Frozen river detection method and device based on hybrid second-order gradient matrix eigenvalue

By using a mixed second-order gradient matrix eigenvalue and direction filter template in infrared images, the problems of slow detection speed and sample dependence of icy rivers in infrared imaging technology are solved, and efficient and accurate river detection is achieved, reducing calculations quantity and false alarm rate.

CN115861341BActive Publication Date: 2024-11-05UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211513359.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-11-29
Publication Date
2024-11-05
Estimated Expiration
2042-11-29

AI Technical Summary

Technical Problem

In the existing infrared imaging technology, the detection speed of icy rivers is slow and requires a large amount of sample data for training and learning, making it difficult to detect effectively, and multi-directional filtering analysis increases calculation time.

Method used

Using a method based on the hybrid second-order gradient matrix eigenvalue, the direction filter template is designed by calculating the eigenvalue and eigenvector of the mixed second-order gradient matrix of the infrared image, reducing the calculation amount and improving detection efficiency.

Benefits of technology

It realizes efficient and accurate detection of frozen rivers, reduces false alarm rates, and avoids the process of multi-directional filtering to enhance rivers, and improves detection speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115861341B_ABST
    Figure CN115861341B_ABST
Patent Text Reader

Abstract

The present invention discloses a frozen river detection method and device based on the eigenvalue of a mixed second-order gradient matrix. Aiming at the problem of interference of high-reflection frozen rivers on the detection of specific small targets in infrared imaging, the present invention aims to detect frozen rivers in infrared images, thereby reducing the false alarm rate of specific small target detection. The method comprises: inputting an original infrared image, calculating a mixed second-order gradient matrix of the input infrared image, constructing a frozen river candidate area map and a river direction vector map according to the eigenvalue and eigenvector of the mixed second-order gradient matrix; designing a filter template of a corresponding direction according to the river direction vector map, calculating the response value of the filter template on the river candidate area map, and obtaining a river area enhanced image; binarizing the river area enhanced image, setting a suitable area size threshold according to prior information and experimental results, eliminating areas with smaller areas, and obtaining the final detection result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image infrared imaging target detection, and in particular to a frozen river detection method and device based on mixed second-order gradient matrix eigenvalues. Background Art

[0002] With the development of infrared detection technology, infrared imaging target detection plays an important role in precision guidance, early warning, reconnaissance, etc. However, some scenes in nature, such as frozen rivers, lakes, cirrus clouds, snow-capped mountains, forest fires, etc., will produce high reflection and strong radiation energy, and have strong visual significance and high brightness response in infrared imaging scene images like specific targets, so they will strongly interfere with the detection of specific targets, resulting in a high false alarm rate.

[0003] Most of the current river detection technologies are based on data from visible light bands, and there is little research on infrared imaging data. At the same time, some river detection methods use machine learning to extract river features, such as texture, local entropy, etc., and input the extracted features into the classifier for training. This type of method requires a large number of samples and has a strong dependence on sample data. However, there is little public infrared image frozen river data, which makes it difficult to support classifier learning. Some detection methods also use active contour models and edge information to detect rivers, and use mathematical morphology methods to further optimize the extracted rivers. In addition, in order to make full use of the directionality of the river, existing methods generally use Gabor transform or shearlet transform to perform multi-scale and multi-directional analysis of images, which will undoubtedly increase the running time of the detection method.

[0004] Therefore, it is necessary to propose an efficient and accurate method for detecting frozen rivers. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of slow river detection speed and the need for a large amount of sample data for training and learning. The present invention provides a frozen river detection method based on the eigenvalues ​​of a mixed second-order gradient matrix, which can achieve high-efficiency and high-precision detection of frozen rivers.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A frozen river detection method based on mixed second-order gradient matrix eigenvalues ​​comprises the following steps:

[0008] Step 1. Obtain an original infrared image I1 containing a frozen river to be processed;

[0009] Step 2. Calculate the mixed second-order gradient matrix of the original infrared image, and obtain the frozen river candidate area map and the river direction vector map according to the eigenvalue and eigenvector of the mixed second-order gradient matrix;

[0010] Step 3. Design a filter template in the corresponding direction according to the river direction vector map obtained in step 2, calculate the response value of the filter template on the river candidate area map, and obtain an enhanced image of the river area;

[0011] Step 4 performs binarization processing on the enhanced image of the river area, sets a suitable area size threshold based on prior information and experimental results, removes areas with smaller areas, and outputs the detection results of frozen rivers.

