A plastering robot feeding detection method and device

Abstract

The present invention relates to a plastering robot feeding detection method and device, comprising: obtaining a real-time original image of the feeding execution end of the plastering robot feeding process; preprocessing the original image, including performing wide-angle distortion reduction, extracting a valid area from the image, and evenly dividing the valid area into multiple unit images along the horizontal direction; for each unit image, obtaining a feeding grade A based on an RGB threshold filter detection and a feeding grade B based on an LSTM network classification model detection; the feeding grade represents the proportion of the cement area in the entire image; calculating the error E = |A-B| between the two feeding grades A and B, comparing the error E with a set discrimination value, and selecting one of A and B as the feeding grade of the unit image; and summing and averaging the feeding grades of all unit images to obtain a final feeding grade. The present invention enables the plastering robot to accurately and quickly identify the feeding situation.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and device for detecting feeding of a plastering robot. Background Art

[0002] Plastering robots are an important type of construction robot, and research has been conducted on their navigation, positioning, task planning, material feeding control, and construction quality inspection. In existing technologies, some types of plastering robots plaster interior walls by mimicking workers pushing and pulling scrapers. However, in construction practice, the supply of cement mortar still requires manual operation or assistance. This results in problems such as the feeding speed being difficult to match the scraper consumption, which directly affects the quality of the plastered wall. Therefore, an automatic feeding system for plastering robots is used to replace manual feeding and better match the scraper's demand for cement mortar. Quickly and accurately determining the current feeding status of the plastering robot is a key step and core difficulty of the automatic feeding system.

[0003] In the prior art, the research and development of automatic feeding systems mainly focuses on the hardware part (storage device, pumping device, feeding pipeline and actuator), and the research and development of the software control part is still insufficient. Patent CN202011179076.7 discloses a method of adjusting the feeding speed by using the flow rate of the feeding pipeline or the feeding pressure and weight of the actuator. However, this method does not directly detect the feeding situation, but only reflects the feeding situation from the side through pressure, weight, etc., and the error is large. Patent CN201620358042.7 discloses an indoor autonomous mobile intelligent plastering machine, which detects the feeding situation through a height detector in the cement mortar hopper. However, the plastering machine can only detect the peak value of the cement mortar, and cannot detect the distribution in the horizontal direction, and cannot solve the problem of local accumulation of raw materials.

[0004] Traditional image processing methods are fast and highly accurate, but they are sensitive to ambient lighting and have poor robustness. Directly applying these methods to the plastering robot's feeding detection system cannot ensure accurate measurement. Therefore, there is currently a lack of technical means to accurately and quickly identify the feeding status of the feeding system, making it impossible to fully automate the feeding of plastering robots. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method and device for detecting the feeding of a plastering robot, the purpose of which is to quickly and accurately automatically identify the feeding status of the plastering robot feeding system.

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

[0007] The present invention provides a plastering robot feeding detection method, comprising:

[0008] Obtain real-time original images of the feeding execution end during the plastering robot feeding process;

[0009] Preprocessing the original image includes restoring wide-angle distortion, extracting an effective area from the image, and evenly dividing the effective area into a plurality of unit images along the horizontal direction; the effective area is the area where cement is mainly distributed;

[0010] For each unit image, feed grade A is obtained based on RGB threshold filter detection, and feed grade B is obtained based on LSTM network classification model detection. The feed grade represents the proportion of the cement area in the entire image. The LSTM network classification model is trained with the image grayscale value integral sequence as input and the feed grade detection type as output. The feed grade detection type is divided according to the proportion of the cement area in the entire image.

[0011] Calculate the error E = |AB| between the two feeding levels A and B, compare the error E with the set discrimination value, and select one of A and B as the feeding level of the unit image based on the comparison result;

[0012] The feed levels of all unit images are summed and averaged to obtain the final feed level.

