Target recognition method for airborne navigation image based on harmony search of spatial entropy

Abstract

Disclosed in the present invention is a target recognition method for an airborne navigation image based on harmony search of spatial entropy, comprising: using position distribution information of an image grayscale space to calculate a distribution entropy value of the image grayscale space, mapping the distribution entropy value into a uniform distribution function, and by means of linear transformation processing including 2D-DCT, transformation domain coefficient weighting, and inverse 2D-DCT, simultaneously enhancing global and local contrast of an original image. By incorporating the harmony search algorithm of the maximum entropy, although the algorithm is similar to the genetic algorithm, a multi-point crossover mode is used in the process of searching for a new solution in the field, so that the new solution is more diversified, and the probability of escaping from a local optimal solution is increased.

Description

A Harmony Search-Based Airborne Navigation Image Target Recognition Method Technical Field

[0001] This invention relates to the field of airborne navigation image processing, and in particular to an airborne navigation image target recognition method based on spatial entropy and harmony search. Background Technology

[0002] In the field of airborne navigation image processing, many target detection methods focus on high-resolution image processing, typically employing deep learning algorithms such as R-CNN (Region with CNN feature), SSD (Single Shot MultiBox Detector), and ANN, as well as multi-scale feature fusion algorithms. However, these methods are often complex in principle and difficult to implement. Therefore, this paper introduces an automatic classification technology for airborne navigation images after processing. By setting up image transformation, feature extraction, and result comparison, it achieves automatic transformation, recognition, and classification of airborne navigation image sets. The entire algorithm system is fully automated under the control of deep learning network technology, and the classification performance of the deep learning network is improved through human intervention and algorithm optimization upgrades.

[0003] The system comprises an image acquisition and preprocessing module, an image recognition module, and an image comparison and classification module. Its working principle is as follows: The image acquisition and preprocessing module is primarily responsible for the initial processing of images. First, the airborne navigation image set is input into the system and converted to grayscale to facilitate subsequent algorithmic recognition and analysis. Then, simple classifications such as viewpoint change and scaling analysis are performed. Next, the image is input into the image recognition module, which is responsible for identifying and extracting features of similar objects in the grayscale image. Probabilistic statistical methods, approximation theory, and other analytical operations are used in conjunction to obtain the feature recognition of the target in the image. Finally, image comparison and classification are performed. Based on a deep neural network learning model, algorithms such as manually assigned variable weights and updates, and simulation prediction analysis are used to automatically classify and identify targets in the image based on their features.

[0004] The aforementioned automatic classification technology for airborne navigation images, based on deep learning network technology, achieves full automation, eliminating reliance on operator experience and the need for repeated parameter adjustments. The processing is relatively simple and efficient. However, considering the use of deep neural networks, the algorithm design remains complex. Furthermore, it fails to address the limited information content and low contrast of small targets in airborne navigation images, as well as the presence of multiple backgrounds and susceptibility to weather conditions. This results in poor adaptability and robustness under such complex environmental conditions.

[0005] In summary, a target recognition method for airborne navigation images based on spatial entropy and harmony search is needed to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an airborne navigation image target recognition method based on spatial entropy and harmony search, aiming to solve the aforementioned problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an airborne navigation image target recognition method based on spatial entropy and harmony search, comprising the following steps:

[0008] Step S1: Convert the original image to grayscale;

[0009] Step S2: Calculate the gray-scale spatial entropy and distribution function of the image based on the two-dimensional spatial histogram, and enhance the contrast of the image;

[0010] Step S3: Transformation and local contrast enhancement: The transform domain coefficients are weighted to enhance the contrast of the local image.

[0011] Step S4: Find the optimal threshold segmentation point. Based on the characteristics of the recorded navigation image and the specific target features, find the optimal threshold segmentation point through an optimal search algorithm.

[0012] Step S5: Output the target recognition and segmentation results. Based on the determined optimal grayscale vector, the optimal threshold segmentation point is obtained, and threshold segmentation is performed accordingly to output the final target recognition and segmentation result image.

[0013] Optionally, the original image is grayscaled in step S1 in the following way:

[0014] Determine the range of pixel values ​​corresponding to each gray level, convert the original image into an image with K gray levels, and calculate the two-dimensional spatial histogram on the spatial grid of the image so that the original image corresponds to the gray level image.