[0012] Furthermore, the step 2 comprises the following steps:

[0013] Step 2.1 Select a Gaussian function with a scale factor of σ1 to perform smoothing filtering on image I1 to obtain image I2, where the value range of σ1 is 0.5 to 3.5;

[0014] Step 2.2 calculates the second-order gradient in the X direction, the second-order gradient in the Y direction, and the cross-second-order gradient of each pixel in the image I2 to obtain a mixed second-order gradient matrix for each pixel in the image I2;

[0015] Step 2.3 Calculate the corresponding eigenvalues ​​and eigenvectors according to the matrix H. The matrix corresponding to the largest eigenvalue is I3, and the matrix corresponding to the smallest eigenvalue is I4. Calculate the candidate area map of frozen rivers I5. The formula is as follows:

[0016]

[0017] Where λ is the scale factor. For 8-bit images, the value range of λ is 2 to 15. abs() means to find the absolute value.

[0018] Step 2.4 uses the inverse tangent function on the eigenvector corresponding to the smallest eigenvalue obtained in step 2.3 to obtain the river direction vector diagram I6.

[0019] Furthermore, the step 3 comprises the following steps:

[0020] Step 3.1 Let the angle value of position (m, n) in image I6 be θ i , design θ i The filter template F under the direction mn , F mn The size is Len×Len, L is θ i The length of the river in the direction ranges from 10 to 20, and σ2 represents the width of the river, ranging from 0.5 to 4;

[0021] For the filter template F mn For any point (x, y) in iThe new coordinates after are (x′, y′), and the filter template F is calculated. mn The weight coefficient of (x, y) is calculated as follows:

[0022]

[0023] F mn Subtract the average value of F from each non-zero pixel in the filter template F to obtain the final filter template F. mn ′;

[0024] Step 3.2 Take the image block P of size Len×Len with (m, n) as the center in image I5 mn , calculate the filter template F mn ′ is the response value on I5, and the river area enhanced image I7 is obtained. The calculation formula is as follows:

[0025]

[0026] For an 8-bit image, the value range of β is 5 to 20.

[0027] The present invention also provides a frozen river detection device based on mixed second-order gradient matrix eigenvalues, comprising the following modules:

[0028] Data acquisition module: acquiring the original infrared image I1 containing the frozen river to be processed;

[0029] Feature calculation module: calculates the mixed second-order gradient matrix of the original infrared image I1, and obtains the frozen river candidate area map and river direction vector map according to the eigenvalue and eigenvector of the mixed second-order gradient matrix;

[0030] Filtering module: Design a filter template in the corresponding direction according to the obtained river direction vector map, calculate the response value of the filter template on the river candidate area map, and obtain the river area enhanced image;

[0031] Output module: Binarize the enhanced image of the river area, set the appropriate area size threshold based on prior information and experimental results, eliminate areas with smaller areas, and output the detection results of frozen rivers.

[0032] In the above device, the feature calculation module is implemented by the following steps:

[0033] Step 2.1 Select a Gaussian function with a scale factor of σ1 to perform smoothing filtering on the original infrared image I1 to obtain image I2, where the value range of σ1 is 0.5 to 3.5;

[0034] Step 2.2 calculates the second-order gradient in the X direction, the second-order gradient in the Y direction, and the cross-second-order gradient of each pixel in the image I2 to obtain a mixed second-order gradient matrix H for each pixel in the image I2;

[0035] Step 2.3 Calculate the corresponding eigenvalues ​​and eigenvectors according to the matrix H. The matrix corresponding to the largest eigenvalue is I3, and the matrix corresponding to the smallest eigenvalue is I4. Calculate the frozen river candidate area map I5 according to I3 and I4. The formula is as follows:

[0036]

[0037] Where λ is the scale factor. For 8-bit images, the value range of λ is 2 to 15. abs() means to find the absolute value.

[0038] Step 2.4 uses the inverse tangent function on the eigenvector corresponding to the smallest eigenvalue obtained in step 2.3 to obtain the river direction vector diagram I6.

[0039] In the above device, the filtering module is specifically implemented by the following steps:

[0040] Step 3.1 Record the angle value of the position (m, n) in the river direction vector diagram I6 as θ i , design θ i The filter template F under the direction mn , F mn The size is Len×Len, L is θ i The length of the river in the direction ranges from 10 to 20, and σ2 represents the width of the river, ranging from 0.5 to 4;

[0041] For the filter template F mn For any point (x, y) in i The new coordinates after are (x′, y′), and the filter template F is calculated. mn The weight coefficient at position (x, y) is calculated as follows:

[0042]

[0043] F mn Subtract F from each non-zero pixel in mn The final filter template F is obtained by averaging mn ′;

[0044] Step 3.2 Take the image block P of size Len×Len with (m, n) as the center in image I5 mn , calculate the filter template F mn′ is the response value on image I5, and the river area enhanced image I7 is obtained. The calculation formula is as follows:

[0045]

[0046] For an 8-bit image, the value range of β is 5 to 20.

[0047] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0048] 1. The present invention obtains the flow direction of the river through the eigenvector of the mixed second-order gradient matrix of each pixel point in the infrared image, avoids the conventional use of filters in multiple directions to enhance the flow of the river, and reduces the amount of calculation.

