A monitoring method for the safe operation of an agricultural product transportation conveyor belt

Abstract

The present invention relates to the technical field of conveyor belt monitoring, and specifically relates to a method for monitoring the safe operation of an agricultural product transportation conveyor belt; obtaining a grayscale image corresponding to the conveyor belt; respectively performing filtering processing on the grayscale image in multiple directions by using filtering kernels of different scales to obtain multiple feature extraction maps, and then fusing them to obtain a saliency map; extracting suspected crack regions in the saliency map, calculating gradient distribution features, random indexes and intensity response values corresponding to each suspected crack region, and further obtaining evaluation indexes, and obtaining real cracks based on the evaluation indexes; constructing a monitoring matrix according to data information, obtaining mutation points of each row of data information, and calculating an abnormality evaluation value when the number of mutation points of any row of data information is greater than a set number; calculating an operation state index according to the evaluation indexes corresponding to each real crack and the abnormality evaluation value, and monitoring the conveyor belt based on the operation state index. The present invention can accurately achieve the purpose of monitoring the conveyor belt.

Description

Technical Field

[0001] The present invention relates to the technical field of conveyor belt monitoring, and particularly relates to a method for monitoring the safe operation of an agricultural product transportation conveyor belt. Background Art

[0002] Conveyor belts for transportation are widely used in multiple industries such as coal mines, chemical industries, agriculture, and metallurgy. Their main advantages include large conveying capacity, adaptability to long transportation distances, economy, safety, and high reliability. Especially in the process of agricultural production and transportation, the conveyor belt is one of the main transportation equipment for agricultural products. During the operation of the conveyor belt, there are accidental situations such as metal objects getting stuck, drums and rollers being jammed, and gangue scratching, which may lead to accidents such as longitudinal tearing of the conveyor belt and large-amplitude vibration of the conveyor belt, affecting the stability of the conveyor belt operation. The belt of the conveyor belt may experience wear and aging, resulting in breakage. Once the conveyor belt breaks and is not repaired and replaced in time, it will seriously affect the agricultural product transportation process and pose unpredictable risks. Therefore, how to detect cracks in the conveyor belt and how to accurately monitor the operation process of the conveyor belt are particularly important.

[0003] In the prior art, the method for detecting cracks in the conveyor belt is as follows: By obtaining the surface image of the conveyor belt and using image processing technology to obtain the cracks in the conveyor belt, that is, performing threshold segmentation on the image to obtain the cracks, so as to monitor the operation process of the conveyor belt. This method has high requirements for the threshold. Once the accuracy of the threshold selection is insufficient, the cracks in the conveyor belt cannot be detected in time, and accurate monitoring of the conveyor belt cannot be achieved. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method for monitoring the safe operation of an agricultural product transportation conveyor belt, and the specific technical solution adopted is as follows:

[0005] Obtain the surface image of the conveyor belt, preprocess the surface image of the conveyor belt to obtain the corresponding grayscale image of the conveyor belt; perform filtering processing on the grayscale image in multiple directions using filtering kernels of different scales to obtain multiple feature extraction maps; fuse the multiple feature extraction maps to obtain a saliency map;

[0006] Extract the straight-line segment areas in the saliency map and record them as suspected crack areas, and obtain the gradient distribution characteristics corresponding to each suspected crack area according to the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel point in the suspected crack area;

[0007] According to the width and length of the minimum circumscribed rectangle corresponding to each suspected crack area and the distance from all its suspected crack pixel points to the edge line of the conveyor belt, obtain the random index corresponding to each suspected crack area;

[0008] The grayscale mean value of all suspected crack pixels corresponding to each suspected crack area is recorded as the intensity response value. Based on the gradient distribution characteristics, random index, and intensity response value corresponding to each suspected crack area, an evaluation index corresponding to each suspected crack area is obtained, and the suspected crack area with the evaluation index greater than the set threshold is recorded as a real crack;

[0009] Collect data information corresponding to different monitoring parameters during the operation of the conveyor belt, construct a monitoring matrix based on the data information, and obtain the mutation points corresponding to each row of data information. When the number of mutation points corresponding to any row of data information is greater than the set number, then according to the maximum value, minimum value, and variance corresponding to this row of data information, calculate the abnormal evaluation value corresponding to this row of data information;

[0010] According to the evaluation index corresponding to each real crack and the abnormal evaluation value, calculate the operation status index of the conveyor belt, and monitor the conveyor belt based on the operation status index.

