A tank multi-medium liquid interface identification method based on feature extraction

Abstract

This invention belongs to the field of multi-media liquid interface recognition technology, and relates to a method for recognizing multi-media liquid interfaces in storage tanks based on feature extraction. The invention generates a circumferentially unfolded image by collecting echo signals from ultrasonic phased array sensors distributed circumferentially on the sidewall of the storage tank. It constructs a feature vector by fusing the absolute value of the gray-level difference in the pixel height direction with the proportion of the sign consistency of the difference in the eight neighboring regions. Candidate edge points are adaptively selected based on the median distance of the feature vector distribution. Furthermore, a candidate set of interface heights is selected based on intervals with local maxima of circumferential continuity and monotonically decreasing towards both sides. Finally, a stable interface is confirmed by benchmark comparison of the morphological overlap sequence of consecutive multi-frame sequences and temporal fluctuation constraints. This method integrates three criteria: spatial distribution sparsity, circumferential continuity, and temporal stability, effectively suppressing interference from bubbles and transient disturbances, and improving the accuracy and robustness of interface recognition under multi-media coexistence conditions.

Description

Technical Field

[0001] This invention belongs to the field of multi-media liquid interface recognition technology, and relates to a method for recognizing multi-media liquid interfaces in storage tanks based on feature extraction. Background Technology

[0002] In the petrochemical and energy storage and transportation sectors, vertical storage tanks are commonly used to store various liquid media with stratified density, such as crude oil and water, or organic solvents and water. Accurately identifying the interfaces between these layers is crucial for material metering and process safety.

[0003] Currently, industrial sites mostly employ acoustic-based detection methods, analyzing changes in echo signals to determine interface locations. However, these methods still face significant limitations in practical applications.

[0004] First, most devices use a single-point or limited measurement point layout, which can only acquire echo information from local locations in the storage tank. This not only makes it difficult to effectively capture the complete distribution pattern of the interface around the tank wall, thus making it impossible to verify whether the interface forms a continuous and complete transition zone, but also makes it difficult to distinguish between the real interface and instantaneous disturbances caused by bubbles, impurities, etc., and the output results are easily affected by interference and change abruptly.

[0005] Secondly, when the acoustic properties of adjacent media are similar, the interface echo characteristics are weak. Traditional methods rely on fixed criteria for identification, which has poor adaptability and is prone to missed detections or misjudgments.

[0006] Especially in complex operating conditions where multiple media coexist, the identification of each interface is often independent, and no correlation constraint is established for the distribution characteristics between layers. When the intermediate layer is thin or its acoustic characteristics are ambiguous, its interface signal is easily masked or interfered with by adjacent strong reflective layers, which significantly reduces the ability of existing methods to stably and accurately extract all interface positions from interwoven echoes. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background technology, a method for identifying multi-media liquid interfaces in storage tanks based on feature extraction is proposed.

[0008] The objective of this invention can be achieved through the following technical solution: a method for identifying multi-media liquid interfaces in storage tanks based on feature extraction, comprising: acquiring echo signals from ultrasonic phased array sensors distributed circumferentially on the sidewall of the storage tank to generate a circumferentially unfolded image, calculating the absolute value of grayscale difference in the height direction of each pixel in the image and the proportion of consistent difference signs in its eight neighboring regions, and combining them into a feature vector.

[0009] Calculate the distribution distance between each feature vector and its local neighborhood vectors. If the distance is greater than the median neighborhood distance, mark the corresponding pixel as a candidate edge point and generate a candidate edge image with white pixels representing the edges.

[0010] Scan the candidate edge image line by line along the height direction, using the proportion of continuous white pixel segments as circumferential continuity, retaining the intervals with local maxima and monotonically decreasing to both sides, eliminating isolated abrupt change points and outputting the candidate set of interface heights.

