CN118261808A - Camouflage target enhancement method - Google Patents

Abstract

The invention discloses a camouflage target enhancement method, which adopts a hyperspectral image acquired by a hyperspectral camera as input, reduces the dimension of an original image through PCA principal component analysis, and then utilizes an SPA continuous projection algorithm to carry out spectral feature selection on 16 feature wave bands obtained by dimension reduction to obtain an optimal wave band combination; the optimal wave band combination is highly positively correlated with the characteristics of a camouflage target, the wave band combination is fused to be used as an R channel, the rest wave bands positively correlated with the background are averaged to be used as a B channel and a G channel, and an RGB pseudo-color image is finally generated; the obtained pseudo-color image marks pixels of the camouflage target as red, and the backgrounds such as vegetation, mountain and the like in the scene are marked as bluish green, so that the effect of enhancing the camouflage target is achieved.

Description

Camouflage target enhancement method

Technical Field

The invention belongs to the technical field of computer vision, and particularly relates to a camouflage target enhancement method.

Background

Camouflage target segmentation is an emerging field of computer vision in recent years, and is also an important branch of computer vision segmentation tasks, and has received attention from many researchers in recent years. Camouflage targets refer to objects which are highly similar to or are blocked by the background, and are usually skillfully fused with the environment, so that the colors, the postures and the like of the objects are highly similar to the environment, and the objects are camouflaged, so that the objects are difficult to find. For example, chameleons living in deserts, polar bears on ice layers, soldiers wearing camouflage clothing, and the like, are all referred to as camouflage targets. And the camouflage target segmentation aims at detecting camouflage targets in the visual scene and separating the camouflage targets from the background to extract useful camouflage target information. Camouflage target segmentation can be used in many applications, in addition to its own scientific research value, in computer vision (for search and rescue work, discovery of rare animals), medical image segmentation (such as polyp segmentation, lung infection segmentation, retinal image segmentation), agriculture (disaster detection, locust detection), art (lifelike mix, entertainment art), military exploration, etc.

The most traditional method is to enhance the image of the object in the scene by a system relying on visible light, generally, an image processing technology and algorithm are applied to improve the visual quality of the object in the scene, and enhance the edge, contrast, brightness and other characteristics of the object so as to better observe and analyze the object. Target enhancement includes adjusting the brightness, contrast, color balance, sharpening, etc. of the image to highlight the target and improve the visibility of the target. Finally, the enhanced target is observed and judged by naked eyes.

Thermal infrared imaging technology is one of the important technologies in the thermal imaging field, and is mainly used for image visualization effect. Thermal infrared images, due to their unique physical characteristics and application scenarios, can create different thermal features in the background, such as temperature differences, distribution patterns of thermal radiation, etc., by thermal imaging techniques. These thermal features may be captured by the thermal imaging device and converted into an image such that the camouflage target appears in the thermal image, resulting in an enhanced camouflage target. In addition, by analyzing features in the thermal image, a comparison can be made with known camouflage patterns to identify potential camouflage targets. Thermal imaging technology is advantageous during night or low light conditions because it can detect thermal radiation from a target without relying on visible light.

The shortcomings of the visible light system are apparent in that the visible light system can only perceive information in the visible spectrum and may not be effectively identified and enhanced for certain camouflage techniques. For example, objects using infrared camouflage material may merge with the background under visible light, which is difficult to distinguish. Visible light systems are sensitive to changes in lighting conditions and environmental background. In low light or complex backgrounds, the enhancement effect of the target may be limited. In addition, the resolution and detail of the visible light image are limited by the device itself, which may result in some subtle target features being lost during the image enhancement process, affecting the accuracy of recognition and analysis. Therefore, when dealing with a camouflage target highly fused with a background, it is difficult for a system relying only on visible light to independently accomplish the camouflage target enhancement task.

