A low-altitude target edge detection method based on PCNN and improved morphology

Abstract

The application discloses a low-altitude target edge detection method based on PCNN and improved morphology, and belongs to the technical field of aerospace detection. The method is as follows: an image containing a low-altitude target is obtained by a photoelectric load on a low-altitude platform, and a filtering algorithm is used to pre-process the image; a PCNN model is used to perform image segmentation on the filtered image, and a region containing the low-altitude target is screened out; the region is extracted to obtain an image to be detected, and the influence of details on edge detection is eliminated; an improved morphological gradient is used to perform edge detection on the obtained image to be detected; the improved morphological gradient is used to suppress random noise, so that the precision of low-altitude target edge detection and the anti-noise capability are improved; a morphological thinning operator is used to thin the obtained low-altitude target edge, the connectivity of small parts of the image is maintained, and the precision of low-altitude target edge detection is improved. The application has the advantages of high detection precision and strong anti-interference capability.

Description

Technical Field

[0001] This invention belongs to the field of aerospace detection technology and relates to a low-altitude target edge detection method based on PCNN and improved morphology. Background Technology

[0002] In recent years, with the gradual opening of low-altitude airspace, many countries have seen the future development trend of low-altitude airspace and have begun to vigorously promote its application, making many low-altitude aircraft a research hotspot. The detection, identification, and tracking of low-altitude targets has always been a challenge for low-altitude detection systems. With the impending opening of low-altitude airspace policies in recent years, solving the problem of low-altitude target detection and identification has become urgent.

[0003] Low-altitude target detection faces challenges such as limited available features, diverse backgrounds, and large target sizes, making it difficult to obtain satisfactory detection results using existing algorithms, and also hindering the acquisition of fine target edges. While advancements in artificial intelligence have led to the development of numerous small target detection methods, a more robust and accurate detection method is needed for tasks like low-altitude target edge detection. This method should offer advantages such as high detection accuracy and strong anti-interference capabilities, be deployable on airborne platforms, and improve the accuracy and robustness of low-altitude target edge detection. Summary of the Invention

[0004] The main objective of this invention is to provide a low-altitude target edge detection method based on pulse-coupled neural networks (PCNN) and improved morphology, which can improve the accuracy of low-altitude target edge detection even in images with significant noise and background interference. This invention has the advantages of high detection accuracy and strong anti-interference capability, and can be deployed on equipment in airspace platforms, improving the accuracy and robustness of low-altitude target edge detection.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] This invention discloses a low-altitude target edge detection method based on PCNN and improved morphology. It acquires images containing low-altitude targets captured by an optoelectronic payload mounted on an airborne platform, and then preprocesses the images using a filtering algorithm. Before image processing, the weights and bias parameters of each node in the PCNN model are initialized. The PCNN model is used to segment the filtered image, filtering out regions containing low-altitude targets and extracting the regions to be detected, thus improving the accuracy and speed of image segmentation, eliminating the influence of details on edge detection, and improving the anti-interference ability of subsequent edge detection. For the obtained image to be detected, an improved morphological gradient is used for edge detection. The morphological gradient filtering operator's ability to suppress random noise is utilized to improve the accuracy and noise resistance of low-altitude target edge detection. A morphological thinning operator is used to thin the obtained low-altitude target edges, maintaining the connectivity of small parts of the image and improving the accuracy of low-altitude target edge detection. This invention has the advantages of high detection accuracy and strong anti-interference ability, and can be deployed on airborne platform equipment, improving the accuracy and robustness of low-altitude target edge detection.

[0007] This invention discloses a low-altitude target edge detection method based on PCNN and improved morphology, comprising the following steps:

[0008] Step 1: Acquire images containing low-altitude targets taken by the optoelectronic payload mounted on the airborne platform, and then preprocess the images using a filtering algorithm.

[0009] As a preferred method, adaptive median filtering can remove noise from images while preserving information about small objects to the maximum extent. Adaptive median filtering dynamically changes the window size of the median filter based on preset conditions to simultaneously achieve noise reduction and detail preservation.