[0015] Optionally, the two-dimensional spatial histogram on the spatial grid of the computed image is:

[0016] Assuming the original image size is H*W, and x(i,j) is the value of each pixel in image X, then x(i,j)... [0, z+], and will As a set of K gray levels existing in image X, where Calculate the gray levels on the spatial grid of X. The two-dimensional spatial histogram is as follows:

[0017] ,

[0018] Where m,n z+, h k (m,n) [0, z+] is the location of the image region in the spatial grid. The number of times grayscale value Xx appears.

[0019] The total number of grids in a 2D histogram is M*N. The number of different gray levels K and their ratios are used to... , ,

[0020] in This indicates taking the nearest integer and generating grayscale values. Two-dimensional spatial histogram At that time, the aspect ratio of the original image is preserved on the spatial grid.

[0021] Optionally, in step S2, the entropy measure of the gray level is calculated as follows:

[0022] ,

[0023] Calculate discrete functions Measuring grayscale Compared to other gray levels And normalize it:

[0024] ,

[0025] ,

[0026] In the formula, K represents the number of gray levels.

[0027] Optionally, in step S2, the distribution function F is calculated. k for:

[0028] ,

[0029] ,

[0030] In the formula , , , .

[0031] Optionally, the transformation and local contrast enhancement in step S3 are performed in the following ways:

[0032] Let the original image and the transformed image of the same size be X and D, respectively. ,

[0033] ,

[0034] ,

[0035] , ,

[0036] Where: the values of k and l respectively represent frequency components,

[0037] Modify the transform domain coefficient d(k, l) to obtain a higher frequency transform coefficient ,

[0038] ,

[0039] ,

[0040] In the formula, is the weighted frequency allocation coefficient after being increased,

[0041] Perform an inverse transform:

[0042] ,

[0043] ,

[0044] Obtain an output image with enhanced global or local contrast.

[0045] Optionally, in step S4, find the optimal threshold segmentation point by the following method:

[0046] Step S41: According to the characteristics of the airborne navigation image and the characteristics of specific targets, preset an initial set of feature thresholds as the gray vector X = (x1, x2,..., xn) of the initial threshold segmentation point, where 1 < n < 128, and the threshold segmentation point is normally distributed;

[0047] Step S42: Under each initial threshold segmentation point, calculate the sum of the entropy values H(n) of the two categories of target and background, generate an initial harmony memory library HM, select one or more harmonies from the harmony library, and perform pitch fine-tuning to generate new harmonies;

[0048] Step S43: Calculate the entropy values of categories A and B at each gray level corresponding to the new harmony vector, take the maximum entropy value as the solution under this vector, and if this value is greater than the value in HM, update the harmony memory library HM;

[0049] Step S44: When the number of times of updating the gray vector reaches the preset number of iterations, or when the sum of the entropy values of categories A and B obtained by solving under the new gray vector reaches the maximum, determine the optimal threshold segmentation point.

[0050] Optionally, in step S43, the entropy values ​​of classes A and B at each gray level corresponding to the new harmony vector are calculated as follows:

[0051] The image grayscale level is divided into two classes, A and B, starting from the grayscale value Th. Class A and B are composed of all pixels in the image with grayscale values ​​in the range [0, Th] and [Th+1, n], respectively.

[0052] , ,

[0053] In the formula: The sum of the gray level probabilities from 1 to Th. P is the sum of the gray level probabilities from Th+1 to n. i Let be the probability distribution of each gray level in the image.

[0054] Optionally, the sum of the entropy values ​​of classes A and B in step S44 is:

[0055] ,

[0056] The harmony memory bank is:

[0057] .

[0058] Optionally, the threshold image segmentation in step S5 is as follows:

[0059] The image is binarized by selecting a threshold to convert it into a black and white image. The gray values ​​in the image are then compared with the threshold. If the gray value is greater than the threshold, the gray value of the pixel is set to 255; otherwise, it is set to 0.

[0060] The beneficial effects of this invention are:

[0061] In this invention, the airborne navigation image is captured from a high altitude, resulting in relatively low contrast, making it difficult to distinguish the target from the background and thus challenging target recognition. By utilizing the positional distribution information of the image's grayscale space, the image's grayscale space distribution entropy value is calculated and mapped to a uniform distribution function. Then, through linear transformation processing using 2D-DCT, transformation system weighting, and inverse 2D-DCT, the global and local contrast of the original image are simultaneously enhanced.