Thermal imaging techniques also have some limitations. For example, if the camouflage target has similar thermal characteristics to the background or camouflage measures for thermal imaging are taken, the effectiveness of the thermal imaging technique may be reduced. Furthermore, thermal imaging techniques are also subject to environmental conditions (e.g., atmospheric diffusion, shielding, etc.). It should be noted that in a real battlefield environment, the heat source is complex, and the environmental temperature is close to the human body temperature, which causes great interference to the thermal imaging technology for detecting the human body camouflage unit.

Disclosure of Invention

The invention aims to solve the following technical problems: the target enhancement effect evaluation index of the visible light system is highly dependent on naked eye judgment, has stronger subjective randomness, and is difficult to deal with flexibly changeable camouflage targets; target enhancements based on thermal imaging techniques are susceptible to complex heat source interference on the battlefield. It is difficult to achieve stable, comprehensive, reliable camouflage target enhancement capability in complex battlefield environments, both in visible light systems and thermal imaging systems.

The invention provides a method for enhancing a camouflage target, which aims to improve the visibility of the camouflage target and highlight camouflage objects which are difficult to distinguish by naked eyes from a complex background environment.

In order to achieve the above purpose, the invention adopts the following technical scheme:

a method of camouflage target enhancement, comprising the steps of:

Step 1: taking a hyperspectral image acquired by a hyperspectral camera as input;

step 2: performing PCA dimension reduction on the collected hyperspectral image data, and selecting the first 16 main components as a group of new feature vectors;

Step 3: carrying out spectral feature selection on 16 feature vectors obtained by PCA dimension reduction by using an SPA continuous projection algorithm, and screening out four feature wave bands as feature variable combinations containing minimum redundant information and minimum collinearity by using the SPA continuous projection algorithm;

step 4: visualizing and analyzing the selected characteristic wave bands to verify the correlation between the selected characteristic wave bands and a camouflage target, drawing a scatter diagram between the selected characteristic wave bands and the camouflage target, observing the relation between the selected characteristic wave bands and the camouflage target, if the obvious correlation between the selected characteristic wave bands and the camouflage target is found, the characteristic wave bands can be considered to be correlated with the camouflage target, otherwise, adjusting parameters of an SPA algorithm, and returning to the step 3;

Step 5: and fusing the optimal band combination which is screened by the SPA and is positively correlated with the camouflage target, and taking the optimal band combination as an R channel for finally outputting a pseudo-color image, wherein the rest band combination is averaged to a G channel and a B channel, so that the pseudo-color image obtained by mapping marks the camouflage target as red, and the rest background part is mainly blue and green, thereby achieving the effect of enhancing the camouflage target.

As a preferred technical solution, the method for enhancing a camouflage target further includes step 6: selecting a region simultaneously containing the enhanced camouflage target and the background as an ROI, converting the region into a Lab color space, and calculating the green-to-red color difference gradient and the color contrast of the ROI as indexes for evaluating the camouflage enhancement effect.

As a preferred technical solution, in step 2, the specific steps of PCA dimension reduction include:

Let the original data matrix X be represented as:

X＝[x₁,x₂,...,x_m]^T (1)；

the linear transformation of X by the transformation matrix a= [ a ₁,a₂,...,a_m]^T gives:

The size of matrix Z is m×n, where i=1, 2,..n, n > m;

The vector mean is zeroed, i.e., E (X) =0, and the variance and covariance of each vector are as in equation (3) (4):

Var(z_i)＝a_i ^T∑a_i(i＝1，2，...，m) (3)；

Cov(z_i,z_j)＝a_i ^T∑a_j(i，j＝1，2，...，m) (4)；

Constraint a ₁ ^Ta₁ =1 for the first principal component z ₁, and the maximum value of Var (z ₂) is satisfied when solving for the second principal component

In principal component analysis, the conditional extremum of z ₁＝z₁ ^T X is the first principal component, the requirement a ₁＝[a₁₁,a₂₁,...,a_m1]^T, where Var (z ₁) is the largest under the constraint of a ₁ ^Ta₁ =1, calculated using the lagrangian multiplier method, as in equation (5):

wherein:

Is a Lagrangian function, and lambda is a parameter; calculating the characteristic equation to obtain a characteristic value and a characteristic vector, and obtaining an ith main component according to z _i＝a_i ^T X, i=1, 2.

