A particle dynamic angle of repose measurement method based on semantic segmentation and contour robust fitting

Abstract

This invention discloses a method for measuring the dynamic rest angle of particles based on semantic segmentation and robust contour fitting, belonging to the fields of image processing, machine vision, and particle mobility measurement technology. The method first acquires a video frame sequence of particles in motion and inputs each frame image into a semantic segmentation network to perform pixel-level segmentation of the particle accumulation region, obtaining an initial binary mask. Then, based on the regional continuity, boundary continuity, and contour position relationship between adjacent frame masks, the initial binary mask undergoes temporal stabilization processing, combined with morphological post-processing to obtain an optimized particle accumulation region. On this basis, continuous outer contours and contour point sets representing the upper edge of particle accumulation are extracted to determine the peak boundary position, dividing the contour point set into a left slope point set and a right slope point set. Sliding window linear regression is performed on the left and right slope point sets respectively to obtain multiple local tilt angles, and representative tilt angles of the left and right slopes are obtained through robust statistics. Furthermore, based on the effective extension length in the x-coordinate direction on both sides of the peak boundary position, the representative tilt angle corresponding to the side with the larger extension length is selected as the dynamic rest angle of the corresponding frame. Temporal statistical filtering or smoothing processing is then performed on the dynamic rest angle results of consecutive frames to obtain continuous output results. This invention can reduce the impact of single-frame missegmentation, boundary jitter, surface burrs and abnormal points on measurement results, and improve the continuity, stability and accuracy of particle dynamic repose angle measurement under complex working conditions.

Description

Technical Field

[0001] This invention relates to the fields of image processing, machine vision, video analysis, and particle flow measurement technology, specifically to a method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting. Background Technology

[0002] The dynamic angle of repose is one of the important parameters characterizing the flow properties of particulate materials and is widely used in pharmaceuticals, chemicals, mining, metallurgy, agriculture and other fields. During the operation of particle conveying, mixing, granulation, drying and rotary equipment, the dynamic angle of repose can reflect the packing characteristics and flow behavior of particles in motion, thus having high engineering application value.

[0003] Existing methods for measuring dynamic angles of repose mainly include manual observation, mechanical structure-assisted measurement, and image processing-based automated measurement. Manual observation relies on the operator's subjective judgment of the particle accumulation surface, resulting in low measurement efficiency, poor repeatability, and insufficient consistency of results. Mechanical structure-assisted measurement typically focuses on the structural design of the experimental setup or the arrangement of the angle acquisition mechanism; while it can improve local measurement conditions, it is difficult to apply to continuous online measurement of particles under dynamic conditions.

[0004] Image processing-based automatic measurement methods typically employ thresholding, edge detection, contour extraction, and linear fitting to obtain the slope angle of particle accumulation surfaces. However, in real-world particle motion video scenarios, particle accumulation areas often suffer from complex backgrounds, brightness fluctuations, local occlusion, particle adhesion, blurred boundaries, contour burrs, and inter-frame boundary jitter, making it difficult for traditional thresholding and edge detection methods to reliably extract accurate and effective particle accumulation boundaries. Furthermore, even if the particle accumulation contour is obtained, directly performing overall linear fitting on all contour points is easily affected by local burrs, surface undulations, and discrete outliers, resulting in significant fluctuations in the obtained repose angle and making it difficult to stably reflect the true tilt state of the overall particle accumulation slope.

[0005] In recent years, semantic segmentation technology has demonstrated strong adaptability in extracting target regions from complex image scenes, and can improve the recognition accuracy of particle accumulation regions to some extent. However, in videos with dynamic particle accumulation, relying solely on single-frame semantic segmentation results may still lead to problems such as insufficient boundary restoration, loss of local details, inter-frame mask jumps, and peak position drift. Meanwhile, the dynamic particle accumulation state typically exhibits significant temporal continuity and asymmetric characteristics of left and right slopes. Without a dedicated processing mechanism for the consistency of consecutive video frames and slope differences, it is difficult to obtain continuous, stable, and representative dynamic repose angle output results.