[0062] In this invention, the maximum entropy harmony search algorithm is used. Although it is similar to the genetic algorithm, a multi-point mating mode is adopted in the process of searching for new solutions in the neighborhood, which makes the new solutions more diverse and increases the possibility of the algorithm escaping local optima.

[0063] In this invention, the combination and application of the concept of entropy value make full use of the grayscale information inherent in the image to simultaneously enhance the global and local contrast of traditional images. Furthermore, the threshold at which the entropy values ​​of the target and the background reach their maximum values ​​is used as the optimal threshold segmentation point, thereby achieving accurate identification and segmentation of specific targets in the image.

[0064] In this invention, the computational load is significantly reduced by combining the maximum entropy harmony search algorithm. The appropriate selection of the grayscale set based on the initial threshold during the search can effectively accelerate the convergence speed of the algorithm and quickly obtain the optimal threshold segmentation point. Attached Figure Description

[0065] Figure 1 is a flowchart of one method of the present invention.

[0066] Figure 2 is a flowchart of the calculation of the distribution function according to the present invention.

[0067] Figure 3 is a flowchart of an optimal threshold point according to the present invention. Detailed Implementation

[0068] To more clearly illustrate the technical solutions in the embodiments of the invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] As shown in Figures 1, 2, and 3, an airborne navigation image target recognition method based on spatial entropy and harmony search includes the following:

[0070] By utilizing the positional distribution information of the image's grayscale space, the image's grayscale space distribution entropy value is calculated to obtain a uniform distribution function. This function is then mapped to a uniform distribution function to enhance the contrast of the global image. Further enhancement of local image contrast is achieved through processing methods such as 2D-DCT, transform domain coefficient weighting, and inverse 2D-DCT. After simultaneous enhancement of both global and local images, an improved harmony search algorithm is used to find the optimal background-image segmentation threshold point, ultimately achieving target detection and recognition. The algorithm implementation steps are as follows:

[0071] The original image is converted to grayscale. Assume the original image size is H*W, and x(i,j) is the pixel value of image X. [0, z+], and will As a set of K gray levels existing in image X, where Calculate the gray levels on the spatial grid of X. The two-dimensional spatial histogram is as follows:

[0072] ,

[0073] Where m,n z+, hk(m,n) [0, z+] is the location of the image region in the spatial grid. The number of times grayscale value xx appears. The total number of grid cells in a 2D histogram is M*N. The number of different grayscale levels K and their ratios are used to determine the number of occurrences. , ,

[0074] in This indicates taking the nearest integer and generating grayscale values. Two-dimensional spatial histogram At the same time, the aspect ratio of the original image is preserved on the spatial grid, thereby preserving the spatial features of the pixels;

[0075] Calculate the gray-level spatial entropy and distribution function of an image: Based on the two-dimensional histogram, calculate the entropy measure of the gray levels.

[0076] ,

[0077] Calculate discrete functions Measuring grayscale Compared to other gray levels The relative importance of K is determined and normalized.

[0078] , ,

[0079] Calculate the cumulative distribution function and map it to the image Y:

[0080] , ,

[0081] in, , , .

[0082] After the global contrast enhancement described above, in order to enhance the local contrast simultaneously, a two-dimensional discrete cosine transform (2D-DCT) is used on the basis of the above processing. After transform domain coefficient weighting and inverse 2D-DCT transformation, an output image with both global and local contrast enhancement is obtained.

[0083] The steps of 2D-DCT transformation, including weighted summation and inverse transformation of transform domain coefficients, are as follows: Assume the original image and the transformed image of the same size generated by the 2D discrete cosine transform are X and D, respectively, where:

[0084] ,

[0085] ,

[0086] ,

[0087] , ,

[0088] Among them, the values of k and l respectively represent frequency components. To increase local contrast, higher frequency transformation coefficients need to be adjusted, that is, the transformation domain coefficient d(k, l) is corrected to obtain higher frequency transformation coefficients. , where is the weighted frequency distribution coefficient after being increased.

[0089] ,

[0090] ,

[0091] Among them, the higher the value of ɑ, the higher the local enhancement degree. The selection of parameters is calculated according to the entropy value estimation formula mentioned above; γ ∈ [0, 1], which determines the degree of local contrast enhancement. When the value is 0, only global enhancement is performed. In this algorithm, its value can be selected as 1. After weighting the transformation domain coefficients, the following formula is used for the inverse 2D-DCT transformation to obtain the output image with both global and local enhancement: .