As a preferred technical solution, in step 3, the spectral matrix X is set to be an N row p column matrix, where: n is the number of samples; p is the number of input spectral bands, and X _k∈R^N is the kth column vector of X;

The SPA continuous projection algorithm comprises the following specific steps:

1) Setting the number of wavelengths M to be extracted, before the first iteration (m=1), the h column X _j of optional X _N×P is denoted as X _k (0), k (0) =j, j e 1, …, p;

2) The position set of the column vector which is not selected is recorded as

3) Respectively calculating the projection of the current column vector x _k (m-1) of the residual column vector x _j (j epsilon S);

P＝x_j-[x_j ^Tx_k(m-1)]x_k(m-1)[x_k(m-1) ^Tx_k(m-1)]^-1，j∈S；

4) Selecting the serial number of the wavelength point corresponding to the projection maximum value to enable

5)If M is less than or equal to M, returning to the step 2); otherwise, ending the algorithm;

The final extracted wavelength variable combination of SPA is { k (M), m=0, …, M-1}, multiple linear regression analysis is carried out on each pair of variable subsets determined by x _j and M, a multiple correction model is established, the prediction standard deviation of a verification set is obtained, and x _j and M corresponding to the minimum RMSEV value are the optimal values.

As a preferred technical solution, in step 6, the image processing method uses the cv2. Cvtdcolor function in the OpenCV library to convert the ROI image into a Lab color space, where the Lab color space contains brightness and two color channels: green-red and blue-yellow, in the Lab color space, the green-to-red color gradient is obtained by calculating the difference of the a-channel of the pixel in the ROI, the color contrast is the average value of the color in the ROI, and compared with the average value of the color in the background, the color gradient and the color contrast are used as quantization indexes of the visibility of the camouflage object in the background environment, reflecting the color difference and the contrast between the camouflage object and the background.

After the technical scheme is adopted, the invention has the following advantages:

(1) Collecting hyperspectral images containing rich spectral information through a hyperspectral camera, and selecting a band combination related to the camouflage target height;

(2) And (3) carrying out image enhancement on the camouflage targets in the scene, highlighting the camouflage targets and improving the visibility of the camouflage targets. Subjective randomness of naked eye judgment is avoided;

(3) The wave band combination related to the camouflage target height is used as an R channel through fusion, the rest of the background is used as a B channel and a G channel after being averaged, a pseudo-color image is generated in the channel mapping mode, the camouflage is marked as red, and the camouflage target is conveniently and rapidly detected;

(4) Taking the image with the enhanced camouflage target as a front work of the subsequent camouflage target detection, and providing more excellent original data for the development of a related algorithm for the subsequent camouflage target identification;

(5) The method has the advantages that a stable camouflage target enhancement effect is obtained by using a small data quantity and a training quantity, the operation speed is high, the instantaneity is good, and the target enhancement performance is excellent.

Drawings

FIG. 1 is a flow chart of steps of a method of camouflage target enhancement;

Fig. 2 is an exemplary diagram of camouflage target enhancement effects.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The invention provides a method for enhancing a camouflage target, which aims at improving the visibility of the camouflage target and highlighting camouflage objects which are difficult to distinguish by naked eyes from a complex background environment; on the other hand, by the method for enhancing the camouflage target, the color difference gradient and the color contrast of the camouflage object and the surrounding background are improved, and convenience is provided for the subsequent camouflage target identification task.