[0010] Step 2: Before image processing, initialize the weights and bias parameters of each node in the PCNN model.

[0011] As a preferred method, an iterative algorithm is used to automatically select the threshold. The setting of the threshold has a significant impact on the processing of subsequent content and needs to be determined precisely.

[0012] Step 3: Use the PCNN model to segment the image after adaptive median filtering, filter out the regions containing low-altitude targets, extract the regions to be detected, improve the accuracy and speed of image segmentation, eliminate the influence of details on edge detection, and thus improve the anti-interference ability of subsequent edge detection.

[0013] The PCNN model consists of an input layer, a coupling layer, and a spike output layer. When processing images, it is a single-layer, two-dimensional, locally connected network. The number of neurons equals the number of pixels in the input image, with a one-to-one correspondence between neurons and pixels. Each neuron is connected to its corresponding pixel and also to its neighboring neurons. The basic PCNN model can be simplified as follows:

[0014]

[0015]

[0016] U ij (n)=F ij (n)(1+βL ij (n))

[0017] E ij (n)=exp(-α E E ij (n-1)+V E V ij (n-1)

[0018]

[0019] in, and α is the decay time constant. E V is the threshold decay time constant. E V is the threshold amplitude coefficient. F and V L β is the connection amplitude coefficient, M is the connection strength, and M is the connection amplitude coefficient. ijkl and W ijkl This is the link weight.

[0020] Step 4: For the obtained image to be detected, perform edge detection using an improved morphological gradient. Utilize the morphological gradient filtering operator's ability to suppress random noise to improve the accuracy and noise resistance of low-altitude target edge detection.

[0021] The definitions of opening and closing operations for grayscale images have the same form as those for binary images. The definitions of opening and closing operations are as follows:

[0022]

[0023]

[0024] Where f is the original image and b is the structuring element. Θ represents the expansion operation, and Θ represents the erosion operation.

[0025] Since the width of random noise may be smaller than the width of the structuring element, this will cause the random noise to be mixed in with the processed image. Therefore, two 3×3 structuring elements are constructed: a cross-shaped structuring element C1 and a cross-shaped structuring element C2.

[0026]

[0027] This makes the structuring elements symmetrical and complementary about the origin, effectively reducing missed detections. Using a weighted combination of these two morphological gradients, the improved morphological gradient filtering operator is:

[0028]

[0029]

[0030] y = c1y × +c2y + +k(max(y × ,y + )-min(y × ,y + ))

[0031] Where c1 and c2 are weighting coefficients, y × and y + Let c1, c2, and k be the improved morphological gradients corresponding to two structuring elements with different shapes, and k be the coefficient of the limit difference between the edge extractions of the two structuring elements after edge extraction. c1, c2, and k are all determined using an adaptive algorithm.

[0032] Step 5: Use morphological thinning operators to thin the obtained low-altitude target edges, maintain the connectivity of small parts of the image, and improve the accuracy of low-altitude target edge detection.

[0033] Mathematical morphological thinning algorithms can be used to obtain the shape features and topological structure of a target image and reduce the amount of redundant information in the image. A typical thinning method is Zhang's fast parallel algorithm. Assuming there is a 3×3 image region, each point is labeled P1, P2, ..., P8, where 1 represents black and 0 represents white. If the center P1 = 1 (black point), P1 is deleted (P1 = 0) when all four conditions are met simultaneously.

[0034] ① 2≤N(P1)≤6;

[0035] ② T(P1)=1;

[0036] ③P2*P4*P8=0 and T(P1)≠1;

[0037] ④P2*P4*P6=0 and T(P4)≠1.

[0038] Mathematical morphological thinning algorithms are iterative algorithms that repeatedly delete pixels that meet certain conditions from the edges of an image, eventually resulting in single-pixel edges. Whether or not a pixel is deleted depends only on the result of the previous iteration, and is independent of the pixel value and its order in the image.