[0092] ,

[0093] After being processed by the above-mentioned 2D-DCT transformation, weighted by the transformation domain function, and inverse 2D-DCT transformation, an image with both global and local contrast enhancement is obtained.

[0094] After obtaining the output contrast-enhanced image, an improved harmony search optimal algorithm will be used to find the optimal threshold segmentation point, so as to accurately segment and identify specific targets in the image:

[0095] First, according to the characteristics of the airborne navigation image and the features of specific targets, an initial feature threshold set is preset as the gray vector X = (x1, x2,..., xn) of the initial threshold segmentation point, where 1 < n < 128, and the threshold segmentation point follows a normal distribution; under each initial threshold segmentation point, there is a corresponding sum H(n) of the entropy values of the two categories of target and background calculated at this threshold point, generating an initial harmony memory bank HM. Among them, the principle and formula for calculating the entropy value are as follows:

[0096] For an image with gray levels n, p1, p2, ..., pn are the probability distributions of each gray level in the image. The gray levels of the image are divided into two classes, A and B, starting from the gray level Th. A and B are composed of all pixels in the image with gray levels in [0, Th] and [Th+1, n], respectively.

[0097] make This represents the sum of the gray level probabilities from 1 to Th. Let represent the sum of gray level probabilities from Th+1 to n. Then, the probability distribution of each gray level pixel in class A within the entire pixel set of class A is: , ,..., The probability distribution of each gray level pixel in class B within the entire class B pixel set is as follows: , ,..., The information entropies H(A) and H(B) for classes A and B are calculated as follows:

[0098] , ,

[0099] The sum of the information entropies of classes A and B is:

[0100] .

[0101] Based on the initial threshold segmentation point set and the corresponding entropy value, HMS harmonic vectors are randomly generated and placed into the harmony memory HM, and the entropy value function f(X) under the corresponding vector is recorded. The generated initial harmony memory HM is in the following form:

[0102] Based on the dichotomy method and empirical considerations, one or more harmonics are selected from the harmony library, and their pitch is fine-tuned to generate new harmonics. Assume each variable in the new harmonic variable X'... For each k=1,2,...,m, the probability of HMCR (Memory Consideration Rate, the probability of selecting a harmony from the harmony library) is taken from the initial harmony library HM. If any value in HM is taken, then there is a probability that 1-HCMR will randomly take any value within the entire initial threshold set. When the new variable is taken from HM, in order to increase the diversity of solutions and expand the search range, the variable needs to be fine-tuned, and the probability of fine-tuning is PAR. Then, the new variable... :

[0103] ,

[0104] In the formula, rand is a random number uniformly distributed between [0,1], and BW is the bandwidth for pitch fine-tuning (generally taken as 1 in this grayscale image threshold). The updated harmony vector can be obtained from the above update criterion.

[0105] Calculate the entropy values ​​of classes A and B under each gray level corresponding to the new harmony vector, and take the largest entropy value as the solution of f(X') under this vector. If this value is greater than f(X') in HM, then... i If the value of ) is changed, it is replaced, indicating that the threshold segmentation effect under the new vector is better, and the harmony memory HM is updated accordingly;

[0106] If the number of times the grayscale vector is updated reaches the preset number of iterations, or the sum of the entropy values ​​of classes A and B under the new grayscale vector reaches the maximum (or the entropy increase is no longer significant), it means that the grayscale value has reached the optimal threshold segmentation point. If the number of iterations has not been reached or the entropy value continues to increase, return to step 6 to update the grayscale vector again, and continue to calculate the entropy value and determine whether the termination condition is met.

[0107] Based on the determined optimal grayscale vector, the optimal threshold segmentation point is obtained and threshold segmentation is performed accordingly, outputting the final target recognition and segmentation result image.

[0108] Thresholding image segmentation is a common image processing algorithm. Its core is image binarization, where a threshold is selected to convert the image into a black-and-white binary image. If the grayscale value of a pixel is greater than a certain threshold, its grayscale value is set to 255; otherwise, it is set to 0. It is commonly used for image segmentation and edge detection. Assuming the optimal threshold point is 's' obtained using the above method, and the original image after grayscale conversion is 'x', then image thresholding can be achieved using the MATLAB plotting function `im2bw(s,x)`.