[0011] Preferably, the method for fusing multiple feature extraction maps to obtain a saliency map is as follows: First, perform a position alignment operation on the feature extraction maps corresponding to different directions at the same scale. According to the pixel values of the pixel points corresponding to the same position, calculate the pixel mean value corresponding to this position to obtain the fused feature extraction map corresponding to this scale; Then, perform saliency processing on the fused feature maps corresponding to different scales respectively to obtain the fused saliency maps corresponding to different scales. Furthermore, assign weights to each fused saliency map, and perform weighted processing on all fused saliency maps according to the weights to obtain the saliency map.

[0012] Preferably, the method for obtaining the gradient distribution characteristics is: Denote the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel point as gradient information. The gradient information corresponding to all suspected crack pixel points corresponding to each suspected crack area constitutes the gradient information set corresponding to this suspected crack area. Denote the gradient information with the same value as the same kind of gradient information, calculate the ratio of the frequency of occurrence of each kind of gradient information in the gradient information set to the number of all gradient information in the gradient information set, and determine the gradient distribution characteristics corresponding to each suspected crack area according to the ratio and the variance value of all gradient direction angles in the gradient information set.

[0013] Preferably, the method for obtaining the random index is:

[0014] Calculate the sum of the distances from all suspected crack pixel points in each suspected crack area to the edge line of the conveyor belt, calculate the ratio of the width to the length of the minimum bounding rectangle, and determine the random index according to the sum and the ratio.

[0015] Preferably, the evaluation index is:

[0016]

[0017] Among them, P k is the evaluation index corresponding to the suspected crack area k, Ran k is the random index corresponding to the suspected crack area k, is the strength response value corresponding to the suspected crack area k; Gray k ' is the value after normalization of the gradient distribution characteristics corresponding to the suspected crack area k, θ k is the included angle formed by the straight line corresponding to the suspected crack area k and the edge line of the conveyor belt; e is the natural constant.

[0018] Preferably, the method for obtaining the mutation points corresponding to each line of data information is: obtaining the edge values corresponding to all data information in the monitoring matrix through a horizontal edge detection operator, and recording the data information with an edge value of 1 as the mutation points, so as to obtain the mutation points corresponding to each line of data information.

[0019] Preferably, the abnormal evaluation value is:

[0020]

[0021] Among them, z c is the abnormal evaluation value corresponding to the c-th line of data information, h c,max is the maximum value corresponding to the c-th line of data information, h c,min is the minimum value corresponding to the c-th line of data information, σ c is the variance corresponding to the c-th line of data information; exp(·) is the exponential function with the natural constant e as the base.

[0022] Preferably, the operation status index is the sum of the evaluation indexes corresponding to all real cracks and the sum of all abnormal evaluation values.

[0023] The embodiments of the present invention have at least the following beneficial effects:

[0024] The present invention uses filter kernels of different scales to filter grayscale images in multiple directions to obtain multiple feature extraction images, which are then fused to obtain a saliency map; the suspected crack area in the saliency map is extracted, the gradient distribution characteristics, random indicators and intensity response values ​​corresponding to each suspected crack area are calculated, and then the evaluation index is obtained, and the real crack is obtained based on the evaluation index. The saliency map can clearly reflect the texture information in the grayscale image, and then accurately obtain the texture of the conveyor belt surface. In the subsequent operation process, the detection accuracy of the real crack is improved, which is convenient for identifying the crack area. The calculation of the evaluation index combines the gradient distribution characteristics, random indicators and intensity response values, takes into account various factors, and can accurately obtain the real crack. At the same time, the present invention also constructs a monitoring matrix, calculates the abnormal evaluation value corresponding to each row of data information, calculates the operating status index according to the evaluation index and the abnormal evaluation value corresponding to each real crack, and monitors the conveyor belt based on the operating status index. The present invention monitors the conveyor belt not only by considering the cracks on the surface of the conveyor belt, but also by considering the data information corresponding to the monitoring parameters of the conveyor belt. It can accurately achieve the purpose of automatic monitoring of the conveyor belt and improve the monitoring accuracy. In summary, the present invention has the advantages of high monitoring accuracy, low system cost and automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 The present invention is a flowchart of the steps of an embodiment of a method for monitoring the safe operation of a conveyor belt for transporting agricultural products. DETAILED DESCRIPTION

[0027] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the scheme proposed according to the present invention, its specific implementation, structure, features and effects are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0028] Unless defined otherwise, 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 belongs.