[0006] Therefore, how to stably extract the particle accumulation area and its upper edge contour in particle motion video scenes, and suppress the measurement deviation caused by boundary jitter, local anomalies and asymmetric slopes under continuous frame conditions, so as to realize the automatic, continuous and robust measurement of the dynamic repose angle of particles, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] The purpose of this invention is to overcome the aforementioned shortcomings of the prior art and provide a method for measuring the dynamic angle of repose of particles based on semantic segmentation and robust contour fitting, thereby improving the continuity, stability, accuracy, and robustness of dynamic angle of repose measurement in particle motion video scenes. To achieve the above objective, this invention adopts the following technical solution:

[0008] A method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting includes:

[0009] S1. Collect video frame sequences of particles in motion;

[0010] S2. Input each frame image in the video frame sequence into the semantic segmentation network, perform pixel-level segmentation on the particle accumulation region, and obtain the initial binary mask for each frame. The encoding end of the semantic segmentation network records the pooling index during downsampling, and the decoding end performs depooling based on the pooling index and fuses it with the feature map of the corresponding layer of the encoding end to restore the spatial details of the particle accumulation boundary.

[0011] S3. Based on the regional continuity, boundary continuity and contour position relationship between the initial binary masks of adjacent frames, the initial binary mask is subjected to time-stabilized processing to obtain a time-stabilized particle accumulation region mask.

[0012] S4. Perform morphological post-processing on the time-stable particle accumulation region mask to fill the region holes, connect adjacent regions and improve boundary continuity to obtain an optimized particle accumulation region.

[0013] S5. Extract a continuous outer contour based on the optimized particle accumulation region, and extract a set of contour points representing the upper edge of particle accumulation from the continuous outer contour.

[0014] S6. Determine the peak boundary position based on the set of contour points, and divide the set of contour points into a left slope point set and a right slope point set using the peak boundary position as the boundary.

[0015] S7. Perform sliding window linear regression on the left slope point set and the right slope point set respectively to obtain multiple local fitting slopes, and convert the local fitting slopes into local tilt angles relative to the horizontal plane.

[0016] S8. Robust statistics are performed on multiple local dip angles corresponding to the left slope and multiple local dip angles corresponding to the right slope to obtain the representative dip angles of the left slope and the right slope. The effective extension length of the left and right slopes in the lateral direction is determined based on the peak boundary position. The representative dip angle corresponding to the side with the larger effective extension length is determined as the dynamic rest angle of the current frame.

[0017] S9. Perform time-series statistical filtering or smoothing on the dynamic repose angle results of consecutive frames to obtain continuous output results of particle dynamic repose angles.

[0018] Furthermore, in step S1, the video frame sequence is acquired by an industrial camera, CMOS camera, area scan camera or other image acquisition device, and the acquisition object is the particle accumulation surface in motion inside a rotary drum, roller or other particle movement device.

[0019] Furthermore, in step S2, the semantic segmentation network adopts an encoder-decoder structure, with the encoding end including a multi-level convolutional downsampling structure and the decoding end including a multi-level upsampling structure corresponding to the encoding end; during each level of upsampling, unpooling is performed using the pooling index recorded during the corresponding level of downsampling, and the unpooling result is concatenated or fused with the corresponding level feature map of the encoding end.

[0020] Furthermore, in step S2, at least some convolutional layers of the decoding end of the semantic segmentation network adopt depthwise separable convolution to reduce the number of parameters and computational load, thereby improving processing efficiency in continuous video processing scenarios.

[0021] Furthermore, in step S2, the semantic segmentation network is trained using a combined loss function, which includes a Dice loss term and a Focal loss term, to reduce the impact of sample imbalance between the granular region and the background region on the segmentation result.

[0022] Furthermore, in step S3, the time-series stabilization process includes at least one of the following methods: performing overlap region consistency correction on the mask results of adjacent frames, performing smooth update on the boundary positions of adjacent frames, performing time-series smoothing on the peak boundary positions, and suppressing abnormal jumps in the dynamic rest angle results of adjacent frames.

[0023] Furthermore, in step S3, when performing overlap region consistency correction on the mask results of adjacent frames, if the overlap ratio between the current frame mask and the previous frame mask is lower than a preset threshold, or if the offset of the current frame boundary relative to the previous frame boundary exceeds a preset range, the current frame mask boundary is smoothed.

[0024] Furthermore, in step S4, the morphological post-processing includes morphological closing operations, preferably using rectangular structuring elements for dilation followed by erosion.