[0049] 2. Since the largest eigenvalue in the mixed second-order gradient matrix in the frozen river area is much larger than the smallest eigenvalue, and the largest eigenvalue is negative, some non-river areas can be eliminated according to this numerical condition, and then the corresponding filter template is designed through the Gaussian-like shape of the river cross-section to enhance the river area, further highlighting the significance of the river. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 is a flow chart of the present invention;

[0052] Figure 2 The original infrared image containing the frozen river in Example 1 of the present invention;

[0053] Figure 3 The frozen river candidate area map in Example 1 of the present invention;

[0054] Figure 4 The river area enhancement map in Example 1 of the present invention;

[0055] Figure 5 This is a diagram of the frozen river detection results in Example 1 of the present invention. DETAILED DESCRIPTION

[0056] To facilitate those skilled in the art to understand the technical content of the present invention, the content of the present invention is described in detail below with reference to the accompanying drawings.

[0057] Example 1

[0058] like Figure 1 As shown, a frozen river detection method based on mixed second-order gradient matrix eigenvalues ​​includes the following steps:

[0059] Step 1. Obtain an original infrared image I1 containing a frozen river to be processed;

[0060] Step 2. Calculate the mixed second-order gradient matrix of the original infrared image, and obtain the frozen river candidate area map and the river direction vector map according to the eigenvalue and eigenvector of the mixed second-order gradient matrix;

[0061] Step 3. Design a filter template in the corresponding direction according to the river direction vector map obtained in step 2, calculate the response value of the filter template on the river candidate area map, and obtain an enhanced image of the river area;

[0062] Step 4 performs binarization processing on the enhanced image of the river area, sets a suitable area size threshold based on prior information and experimental results, removes areas with smaller areas, and outputs the detection results of frozen rivers.

[0063] On the basis of the above technical solution, step 2 comprises the following steps:

[0064] Step 2.1 Select a Gaussian function with a scale factor of σ1 to perform smoothing filtering on image I1 to obtain image I2, where the value range of σ1 is 0.5 to 4;

[0065] Step 2.2 calculates the second-order gradient in the X direction, the second-order gradient in the Y direction, and the cross-second-order gradient of each pixel in the image I2 to obtain a mixed second-order gradient matrix for each pixel in the image I2;

[0066] Step 2.3 Calculate the corresponding eigenvalues ​​and eigenvectors according to the matrix H. The matrix corresponding to the largest eigenvalue is I3, and the matrix corresponding to the smallest eigenvalue is I4. Calculate the candidate area map of frozen rivers I5. The formula is as follows:

[0067]

[0068] Where λ is the scale factor. For 8-bit images, the value range of λ is 2 to 15. abs() means to find the absolute value.

[0069] Step 2.4 uses the inverse tangent function on the eigenvector corresponding to the smallest eigenvalue obtained in step 2.3 to obtain the river direction vector diagram I6.

[0070] On the basis of the above technical solution, step 3 comprises the following steps:

[0071] Step 3.1 Let the angle value of position (m, n) in image I6 be θ i , design θi The filter template F under the direction mn , F mn The size is Len×Len, L is θ i The length of the river in the direction ranges from 10 to 20, and σ2 represents the width of the river, ranging from 0.5 to 4;

[0072] For the filter template F mn For any point (x, y) in i The new coordinates after are (x′, y′), and the filter template F is calculated. mn The weight coefficient of (x, y) is calculated as follows:

[0073]

[0074] F mn Subtract the average value of F from each non-zero pixel in the filter to obtain the final filter template F. mn ′;

[0075] Step 3.2 In image I5, take (m, n) as the center and take the size The image block P mn , calculate the filter template F mn ′ is the response value on I5, and the river area enhanced image I7 is obtained. The calculation formula is as follows:

[0076]

[0077] For 8-bit images, the value range of β is 5 to 20.

[0078] In the above scheme, by using the eigenvector corresponding to the minimum eigenvalue of the mixed second-order gradient matrix, the direction of the smallest curvature change corresponding to each pixel point is calculated, and the river pixel point corresponds to the flow direction of the river. Then, the corresponding filter template is designed through the Gaussian-like shape of the river cross-section to enhance the river area, further highlighting the significance of the river, avoiding the conventional process of using filters in multiple directions to enhance the river, and reducing the amount of calculation.