As shown in fig. 1, a method for enhancing camouflage targets of the present invention includes the steps of:

step 1: the hyperspectral image acquired by the hyperspectral camera is used as input.

In this embodiment, the operating system used in the software deployment environment is Linux Ubuntu 5.10.120-tegra, and the architecture is a 64-bit ARM architecture aarch.

Linux and Ubuntu are open-source operating systems, and have a huge open-source software ecosystem. After the hyperspectral image is acquired, various open source tools and libraries are required to process and analyze the image data, such as OpenCV, numPy, sciPy and the like. Under the Linux Ubuntu environment, the open source tools can be acquired and used, and convenience is provided for realizing camouflage target enhancement tasks.

The 64-bit ARM architecture aarch provides a larger memory addressing space and higher computing power, and is suitable for processing hyperspectral image data with larger data volume. Furthermore, the aarch64,64 architecture is widely used in embedded systems, which provides support for the implementation of camouflage target enhanced functionality on more mobile devices.

Systems that rely on visible light respond only to spectral features within a limited spectral band and cannot collect camouflage targets that are highly fused to the background. The hyperspectral image collected by the hyperspectral camera contains rich spectral information, and the extra spectral information enables the hyperspectral image to better distinguish different objects and materials. In the technology of camouflage recognition enhancement, vegetation and camouflage have different reflection characteristics in spectrum due to differences in materials, which is one of the key reasons why they can be resolved by hyperspectral images. Vegetation generally has higher reflectivity in the visible light and near infrared spectrum ranges, camouflage generally has different spectral responses, and by analyzing the differences of the two spectra in different wave bands, a specific wave band combination can be selected, so that better distinguishing effect can be obtained. In addition, when the camouflage target is far from the detection device, the reliability of the spatial dimension information is generally difficult to ensure, and the information of the spectral dimension is not lost due to the increase of the target distance.

Step 2: PCA dimension reduction is carried out on the collected hyperspectral image data, and the first 16 principal components are selected as a group of new feature vectors.

The band information of different substances is seriously overlapped, so that the full-band spectrum contains a large amount of redundant information and noise, thereby influencing the enhancement accuracy of the camouflage target. PCA principal component analysis is a commonly used dimension reduction technique that can convert raw data into a new set of variables, called principal components, by linear transformation. Each principal component is a linear combination of the individual bands in the original data and is ordered from large to small in variance. By selecting the first few principal components, most of variance information in the original data can be reserved, so that the data dimension reduction is realized. Assume that the original data matrix X can be expressed as:

X＝[x₁,x₂,...,x_m]^T (1)；

the linear transformation of X by the transformation matrix a= [ a ₁,a₂,...,a_m]^T gives:

The size of matrix Z is m×n, where i=1, 2,..n, n > m. The vector mean is zeroed, i.e., E (X) =0, and the variance and covariance of each vector are as in equation (3) (4):

Var(z_i)＝a_i ^T∑a_i(i＝1，2，...，m) (3)；

Cov(z_i,z_j)＝a_i ^T∑a_j(i，j＝1，2，...，m) (4)；

In order to retain the m vector information in the original data, when the first principal component z ₁ is constrained to have a ₁ ^Ta₁ =1, and then the second principal component is solved, it is required that z ₂ does not include the first principal component and includes the information of the remaining vectors as much as possible, that is, when Var (z ₂) is satisfied and reaches the maximum value, cov (z ₂,z₁)＝a₂ ^T∑a₁ =0.

In principal component analysis, the conditional extremum of z ₁＝a₁ ^T X is the first principal component, the requirement a ₁＝[a₁₁,a₂₁,...,a_m1]^T, where Var (z ₁) is the largest under the constraint of a ₁ ^Ta₁ =1, calculated using the lagrangian multiplier method, as in equation (5):

wherein:

as a lagrangian function, λ is a parameter. Calculating the characteristic equation to obtain a characteristic value and a characteristic vector, and obtaining an i-th main component according to z _i＝a_i ^T X, i=1, 2.