[0029] See also Figure 1, which shows a step flowchart of a method for monitoring the safe operation of an agricultural product transportation conveyor belt provided by an embodiment of the present invention. The method includes the following steps:

[0030] Step 1, obtain the conveyor belt surface image, preprocess the conveyor belt surface image to obtain the grayscale image corresponding to the conveyor belt; perform filtering processing on the grayscale image in multiple directions using filtering kernels of different scales to obtain multiple feature extraction maps; fuse the multiple feature extraction maps to obtain a saliency map.

[0031] Specifically, first, use a camera to collect images of the conveyor belt to obtain the conveyor belt surface image for detecting the surface condition of the conveyor belt. During the process of collecting conveyor belt images by the camera, it is necessary to ensure that the conveyor belt surface image can be completely collected. Therefore, set the camera at the starting end of the conveyor belt and collect the conveyor belt surface image in real time during the operation of the conveyor belt. The specific position deployment of the camera and the setting of camera parameters, etc., can be set by the implementer according to the actual situation; at the same time, the time interval between two adjacent detection moments for collecting the conveyor belt surface image by the camera and the time interval between two adjacent time periods should be consistent with the time interval between two adjacent detection moments and the time interval between two adjacent time periods corresponding to the data information of different monitoring parameters collected by the conveyor belt during operation. In this embodiment, the continuous multi-frame conveyor belt surface images collected at multiple detection moments in each time period are stitched and fused to obtain the stitched image corresponding to each time period. The method of stitching and fusing is a well-known technology, and the implementer can select the corresponding method according to the specific situation.

[0032] Then analyze the obtained stitched image to analyze the crack problem on the conveyor belt surface, identify and extract the crack area to characterize the abnormal condition on the conveyor belt surface. For the collected conveyor belt images, considering that other irrelevant areas will interfere with the identification of the crack area and there is a certain correlation between the subsequent analysis of the crack area distribution and the two-edge position information of the conveyor belt, therefore, first extract the target area of the stitched image, that is, obtain the grayscale image of the conveyor belt. Specifically, perform grayscale processing on the stitched image to obtain the grayscale image corresponding to the stitched image, and identify the conveyor belt area through the trained semantic segmentation network, and then obtain the grayscale image corresponding to the conveyor belt. Manually set the pixel points in the conveyor belt area to 1 and set the grayscale values of other pixel points to 0 to obtain a label image. Use a large number of label images and the grayscale image corresponding to the stitched image to train the semantic segmentation network to obtain the trained semantic segmentation network. The specific semantic segmentation process of the semantic segmentation network is a well-known technology in the prior art and is not within the protection scope of the present invention and will not be elaborated.

[0033] Further, to avoid problems such as blurred and discontinuous edge extraction during semantic segmentation, in this embodiment, based on the semantic segmentation effect diagram obtained through the semantic segmentation network (the semantic segmentation effect diagram is a binary image, and the pixel points with a pixel value of 1 in the semantic segmentation effect diagram are the pixel points corresponding to the conveyor belt), the set of pixel points corresponding to the upper and lower edges of the conveyor belt is obtained and denoted as U u and U d , and two straight lines are respectively fitted through the pixel position information of the pixel points in their respective pixel point sets to obtain the upper edge line Y u and the lower edge line Y d . In this embodiment, the RANSAC algorithm is used for fitting. Furthermore, the pixel values of the pixel points included in the upper edge line Y u and the lower edge line Y d in the semantic segmentation effect diagram are set to 1 to complete the correction of the semantic segmentation effect diagram. The corrected semantic segmentation effect diagram is used as a mask to perform a multiplication operation with the grayscale image to obtain the grayscale image corresponding to the conveyor belt. The RANSAC algorithm is a well-known technology and will not be elaborated here.

[0034] In order to be able to more quickly and accurately extract the crack area on the surface of the conveyor belt, in this embodiment, the grayscale image is filtered using filtering kernels of different scales in multiple directions to obtain multiple feature extraction maps; the multiple feature extraction maps are fused to obtain a saliency map.

[0035] Specifically, the Gabor filter is used to filter the grayscale image. One filtering kernel of a scale can obtain one feature extraction map in each direction. For example, a filtering kernel with a scale of 3×3 can obtain one feature extraction map in the 0° direction. Similarly, a filtering kernel with a scale of 3×3 can also obtain one feature extraction map in the 45° direction. In this embodiment, the grayscale image is filtered using 3 filtering kernels of different scales in 8 directions. The scales of the filtering kernels are 3×3, 5×5, and 7×7 respectively. The 8 directions correspond to the 8-neighborhood directions of the pixel points, that is, a total of 24 feature extraction maps are obtained. The implementer can select the scale of the filtering kernel, the number of filtering kernels, and the number of directions according to the actual situation during the actual operation. Using the Gabor filter to filter the grayscale image is a well-known technology and is not within the protection scope of the present invention, so it will not be elaborated here.