[0013] Further technical solutions are:

[0014] The LSTM network classification model-based detection method for obtaining the feed grade includes:

[0015] Grayscale the unit image, calculate the grayscale value of each pixel and normalize it to the interval [0,1] to obtain a grayscale image;

[0016] Calculate the grayscale integral sequence GIS = [GI1 GI2 GI3…GI N ], where the grayscale value integral of each column of pixels in the horizontal direction of the grayscale image is Gray is the grayscale value of each pixel, H is the height of the grayscale image, and subscript N is the horizontal length of the grayscale image;

[0017] The grayscale integral sequence GIS is used as the input of the LTSM network model to directly output the feed grade detection type;

[0018] The feed grade detection type is characterized by the ratio of the cement area in the unit image to the total area. During model training, the ratio is labeled and divided into several categories according to the set class interval.

[0019] The discrimination value is set according to the class interval, and the discrimination value is smaller than the class interval.

[0020] The error E is compared with the set discrimination value. When the error E is less than the discrimination value, it is considered that the feed levels detected by the two methods are consistent, and A is selected as the feed level of the unit image; when the error E is greater than the discrimination value, it is considered that the feed levels detected by the two methods are inconsistent, and B is selected as the feed level detection result of the unit image.

[0021] The method of obtaining the feed level based on RGB threshold filter detection includes:

[0022] Perform distribution statistics analysis on the RGB three-channel component values ​​of the unit image to obtain the RGB three-channel component value characteristics of the cement distribution area in the image;

[0023] Setting filtering conditions of the RGB threshold filter according to the RGB three-channel component value characteristics;

[0024] The filtering conditions are used to judge each pixel one by one to obtain the percentage of cement pixels in the entire image, which is the feed grade A of the unit image.

[0025] The effective area is a rectangular area, and the extraction method is:

[0026] The upper limit of the area is the maximum height that the cement mortar on the scraper can reach during the previous plastering operations, and a certain margin must be retained upwards;

[0027] The lower limit of the area is the upper edge of the scraper;

[0028] The left and right limits of the area are the left and right edges of the scraper.

[0029] When preprocessing the original image, before extracting the effective area from the image, the method further includes: adjusting the image to be horizontal according to a set reference.

[0030] The present invention also provides a plastering robot feeding detection device, comprising:

[0031] An image acquisition unit is used to acquire real-time original images during the scraper feeding process of the plastering robot;

[0032] An image preprocessing unit is used to preprocess the original image, including restoring wide-angle distortion, extracting an effective area from the image, and evenly dividing the effective area into a plurality of unit images along the horizontal direction; the effective area is the area where cement is mainly distributed;

[0033] A grade detection unit is configured to obtain, for each unit image, a feed grade A based on an RGB threshold filter and a feed grade B based on an LSTM network classification model; the feed grade represents the proportion of the cement area in the entire image; the LSTM network classification model is trained to take an image grayscale value integral sequence as input and a feed grade detection type as output; the feed grade detection type is divided according to the proportion of the cement area in the entire image; the error E = |AB| between the two feed grades A and B is calculated, the error E is compared with a set discrimination value, and based on the comparison result, one of A and B is selected as the feed grade of the unit image;

[0034] Output unit, add and average the feed levels of all unit images to obtain the final feed level.

[0035] The beneficial effects of the present invention are as follows:

[0036] The method of the present invention is developed based on real-time images from the feeding execution end. This real-time image directly reflects the feeding status, providing a more intuitive picture than indirect measurement variables such as altimeters and load weight, while avoiding data distortion caused by indirect variables. This improves detection accuracy and enhances practicality by leveraging input information.

[0037] The method of the present invention combines the advantages of the LSTM classification network and the RGB threshold filter, and improves the detection accuracy under a good construction environment. The RGB threshold filter calculates the feed grade by judging the pixels in the image one by one. Under a good construction environment, the detection results of the LSTM classification network and the RGB threshold filter are similar. The final detection result is based on the detection result of the RGB threshold filter, and the detection accuracy is improved to the pixel level. At the same time, the present invention can maintain detection accuracy under harsh construction environments and abnormal conditions (such as cement splashing). The neural network has the advantages of strong robustness and high adaptability because the samples are widely distributed and the calculation rules are not artificially set for internal relationships. When the detection results of the LSTM classification network and the RGB threshold filter differ greatly, the final detection result is based on the detection result of the LSTM classification network, which makes up for the situation where traditional image processing methods have misjudgments in harsh construction environments such as poor lighting conditions and dust obstruction. It avoids the misjudgment of the feed grade and improves the robustness and adaptability of the detection algorithm.