[0025] Further, in step S5, the extraction of the continuous outer contour includes edge extraction or contour tracking of the morphologically post-processed mask boundary to obtain the outermost continuous boundary of the particle accumulation region.

[0026] Furthermore, in step S5, the extraction of the continuous outer contour includes edge extraction of the mask boundary after morphological post-processing, and the edge extraction preferably adopts a gradient-based edge detection method.

[0027] Furthermore, in step S6, the peak boundary position is determined by the longitudinal extreme value region of the contour point set, and the lateral center position corresponding to the longitudinal extreme value region is taken as the peak boundary position, so as to reduce the influence of local anomalies on peak positioning.

[0028] Further, in step S7, the sliding window linear regression includes: sorting the contour point set according to the horizontal coordinate order; dividing the left slope point set and the right slope point set into windows according to the preset window width and sliding step size; performing linear regression on the contour points in each window to obtain the local fitting slope of the corresponding window; and converting the local fitting slope into a local tilt angle relative to the horizontal plane.

[0029] Further, in step S7, the window width is determined proportionally based on the total number of contour points, and the sliding step is less than the window width, preferably 3% to 10% of the total number of contour points on the corresponding slope, more preferably 5%; the sliding step is less than the window width, preferably 1 / 2 to 1 / 4 of the window width, more preferably 1 / 3.

[0030] Furthermore, in step S8, when performing robust statistics on multiple local dip angles corresponding to the left slope and multiple local dip angles corresponding to the right slope, at least one of the following methods is used to obtain the representative dip angle of the corresponding slope: median, truncated mean, or weighted median, with the median being preferred.

[0031] Furthermore, in step S8, based on the peak boundary position, the effective extension lengths of the left slope point set and the right slope point set in the lateral coordinate direction are determined respectively, and the representative tilt angle corresponding to the side with the larger effective extension length is used as the dynamic rest angle of the current frame, so as to enhance the ability to represent the geometric features of the dominant accumulation slope in the current frame.

[0032] Furthermore, in step S9, the temporal statistical filtering or smoothing of the dynamic rest angle results of continuous frames adopts at least one of the following methods: moving average, median filtering, weighted smoothing, or abnormal frame removal.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] This invention achieves automatic and continuous measurement of dynamic rest angles in particle motion video scenarios by collaboratively designing semantic segmentation of particle accumulation regions, temporal stability of segmentation results, boundary morphology optimization, upper edge contour extraction, peak boundary, local linear regression of slope surface, robust statistics of local tilt angles, and temporal smoothing of dynamic results.

[0035] This invention reduces the impact of single-frame missegmentation, local boundary jitter, and short-term peak position drift on subsequent contour analysis and angle calculation by performing time-series stabilization processing on the initial binary mask of continuous frames, thereby improving the continuity and consistency of the particle accumulation region boundary in the time dimension.

[0036] This invention reduces the interference of local burrs, surface undulations, and discrete outliers on the overall slope inclination estimation by performing sliding window linear regression on the left and right slopes respectively and robust statistics on local inclination angles, thereby improving the stability and robustness of dynamic repose angle measurement results.

[0037] This invention enhances the ability of measurement results to characterize the current dominant accumulation slope under dynamic asymmetric accumulation conditions by determining the representative dip angle corresponding to the dominant slope based on the lateral extension length on both sides of the peak, thereby improving the representativeness of the dynamic repose angle results.

[0038] This invention is applicable to particle dynamic measurement scenarios in rotary drums, rollers, granulation equipment, mixing equipment, and other particle movement devices, and has good engineering application value. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the process for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting, as provided in an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of the semantic segmentation network structure in an embodiment of the present invention.

[0041] Figure 3 This is a schematic diagram of the segmentation results of particle accumulation regions in consecutive video frames in an embodiment of the present invention.

[0042] Figure 4 This is a schematic diagram comparing the mask boundary effect before and after timing stabilization processing in an embodiment of the present invention.

[0043] Figure 5 This is a schematic diagram of the optimized particle packing outer contour and upper edge contour point set extraction, peak boundary and left and right slope division in an embodiment of the present invention.

[0044] Figure 6 This is a schematic diagram of obtaining the local tilt angle using sliding window linear regression in an embodiment of the present invention.