[0036] The method for fusing multiple feature extraction images to obtain a saliency map is specifically as follows: first, the feature extraction images corresponding to different directions at the same scale are aligned, and the pixel mean corresponding to the position is calculated according to the pixel value of the pixel point corresponding to the same position to obtain a fused feature extraction image corresponding to the scale; then, the fused feature images corresponding to different scales are saliency processed respectively to obtain fused saliency maps corresponding to different scales, and then weights are assigned to each fused saliency map, and all fused saliency maps are weighted according to the weights to obtain a saliency map.

[0037] Specifically, the Itti saliency analysis algorithm is selected to perform saliency processing on the fused feature map. The implementer may select other saliency analysis algorithms, such as the AC algorithm, the HC algorithm, etc. The saliency analysis algorithm is an existing well-known algorithm and will not be elaborated in detail. Then, when assigning weights to each fused saliency map, the same weight is assigned, that is, there are three fused saliency maps in total. In this embodiment, the weight assigned to each fused saliency map is 1 / 3. In actual operation, different weights may be set for each fused saliency map according to the different information contained in each fused saliency map.

[0038] It should be noted that by filtering the grayscale image in multiple directions using filter kernels of different scales, the texture information in the grayscale image can be extracted and the texture of the conveyor belt surface can be obtained. The significant processing can enhance the acquired texture information, improve the detection accuracy of the crack area in the subsequent operation process, and facilitate the identification of the crack area.

[0039] Step 2: extract the straight line segment area in the saliency map and record it as the suspected crack area, and obtain the gradient distribution characteristics corresponding to each suspected crack area according to the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel point in the suspected crack area.

[0040] Since most cracks are in the form of straight lines, the Hough line detection algorithm is used to extract the straight line segment area in the saliency map and record it as the suspected crack area. The Hough line detection algorithm is a well-known technology and the specific process will not be repeated here.

[0041] Considering that in the actual analysis scenario, there are still many lines on the surface of the conveyor belt due to its own uneven surface distribution and some scratches generated during the operation. These scratches do not actually cause actual harm to the conveyor belt. Scratches and lines formed by the uneven surface distribution of the conveyor belt itself are collectively referred to as false cracks. Due to the high similarity between false cracks and real cracks, it is impossible to determine which are real cracks and which are false cracks only through the Hough line detection algorithm. Therefore, further analysis is carried out on the suspected crack areas.

[0042] Specifically, according to the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel in the suspected crack area, the gradient distribution characteristics corresponding to each suspected crack area are obtained.

[0043] The method for obtaining the gradient distribution characteristics is as follows: Denote the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel as gradient information i = 1, 2, ……, N k , where is the gradient amplitude corresponding to the i-th suspected crack pixel in the suspected crack area k, is the gradient direction angle corresponding to the i-th suspected crack pixel in the suspected crack area k, and N k is the total number of suspected crack pixels in the suspected crack area k. The method for obtaining the gradient amplitude and gradient direction angle corresponding to the suspected crack pixel is a well-known technology; the gradient information corresponding to all the suspected crack pixels in each suspected crack area constitutes the gradient information set corresponding to the suspected crack area. Denote the gradient information with the same value as the same kind of gradient information, calculate the ratio of the frequency of each kind of gradient information appearing in the gradient information set to the number of all gradient information in the gradient information set, and determine the gradient distribution characteristics corresponding to each suspected crack area according to the ratio and the variance value of all gradient direction angles in the gradient information set; the formula expression of the gradient distribution characteristics is:

[0044]

[0045] where Gray k is the gradient distribution characteristic corresponding to the suspected crack area k, is the variance value of all gradient direction angles in the suspected crack area k, w v is the ratio of the frequency of the v-th kind of gradient information appearing in the gradient information set to the number of all gradient information in the gradient information set, V k is the number of types of gradient information; e is the natural constant; lg w v is the logarithm with the natural constant 10 as the base of w v .

[0046] Furthermore, perform normalization processing on the gradient distribution characteristics to ensure that the values of the gradient distribution characteristics are in the interval (0, 1), which is convenient for calculating the evaluation indexes of each suspected crack area later.