[0109] Alternative Option 1:

[0110] For low-contrast images, the grayscale distribution is too concentrated in a region with a very small dynamic range, and the grayscale changes between the target and the background are not significant, making it difficult to segment the target and the background. To address this, linear or nonlinear transformations are usually used to expand the dynamic range of grayscale, or histogram equalization is used to make the grayscale distribution fill the entire grayscale level space, thereby improving the contrast of the image.

[0111] This algorithm is based on the grayscale information of an image. First, after converting the image to grayscale, the mean grayscale value of a rectangular region centered on a certain point in the image is taken and replaced with the original grayscale value. Then, the grayscale variance image of the original image is calculated, thereby enhancing the image contrast. Next, grayscale partitioning is performed, classifying the image according to the pixel grayscale attributes. A set of grayscale values ​​is selected to divide it into different grayscale regions, and the variance within each interval is minimized, thus dividing the grayscale of the entire image into several regions. Finally, regions with the same grayscale are connected together, and median filtering is used to remove noise and connect regions with the same grayscale, thereby achieving the segmentation of the target and the background.

[0112] This method is simple in principle, easy to implement, and convenient to calculate. It makes full use of the grayscale information of the image and eliminates the influence of low image contrast and background noise. It completes the segmentation of the image target by only calculating the mean and variance of the grayscale, dividing the grayscale variance interval and median filtering.

[0113] Alternative Option 2:

[0114] Thresholding based on maximum entropy is one of the most commonly used algorithms in image segmentation. It utilizes entropy, a concept in information theory used to measure the amount of information: the more uniform the distribution, the greater the entropy. In image processing, this involves fully utilizing the grayscale information of the image. After converting the image to grayscale, all pixels are divided into two categories—target and background—according to a certain threshold. When both tend to be uniformly distributed, the target entropy and background entropy also tend to be at their maximum, resulting in the best separation effect between the target and background.

[0115] This approach involves directly converting the original image to grayscale, then calculating entropy based on a preset threshold. The target and background entropies are calculated by iterating through all grayscale values ​​until their sum is maximized; this threshold is the optimal threshold segmentation point. Image segmentation based on this optimal threshold point allows for target identification and segmentation.

[0116] This method is simple to implement, easy to operate, and has stable performance. It only requires converting the image to grayscale, calculating the entropy value at each grayscale level, and finding the grayscale value that maximizes the entropy value. This eliminates the need to design an algorithm to search for the optimal threshold, thus greatly reducing the complexity of image segmentation algorithms and making it widely used in practice.

[0117] Airborne navigation images are taken from a high altitude, resulting in relatively low contrast, making it difficult to distinguish the target from the background and thus challenging target recognition. By utilizing the positional distribution information of the image's grayscale space, the image's grayscale space distribution entropy value is calculated and mapped to a uniform distribution function. Then, through 2D-DCT, transformation system weighting, and inverse 2D-DCT linear transformation processing, the global and local contrast of the original image are simultaneously enhanced.

[0118] The application of the maximum entropy harmony search algorithm, although similar to the genetic algorithm, adopts a multi-point mating mode in the process of searching for new solutions in the neighborhood, which makes the new solutions more diverse and increases the possibility of the algorithm escaping local optima.

[0119] The combination and application of the concept of entropy fully utilizes the grayscale information inherent in the image to simultaneously enhance the global and local contrast of traditional images. Furthermore, the threshold at which the entropy values ​​of the target and the background reach their maximum values ​​is used as the optimal threshold segmentation point, enabling accurate identification and segmentation of specific targets in the image.

[0120] By combining the maximum entropy harmony search algorithm, the computational cost is significantly reduced. When searching, the gray set based on the initial threshold is appropriately selected, which can effectively accelerate the convergence speed of the algorithm and quickly find the optimal threshold segmentation point.

[0121] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An aerial vehicle navigation image target recognition method based on spatial entropy and harmony search, characterized in that, The method comprises the following steps: Step S1: original image gray scale, converted into a gray image; Step S2: according to the two-dimensional space histogram calculation image gray space entropy and distribution function, contrast enhancement of graphics; Step S3: transformation and local contrast enhancement, the transformed domain coefficient is weighted, and the contrast of the local image is enhanced; Step S4: find the best threshold segmentation point, according to the characteristics of the recorded navigation image and the characteristics of the specific target, the best threshold segmentation point is found by searching the optimal algorithm; Step S5: target recognition and segmentation result output, according to the determined optimal gray vector, the best threshold segmentation point is obtained, and the threshold segmentation is carried out according to the best threshold segmentation point, and the final target recognition and segmentation result image is output.