The hyperspectral image acquired in the step 1 contains a large amount of spectrum information, but certain information redundancy exists, so that challenges are caused to a camouflage target identification task which needs to process a large amount of spectrum information in a short time, linear combination of original spectrum bands is obtained by dimension reduction through PCA main component analysis, important main components are selected according to importance ranking, and the effects of removing redundant information and retaining main information can be achieved.

Step 3: and carrying out spectral feature selection on the 16 feature vectors obtained by reducing the dimension of the PCA by using an SPA continuous projection algorithm. Four characteristic wave bands are screened out through the algorithm to be used as characteristic variable combinations with minimum redundant information and minimum collinearity.

The SPA continuous projection algorithm provides a band selection method with minimal redundant information. Let the spectrum matrix X be an N row p column matrix. Wherein: n is the number of samples; p is the number of spectral bands input and X _k∈R^N is the kth column vector of X. The method comprises the following specific steps:

1) The number of wavelengths to be extracted M is set, and before the first iteration (m=1), the j-th column X _j of optional X _N×P is denoted as X _k (0), k (0) =j, j e 1.

2) The position set of the column vector which is not selected is recorded as

3) Respectively calculating projections of the remaining column vectors x _j (j e S) and the current column vector x _k(m-1)

P＝x_j-[x_j ^Tx_k(m-1)]x_k(m-1)[x_k(m-1) ^Tx_k(m-1)]^-1，j∈S

4) Selecting the serial number of the wavelength point corresponding to the projection maximum value to enable

5)If M is less than or equal to M, returning to the step 2); otherwise, the algorithm is ended.

The final extracted wavelength variable combination of SPA is { k (M), m=0, …, M-1}. Multiple linear regression analysis (MLR) was performed on each pair of x _j and M-determined variable subsets, and a multiple correction model was built to obtain the predicted standard deviation of the validation set (RMSEV), with x _j and M corresponding to the smallest RMSEV values being the optimal values.

The SPA screening band can extract the spectral characteristics of the target, so that the difference between the target and the background is more obvious. By selecting the band that most discriminates the object, the visibility of the object can be enhanced, thereby improving the recognition performance of the camouflage object.

And in the hyperspectral image acquisition and processing process, PCA (principal component analysis) degradation and SPA (space imaging) spectral feature selection are sequentially carried out on data acquired by a hyperspectral camera. The principal component of the first 16 is selected by utilizing a PCA principal component analysis method, and then the optimal characteristic variable combination containing 4 characteristic wave bands is obtained through SPA+MLR. The method ensures that hyperspectral data with large data volume and large calculation amount and high training cost can greatly reduce information redundancy while keeping main characteristics, simplifies the calculation process, ensures better performance and creates conditions for enhancing real-time performance of camouflage identification.

Step 4: the selected characteristic bands are visualized and analyzed to verify their correlation with the camouflage target. And drawing a scatter diagram between the selected characteristic wave band and the camouflage target, and observing the relation between the selected characteristic wave band and the camouflage target. If a significant correlation is found between the selected characteristic bands and the camouflage target, then these characteristic bands may be considered to be correlated with the camouflage target. Otherwise, parameters (neighborhood size and similarity measure) of the SPA algorithm are adjusted, and the step 3 is returned.

Scatter plots are a common visualization method for observing the relationship between two variables. When a scatter plot is drawn, the value of one variable is taken as the abscissa and the value of the other variable is taken as the ordinate, each data point representing a sample. By observing the distribution of the data points in the coordinate system, it can be preliminarily determined whether there is a correlation or trend between the two variables.

The scatter plot drawn may help assess the discrimination capability of different bands, i.e., whether they are able to effectively discriminate between camouflage targets and backgrounds. If the reflectivity of a certain wave band has obvious difference between the camouflage target and the background, the separation degree between the camouflage target and the background can be seen in the scatter diagram, and the data points are distributed more discretely.