[0047] Considering that when there is a real crack on the conveyor belt surface, the gradient information of the crack pixels corresponding to the cracks in the same range remains consistent, that is, the number of types of gradient information in the corresponding gradient information set is not large, and the ratio of the frequency of each kind of gradient information appearing in the gradient information set to the number of all gradient information in the gradient information set should be large. At the same time, It can characterize the fluctuation of all suspected crack pixels in the suspected crack area k. The greater the fluctuation, the less likely it is to be a real crack. Based on this, the gradient distribution characteristics corresponding to each suspected crack area are calculated as a reference factor for subsequent judgment of whether each suspected crack area is a real crack. When the gradient distribution characteristics are greater, it is considered that the gradient distribution in the suspected crack area is more complex, that is, the gradient information is more inconsistent, and the possibility that the suspected crack area is a real crack is smaller.

[0048] Step 3: According to the width and length of the minimum circumscribed rectangle corresponding to each suspected crack area and the distances from all suspected crack pixels to the conveyor belt edge line, obtain the random index corresponding to each suspected crack area.

[0049] Since the growth of cracks is random, scratches are mostly one-time defects caused acutely, and the growth of crack morphologies on the conveyor belt surface is relatively random. Therefore, by constructing a random index, it is used to analyze the defect distribution in the suspected crack area.

[0050] The method for obtaining the random index is as follows: Calculate the sum of the distances from all suspected crack pixels in each suspected crack area to the conveyor belt edge line, calculate the ratio of the width and length of the minimum circumscribed rectangle, and determine the random index according to the sum and the ratio. The formula expression of the random index is:

[0051]

[0052] Among them, Ran k is the random index corresponding to the suspected crack area k, w k is the width of the minimum circumscribed rectangle corresponding to the suspected crack area k; h k is the length of the minimum circumscribed rectangle corresponding to the suspected crack area k; d i,k is the distance from the i-th suspected crack pixel in the suspected crack area k to the conveyor belt edge line, and N k is the total number of suspected crack pixels in the suspected crack area k.

[0053] is the ratio of the width and length of the minimum circumscribed rectangle corresponding to the suspected crack area k. The ratio can characterize the morphology corresponding to the suspected crack area. The smaller the ratio, the greater the possibility that the morphology of the suspected crack area is strip-shaped. According to the characteristic that most real cracks are strip-shaped, the possibility that the corresponding suspected crack area is a real crack is greater; d i,kCharacterize the distance from the suspected crack pixel points to the conveyor belt edge line, where the conveyor belt edge line includes the upper edge line and the lower edge line. When calculating, uniformly select the upper edge line or the lower edge line of the conveyor belt as the conveyor belt edge line. The greater the distance, the more the suspected crack area is distributed in the middle position of the conveyor belt, and the corresponding random index is greater. The greater the random index, the greater the possibility that the corresponding suspected crack area is a real crack.

[0054] Step 4: Denote the gray mean value of all suspected crack pixel points corresponding to each suspected crack area as the intensity response value. Based on the gradient distribution characteristics, random index, and intensity response value corresponding to each suspected crack area, obtain the evaluation index corresponding to each suspected crack area, and denote the suspected crack area with the evaluation index greater than the set threshold as a real crack.

[0055] Since when there is a crack on the conveyor belt surface, the crack has a certain depth, and the gray value at the crack is relatively low compared to defects such as scratches during image acquisition. Therefore, calculate the gray mean value of all suspected crack pixel points corresponding to each suspected crack area and denote it as the intensity response value. The smaller the intensity response value, the greater the possibility that the corresponding suspected crack area is a real crack.

[0056] Then, based on the gradient distribution characteristics, random index, and intensity response value corresponding to each suspected crack area, obtain the evaluation index corresponding to each suspected crack area.

[0057] The evaluation index is:

[0058]

[0059] where P k is the evaluation index corresponding to the suspected crack area k, Ran k is the random index corresponding to the suspected crack area k, is the intensity response value corresponding to the suspected crack area k; Gray k ' is the value after normalizing the gradient distribution characteristics corresponding to the suspected crack area k, θ k is the angle formed by the straight line corresponding to the suspected crack area k and the conveyor belt edge line; e is the natural constant.

[0060] The straight line corresponding to the suspected crack area k is obtained by fitting the suspected crack pixel points corresponding to the suspected crack area. The fitting method is a well-known technology and will not be elaborated here. Considering that the conveyor belt has a certain tensile force in the running direction during operation, therefore, the directions of most tearing cracks on the conveyor belt surface are generally perpendicular to the running direction of the conveyor belt, that is, introducing the angle θ k formed by the straight line corresponding to the suspected crack area and the conveyor belt edge line for calculating the evaluation index can obtain real cracks more accurately.