[0045] Figure 7 This is a schematic diagram of the dynamic repose angle continuous output working page in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Example 1: Overall Method Flow

[0048] like Figure 1 As shown, this embodiment provides a method for measuring the dynamic angle of repose of particles based on semantic segmentation and robust contour fitting. It is applicable to the automatic and continuous measurement of the dynamic angle of repose of particles in motion in rotary drums, rollers, granulation equipment, mixing equipment or other particle motion devices, and includes the following steps.

[0049] Step S1: Collect video frame sequences of particles in motion.

[0050] In this embodiment, an industrial camera, CMOS camera, area scan camera, or other image acquisition device can be used to continuously capture the particle movement process to obtain a video frame sequence containing the particle accumulation region and the contour of the accumulation surface. The acquired images can be grayscale images, color images, or pre-processed images. If necessary, the original image can be pre-processed by cropping the region of interest, normalizing brightness, or suppressing noise to reduce background interference.

[0051] Step S2: Input each frame image in the video frame sequence into the semantic segmentation network to obtain the initial binary mask of the particle accumulation region.

[0052] The semantic segmentation network is used to distinguish the particle accumulation region from the background region at the pixel level. This step enables the stable extraction of the main particle accumulation region under conditions of complex background, varying brightness, local particle adhesion, and blurred boundaries, providing a foundation for subsequent contour analysis.

[0053] Step S3: Perform time-stabilization processing on the initial binary mask of each frame to obtain a time-stabilized particle accumulation region mask.

[0054] In dynamic particle motion videos, there is a strong temporal correlation between consecutive frames. Even if the segmentation result of a single frame is generally correct, boundary jumps, mask jitter, or short-term peak position drift may still occur. To address this, this embodiment performs temporal stabilization correction on the current frame mask based on the overlapping area relationship between masks of adjacent frames, the continuity of boundary positions, and the relationship of peak candidate position changes, thereby reducing the impact of instantaneous missegmentation on subsequent measurements. Experimental verification in this embodiment shows that when the window width is less than 3%, the number of points within the window is too small, and linear regression is easily affected by noise, leading to severe fluctuations in local tilt angles; when the window width is greater than 10%, the window coverage is too large, failing to effectively capture local changes in the slope, thus losing the purpose of the sliding window. A step size of 1 / 3 is chosen to obtain sufficiently dense local tilt angle sampling points while ensuring computational efficiency, providing a sufficient data foundation for subsequent robust statistics.

[0055] Step S4: Perform morphological post-processing on the time-stable particle accumulation region mask to obtain the optimized particle accumulation region.

[0056] The morphological post-processing may include closing operations to fill holes in regions, connect adjacent regions, and improve boundary continuity. If necessary, boundary smoothing or local patching operations may be added based on the actual image quality to further improve the reliability of contour extraction.

[0057] Step S5: Extract the continuous outer contour based on the optimized particle accumulation region, and extract the contour point set representing the upper edge of the particle accumulation from it.

[0058] In this embodiment, edge extraction, boundary tracking, or contour search can be used to obtain the outermost continuous boundary of the particle accumulation region, and further filter to obtain a set of contour points representing the upper edge of the accumulation, which serves as the data basis for subsequent peak boundary and slope fitting.

[0059] Step S6: Determine the peak boundary position based on the contour point set, and divide the contour point set into a left slope point set and a right slope point set.

[0060] In this embodiment, the peak position is preferably determined by the longitudinal extreme value region of the contour point set, and the lateral center position corresponding to the extreme value region is used as the peak boundary position to reduce the influence of a single abnormal high point or local burr on the peak positioning.

[0061] Step S7: Perform sliding window linear regression on the point sets of the left slope and the right slope respectively to obtain multiple local fitted slopes, and convert the local fitted slopes into local tilt angles relative to the horizontal plane.

[0062] Since the upper edge of particle accumulation usually has local burrs, ripples, and discrete outliers, directly performing global linear fitting on the entire contour can easily lead to the fitting results being influenced by local outlier areas. By dividing the left and right slopes into sliding windows and performing local linear regression within the windows, the slope contour can be decomposed into multiple local linear segments, thus obtaining a more robust local dip angle distribution.