[0011] For each candidate height in the candidate set of interface heights, the edge pixel distribution of the corresponding row is extracted in the candidate edge images of multiple consecutive frames. The morphological overlap between adjacent frames is calculated. When the overlap sequence is consistently not lower than the baseline and the fluctuation is limited, the height is confirmed as a stable interface, thus obtaining the final interface height position of each medium.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention constructs a circumferential unfolded image and extracts pixel-level feature vectors, and integrates the absolute value of gray difference in the height direction with the consistent ratio of the eight neighboring symbols, effectively characterizing the spatial extension characteristics of the interface in the circumferential direction of the tank wall. This solves the defect that single-point measurement cannot verify the circumferential continuity of the interface, and makes it possible to distinguish the continuous edges formed by the real interface from the isolated artifacts caused by local disturbances such as bubbles and impurities. This achieves the suppression of instantaneous interference in the candidate edge recognition process and improves the anti-interference ability of interface detection.

[0013] (2) This invention calculates the distribution distance between the feature vector and its local neighborhood and uses the median neighborhood distance as the adaptive discrimination criterion, thus avoiding dependence on a fixed threshold. It solves the problems of missed detection and misjudgment caused by weak echo features and insufficient adaptability of traditional fixed criteria when the acoustic characteristics of adjacent media are close. It achieves the technical effect of being able to stably identify edges even under conditions of fluctuating medium characteristics or blurred interfaces, and enhances the adaptability to complex medium systems.

[0014] (3) This invention integrates the three criteria of spatial distribution sparsity, circumferential continuity and temporal stability to form a logical connection, and uses the monotonic decay characteristic of circumferential continuity in the height direction to screen candidate interfaces, thereby indirectly establishing spatial correlation constraints between layers. This allows the intermediate layer interfaces to be effectively extracted under the dual verification of spatial distribution and temporal evolution, improving the completeness and accuracy of identification under multi-media coexistence conditions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the steps of a multi-media liquid interface identification method for storage tanks based on feature extraction, as described in this invention.

[0017] Figure 2 This is a flowchart of the method for obtaining candidate edge images in this invention.

[0018] Figure 3 This is a flowchart of the method for obtaining the candidate set of interface height in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] The following description, in conjunction with the accompanying drawings, details a specific scheme for a multi-media liquid interface identification method for storage tanks based on feature extraction, provided by the present invention.

[0022] Please see Figure 1 As shown, the present invention provides a method for identifying multi-media liquid interfaces in storage tanks based on feature extraction, including: S1, acquiring echo signals from ultrasonic phased array sensors distributed circumferentially on the sidewall of the storage tank to generate a circumferentially unfolded image, calculating the absolute value of grayscale difference in the height direction of each pixel in the image and the proportion of consistent difference signs in its eight neighborhoods, and combining them into a feature vector.

[0023] Considering that the multi-media system inside a large vertical storage tank forms horizontal stratification under the action of gravity, and that the acoustic impedance characteristics of each layer of media are different, causing ultrasonic waves to be reflected at the interface; at the same time, in order to fully obtain the distribution pattern of the interface in the circumferential direction of the tank wall and avoid the safety risks of invasive measurement, circumferentially arranged sensors are used to penetrate the tank wall for detection in a non-contact manner.

[0024] In one specific embodiment, multiple ultrasonic phased array sensors installed on the side wall of the storage tank and evenly distributed in a circumferential direction are used to synchronously acquire echo signals during the scanning cycle.

[0025] The echo signals from each sensor are independently processed by beamforming to generate a local B-scan image with the tank height as the vertical axis and the radial depth as the horizontal axis.

[0026] All local B-scan images are stitched together according to the sensor's installation angle on the tank wall to form a two-dimensional circumferential unfolded image with the tank height as the vertical axis and the circumferential unfolding angle as the horizontal axis. The pixel grayscale values ​​of this image represent the acoustic impedance characteristics of the corresponding spatial location, causing the medium interface to appear as a grayscale edge structure extending circumferentially in this image.

[0027] Since the multi-media interface appears as a region of abrupt gray-level change along the horizontal direction in a two-dimensional circumferential unfolded image, this change usually exhibits obvious gray-level step characteristics in the height direction.

