Hydrophone visual positioning algorithm based on color fusion

Abstract

The invention relates to the technical field of underwater sound measurement and calibration, in particular to a hydrophone visual positioning algorithm based on color fusion, which comprises the following steps: acquiring RGB images of a hydrophone and a vibrating diaphragm; converting the image into a Lab color space, and separating brightness and chrominance information; respectively carrying out edge detection on the brightness channel and the chrominance channel; a fusion weight is dynamically calculated according to the edge intensity of each channel, weighted fusion and logic optimization are carried out on multi-channel edge detection results, and brightness and color edge features are comprehensively utilized; performing morphological processing on the fused edge to optimize the contour quality; according to the method, the integrity and the accuracy of edge detection are enhanced by fusing color information, personal errors are eliminated, high precision of visual positioning in the hydrophone calibration process is realized, and the overall performance and the reliability of a calibration system are improved.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic measurement and calibration technology, specifically to a visual positioning method for hydrophone calibration, and more particularly to a contour enhancement detection algorithm based on color fusion, which is suitable for high-precision visual positioning between the hydrophone and the diaphragm in a laser interferometric hydrophone calibration system. Background Technology

[0002] In the field of underwater acoustic metrology, hydrophone calibration is typically performed using laser interferometry. This calibration method requires precise measurement of the relative position between the hydrophone and the vibrating diaphragm in the calibration device to ensure the accuracy of sound pressure transmission and the reliability of the calibration results. Traditional calibration processes usually involve acquiring images of the hydrophone and diaphragm using a visual positioning system and then measuring their positions based on these images.

[0003] Currently, common visual positioning solutions typically employ high-resolution industrial cameras to capture images of the hydrophone and diaphragm. The images are then converted from pixel size to actual physical dimensions, and after image magnification, operators manually mark key points such as the hydrophone edge and diaphragm center. Finally, the distance between the two is calculated. This method's positioning accuracy is highly dependent on the operator's experience and judgment, introducing significant human error and increasing measurement uncertainty, making it difficult to meet the requirements of high-precision hydrophone calibration.

[0004] Furthermore, existing edge detection algorithms for industrial visual positioning are mostly based on grayscale image processing, such as the Canny operator and the Sobel operator. These methods primarily utilize the image's brightness information for edge extraction. For objects with significant color contrast but insignificant brightness differences, their edge detection performance is often poor, easily losing important color boundary information. In the practical scenario of hydrophone calibration, the hydrophone, its background, diaphragm, etc., may exhibit specific color characteristics. Relying solely on edge detection via the brightness channel cannot fully utilize the color information in the image, resulting in incomplete contour extraction and limited positioning accuracy.

[0005] Therefore, there is an urgent need to propose a technical solution that can comprehensively utilize image color information and contour features, reduce human intervention, and improve visual positioning accuracy and robustness. Summary of the Invention

[0006] This invention aims to address the technical problems in existing hydrophone visual positioning methods, such as large subjective errors caused by manual marking, reliance on operator experience, and incomplete contour extraction and insufficient positioning accuracy due to the incomplete utilization of color information in traditional edge detection algorithms. This invention provides a contour enhancement detection algorithm based on color fusion. By fusing edge information from luminance and chrominance channels, it achieves high-precision, automated extraction of the hydrophone and diaphragm contours, thereby improving the positioning accuracy and reliability of the entire calibration system.

[0007] To achieve the above objectives, this invention provides a hydrophone visual localization algorithm based on color fusion, comprising the following steps:

[0008] Step S1: Image Acquisition and Preprocessing

[0009] An industrial camera fixed to a vision positioning mechanism acquires color images of the hydrophone under test and its vibrating diaphragm. The image format is RGB.

[0010] Step S2: Color Space Conversion

[0011] The acquired RGB image is converted to the Lab color space to separate the image's luminance and color information. The specific conversion process includes:

[0012] RGB to XYZ conversion: Converting normalized RGB values ​​to the CIE XYZ color space using a linear transformation matrix;

[0013] XYZ to Lab conversion: Using a non-linear transformation function, the XYZ values ​​are converted to the CIE Lab color space to obtain the luminance channel L and the chrominance channels a* and b*.