[0063] Step S8: Robust statistics are performed on multiple local tilt angles corresponding to the left slope and multiple local tilt angles corresponding to the right slope in each frame to obtain the representative tilt angle of the left slope and the representative tilt angle of the right slope in the corresponding frame. The dynamic rest angle of the corresponding frame is determined by combining the lateral extension characteristics of the left and right slopes on both sides of the peak boundary.

[0064] In this embodiment, the median is preferably used as the representative inclination angle of the left and right slopes. Furthermore, using the peak boundary as a reference, the effective extension lengths of the left and right slopes in the lateral direction are calculated separately, and the representative inclination angle corresponding to the side with the larger effective extension length is used as the dynamic rest angle of the current frame. This method is more effective in reflecting the actual tilt state of the dominant accumulation slope in the current frame when there is a significant asymmetrical distribution of the left and right slopes during dynamic accumulation.

[0065] Step S9: Perform time-series statistical filtering or smoothing on the dynamic rest angle results of consecutive frames to obtain continuous output results of dynamic rest angles.

[0066] In this embodiment, moving average, median filtering, weighted smoothing, or abnormal frame removal can be used to suppress local instantaneous jumps, thereby obtaining a smoother and more stable dynamic rest angle time series.

[0067] Example 2: Implementation of Semantic Segmentation Network

[0068] like Figure 2 As shown, in this embodiment, the semantic segmentation network adopts an encoder-decoder structure. The encoder uses a multi-level convolutional downsampling structure, recording the pooling index during max pooling; the decoder uses a multi-level upsampling structure corresponding to the encoder, first performing depooling using the corresponding pooling index during each decoding stage to recover spatial location information, and then concatenating or fusing it with the feature map of the corresponding level of the encoder to enhance the ability to recover particle stacking boundaries and local details.

[0069] In a preferred implementation, the encoder employs a VGG-based convolutional coding network, and at least some convolutional layers in the decoder use depthwise separable convolutions to reduce the number of parameters and computational load, thereby improving the processing efficiency for consecutive video frames. The output layer can use 1×1 convolutions to obtain pixel-level classification results.

[0070] To improve segmentation accuracy under imbalanced particle and background region conditions, a combined loss function consisting of Dice and Focal loss terms is preferred during model training. By combining region overlap optimization with a focus mechanism for hard-to-classify samples, the accuracy and stability of particle accumulation region extraction can be further enhanced.

[0071] Example 3: Timing Stabilization Processing Method

[0072] like Figure 3 and Figure 4 As shown, in this embodiment, temporal stabilization processing is performed on the initial binary mask of consecutive video frames to improve the consistency of consecutive frame segmentation results.

[0073] In one implementation, an overlap region analysis is performed between the current frame mask and the previous frame mask. When the overlap ratio between the current frame mask and the previous frame mask is lower than a preset threshold, or when the offset of the current frame boundary relative to the previous frame boundary exceeds a preset range, the boundary of the current frame mask is smoothed and corrected.

[0074] In another implementation, the peak boundary position detected by continuous frames is smoothed over time to avoid unstable division of the left and right slopes due to drastic changes in the peak position in a short period of time.

[0075] In another implementation, abnormal jump detection can be performed on the dynamic rest angle sequence initially calculated from consecutive frames, and abnormal frames that exceed a preset threshold can be removed, replaced, or smoothed.

[0076] Through the above-mentioned timing stabilization processing, the impact of single-frame missegmentation, local boundary jitter, short-term peak position drift, and instantaneous abnormal frames on the continuous measurement results of particle dynamic repose angle can be effectively reduced.

[0077] Example 4: Contour extraction and robust fitting of slope surfaces

[0078] like Figures 5 to 6 As shown, after obtaining a temporally stable and morphologically optimized particle accumulation region, this embodiment further extracts the continuous outer contour and upper edge contour point set, and performs robust fitting of the slope surface.

[0079] First, the upper edge contour point set is extracted based on the outermost continuous boundary of the particle accumulation region. Then, the peak boundary position is determined based on the longitudinal extreme value region, and the upper edge contour is divided into the left slope and the right slope accordingly. Next, the left and right slope point sets are divided into sliding windows according to the horizontal coordinate order. Linear regression is performed on the contour points in each window to obtain the local fitting slope of the corresponding window, and then converted into a local tilt angle.