[0028] However, during the operation of storage tanks, factors such as noise caused by temperature gradients, reflections from internal components, and medium mixing transition zones may exist, resulting in blurred characteristics at the interface edges. Traditional segmentation methods that rely on a single fixed criterion are difficult to effectively distinguish between the real interface and interference signals. Therefore, it is necessary to construct a composite feature that can simultaneously characterize the edge response intensity and local connectivity.

[0029] In one specific embodiment, for each pixel in the circumferentially unfolded image, the difference between its grayscale value and the grayscale value of its adjacent pixel in the image height direction is calculated to obtain a signed grayscale difference value. The absolute value of this difference is then taken to obtain the absolute grayscale difference value of the pixel in the height direction. This absolute difference value characterizes the edge response intensity of the pixel location in the height direction of the storage tank.

[0030] Simultaneously, the algebraic sign (positive or negative) of the grayscale difference values ​​in the height direction of the pixel and all pixels in its eight neighborhoods is obtained. The number of pixels in the eight neighborhoods with the same difference sign as the central pixel is counted, and this number is divided by eight to obtain the proportion of pixels with the same difference sign in the eight neighborhoods. This proportion represents the connectivity consistency of the edge direction at this location within the local region.

[0031] Finally, the absolute value of the grayscale difference of each pixel is combined with the sign of the difference in its eight neighborhoods in a consistent ratio to form a two-dimensional feature vector.

[0032] Please see Figure 2 As shown, S2 calculates the distribution distance between each feature vector and its local neighborhood vector. If the distance is greater than the median neighborhood distance, the corresponding pixel is marked as a candidate edge point, and a candidate edge image is generated with white pixels representing the edges.

[0033] Since the real interface edge is usually sparsely distributed in the feature space, while the feature vector distribution of uniform medium region or smooth transition region is relatively dense, if a fixed threshold is used for edge determination, it is easily affected by the fluctuation of the acoustic properties of the medium.

[0034] Therefore, by constructing a feature vector field from the feature vectors of all pixels, for each feature vector in the field, a local neighborhood is formed by selecting other feature vectors within a square neighborhood of a preset size, centered on the spatial position of that feature vector in the circumferentially unfolded image. In this embodiment, the preset size is set to 5×5 pixels, but the implementer can set it according to specific circumstances.

[0035] Calculate the Euclidean distance between the feature vector and each feature vector in its local neighborhood to obtain the distribution distance set, and take the median of the distribution distance set as the neighborhood median.

[0036] When the distribution distance of the feature vector is greater than the median neighborhood distance, it indicates that the feature vector has differences in the local region. Therefore, the corresponding pixel in the circumferential unfolded image is marked as a white pixel in the candidate edge image to represent the candidate edge point.

[0037] Otherwise, they are marked as black pixels to represent non-edge areas, thus forming a candidate edge image in which candidate edge points are represented by white pixels.

[0038] Please see Figure 3 As shown, S3 scans the candidate edge image line by line along the height direction, using the proportion of continuous white pixel segments as circumferential continuity, retaining the intervals with local maxima that monotonically decrease to both sides, eliminating isolated abrupt change points and outputting the candidate set of interface height.

[0039] Considering that the actual multi-media interface inside the storage tank is physically a continuous separating surface, in the candidate edge image, the white pixels, i.e., candidate edge points, corresponding to the height of the actual interface, should cluster in the circumferential direction (i.e., the image row direction) to form a long continuous pixel segment. Conversely, false edge points generated by noise, bubbles, or transient interference are often sparsely distributed, broken, or scattered, with low circumferential continuity.

[0040] Therefore, by quantifying and analyzing the circumferential continuity of the upper edge points of each height row, and identifying the variation pattern of this continuity in the height direction, it is possible to distinguish between the continuous region where the real interface is located and the noise abrupt change point.

[0041] In one specific embodiment, by scanning the candidate edge image line by line along the height direction, for each image line corresponding to a height position, a pixel segment consisting of consecutive white pixels in that line is identified.