[0061] When calculating the evaluation index, not only the gradient distribution characteristics, random index and intensity response value corresponding to the suspected defect area are combined, but also the included angle formed by the straight line corresponding to the suspected crack area and the conveyor belt edge line is introduced, taking into account various factors, and the real crack can be accurately obtained. The evaluation index Ran k represents the possibility that the suspected crack area is a real crack; according to step 3, the larger the random index, the greater the possibility that the corresponding suspected crack area is a real crack, that is, the random index and the evaluation index show a positive correlation; Gray k The larger ', the more complex the gradient distribution in the suspected crack area and the more inconsistent the gradient information, the smaller the possibility that the corresponding suspected crack area is a real crack, that is, Gray k ' and the evaluation index show a negative correlation; the intensity response value The larger it is, the higher the gray value of the suspected crack area, indicating that the possibility that the corresponding suspected crack area is a real crack is lower, that is, the intensity response value and the evaluation index show a negative correlation.

[0062] So far, the evaluation indexes corresponding to each suspected crack area are obtained. Compare the size of the evaluation index and the set threshold. Mark the suspected crack area where the evaluation index is greater than the set threshold as a real crack, and mark the suspected crack area where the evaluation index is less than or equal to the set threshold as a false crack. The false crack does not participate in the monitoring of the conveyor belt state in the subsequent process.

[0063] In this embodiment, in order to obtain real cracks more quickly, the evaluation index is first normalized so that the value of the normalized evaluation index is between 0 and 1. Then compare the size of the normalized evaluation index and the set threshold. Mark the suspected crack area where the normalized evaluation index is greater than the set threshold as a real crack, and mark the suspected crack area where the normalized evaluation index is less than the set threshold as a false crack. The value of the set threshold is 0.6. In the specific implementation process, the implementer can adjust the value of the set threshold according to the actual situation.

[0064] Step 5, collect the data information corresponding to different monitoring parameters during the operation of the conveyor belt, and construct a monitoring matrix based on the data information to obtain the mutation points corresponding to each row of data information. When the number of mutation points corresponding to any row is greater than the set number, calculate the abnormal evaluation value corresponding to the data information of that row according to the maximum value, minimum value and variance of the data information of that row.

[0065] To analyze and identify abnormal conditions of invisible factors during the operation of the conveyor belt, data information corresponding to different monitoring parameters during the operation of the conveyor belt is collected in real time. The monitoring parameters include but are not limited to conveyor belt frequency, conveyor belt tension, and conveyor belt operating power, etc. The data information corresponding to different monitoring parameters is obtained by corresponding sensors or data acquisition devices. Denote the number of monitoring parameters as C, and the implementer selects the value of C according to the specific situation. Considering that most of the collected data information is continuous and not convenient for analysis, therefore, the collected data information is discretized. For each monitoring parameter, in this embodiment, the time interval between two adjacent detection times is set as T = 1s, and the time interval between two adjacent time periods is F = 10min, that is, the monitoring and identification of the conveyor belt operation status is carried out every ten minutes, and s data information corresponding to each monitoring parameter is collected every 10min. In this embodiment, s = 300. Based on the data information corresponding to each obtained monitoring parameter, a monitoring matrix is established:

[0066]

[0067] where H is the monitoring matrix, and h 11 is the data information corresponding to the first monitoring parameter at the first detection time, and h 1s is the data information corresponding to the first monitoring parameter at the s-th detection time, and h 21 is the data information corresponding to the second monitoring parameter at the first detection time, and h 2s is the data information corresponding to the second monitoring parameter at the s-th detection time, and h C1 is the data information corresponding to the C-th monitoring parameter at the first detection time, and h Cs is the data information corresponding to the C-th monitoring parameter at the s-th detection time.

[0068] Each monitoring parameter corresponds to a row of data information in the monitoring matrix. The monitoring matrix is normalized to ensure that the value of each data information is in the interval (0, 1). According to the monitoring matrix, it can be known that each row of data information represents a set of data information corresponding to the same monitoring parameter at different detection times within a time period. When the conveyor belt is operating normally, the data information corresponding to the same monitoring parameter at different detection times within a time period should be roughly the same. Based on this, the mutation points corresponding to each row of data information are obtained. Specifically, the horizontal edge detection operator Sobel 水平 is used to obtain the edge values corresponding to all data information in the monitoring matrix, and the data information with an edge value of 1 is recorded as a mutation point, and then the mutation points corresponding to each row of data information are obtained, and the number of mutation points corresponding to each row of data information is counted. Of course, not every row of data information has mutation points. If there are no mutation points in one row of data information, the number of mutation points corresponding to that row of data information is 0.