[0080] In a preferred implementation, the window width is determined proportionally to the total number of corresponding slope contour points, preferably 3% to 10% of the total number of corresponding slope contour points, more preferably 5%; the sliding step size is set to 1 / 2 to 1 / 4 of the window width, more preferably 1 / 3. For multiple local dip angles obtained for the left and right slopes respectively, the median is preferably used as the representative dip angle of the slope to reduce the influence of local outliers, surface burrs, and short-range fluctuations on the final measurement results.

[0081] Furthermore, using the peak boundary as a reference, the effective lateral extension lengths of the left and right slopes are calculated separately, and the representative inclination angle corresponding to the side with the larger lateral extension length is determined as the dynamic repose angle of the current frame. This method is particularly suitable for dynamic measurement scenarios where there is significant asymmetric accumulation on the left and right slopes, and helps to enhance the characterization ability of the current dominant slope geometry.

[0082] Example 5: Continuous Output Scenario

[0083] like Figure 7 As shown, the method of this invention can be applied to continuous video measurement scenarios of particles in rotary drums, rollers, granulation equipment, mixing equipment, or other moving devices. For equipment operation monitoring involving particle conveying, mixing, tumbling, granulation, and drying processes in fields such as pharmaceuticals, chemicals, mining, metallurgy, and agriculture, this invention can provide continuous, automatic, and stable dynamic angle of repose output results, providing technical support for particle flowability evaluation, process optimization, and process control.

[0084] The above description is merely a preferred embodiment of the present invention and 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 method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting, characterized in that, Includes the following steps: S1. Collect video frame sequences of particles in motion; S2. Input each frame image in the video frame sequence into the semantic segmentation network, perform pixel-level segmentation on the particle accumulation region, and obtain the initial binary mask for each frame. S3. Perform time-stabilization processing on the initial binary mask to obtain a time-stabilized particle accumulation region mask. The timing stabilization process includes: calculating the overlap ratio between the current frame mask and the previous frame mask; when the overlap ratio is lower than a preset threshold, smoothing the boundary of the current frame mask according to the previous frame mask. S4. Perform morphological post-processing on the time-stable particle accumulation region mask to obtain the optimized particle accumulation region. S5. Extract the outermost continuous contour from the optimized particle accumulation region, and extract the upper edge contour point set located at the top of the accumulation region from the outermost continuous contour. S6. Determine the peak boundary position in the upper edge contour point set, and divide the upper edge contour point set into a left slope point set and a right slope point set using the peak boundary position as the boundary. S7. Perform sliding window linear regression on the left slope point set and the right slope point set respectively to obtain multiple local dip angles; perform robust statistics on the multiple local dip angles corresponding to each slope point set to obtain the representative dip angle of the slope. S8. Determine the effective extension length of the left slope point set and the right slope point set in the horizontal coordinate direction, and determine the representative tilt angle of the slope corresponding to the larger effective extension length as the dynamic rest angle of the current frame.

2. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S2, the semantic segmentation network adopts an encoder-decoder structure. The encoder records the pooling index during downsampling, and the decoder uses the pooling index to perform depooling and fuses the depooling result with the feature map of the corresponding layer of the encoder.

3. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S3, the time-series stabilization process further includes: performing time-series filtering on the peak candidate positions extracted from the masks of each frame.

4. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S5, the upper edge contour point set is the portion located at the top of the accumulation region and representing the upper edge of particle accumulation, extracted from the outermost continuous contour.

5. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S6, the peak boundary position is determined by identifying one or more points with the largest ordinate values ​​in the upper edge contour point set as the extreme value region, and the geometric center or lateral center of the extreme value region is taken as the peak boundary position.

6. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S7, the sliding window linear regression includes: sorting the slope point set according to the horizontal coordinate order; dividing the slope into a window of preset width and a step size smaller than the preset width; performing linear regression on the contour points in each window to obtain the local fitting slope, and then converting it into the local tilt angle.

7. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S7, when performing robust statistics on multiple local dip angles, the representative dip angle is obtained by using at least one of the following methods: median, truncated mean, or weighted median.

8. The method for measuring the dynamic repose angle of particles based on semantic segmentation and robust contour fitting according to claim 1, characterized in that, In step S8, the effective extension length is determined by the lateral distance between the peak boundary position and the outermost effective contour point of the corresponding slope.

Images