[0038] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The figure is a structural diagram of the feeding system of the plastering robot in the prior art.

[0040] Figure 2 Schematic diagram of the flow of the feeding detection method according to an embodiment of the present invention.

[0041] Figure 3 This is the image after distortion restoration according to an embodiment of the present invention.

[0042] Figure 4 This is a statistical diagram of the distribution of RGB three-channel component values ​​in an embodiment of the present invention.

[0043] Figure 5 Schematic diagram of the structure of the LSTM network classification model according to an embodiment of the present invention.

[0044] In the figure: 1. Feed pump; 2. Bracket; 3. Scraper; 4. Guide rail; 5. Feed head; 6. Feed pipe. DETAILED DESCRIPTION

[0045] The specific embodiments of the present invention are described below with reference to the accompanying drawings.

[0046] The existing plastering robot mainly consists of a mobile chassis, a plastering actuator and a feeding system. Figure 1 The feeding system consists of a feeding pump 1, a feeding pipe 6, a feeding head 5 (material trough) and its guide rail 4, a scraper 3 and a bracket 2. The feeding head 5 can move horizontally along the guide rail 4, and the scraper 3 is fixed to the end of the guide rail 4. The feeding head 5 and its guide rail 4 and scraper 3 can be synchronously raised and lowered in the vertical direction along the bracket. Among them, the feeding head 5 and the scraper 3 constitute the feeding execution end. Cement mortar needs to be fed during the rising process of the scraper 3 to complete the wall painting. During operation, the cement mortar is first transported to the feeding pipe 6 by the feeding pump 1, and then sprayed out by the feeding head 5 at the end of the feeding pipe 6. The feeding head 5 reciprocates horizontally along the guide rail 4, thereby realizing the feeding of the scraper 3 during its upward work.

[0047] See also Figure 2 This application is a real-time detection method based on machine vision that can be used for the feeding status of cement mortar when the feeding execution end of a plastering robot is working upstream, including the following process:

[0048] S1. Acquire real-time original images during the feeding process of the plastering robot feeding execution end.

[0049] Specifically, the original image can be directly obtained by an industrial wide-angle camera (or a frame can be captured from a real-time video), and the format can be png, with a size of 480*640 pixels. Figure 1 The upper left side of the feed head 5 is photographed from above, and the feed head (feed trough) and the scraper are photographed at the same time.

[0050] S2. Preprocessing the original image, including restoring wide-angle distortion, extracting a valid area from the image, and evenly dividing the valid area into a plurality of unit images along the horizontal direction.

[0051] Specifically, the industrial wide-angle camera is calibrated to obtain its intrinsic parameter matrix and distortion parameters, and then the undistort function is used to restore the distortion of the original image.

[0052] S21. Camera calibration: Use Zhang Zhengyou calibration method to calibrate the industrial wide-angle camera (take standard calibration pictures at different spatial positions, the side length of the black and white square in the standard calibration picture is 20 mm) to obtain the camera intrinsic parameter matrix and distortion parameters. The intrinsic parameter matrix M contains the intrinsic focal length f x 、f y The coordinates of the origin of the image coordinate system in the pixel coordinate system are (u0, v0). The distortion parameter matrix D contains five parameters: radial distortion parameters k1, k2, k3 and tangential distortion parameters p1, p2.

[0053]

[0054] D=[k1 k2 k3 p1 p2]

[0055] In this embodiment, the camera parameters are exported in XML file format. The intrinsic parameter matrix M and the distortion parameter matrix D are as follows:

[0056]

[0057] D=[-0.3872392881686612 0.0913128378192669 0 0 0]

[0058] Use the undistort function in the OpenCV library to restore the original image to a wide-angle distortion to compensate for the radial and tangential distortion of the wide-angle lens. Figure 3 .

[0059] S22. Rotate the image after wide-angle distortion restoration to a horizontal state: Using the bottom edge of the scraper as a reference, rotate the image to a horizontal state so that the image maintains the same horizontal reference.

[0060] The image tilt is caused by an undesirable camera installation position. If the camera is installed on the central axis of the scraper and is kept horizontal with the scraper, this step can be skipped. In this embodiment, the image is rotated 8 degrees counterclockwise.