Step 5: and fusing the optimal band combination which is screened by the SPA and is highly positively correlated with the camouflage target, and taking the optimal band combination as an R channel for finally outputting a pseudo-color image, wherein the rest band combination is averaged to a G channel and a B channel, so that the obtained pseudo-color image marks the camouflage target as red, and the rest background part is mainly blue and green, thereby achieving the effect of enhancing the camouflage target.

After the selected characteristic wave bands are indexed to the corresponding gray level images, pixel values of the images are assigned to red channels of the RGB images, pixel values of unselected characteristic wave band images are evenly distributed to blue and green channels of the RGB images, the pixel values of each image are divided by the number of unselected characteristic wave bands through traversing the unselected characteristic wave band images, and then the pixel values are added to the blue and green channels of the RGB images, so that a pseudo-color image is obtained. The pseudo-color map marks the camouflage target as red, and the rest background part is mainly blue and green, so that the camouflage target enhancement effect is achieved, and the enhancement effect is shown in fig. 2.

Taking these bands, which are positively correlated to the camouflage target height, as the R-channel of the pseudo-color map, allows the camouflage target to be displayed in red in the image, since red is generally considered to be one of the colors most noticeable to the human eye. And the rest wave bands are combined and averaged to the two channels G and B, the background part can be mainly blue and green, and color distribution which is obviously compared with the camouflage target is formed, so that the camouflage target is highlighted, the camouflage target is more obvious in a pseudo-color image, the visibility of the target is improved, and the effect of enhancing the camouflage target is achieved.

Step 6: selecting a region simultaneously containing the enhanced camouflage target and the background as an ROI, converting the region into a Lab color space, and calculating the green-to-red color difference gradient and the color contrast of the ROI as indexes for evaluating the camouflage enhancement effect.

The image processing method uses the cv2.cvtColor function in the OpenCV library to convert the ROI image into Lab color space. The Lab color space contains luminance (L) and two color channels: green-red (a) and blue-yellow (b). In Lab color space, the green-to-red color gradient can be obtained by calculating the difference in the a-channel of pixels in the ROI, and the color contrast is calculated as the average of the colors in the ROI and compared with the average of the colors in the background. The color difference gradient and the color contrast can be used as quantization indexes of the visibility of the camouflage target in the background environment, and reflect the color difference and the contrast between the camouflage target and the background.

By calculating the color gradient and the color contrast, the degree of camouflage enhancement effect can be quantified. A larger color gradient and color contrast generally means that the camouflaged object is more noticeable in the pseudo-color image, thereby improving the visibility and recognition of the object.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Furthermore, it should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a separate embodiment, and that this description is provided for clarity only, and that the disclosure is not limited to the embodiments described in detail below, and that the embodiments described in the examples may be combined as appropriate to form other embodiments that will be apparent to those skilled in the art.

Claims (5)

1. A method of camouflage target enhancement, comprising the steps of:

Step 1: taking a hyperspectral image acquired by a hyperspectral camera as input;

step 2: performing PCA dimension reduction on the collected hyperspectral image data, and selecting the first 16 main components as a group of new feature vectors;

Step 3: carrying out spectral feature selection on 16 feature vectors obtained by PCA dimension reduction by using an SPA continuous projection algorithm, and screening out four feature wave bands as feature variable combinations containing minimum redundant information and minimum collinearity by using the SPA continuous projection algorithm;

step 4: visualizing and analyzing the selected characteristic wave bands to verify the correlation between the selected characteristic wave bands and a camouflage target, drawing a scatter diagram between the selected characteristic wave bands and the camouflage target, observing the relation between the selected characteristic wave bands and the camouflage target, if the obvious correlation between the selected characteristic wave bands and the camouflage target is found, the characteristic wave bands can be considered to be correlated with the camouflage target, otherwise, adjusting parameters of an SPA algorithm, and returning to the step 3;

Step 5: and fusing the optimal band combination which is screened by the SPA and is positively correlated with the camouflage target, and taking the optimal band combination as an R channel for finally outputting a pseudo-color image, wherein the rest band combination is averaged to a G channel and a B channel, so that the pseudo-color image obtained by mapping marks the camouflage target as red, and the rest background part is mainly blue and green, thereby achieving the effect of enhancing the camouflage target.