[0039] Beneficial effects:

[0040] 1. This invention discloses a low-altitude target edge detection method based on PCNN and improved morphology. It uses a simplified PCNN network to filter out regions where low-altitude targets exist, thereby improving the accuracy and speed of image segmentation, eliminating the influence of details on edge detection, and thus improving the anti-interference ability of subsequent edge detection.

[0041] 2. The present invention discloses a low-altitude target edge detection method based on PCNN and improved morphology. It uses improved morphology for edge detection, utilizes detection operators with different structures to improve the accuracy of low-altitude target edge detection, and utilizes the characteristics of morphological gradient filtering operators to suppress random noise, thereby improving the noise resistance of low-altitude target edge detection.

[0042] 3. The present invention discloses a low-altitude target edge detection method based on PCNN and improved morphology, which uses morphological thinning operators to thin the obtained low-altitude target edges, maintains the connectivity of small parts of the image, and improves the accuracy of low-altitude target edge detection. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating a low-altitude target edge detection method based on PCNN and improved morphology in this invention.

[0044] Figure 2 This is a schematic diagram of the Pulse Coupled Neural Network (PCNN) structure in this invention;

[0045] Figure 3 This is a schematic diagram of the improved morphological structure in this invention;

[0046] Figure 4 This is a diagram showing the results of low-altitude target edge detection based on PCNN and improved morphology in this invention. Detailed Implementation

[0047] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings. However, those skilled in the art will understand that the following embodiments are for illustrative purposes only and should not be considered as limiting the scope of the invention.

[0048] Reference Figure 1 This embodiment discloses a lightweight method for detecting unknown targets in deep space, and the specific implementation steps are as follows:

[0049] Step 1: Acquire images containing low-altitude targets taken by the optoelectronic payload mounted on the airborne platform, and then preprocess the images using a filtering algorithm.

[0050] Adaptive median filtering can remove noise from images while preserving information about small objects to the maximum extent. Adaptive median filtering dynamically changes the window size of the median filter based on preset conditions to balance noise reduction and detail preservation.

[0051] The following is a detailed description of the adaptive median filter algorithm:

[0052] The following symbols are predefined:

[0053] S xy The effective area of ​​the filter is the region covered by the filter window, and the center point of this region is the pixel in the y-th row and x-th column of the image.

[0054] Z min :S xy The smallest grayscale value;

[0055] Z max :S xy The largest grayscale value in the middle;

[0056] Z med :S xy The median of all grayscale values;

[0057] Z xy : Represents the grayscale value of the pixel in the y-th row and x-th column of the image;

[0058] S max :S xy The maximum allowed window size.

[0059] Adaptive median filtering is divided into two processes: Process 1 and Process 2.

[0060] Process 1: A1 = Z med -Z min A2 = Z med -Z max If A1 > 0 and A2 < 0, then proceed to process two. Otherwise, increase the window size; if the increased size ≤ S max Repeat step one if necessary; otherwise, output Z directly. med .

[0061] Process 2: B1 = Z xy -Z min B2 = Z xy -Z max If B1 > 0 and B2 < 0, then output Z. xy Otherwise, output Z.med .

[0062] Step 2: Before image processing, initialize the weights and bias parameters of each node in the PCNN model.

[0063] As a preferred method, an iterative algorithm is used to automatically select the threshold. The setting of the threshold has a significant impact on the processing of subsequent content and needs to be determined precisely.

[0064] The algorithm for finding the optimal threshold is as follows:

[0065] (1) Choose an initial estimate of θ, which can be the minimum gray value in the image;

[0066] (2) When using θ as the threshold, the region with gray level greater than θ is defined as Q1, and the region with gray level less than θ is defined as Q2;

[0067] (3) Calculate the average gray values ​​t1 and t2 of the pixels in regions Q1 and Q2;

[0068] (4) Obtain a new threshold:

[0069] θ = 1 / 2(t1 + t2)

[0070] (5) Continue until θ achieves the optimal effect or reaches the specified number of iterations, then use θ as the final threshold result. Otherwise, repeat steps (2) to (4) until the requirements are met.