[0061] S23, extracting the effective area: The effective area is determined according to the actual operation situation. Figure 3As shown, the effective area extracted in this embodiment is a rectangle (shown in the rectangular box in the figure), the upper limit of which is the maximum height that the cement mortar can reach in all previous plastering operations, and a certain margin must continue to be retained upwards, the lower limit is the upper edge of the scraper, and the left and right limits are the left and right edges of the scraper.

[0062] In this embodiment, the coordinates of the lower left vertex of the rectangular effective area are (20, 100), and the coordinates of the upper right vertex are (470, 200).

[0063] The purpose of extracting the effective area is to extract the area where cement is mainly distributed from the image.

[0064] S24: Divide the rectangular effective area into multiple unit images in the horizontal direction. In this embodiment, the image is divided into three unit images: left, center, and right. The format size of each unit image is 150*100 pixels.

[0065] The purpose of evenly segmenting the image is to avoid localized accumulation of cement mortar, which would hinder the subsequent neural network model construction from making a more appropriate judgment on the entire image, resulting in inaccurate labeling. This also prevents interference with the overall feed grade judgment of the image.

[0066] S3. For each unit image, obtain feed grade A based on RGB threshold filter detection and obtain feed grade B based on LSTM network classification model detection.

[0067] S31, obtaining a feed level based on RGB threshold filter detection, including:

[0068] The distribution statistics of the RGB three-channel component values ​​of each unit image are analyzed to obtain the RGB three-channel component value characteristics of the cement area. Figure 4 As shown in the figure, statistics show that in areas where the RGB component values ​​are greater than 100, the frequency of the three channel component values ​​is basically the same, representing a non-cement area. The area where the RGB component values ​​are less than 100 is the cement distribution area. As can be seen from the figure, in the cement area, the B channel component value frequency curve is shifted to the left by approximately 20 unit component values. In other words, the peak shift of the B channel component value frequency curve is the main characteristic of cement pixels. Based on this, the following filtering conditions are designed in this embodiment to filter out cement pixels:

[0069]

[0070] Under the above filtering conditions, the RGB threshold filter can determine whether the current pixel is a cement-characteristic pixel. This filter is then applied to each pixel, counting the number of cement-characterized pixels in the image. The percentage of cement pixels in the entire image represents the cement grade for that unit image. In this example, the grade for the three unit images is 24.6733%, 6.96667%, and 11.7467%, respectively.

[0071] S32. Obtaining the feed grade based on LSTM network classification model detection, including:

[0072] The LSTM network classification model is obtained through training, taking the image grayscale value integral sequence as input and the feed grade detection type as output; the feed grade detection type is divided according to the proportion of the cement area in the entire image, specifically:

[0073] Grayscale the unit image, calculate the grayscale value (Gray) of each pixel and normalize it to the interval [0,1] to obtain a grayscale image. The calculation formula is: Gray = (R*0.3+G*0.59+B*0.11) / 255, where R, G, and B are the values ​​of the red, green, and blue color channels of each pixel respectively.

[0074] Calculate the grayscale integral sequence GIS = [GI1 GI2 GI3…GI N ], where the grayscale value integral of each column of pixels in the horizontal direction of the grayscale image is Gray is the grayscale value of each pixel, H is the height of the grayscale image, and subscript N is the horizontal length of the grayscale image;

[0075] The grayscale integral sequence GIS is used as the input of the LTSM model to directly output the feed grade detection type;

[0076] The feed grade detection type is characterized by the percentage of the cement area in the unit image. During model training, the percentage is labeled and divided into several categories according to a set class interval (10 in this embodiment).

[0077] In this embodiment, the feed grades are divided into five categories: 10, 20, 30, 40, and 50, which respectively represent that the cement area accounts for 10%, 20%, 30%, 40%, and 50% of the total map area.

[0078] For details, see Figure 5The input layer node of the LTSM network is 1, and the middle hidden layer contains 300 hidden units. In order to improve the speed, each hidden unit only outputs the output value (vector) at the last moment; the last hidden unit is followed by a fully connected layer (FC) to reduce the data dimension to the number of output categories; then it is connected to the softmax layer, which obtains the probability of each category; finally, the classification output layer outputs the final classification result, that is, the classification result with the highest probability.