[0069] The number of mutation points can comprehensively reflect the overall abnormality of the corresponding row of data information. When the number of mutation points is extremely small, it is considered that the mutation points are isolated noise data existing in the process of data information acquisition of the monitoring parameters, that is, it is considered that the mutation points are not formed due to the abnormality of the corresponding monitoring parameters during the operation of the conveyor belt; when the number of mutation points is large, it is considered that the corresponding monitoring parameters are more likely to be abnormal. Therefore, compare the number of mutation points with the set number.

[0070] When the number of mutation points corresponding to any row of data information is greater than the set number, an abnormality evaluation value corresponding to the row of data information is calculated according to the maximum value, minimum value and variance corresponding to the row of data information. The value of the set number is 5, and the implementer can adjust the value of the set number according to the actual situation.

[0071] The abnormality evaluation value is:

[0072]

[0073] where z c is the abnormality evaluation value corresponding to the c-th row of data information, h c,max is the maximum value corresponding to the c-th row of data information, h c,min is the minimum value corresponding to the c-th row of data information, σ c is the variance corresponding to the c-th row of data information; exp(·) is the exponential function with the natural constant e as the base.

[0074] The larger the abnormality evaluation value z c , the higher the possibility that the c-th row of data information is abnormal. σ c (h c,max -h c,min ) represents the degree of fluctuation of the c-th row of data information, h c,max -h c,min represents the difference between the maximum value and the minimum value corresponding to the c-th row of data information. The larger the difference, the higher the degree of fluctuation of the c-th row of data information. The larger σ c , the higher the degree of fluctuation of the c-th row of data information. Since under the condition that the conveyor belt is operating normally, the data information corresponding to the same monitoring parameter at different detection times within a time period should be roughly the same, so the larger the abnormality evaluation value z c , the higher the possibility that the c-th row of data information is abnormal.

[0075] Step 6, calculate the operation state index of the conveyor belt according to the evaluation indexes corresponding to each real crack and the abnormality evaluation value, and monitor the conveyor belt based on the operation state index.

[0076] The operating status index is the sum of the evaluation indices corresponding to all real cracks and the sum of all abnormal evaluation values. The formula expression of the operating status index is as follows:

[0077]

[0078] Wherein, is the operating status index, P m is the evaluation index corresponding to the m-th real crack, M is the number of real cracks, z c is the abnormal evaluation value corresponding to the data information of the c-th row, and C is the number of rows in the monitoring matrix.

[0079] The operating status index characterizes the possibility of abnormalities occurring during the operation of the conveyor belt. Therefore, the larger the operating status index, the more likely the conveyor belt is to have abnormalities. The more the number of real cracks, the worse the operating status of the conveyor belt; the larger the operating status index; the larger the evaluation index corresponding to the real crack, the more obvious the crack state of the real crack is characterized, the worse the operating status of the conveyor belt, and the larger the operating status index. The larger the abnormal evaluation value, the more likely the corresponding monitoring parameter is to have an abnormality. Correspondingly, the more likely the conveyor belt is to have an abnormality, that is, the larger the operating status index.

[0080] It should be noted that when the number of mutation points corresponding to the data information of any row is less than or equal to the set number, the abnormal evaluation value corresponding to the data information of that row is 0.

[0081] Furthermore, the operating status index is normalized so that the value of the operating status index is in the interval (0, 1). Compare the size of the operating status index with the safe operating threshold. When the operating status index is greater than the safe operating threshold, it is considered that the operating condition of the conveyor belt is poor at this time, and a warning prompt is issued to promptly inform the monitoring and management center and prompt relevant personnel to perform maintenance on the conveyor belt to avoid the occurrence of dangerous accidents due to the poor operating condition of the conveyor belt during operation; when the operating status index is less than or equal to the safe operating threshold, it is considered that the operating condition of the conveyor belt is good at this time, and no warning prompt is issued; thus, the task of monitoring the conveyor belt is completed through the operating status index.