[0071] Step 3: Use the PCNN model to segment the image after adaptive median filtering, filter out the regions containing low-altitude targets, extract the regions to be detected, improve the accuracy and speed of image segmentation, eliminate the influence of details on edge detection, and thus improve the anti-interference ability of subsequent edge detection.

[0072] The PCNN model consists of an input layer, a coupling layer, and a spike output layer. When processing images, it is a single-layer, two-dimensional, locally connected network. The number of neurons equals the number of pixels in the input image, with a one-to-one correspondence between neurons and pixels. Each neuron is connected to its corresponding pixel and also to its neighboring neurons. The basic PCNN model can be simplified as follows:

[0073]

[0074]

[0075] U ij (n)=F ij (n)(1+βL ij (n))

[0076] Eij (n)=exp(-α E E ij (n-1)+V E V ij (n-1)

[0077]

[0078] in, and α is the decay time constant. E V is the threshold decay time constant. E V is the threshold amplitude coefficient. F and V L β is the connection amplitude coefficient, M is the connection strength, and M is the connection amplitude coefficient. ijkl and W ijkl This is the link weight.

[0079] Step 4: For the obtained image to be detected, perform edge detection using an improved morphological gradient. Utilize the morphological gradient filtering operator's ability to suppress random noise to improve the accuracy and noise resistance of low-altitude target edge detection.

[0080] The definitions of opening and closing operations for grayscale images have the same form as those for binary images. The definitions of opening and closing operations are as follows:

[0081]

[0082]

[0083] Where f is the original image and b is the structuring element. Θ represents the expansion operation, and Θ represents the erosion operation.

[0084] Since the width of random noise may be smaller than the width of the structuring element, this will cause the random noise to be mixed in with the processed image. Therefore, two 3×3 structuring elements are constructed: a cross-shaped structuring element C1 and a cross-shaped structuring element C2.

[0085]

[0086] This makes the structuring elements symmetrical and complementary about the origin, effectively reducing missed detections. Using a weighted combination of these two morphological gradients, the improved morphological gradient filtering operator is:

[0087]

[0088]

[0089] y = c1y × +c2y+ +k(max(y × ,y + )-min(y × ,y + ))

[0090] Where c1 and c2 are weighting coefficients, y × and y + Let c1, c2, and k be the improved morphological gradients corresponding to two structuring elements with different shapes, and k be the coefficient of the limit difference between the edge extractions of the two structuring elements after edge extraction. c1, c2, and k are all determined using an adaptive algorithm.

[0091] Step 5: Use morphological thinning operators to thin the obtained low-altitude target edges, maintain the connectivity of small parts of the image, and improve the accuracy of low-altitude target edge detection.

[0092] Mathematical morphological thinning algorithms can be used to obtain the shape features and topological structure of a target image and reduce the amount of redundant information in the image. A typical thinning method is Zhang's fast parallel algorithm. Assuming there is a 3×3 image region, each point is labeled P1, P2, ..., P8, where 1 represents black and 0 represents white. If the center P1 = 1 (black point), P1 is deleted (P1 = 0) when all four conditions are met simultaneously.

[0093] ⑤ 2≤N(P1)≤6;

[0094] ⑥ T(P1)=1;

[0095] ⑦ P2*P4*P8=0 and T(P1)≠1;

[0096] ⑧P2*P4*P6=0 and T(P4)≠1.

[0097] This algorithm is an iterative algorithm that repeatedly deletes pixels that meet certain conditions from the image edges, eventually resulting in single-pixel edges. Whether a pixel is deleted depends only on the result of the previous iteration and is independent of the pixel value and its order in the image.