[0079] In this embodiment, the feed levels of the three unit images are 30, 10 and 10 respectively.

[0080] The LTSM classification network has good robustness, and the number of classifications of the LTSM classification network is limited under the condition of ensuring good judgment accuracy. Since the number of cement mortar feeding grade classifications is small when the plastering robot is working, the LTSM classification network is suitable for application in the present invention.

[0081] S4. Calculate the error E = |AB| between the two feeding levels A and B, compare the error E with the set discrimination value, and select one of A and B as the feeding level of the unit image based on the comparison result;

[0082] The feed levels of all unit images are summed and averaged to obtain the final feed level.

[0083] Specifically, the errors E of the three unit images in this embodiment are 5.3257, 3.03333, and 1.7467, respectively.

[0084] The feed grade detection method based on RGB threshold filters is sensitive to ambient lighting and has poor robustness, but it is fast and highly accurate. The LSTM network-based plastering robot feed grade classification algorithm has good robustness, but it suffers from a conflict between judgment precision (number of classifications) and accuracy. The number of classifications required to ensure good accuracy is limited. In summary, the following data fusion rules are set:

[0085] When the error E is less than the discrimination value 5, it is considered that the feeding grades detected by the two methods are consistent, and the final feeding grade is the feeding grade A with higher accuracy.

[0086] When the error E is greater than the discrimination value 5, it is considered that the feeding levels detected by the two methods are inconsistent, and the final feeding level is the feeding level B with higher robustness.

[0087] The discriminant value of 5 is selected as half of the class interval of 10 in the LSTM network classification model.

[0088] According to the above rules, the final feeding levels of the three unit images are 30, 6.96667 and 11.7467.

[0089] S5. Calculate the feed level D of the overall image:

[0090] The final feed levels of the three unit images are summed and averaged to obtain the feed level of the entire image. The calculation formula is: D = (30 + 6.96667 + 11.7467) / 3 = 16.2378.

[0091] The discrimination rules of this embodiment have been verified to be reasonable and effective through experiments. In harsh construction environments and abnormal situations (such as cement splashing), the detection results of the feed level detection method based on the RGB threshold filter will be seriously inconsistent with the actual situation. Due to the good robustness of the LSTM classification network and the wide distribution of samples, its detection results are basically consistent with the actual situation, as shown in the following table:

[0092]

[0093] Therefore, the final detection result is based on the detection result of the LSTM classification network, which avoids the misjudgment of the feeding level and improves the overall robustness and adaptability of the detection algorithm.

[0094] This application proposes a plastering robot feeding detection method that integrates LSTM classification network and RGB threshold filter based on machine vision method. The LTSM classification network has good robustness, but its classification number is limited under the condition of ensuring good judgment accuracy. When the plastering robot is working, the number of cement mortar feeding grade classifications is relatively small, so the LTSM classification network is suitable for application in this application. Therefore, this application combines the advantages of LTSM classification network and traditional image processing method (RGB threshold filter), realizes accurate and rapid identification of the plastering robot feeding situation, lays the foundation for the automatic control of the plastering robot feeding, and helps the plastering robot to become practical.