2. The method of claim 1, wherein the method is based on spatial entropy and harmony search for aerial navigation image target recognition. The original image gray scale in step S1 is converted into a gray image by the following method: Determine the pixel value range corresponding to each gray level, convert the original image into a K gray level image, and calculate the two-dimensional space histogram on the space grid of the image, so that the original image corresponds to the gray image.

3. The method of claim 2, wherein the method further comprises: The two-dimensional space histogram on the space grid of the image is: Assume the original image size H*W, x(i,j) is each pixel value of image X, x(i,j) [0, z+] and the set of K gray levels present as image X, where , the spatial grid of X is calculated The two-dimensional space histogram of is: ， wherein m, n z+, h k (m, n) [0, z+] is positioned in the spatial grid in the image region The number of times of gray Xx appears, The total number of grids on the two-dimensional histogram is M*N, and the number of different gray levels K and the ratio ， ， wherein represents to take the nearest integer, in forming the gray scale two-dimensional spatial histogram of the detected features When, the aspect ratio of the original image is protected on the space grid.

4. The method of claim 1, wherein the method is based on spatial entropy and harmony search for aerial navigation image target recognition. The entropy measure of the gray level in step S2 is: ， Computing a discrete function measuring the gray scale relative to the remaining gray levels And normalized: ， ， In the formula, K is the number of gray levels.

5. The method of claim 1, wherein the method is based on spatial entropy and harmony search for aerial navigation image target recognition. The distribution function F calculated in the step S2 is: k F = 1 - exp(-x) ， ， In the formulae ， ， ， 。 6. The method of claim 1, wherein the method is based on spatial entropy and harmony search for aerial onboard navigation image target recognition. In step S3, the transformation and local contrast enhancement are carried out by the following method: An original picture and a transform picture of the same size generated by a transform of the picture are set as X and D, respectively, ， ， ， ， ， Wherein: the values of k and l respectively represent the frequency components, The transform domain coefficients d(k,l) are modified to obtain higher frequency transform coefficients ， ， ， In the formulae, The weighted frequency distribution coefficient after the adjustment is, Inverse transformation is carried out: ， ， The output image with enhanced global or local contrast is obtained.

7. The method of claim 1, wherein the method further comprises: In step S4, the best threshold segmentation point is found by the following method: Step S41: according to the characteristics of the airborne navigation image and the characteristics of the specific target, the initial feature threshold set is preset as the initial threshold segmentation point gray vector X=(x1,x2,...,xn), wherein 1<n<128, and the threshold segmentation point is normally distributed; Step S42: under each initial threshold segmentation point, the sum of the entropy values of the target and the background H(n) is calculated, the initial and sound memory library HM is generated, one or more tones are selected from the and sound library, the tone is fine tuned, and a new and sound is generated; Step S43: the entropy values of A and B classes under each gray level corresponding to the new and sound vector are calculated, the maximum entropy value is taken as the solution under this vector, and if the value is greater than the value in HM, the and sound memory library HM is updated; Step S44: when the number of times of updating the gray vector reaches the preset iteration number, or the sum of the entropy values of A and B classes under the new gray vector reaches the maximum, the best threshold segmentation point is determined.

8. The method of claim 7, wherein the method further comprises: The entropy values of A and B classes under each gray level corresponding to the new and sound vector in step S43 are: It is set to divide the image gray level from Th into A and B classes, wherein A and B are composed of all pixels with gray level in [0, Th] and [Th+1, n] in the image, ， ， In the formulae: the sum of the probabilities of the gray levels from 1 to Th, P is the sum of the probabilities of the gray levels from Th+1 to n i P is the probability distribution of the gray levels in the image.

9. The method of claim 8, wherein the method further comprises: The sum of the entropy values of A and B classes in step S44 is: ， The and sound memory library is: 。 10. The method of claim 1, wherein the method is based on spatial entropy and harmony search for aerial onboard navigation image target recognition. The threshold image segmentation in step S5 is: The image is binarized by selecting a threshold value to convert the image into black and white, and judging the size of the gray value in the image and the threshold value, if the gray value is greater than the threshold value, the gray value of the pixel is set to 255, otherwise it is set to 0.

Images