[0098] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is used to explain the present invention. It is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A low-altitude target edge detection method based on PCNN and improved morphology, characterized in that: Includes the following steps, Step 1: Acquire images containing low-altitude targets captured by the optoelectronic payload mounted on the airborne platform, and then preprocess the images using a filtering algorithm; Step 2: Before image processing, initialize the weights and bias parameters of each node in the PCNN model; Step 3: Use the PCNN model to segment the image after adaptive median filtering, filter out the regions containing low-altitude targets, extract the regions to obtain the image to be detected, improve the accuracy and speed of image segmentation, and eliminate the influence of details on edge detection; In step three, The PCNN model consists of an input layer, a coupling layer, and a spike output layer. When processing images, the PCNN model is a single-layer, two-dimensional, locally connected network. The number of neurons equals the number of pixels in the input image, with a one-to-one correspondence between neurons and pixels. Each neuron is connected to its corresponding pixel and also to its neighboring neurons. The basic PCNN model can be simplified as follows: YOU ij (n)=F ij (n)(1+βL ij (a)) E ij (n)=exp(-α E )E ij (n-1)+V E V ij (n-1) in, and α is the decay time constant. E V is the threshold decay time constant. E V is the threshold amplitude coefficient. F and V L β is the connection amplitude coefficient, M is the connection strength, and M is the connection amplitude coefficient. ijkl and W ijkl Link weight; Step 4: For the obtained image to be detected, edge detection is performed using an improved morphological gradient. The morphological gradient filtering operator is used to suppress random noise, thereby improving the accuracy and noise resistance of low-altitude target edge detection. The definitions of opening and closing operations for grayscale images have the same form as those for binary images; the definitions of opening and closing operations are as follows: Where f is the original image and b is the structuring element. For dilation, Θ is the dilation operation; for erosion, Θ is the erosion operation. Since the width of random noise may be smaller than the width of the structuring element, this will cause the random noise to be mixed in with the processed image. Therefore, two 3×3 structuring elements are constructed: a cross-shaped structuring element C1 and a cross-shaped structuring element C2. This makes the structuring elements symmetrical and complementary about the origin, effectively reducing missed detections. Using a weighted combination of these two morphological gradients, the improved morphological gradient filtering operator is: y=c1y × +c2y + +k(max(y × ,and + )-min(y × ,and + )) Where c1 and c2 are weighting coefficients, y × and y + The improved morphological gradients are defined for two structuring elements with different shapes, and k is the coefficient of the limit difference between the edge extractions of the two structuring elements after edge extraction. c1, c2 and k are all determined using an adaptive algorithm. Step 5: Use morphological thinning operators to thin the obtained low-altitude target edges, maintain the connectivity of small parts of the image, and improve the accuracy of low-altitude target edge detection.

2. The low-altitude target edge detection method based on PCNN and improved morphology as described in claim 1, characterized in that: Adaptive median filtering can remove noise from an image while preserving small object information to the maximum extent. Adaptive median filtering dynamically changes the window size of the median filter according to preset conditions to simultaneously achieve the effects of noise reduction and detail preservation.

3. The low-altitude target edge detection method based on PCNN and improved morphology as described in claim 1, characterized in that: The threshold is automatically selected using an iterative algorithm.

4. The low-altitude target edge detection method based on PCNN and improved morphology as described in claim 1, characterized in that: Morphological thinning algorithms can be used to obtain the shape features and topological structure of a target image and reduce the amount of redundant information in the image. A typical thinning method is Zhang's fast parallel algorithm. Given a 3×3 image region, each point in it is labeled as P1, P2, ..., P8, where 1 represents black and 0 represents white. If the center P1 = 1, P1 is a black point. When all four conditions are met simultaneously, P1 is deleted, and P1 changes from black to white. ①2≤N(P1)≤6; ②T(P1)=1; ③P2*P4*P8=0 and T(P1)≠1; ④P2*P4*P6=0 and T(P4)≠1; Mathematical morphology thinning algorithms are iterative algorithms that repeatedly delete pixels that meet certain conditions from the edges of an image to obtain single-pixel edges. Whether a pixel is deleted depends solely on the result of the previous iteration, regardless of the pixel value or its order in the image.

Images