[0014] Chroma channel normalization: To facilitate subsequent image processing, the numerical ranges of chroma channels a and b are normalized to the interval [0, 255], resulting in a. norm and b norm ;

[0015] Step S3: Multi-channel edge detection

[0016] For the luminance channel L* and the normalized chrominance channel a norm b norm The Canny edge detection algorithm is executed independently to obtain initial binary edge maps for the three channels. Canny detection includes the following sub-steps:

[0017] Gaussian filtering: Smooths the image of each channel using a Gaussian kernel with a specified standard deviation σ to suppress noise;

[0018] Gradient calculation: The Sobel operator is used to calculate the magnitude and direction of the gradient in the horizontal and vertical directions of the image;

[0019] Non-maximum suppression: Search for local maxima of the gradient magnitude along the gradient direction to refine the edges;

[0020] Dual threshold processing and edge connection: Set high and low thresholds to divide pixels into strong edge points, weak edge points and non-edge points, and retain the weak edge points connected to the strong edges as the final edge points;

[0021] Step S4: Edge Blending

[0022] Edge information detected from the three channels is fused to comprehensively utilize both luminance and color edges. The fusion process includes:

[0023] Channel weight calculation: Based on the average gradient magnitude of each channel image (reflecting the overall edge strength of that channel), the fusion weights of the edge maps of each channel are dynamically assigned. w L* , w a* , w b* The weighting principle is that channels with higher edge strength have a larger proportion in the fusion process.

[0024] Weighted fusion: This involves combining three binary edge maps with their corresponding weights. w L* , w a* , w b* After multiplying and adding, we obtain the weighted fused edge intensity map E. weighted ;

[0025] Binarization and Logical Optimization: For Weighted Fusion Graph E weighted Threshold binarization is performed to obtain a preliminary fused edge map. To further ensure that no significant edges in each channel are lost, a logical OR operation is performed between this preliminary fused edge map and the edge maps of the three original channels to obtain the optimized fused edge map E. fused ;

[0026] Step S5: Morphological Post-processing

[0027] For the merged edge map E fused Morphological operations are performed to optimize edge continuity and smoothness, remove minor noise, and fill edge gaps; specifically including:

[0028] Dilation operation: Dilates the edge map using structuring elements of specific shapes and sizes (such as circles) to connect adjacent edge breaks and fill small holes;

[0029] Erosion operation: Erosion is performed on the dilated image using the same or different structuring elements to eliminate burrs that may be introduced by the dilation and restore the original approximate shape of the edges;

[0030] By combining dilation and erosion, such as a "closing operation" of dilation followed by erosion, the final continuous, smooth, and complete binary edge map E is obtained. final ;

[0031] Step S6: Contour extraction and localization calculation

[0032] From the final edge map E final The outer contours of the hydrophone housing and the vibrating diaphragm are extracted. By calculating feature points such as the geometric center of the contour, the center of the smallest circumcircle, or the center of the fitted ellipse, and combining the camera calibration parameters, the actual distance between the hydrophone and the diaphragm is calculated, thus completing the visual positioning.

[0033] Compared with the prior art, the beneficial effects of the present invention are: by automatically extracting the contour and calculating the position through the algorithm, human error is avoided and the reliability of the calibration results is improved; by introducing color information, namely the Lab chromaticity channel, into the edge detection process, and through multi-channel edge detection and fusion, edges that are not significant in the luminance channel but have strong contrast in the color channel can be effectively extracted, thereby obtaining a more accurate object contour. Attached Figure Description

[0034] Figure 1 The original image (a) and the single-channel binary edge map E of this invention are shown. channel (b);

[0035] Figure 2 E is the single-channel binary edge map of this invention. channel (a) Figure E after weighted fusion weighted (b);

[0036] Figure 3 Figure E is the weighted fusion result of this invention. weighted (a) Final processed image E after morphological dilation final (b). Detailed Implementation

[0037] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, so that those skilled in the art can more clearly understand how to practice the present invention. Although the present invention has been described in conjunction with its preferred embodiments, these embodiments are merely illustrative and not intended to limit the scope of the invention.