2. A method of camouflage target enhancement as claimed in claim 1 and further comprising the step of 6: selecting a region simultaneously containing the enhanced camouflage target and the background as an ROI, converting the region into a Lab color space, and calculating the green-to-red color difference gradient and the color contrast of the ROI as indexes for evaluating the camouflage enhancement effect.

3. A method of camouflage target enhancement as claimed in claim 1, wherein in step 2, the specific step of PCA dimension reduction comprises:

Let the original data matrix X be represented as:

X＝[x₁,x₂,...,x_m]^T (1)；

the linear transformation of X by the transformation matrix a= [ a ₁,a₂,...,a_m]^T gives:

The size of matrix Z is m×n, where i=1, 2,..n, n > m;

The vector mean is zeroed, i.e., E (X) =0, and the variance and covariance of each vector are as in equation (3) (4):

When the first principal component z ₁ is constrained to have a ₁ ^Ta₁ =1 and the second principal component is solved, when Var (z ₂) is satisfied to take the maximum value,

In principal component analysis, the conditional extremum of z ₁＝a₁ ^T X is the first principal component, the requirement a ₁＝[a₁₁,a₂₁,...,a_m1]^T, where Var (z ₁) is the largest under the constraint of a ₁ ^Ta₁ =1, calculated using the lagrangian multiplier method, as in equation (5):

wherein:

Is a Lagrangian function, and lambda is a parameter; calculating the characteristic equation to obtain a characteristic value and a characteristic vector, and obtaining an ith main component according to z _i＝a_i ^T X, i=1, 2.

4. The method of claim 1, wherein in step 3, the spectrum matrix X is set to be an N-row p-column matrix, wherein: n is the number of samples; p is the number of input spectral bands, and X _k∈R^N is the kth column vector of X;

The SPA continuous projection algorithm comprises the following specific steps:

1) Setting the number of wavelengths M to be extracted, and before the first iteration (m=1), the j-th column X _j of the optional X _N×P is denoted as X _k(0), k (0) =j, j e1, …, p;

2) The position set of the column vector which is not selected is recorded as

3) Respectively calculating the projection of the current column vector x _k(m-1) of the residual column vector x _j (j epsilon S);

P＝x_j-[x_j ^Tx_k(m-1)]x_k(m-1)[x_k(m-1) ^Tx_k(m-1)]^-1，j∈S；

4) Selecting the serial number of the wavelength point corresponding to the projection maximum value to enable

5)If M is less than or equal to M, returning to the step 2); otherwise, ending the algorithm;

The final extracted wavelength variable combination of SPA is { k (M), m=0, …, M-1}, multiple linear regression analysis is carried out on each pair of variable subsets determined by x _j and M, a multiple correction model is established, the prediction standard deviation of a verification set is obtained, and x _j and M corresponding to the minimum RMSEV value are the optimal values.

5. A method of camouflage target enhancement as claimed in claim 2, wherein in step 6, the image processing method uses the cv2.cvttcolor function in the OpenCV library to convert the ROI image into Lab color space containing luminance and two color channels: green-red and blue-yellow, in the Lab color space, the green-to-red color gradient is obtained by calculating the difference of the a-channel of the pixel in the ROI, the color contrast is the average value of the color in the ROI, and compared with the average value of the color in the background, the color gradient and the color contrast are used as quantization indexes of the visibility of the camouflage object in the background environment, reflecting the color difference and the contrast between the camouflage object and the background.