[0042] When there are white pixels in the row, the ratio of the number of pixels in the longest continuous white pixel segment to the total number of pixels in the row is calculated as the circumferential continuity at that height position. This continuity represents the spatial extension integrity of the interface edge in the direction of the tank wall. The higher the circumferential continuity, the more continuous and complete the edge distribution in the direction of the tank wall at that height position is, which may correspond to the real interface. The lower the circumferential continuity, the more broken and sparse the edge distribution is, which may be caused by local noise or transient disturbances.

[0043] If there are no white pixels in a certain height row, set the circumferential continuity at that position to zero.

[0044] Arrange the circumferential continuity of each height position in height order to form a continuous sequence. Traverse this sequence and identify the height positions whose circumferential continuity is greater than the circumferential continuity of their directly adjacent previous and next height positions as candidate interface positions.

[0045] Centered on the candidate interface position, extend upwards and downwards along the height direction to check the circumferential continuity of adjacent height positions: when the circumferential continuity of adjacent height positions in the extended direction is less than the circumferential continuity of the previous position, and the algebraic signs of the continuity difference between adjacent positions are consistent, i.e., both are negative, it is determined that the adjacent height position meets the monotonically decreasing condition and is included in the retention interval.

[0046] The extension in that direction is terminated when the circumferential continuity at the effective measurement height boundary of the storage tank, or when the circumferential continuity at adjacent height locations does not satisfy the monotonically decreasing condition.

[0047] The retention interval is composed of the candidate interface position and all continuous height positions that satisfy the above monotonically decreasing condition, which are obtained by extending the candidate interface position upwards and downwards.

[0048] For positions in a continuous sequence where the circumferential continuity is less than that of the adjacent height positions on both sides, or where the sign of the continuity difference between the position and its adjacent positions is reversed, these are identified as isolated mutation points and removed.

[0049] Finally, output all height positions contained within the reserved intervals to form a candidate set of interface heights.

[0050] S4. For each candidate height in the candidate set of interface height, extract the edge pixel distribution of the corresponding row in the candidate edge images of multiple consecutive frames, calculate the morphological overlap between adjacent frames, and when the overlap sequence is consistently not lower than the baseline and the fluctuation is limited, confirm that the height is a stable interface, thereby obtaining the final interface height position of each medium.

[0051] Considering that the physical properties of the actual medium interface inside the storage tank are relatively stable, while the morphology and position of pseudo-edges generated by transient disturbances become random and unstable in consecutive multi-frame images, the spatial characteristics of a single frame image alone cannot completely distinguish this type of dynamic disturbance.

[0052] Therefore, by extracting the edge pixel distribution of candidate heights from multiple consecutive frames of images, calculating the morphological overlap sequence, and combining the baseline value and fluctuation threshold to screen stable interfaces, the final interface height position of each medium is obtained.

[0053] In one specific embodiment, firstly, for any candidate height position in the interface height candidate set, the image row corresponding to the height is located in each frame of candidate edge image, and the binarized state of all pixels in the height row is extracted (white represents candidate edge points, black represents non-edge points), forming a one-dimensional binary distribution sequence as the edge pixel distribution of the candidate height in the current frame.

[0054] Next, obtain the edge pixel distribution at the same candidate height position in the current frame and the previous frame, and count the number of points that are white pixels at the same circumferential position in the two distribution sequences as the number of overlapping pixels.

[0055] Dividing the number of overlapping pixels by the total number of white pixels in the frame with more white pixels yields the normalized morphological overlap between adjacent frames. Here, morphological overlap represents the temporal similarity of the spatial distribution patterns of edges at the same height. A higher morphological overlap indicates a more consistent edge distribution between adjacent frames, potentially corresponding to a stable, real interface; a lower morphological overlap indicates more drastic changes in edge distribution, possibly caused by transient disturbances.

[0056] It should be noted that when the total number of white pixels in the candidate height rows of two frames is zero, the height position is not included in the shape overlap calculation and is directly excluded from the interface height candidate set.