[0038] See Figure 1-3 One embodiment of the present invention provides a hydrophone visual localization algorithm based on color fusion. The present invention mainly uses a contour enhancement detection algorithm based on color fusion, employing algorithms such as RGB color space conversion, Gaussian filtering, gradient calculation and edge thinning, dual thresholding, edge fusion, and morphological processing. The color fusion algorithm enhances the accuracy and precision of edge detection. The steps include:

[0039] Step 1: Color Space Conversion

[0040] An industrial camera fixed to a vision positioning mechanism acquires color images of the hydrophone under test and its diaphragm in RGB format. The RGB images are then converted to Lab color space, separating the luminance channel (L*) and chrominance channels (a*, b*), providing a foundation for multi-channel edge detection, including:

[0041] 1.1 RGB to XYZ space conversion

[0042] Let R be the normalized value of RGB pixels. norm G norm B norm (Range [0,1]), the conversion formula is:

[0043] (1)

[0044] Where X, Y, and Z are CIE standard colorimetric values, and the reference white point is taken from a D65 illuminant (X). n =95.047, Y n =100, Z n =108.883);

[0045] 1.2 XYZ to Lab space conversion

[0046] 1) Luminance channel (range [0,100]):

[0047] (2)

[0048] 2) Red-Green Lane (range [-128, 127]):

[0049] (3)

[0050] 3) Blue-Yellow Channel (range [-128, 127]):

[0051] (4)

[0052] The nonlinear transformation function f(t) is:

[0053] (5)

[0054] 1.3 Chroma Channel Normalization

[0055] For ease of subsequent processing, a* and b* are normalized to [0, 255].

[0056] (6)

[0057] (7)

[0058] Step 2: Multi-channel Canny edge detection

[0059] For L* and the normalized chroma channel a norm b norm The three channels perform Canny edge detection independently, and the process includes Gaussian filtering, gradient calculation, non-maximum suppression, and double thresholding.

[0060] 2.1 Gaussian Filtering

[0061] The Gaussian kernel with a specified standard deviation σ is used to smooth the images of each channel and suppress noise. The Gaussian kernel formula (size (2k+1)×(2k+1), where k=2, i.e., 5×5 kernel):

[0062] (8)

[0063] Where σ = 1.2 (empirical optimal value), convolution operation formula:

[0064] (9)

[0065] 2.2 Gradient Calculation

[0066] The Sobel operator is used to compute the image in the horizontal (G) direction. x ) and vertical (G) y Gradient magnitude and direction in the direction of gradient:

[0067] Sobel convolution kernel:

[0068] (10)

[0069] (11)

[0070] Gradient magnitude G and direction θ:

[0071] (12)

[0072] (13)

[0073] 2.3 Nonmaximum suppression

[0074] Compare the gradient magnitudes of adjacent pixels along the gradient direction θ, retain only local maxima, and refine the edges:

[0075] 1) If θ≈0° (horizontal edge): Compare (x,y-1) and (x,y+1) and retain the pixel with the largest G;

[0076] 2) If θ≈45° (sloping upper edge): Compare (x-1,y+1) and (x+1,y-1), and keep the pixel with the largest G;

[0077] 3) The same applies to the other directions (90°, 135°), and the final refined edge map Inms is obtained.

[0078] 2.4 Dual Threshold Processing

[0079] Let the maximum value of the gradient be G. max Adaptively set low threshold T l and high threshold T h :

[0080] T l =0.15×G max (14)

[0081] T h =0.35×G max (15)

[0082] If G ≥ T h Strong edge (reserved, denoted as 1);

[0083] If T l < G < T h Weak edge (only the part connected to the strong edge is retained, denoted as 1);

[0084] If G ≤ T l Noise (removed, recorded as 0);

[0085] Finally, a single-channel binary edge map E is obtained. channel ,like Figure 1 .

[0086] Step 3, Edge Blending

[0087] By using edge maps from three channels—weighted calculation, weighted fusion, and logical OR integration—both brightness and color edges are preserved.

[0088] 3.1 Channel Weight Calculation

[0089] Weights are assigned based on the average gradient magnitude of each channel (reflecting the overall edge strength of that channel), assuming... Let the average gradient magnitude of each channel be denoted by the weighting formula: w L* =0.2、 w a* =0.6、 w b*=0.2; The weight allocation principle is that the channel with higher edge strength has a larger proportion in the fusion; (Note: This is an empirical optimization value for the spherical scene captured in this shooting, and can be adjusted according to the scene)

[0090] 3.2 Weighted Fusion

[0091] Overlay the edge maps of the three channels according to their weights:

[0092] (16)

[0093] like Figure 2 E is a single-channel binary edge map. channel Comparison with the image after weighted fusion processing;

[0094] 3.3 Binarization and Logical OR Optimization

[0095] The weighted result is binarized (threshold set to 1 / 4), and a logical OR is performed with the edge maps of each channel to ensure that no key edges are lost.