[0079] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention. It should be understood that the scope of protection of the present invention is not limited to the specific statements and embodiments. For those skilled in the art, the present invention may have various modifications and variations. Any modification, replacement, improvement, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A frozen river detection method based on mixed second-order gradient matrix eigenvalues, characterized in that: The method comprises the following steps: Step 1. Obtain an original infrared image I1 containing a frozen river to be processed; Step 2. Calculate the mixed second-order gradient matrix of the original infrared image I1, and obtain the frozen river candidate area map and the river direction vector map according to the eigenvalue and eigenvector of the mixed second-order gradient matrix; Step 3. Design a filter template in the corresponding direction according to the river direction vector map obtained in step 2, calculate the response value of the filter template on the river candidate area map, and obtain an enhanced image of the river area; Step 4 performs binarization processing on the river area enhanced image, sets a suitable area size threshold based on prior information and experimental results, removes areas with smaller areas, and outputs the detection results of frozen rivers; The step 2 comprises the following steps: Step 2.1 Select the scale factor as The original infrared image I1 is smoothed by a Gaussian function to obtain image I2. The value range of is 0.5~3.5; Step 2.2 calculates the second-order gradient in the X direction, the second-order gradient in the Y direction, and the cross-second-order gradient of each pixel in the image I2 to obtain a mixed second-order gradient matrix H for each pixel in the image I2; Step 2.3 Calculate the corresponding eigenvalues ​​and eigenvectors according to the matrix H. The matrix corresponding to the largest eigenvalue is I3, and the matrix corresponding to the smallest eigenvalue is I4. Calculate the frozen river candidate area map I5 according to I3 and I4. The formula is as follows: Where λ is the scale factor. For 8-bit images, the value range of λ is 2~15, and abs() means to find the absolute value; Step 2.4 uses the inverse tangent function on the eigenvector corresponding to the smallest eigenvalue obtained in step 2.3 to obtain the river direction vector diagram I6.

2. The frozen river detection method based on mixed second-order gradient matrix eigenvalues ​​according to claim 1 is characterized in that: The step 3 comprises the following steps: Step 3.1 Record the position of the river direction vector diagram I6 as The angle value is ,design Filter template under direction , The size is , , L is The length of the river in the direction, the value range is 10~20, Represents the width of the river, with a value range of 0.5~4; For filter template Any point in , remember the rotation angle The new coordinates are , calculate the filter template The middle position is The weight coefficient is calculated as follows: right Subtract each non-zero pixel from The final filter template is obtained by averaging ; Step 3.2 In image I5, Center, size Image blocks , calculate the filter template The response value on image I5 is used to obtain the river area enhanced image I7. The calculation formula is as follows: Among them, for 8-bit images, The value range is 5~20.

3. A frozen river detection device based on mixed second-order gradient matrix eigenvalues, characterized in that: Includes the following modules: Data acquisition module: acquiring the original infrared image I1 containing the frozen river to be processed; Feature calculation module: calculates the mixed second-order gradient matrix of the original infrared image I1, and obtains the frozen river candidate area map and river direction vector map according to the eigenvalue and eigenvector of the mixed second-order gradient matrix; Filtering module: Design a filter template in the corresponding direction according to the obtained river direction vector map, calculate the response value of the filter template on the river candidate area map, and obtain the river area enhanced image; Output module: Binarize the enhanced image of the river area, set the appropriate area size threshold based on prior information and experimental results, remove the area with smaller area, and output the detection results of the frozen river; The feature calculation module implementation includes the following steps: Step 2.1 Select the scale factor as The original infrared image I1 is smoothed by a Gaussian function to obtain image I2. The value range of is 0.5~3.5; Step 2.2 calculates the second-order gradient in the X direction, the second-order gradient in the Y direction, and the cross-second-order gradient of each pixel in the image I2 to obtain a mixed second-order gradient matrix H for each pixel in the image I2; Step 2.3 Calculate the corresponding eigenvalues ​​and eigenvectors according to the matrix H. The matrix corresponding to the largest eigenvalue is I3, and the matrix corresponding to the smallest eigenvalue is I4. Calculate the frozen river candidate area map I5 according to I3 and I4. The formula is as follows: Where λ is the scale factor. For 8-bit images, the value range of λ is 2~15, and abs() means to find the absolute value; Step 2.4 uses the inverse tangent function on the eigenvector corresponding to the smallest eigenvalue obtained in step 2.3 to obtain the river direction vector diagram I6.

4. The frozen river detection method based on mixed second-order gradient matrix eigenvalues ​​according to claim 3 is characterized in that: The specific implementation of the filtering module includes the following steps: Step 3.1 Record the position of the river direction vector diagram I6 as The angle value is ,design Filter template under direction , The size is , , L is The length of the river in the direction, the value range is 10~20, Represents the width of the river, with a value range of 0.5~4; For filter template Any point in , remember the rotation angle The new coordinates are , calculate the filter template The middle position is The weight coefficient is calculated as follows: right Subtract each non-zero pixel from The final filter template is obtained by averaging ; Step 3.2 In image I5, Center, size Image blocks , calculate the filter template The response value on image I5 is used to obtain the river area enhanced image I7. The calculation formula is as follows: Among them, for 8-bit images, The value range is 5~20.