[0082] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for monitoring the safe operation of an agricultural product transportation conveyor belt, characterized in that, the method comprises the following steps: Obtain the image of the conveyor belt surface, preprocess the conveyor belt surface image to obtain the grayscale image corresponding to the conveyor belt; respectively perform filtering processing on the grayscale image in multiple directions by using filtering kernels of different scales to obtain multiple feature extraction maps; fuse the multiple feature extraction maps to obtain a saliency map; Extract the straight-line segment areas in the saliency map and record them as suspected crack areas, and obtain the gradient distribution characteristics corresponding to each suspected crack area according to the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel point in the suspected crack area; According to the width and length of the minimum circumscribed rectangle corresponding to each suspected crack area and the distances from all suspected crack pixel points in it to the conveyor belt edge line, obtain the random index corresponding to each suspected crack area; Record the grayscale mean value of all suspected crack pixel points corresponding to each suspected crack area as the intensity response value, and based on the gradient distribution characteristics, random index and intensity response value corresponding to each suspected crack area, obtain the evaluation index corresponding to each suspected crack area, and record the suspected crack area corresponding to the evaluation index greater than the set threshold as a real crack; Collect the data information corresponding to different monitoring parameters during the operation of the conveyor belt, construct a monitoring matrix based on the data information, obtain the mutation points corresponding to each row of data information, and when the number of mutation points corresponding to any row of data information is greater than the set number, then calculate the abnormal evaluation value corresponding to this row of data information according to the maximum value, minimum value and variance corresponding to this row of data information; Calculate the operation state index of the conveyor belt according to the evaluation index corresponding to each real crack and the abnormal evaluation value, and monitor the conveyor belt based on the operation state index; The method for obtaining the gradient distribution characteristics is: record the gradient amplitude and gradient direction angle corresponding to each suspected crack pixel point as gradient information, the gradient information corresponding to all suspected crack pixel points corresponding to each suspected crack area constitutes the gradient information set corresponding to this suspected crack area, record the gradient information with the same value as the same kind of gradient information, calculate the ratio of the frequency of each kind of gradient information appearing in the gradient information set to the number of all gradient information in the gradient information set, and determine the gradient distribution characteristics corresponding to each suspected crack area according to the ratio and the variance value of all gradient direction angles in the gradient information set; The formula expression of the gradient distribution characteristics is: Among them, Gray k is the gradient distribution feature corresponding to the suspected crack area k, is the variance value of all gradient direction angles in the suspected crack area k, w v is the ratio of the frequency of occurrence of the v-th gradient information in the gradient information set to the number of all gradient information in the gradient information set, V k is the number of types of gradient information; e is the natural constant; lg w v is the logarithm of w with the natural constant 10 as the base v . The method for obtaining the random index is: Calculate the sum of the distances from all suspected crack pixel points in each suspected crack area to the conveyor belt edge line, calculate the ratio of the width and length of the minimum circumscribed rectangle, and determine the random index according to the sum and the ratio; The formula expression of the random index is: Among them, Ran k is the random index corresponding to the suspected crack area k, w k is the width of the minimum circumscribed rectangle corresponding to the suspected crack area k; h k is the length of the minimum circumscribed rectangle corresponding to the suspected crack area k; d i,k is the distance from the i-th suspected crack pixel point in the suspected crack area k to the conveyor belt edge line, N k is the total number of suspected crack pixel points in the suspected crack area k; The evaluation index is: Among them, P k is the evaluation index corresponding to the suspected crack area k, Ran k is the random index corresponding to the suspected crack area k, is the strength response value corresponding to the suspected crack area k; Gray k ' is the value after normalization of the gradient distribution characteristics corresponding to the suspected crack area k, θ k is the included angle formed by the straight line corresponding to the suspected crack area k and the edge line of the conveyor belt; e is the natural constant; The abnormal evaluation value is: where z c is the anomaly evaluation value corresponding to the data information of the c-th row, h c,max is the maximum value corresponding to the data information of the c-th row, h c,min is the minimum value corresponding to the data information of the c-th row, σ c is the variance corresponding to the data information of the c-th row; exp(·) is the exponential function with the natural constant e as the base; The operation state index is the sum of the evaluation indexes corresponding to all real cracks and the sum of all abnormal evaluation values.

2. The method for monitoring the safe operation of an agricultural product transportation conveyor belt according to claim 1, characterized in that, The method for fusing multiple feature extraction maps to obtain a saliency map is specifically as follows: First, perform a position alignment operation on the feature extraction maps corresponding to different directions at the same scale, calculate the pixel mean corresponding to the pixels at the same position according to the pixel values of the pixels at the same position, and obtain the fused feature extraction map corresponding to this scale; Then, perform saliency processing on the fused feature maps corresponding to different scales respectively to obtain the fused saliency maps corresponding to different scales. Furthermore, assign weights to each fused saliency map, and perform weighted processing on all fused saliency maps according to the weights to obtain the saliency map.

3. The method for monitoring the safe operation of an agricultural product transportation conveyor belt according to claim 1, characterized in that, The method for obtaining the mutation points corresponding to each row of data information is as follows: Obtain the edge values corresponding to all data information in the monitoring matrix through a horizontal edge detection operator, and record the data information with an edge value of 1 as the mutation point, thereby obtaining the mutation points corresponding to each row of data information.

Images