[0096] (17)

[0097] (18)

[0098] Step 4: Morphological optimization

[0099] Edge continuity and purity are optimized through expansion (filling gaps) and corrosion (removing noise).

[0100] 4.1 Morphological expansion

[0101] The edge image is expanded using a circular structuring element S (radius 2), with the following formula:

[0102] (19)

[0103] (The number of iterations is 2, that is, the expansion operation is performed twice);

[0104] 4.2 Morphological Corrosion

[0105] The expansion result is eroded using the same structural element S to eliminate burrs that may be introduced by the expansion and restore the original approximate shape of the edges. The formula is as follows:

[0106] (20)

[0107] (The number of iterations is 1, that is, the erosion operation is performed once).

[0108] like Figure 3 , is the weighted fusion result of graph E weightedAfter binarization, logistic OR optimization, and morphological dilation, the final continuous, smooth, and complete binary edge map E is obtained. final .

[0109] Step 5: Contour extraction and localization calculation

[0110] From the final edge map E final The outer contours of the hydrophone housing and the vibrating diaphragm are extracted. By calculating feature points such as the geometric center, the center of the smallest circumcircle, or the center of the fitted ellipse of the contour, and combining the camera calibration parameters (the ratio between pixel size and actual physical size), the actual distance between the hydrophone and the diaphragm is calculated, thus completing visual positioning. By combining industrial camera imaging with the above algorithm, visually accurate positioning based on color fusion enhancement visual contour detection algorithm can be systematically achieved, significantly improving the integrity and accuracy of edge detection and visual positioning in single-color scenes with regular objects.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hydrophone visual localization algorithm based on color fusion, characterized in that, Includes the following steps: S1: Acquire color images containing the hydrophone under test and the diaphragm; S2: Convert the color image from the RGB color space to the Lab color space to obtain a luminance channel image and at least one chrominance channel image; S3: Perform edge detection on the luminance channel image and each of the chrominance channel images respectively to obtain multiple corresponding single-channel edge images; S4: Based on the edge intensity information of each channel image, assign fusion weights to the corresponding single-channel edge images and perform weighted fusion to obtain fused edge images; S5: Perform morphological processing on the fused edge image to obtain an optimized target contour image; S6: Extract the contours of the hydrophone and the vibrating diaphragm from the target contour image, and calculate the relative position between them based on the contours.

2. The hydrophone visual localization algorithm based on color fusion according to claim 1, characterized in that, In step S2, converting the color image from the RGB color space to the Lab color space specifically includes: Convert RGB images to the CIE XYZ color space; Convert the XYZ color space values ​​to the CIE Lab color space to obtain the luminance channel. L *and chroma channels a *、 b *; For the chroma channel a *、 b * Normalize the values ​​to make them suitable for image processing.

3. The hydrophone visual localization algorithm based on color fusion according to claim 1 or 2, characterized in that, In step S3, the edge detection adopts the Canny edge detection algorithm, which includes Gaussian filtering, gradient calculation, non-maximum suppression and double threshold processing steps.

4. The hydrophone visual localization algorithm based on color fusion according to claim 1, characterized in that: In step S4, the allocation of fusion weights based on the edge intensity information of each channel image specifically involves: calculating the gradient magnitude statistical characteristics of each original channel image; The fusion weight value of each channel is positively correlated with the corresponding gradient magnitude statistical characteristic value.

5. The hydrophone visual localization algorithm based on color fusion according to claim 4, characterized in that, The gradient magnitude statistical feature is the overall average gradient magnitude of the corresponding channel image, or the average gradient magnitude of a predetermined high percentage interval in the gradient magnitude histogram.

6. The hydrophone visual localization algorithm based on color fusion according to claim 1, characterized in that, In step S4, the weighted fusion specifically includes: Each single-channel edge image is multiplied by its corresponding fusion weight and then superimposed to generate a weighted fused image; The weighted fused image is subjected to threshold binarization to obtain a preliminary fusion result.

7. The hydrophone visual localization algorithm based on color fusion according to claim 6, characterized in that, Step S4, following the weighted fusion, also includes a logic optimization step: The preliminary fusion result is logically ORed with each of the single-channel edge images to generate the final fused edge image, ensuring that all significant edges detected in each channel are preserved.

8. The hydrophone visual localization algorithm based on color fusion according to claim 1, characterized in that, In step S5, the morphological processing includes a combination of dilation and erosion operations to connect edge breakpoints, remove noise, and smooth the contour.

Images