal-to-noise ratio is calculated using the following formula. : 。 5. The characteristic spectral band selection method for target detection in hyperspectral remote sensing images according to claim 4, characterized in that, The target region is the first side length of The length of the second side is A rectangular area; The first side length of the background region is The length of the second side is ; in, =3~5; The target area is located in the center of the background area.

6. A characteristic spectral band selection system for target detection in hyperspectral remote sensing images, characterized in that, include: The signal-to-noise ratio standard value extraction module is used to acquire N hyperspectral remote sensing image data collected by the same satellite. Each hyperspectral remote sensing image data includes M spectral bands; where N and M are both positive integers greater than 1. Furthermore, it is used to perform target detection on each spectral band in each hyperspectral remote sensing image data; based on the target detection results, it determines whether the identification is effective. If the preset effective identification conditions are met, it is recorded as an effective identification event, and the corresponding signal-to-clutter ratio is calculated, and then statistically obtained. One valid identification event; And, for use based on the For each valid identified event, determine the signal-to-noise ratio (SNR) and establish a standard SNR value; ensure that the SNR is not lower than the specified value. The signal-to-noise ratio of the effectively identified event is higher than the aforementioned signal-to-noise standard value; The characteristic spectral band sensitivity analysis module is used to analyze the corresponding number of valid identified events for each spectral band i. The proportion of valid identification events with a signal-to-noise ratio higher than the aforementioned signal-to-noise standard value. ;if If so, the spectral segment is retained; and all retained spectral segments are used as the initial selected spectral segments; Where i is the spectral band number; ; This is the redundancy ratio, and ; And, to adaptively adjust the initial selected spectral band boundaries to obtain the final spectral band range; The target recognition metrics include at least one of target recognition rate and false alarm rate.

7. The characteristic spectral band selection system for target detection in hyperspectral remote sensing images according to claim 6, characterized in that, The signal-to-noise ratio standard value extraction module includes: The coordinate graph generation module is used to create spectral band-signal-noise ratio coordinate graphs; The spectral band-signal-noise ratio coordinate graph is a two-dimensional rectangular coordinate system, with the horizontal axis representing the spectral band number and the vertical axis representing the signal-noise ratio. The characterization point marking module is used to draw characterization points representing valid identification events at corresponding coordinates in the spectrum-signal-noise ratio coordinate graph based on the spectrum and signal-to-noise ratio corresponding to each valid identification event. The signal-to-noise ratio (SNR) standard value determination module is used to determine a SNR standard line parallel to the x-axis; such that the proportion of characterization points located above the SNR standard line among all characterization points is . ; Obtain the ordinate value of the signal-to-noise ratio standard line, and use it as the determined signal-to-noise ratio standard value.

8. The characteristic spectral band selection system for target detection in hyperspectral remote sensing images according to claim 6 or 7, characterized in that, The signal-to-noise ratio standard value extraction module also includes: The adaptive adjustment module is used for: Target detection was performed on all hyperspectral remote sensing image data based on the initially selected spectral band range, and the baseline target detection rate and baseline false alarm rate were obtained respectively. The boundaries of the initially selected spectral band range are adjusted respectively, and after each adjustment, target detection is performed on all hyperspectral remote sensing image data based on the adjusted spectral band range to obtain the current target detection rate and the current false alarm rate. The adjustment is to expand or shrink the spectral range by a set spectral range step size; If the current target detection rate is less than the benchmark target detection rate, or the current false alarm rate is greater than the benchmark false alarm rate, then the target recognition index is determined to have deteriorated relative to the initial spectral range, and the adjustment is terminated. The most recent spectral range before the target recognition index deteriorated will be used as the final selected spectral range.

9. The characteristic spectral band selection system for target detection in hyperspectral remote sensing images according to claim 8, characterized in that, The signal-to-noise ratio standard value extraction module also includes: The signal-to-noise ratio calculation module is used for: Process hyperspectral remote sensing image data into two-dimensional remote sensing images; Based on the labeled target region, the average value of the target pixels is calculated. ; The average value of background pixels is calculated based on the background region. and background pixel standard deviation ; The background region is the region in the two-dimensional remote sensing image excluding the target region; The signal-to-noise ratio is calculated using the following formula. : 。 10. The characteristic spectral band selection system for target detection in hyperspectral remote sensing images according to claim 9, characterized in that, The target region is the first side length of The length of the second side is A rectangular area; The first side length of the background region is The length of the second side is ; in, =3~5; The target area is located in the center of the background area.

Images