[0057] For each candidate height position, obtain the morphological overlap degree between all adjacent frames calculated within 10 consecutive frames to form a morphological overlap degree sequence, and calculate the mean of this sequence as a reference benchmark for evaluating its stability.

[0058] When the morphological overlap of at least three consecutive frames in the sequence is greater than or equal to the reference benchmark, it indicates that the edge distribution at that height position maintains a high degree of similarity, which is consistent with the temporal persistence of a stable interface.

[0059] Meanwhile, when the absolute value of the difference in morphological overlap between adjacent frames does not exceed the median of the absolute values ​​of all adjacent differences in the sequence, it indicates that the overlap sequence fluctuates gently and the edge distribution pattern remains stable in time. This eliminates drastic jumps caused by operating condition disturbances and confirms that the candidate height position is a stable interface.

[0060] Finally, all candidate height positions that are confirmed as stable interfaces are output as the final interface height positions for each medium.

[0061] In summary, this invention first acquires circumferentially distributed ultrasonic echo signals and stitches them together to generate a circumferentially unfolded image. Then, it calculates the absolute value of the gray-level difference in the height direction of each pixel and the proportion of consistent difference signs in its eight neighborhoods, combining them into a feature vector. Next, it calculates the distribution distance between each feature vector and its local neighborhood vectors, and generates a candidate edge image by comparing it with the median neighborhood distance.

[0062] Then, the image is scanned line by line along the height direction. The proportion of continuous white pixel segments is used as the circumferential continuity. Intervals with local maxima and adjacent monotonically changing values ​​are filtered out. Isolated abrupt change points are removed, and a candidate set of interface heights is output. Finally, for each candidate height in the set, the edge pixel distribution of the corresponding row in consecutive frames is extracted, and the morphological overlap between adjacent frames is calculated. When the overlap sequence is consistently not lower than the reference benchmark based on the median and the fluctuation is limited, it is confirmed as a stable interface, and the final interface height position of each medium is output.

[0063] This method integrates three criteria: spatial distribution sparsity, circumferential continuity, and temporal stability, effectively suppressing local disturbances and achieving reliable identification of multi-media interfaces.

[0064] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0065] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0066] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0068] Finally, 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying multi-media liquid interfaces in storage tanks based on feature extraction, characterized in that, include: The echo signals from the ultrasonic phased array sensors distributed circumferentially on the sidewall of the storage tank are collected to generate a circumferentially unfolded image. The absolute value of the gray-level difference in the height direction of each pixel in the image and the proportion of the difference sign consistency in its eight neighborhoods are calculated and combined into a feature vector. Calculate the distribution distance between each feature vector and its local neighborhood vectors. If the distance is greater than the median neighborhood distance, mark the corresponding pixel as a candidate edge point and generate a candidate edge image with white pixels representing the edges. Scan the candidate edge image line by line along the height direction, take the proportion of continuous white pixel segments as the circumferential continuity, retain the intervals with local maxima and monotonically decaying to both sides, remove isolated abrupt change points and output the interface height candidate set; For each candidate height in the candidate set of interface heights, the edge pixel distribution of the corresponding row is extracted in the candidate edge images of multiple consecutive frames. The morphological overlap between adjacent frames is calculated. When the overlap sequence is consistently not lower than the baseline and the fluctuation is limited, the height is confirmed as a stable interface, thus obtaining the final interface height position of each medium.

2. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, The method for obtaining the circumferentially unfolded image is as follows: The echo signals output by each ultrasonic phased array sensor that is uniformly distributed circumferentially on the side wall of the tank are acquired within a single scan cycle. Beamforming is performed on each echo signal to generate a local B-scan image with the tank height as the vertical axis and the radial depth as the horizontal axis. The local B-scan images are stitched together sequentially along the circumferential installation angle of the sensor to form a circumferential unfolded image with the tank height as the vertical axis and the circumferential unfolding angle as the horizontal axis.

3. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, Specifically, the absolute value of the gray-level difference in the height direction of each pixel in the image is calculated as follows: For each pixel in the circumferentially unfolded image, calculate the difference between its gray value and the gray value of its adjacent previous pixel in the image height direction to obtain the signed gray-level difference value. Take the absolute value of the grayscale difference as the absolute value of the grayscale difference in the height direction.

4. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 3, characterized in that, The calculation of the sign consistency ratio within its eight neighborhoods is specifically as follows: Obtain the algebraic sign of the grayscale difference value of the pixel in the height direction; Obtain the algebraic sign of the grayscale difference value of each pixel in the height direction within the eight neighborhoods of the given pixel; Count the number of pixels in the eight-neighborhood that have the same difference sign as the pixel, and divide the number by eight to get the ratio value.

5. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, The method for obtaining the candidate edge image is as follows: Obtain an eigenvector field composed of multiple eigenvectors; For each feature vector in the feature vector field, other feature vectors are selected within the spatial neighborhood of the feature vector centered on its spatial position in the circumferential unfolded image to form a local neighborhood. Calculate the Euclidean distance between the feature vector and each feature vector in the local neighborhood to obtain the distribution distance set, and take the median of the distribution distance set as the neighborhood median; When the distribution distance of the feature vector is greater than the median neighborhood distance, the corresponding pixel in the circumferentially unfolded image is marked as a white pixel in the candidate edge image; otherwise, it is marked as a black pixel, thus forming a candidate edge image in which white pixels represent candidate edge points.

6. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, The method for obtaining the circumferential continuity is as follows: Scan line by line along the height direction of the candidate edge image, and for each image line corresponding to a height position, identify the pixel segment consisting of consecutive white pixels in that line; When there are white pixels in the row, the ratio of the number of pixels in the longest consecutive white pixel segment to the total number of pixels in the row is calculated as the circumferential continuity at that height position. If there are no white pixels in the row, set the circumferential continuity at that height position to zero.

7. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, The method for obtaining the candidate set of interface heights is as follows: Arrange the circumferential continuity of each height position in order of height to form a continuous sequence; Traverse the sequence and identify the height positions where the circumferential continuity is a local maximum as candidate interface positions; Centered on the candidate interface location, extend to both sides along the height direction. When the circumferential continuity of adjacent height locations along the extension direction is monotonically decreasing and the sign of the continuity difference remains consistent, the height location that meets the condition is included in the retention interval. Isolated abrupt changes where the circumferential continuity is less than that of the adjacent positions on both sides or where the sign of the continuity difference is reversed are removed, and the height positions within the retained interval are output to form a candidate set of interface heights.

8. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, The method for obtaining the edge pixel distribution is as follows: For any candidate height position in the candidate height set, locate the height row corresponding to that candidate height position in the candidate edge image; Extract the binary state of all pixels in the height row to form a one-dimensional distribution sequence consisting of white and black pixels. Use this distribution sequence as the edge pixel distribution at the candidate height position.

9. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 1, characterized in that, The method for calculating the morphological overlap between adjacent frames is as follows: Obtain the edge pixel distribution at the same candidate height position in the current frame and the previous frame; The number of locations where both frames contain white pixels is counted, and this number is taken as the number of overlapping pixels. Divide the number of overlapping pixels by the total number of white pixels in the frame with more white pixels to obtain the morphological overlap between adjacent frames.

10. The method for identifying multi-media liquid interfaces in storage tanks based on feature extraction as described in claim 9, characterized in that, The method for obtaining the final interface height position of each medium is as follows: For each candidate height position, obtain its corresponding morphological overlap sequence in multiple consecutive candidate edge images; The mean of the sequence is calculated as a reference benchmark. When the morphological overlap of at least three consecutive frames in the sequence is greater than or equal to the reference benchmark, and the absolute value of the difference in morphological overlap between adjacent frames does not exceed the median of the absolute values ​​of all adjacent differences in the sequence, the candidate height position is confirmed as a stable interface. Output all candidate height positions that are confirmed as stable interfaces, as the final interface height positions for each medium.

Images