[0095] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will be able to modify the technical solutions described in the foregoing embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A plastering robot feeding detection method, characterized in that: include: Obtain real-time original images of the feeding execution end during the plastering robot feeding process; Preprocessing the original image includes restoring wide-angle distortion, extracting an effective area from the image, and evenly dividing the effective area into a plurality of unit images along the horizontal direction; the effective area is the area where cement is mainly distributed; For each unit image, feed grade A is obtained based on RGB threshold filter detection, and feed grade B is obtained based on LSTM network classification model detection. The feed grade represents the proportion of the cement area in the entire image. The LSTM network classification model is trained with the image grayscale value integral sequence as input and the feed grade detection type as output. The feed grade detection type is divided according to the proportion of the cement area in the entire image. Calculate the error between two feed levels A and B , compare the error E with the set discrimination value, and select one from A and B as the feeding level of the unit image according to the comparison result; The feed levels of all unit images are summed and averaged to obtain the final feed level; The LSTM network classification model-based detection method for obtaining the feed grade includes: Grayscale the unit image, calculate the grayscale value of each pixel and normalize it to the interval [0,1] to obtain a grayscale image; Calculate the grayscale integral sequence , where the grayscale value integral of each column of pixels in the horizontal direction of the grayscale image is , Gray is the grayscale value of each pixel, H is the height of the grayscale image, subscript N is the horizontal length of the grayscale image; Grayscale integral sequence GIS As the input of the LTSM network model, the feed level detection type is directly output; The feed grade detection type is characterized by the percentage of the cement area in the unit image. During model training, the percentage is labeled and divided into several categories according to the set class interval. The discrimination value is set according to the class interval, and the discrimination value is smaller than the class interval; The error E is compared with a set discrimination value. When the error E is less than the discrimination value, it is considered that the feed levels detected by the two methods are consistent, and A is selected as the feed level of the unit image. When the error E is greater than the discrimination value, it is considered that the feed levels detected by the two methods are inconsistent, and B is selected as the feed level detection result of the unit image. The method of obtaining the feed level based on RGB threshold filter detection includes: Perform distribution statistics analysis on the RGB three-channel component values ​​of the unit image to obtain the RGB three-channel component value characteristics of the cement distribution area in the image; Setting filtering conditions of the RGB threshold filter according to the RGB three-channel component value characteristics; The filtering conditions are used to judge each pixel one by one to obtain the percentage of cement pixels in the entire image, which is the feed grade A of the unit image.

2. The plastering robot feeding detection method according to claim 1, characterized in that: The effective area is a rectangular area, and the extraction method is: The upper limit of the area is the maximum height that the cement mortar on the scraper can reach during the previous plastering operations, and a certain margin must be retained upwards; The lower limit of the area is the upper edge of the scraper; The left and right limits of the area are the left and right edges of the scraper.

3. The plastering robot feeding detection method according to claim 1, characterized in that: When preprocessing the original image, before extracting the effective area from the image, the method further includes: adjusting the image to be horizontal according to a set reference.

4. A plastering robot feeding detection device, characterized in that: include: An image acquisition unit is used to acquire real-time original images during the scraper feeding process of the plastering robot; An image preprocessing unit is used to preprocess the original image, including restoring wide-angle distortion, extracting an effective area from the image, and evenly dividing the effective area into a plurality of unit images along the horizontal direction; the effective area is the area where cement is mainly distributed; The grade detection unit is used to obtain the feed grade A based on the RGB threshold filter detection and the feed grade B based on the LSTM network classification model detection for each unit image; the feed grade represents the proportion of the cement area in the whole image; the LSTM network classification model is trained with the image gray value integral sequence as input and the feed grade detection type as output; the feed grade detection type is divided according to the proportion of the cement area in the whole image; the error between the two feed grades A and B is calculated , compare the error E with the set discrimination value, and select one from A and B as the feeding level of the unit image according to the comparison result; Output unit, add and average the feed levels of all unit images to obtain the final feed level; The LSTM network classification model-based detection method for obtaining the feed grade includes: Grayscale the unit image, calculate the grayscale value of each pixel and normalize it to the interval [0,1] to obtain a grayscale image; Calculate the grayscale integral sequence , where the grayscale value integral of each column of pixels in the horizontal direction of the grayscale image is , Gray is the grayscale value of each pixel, H is the height of the grayscale image, subscript N is the horizontal length of the grayscale image; Grayscale integral sequence GIS As the input of the LTSM network model, the feed level detection type is directly output; The feed grade detection type is characterized by the percentage of the cement area in the unit image. During model training, the percentage is labeled and divided into several categories according to the set class interval. The discrimination value is set according to the class interval, and the discrimination value is smaller than the class interval; The error E is compared with a set discrimination value. When the error E is less than the discrimination value, it is considered that the feed levels detected by the two methods are consistent, and A is selected as the feed level of the unit image. When the error E is greater than the discrimination value, it is considered that the feed levels detected by the two methods are inconsistent, and B is selected as the feed level detection result of the unit image. The method of obtaining the feed level based on RGB threshold filter detection includes: Perform distribution statistics analysis on the RGB three-channel component values ​​of the unit image to obtain the RGB three-channel component value characteristics of the cement distribution area in the image; Setting filtering conditions of the RGB threshold filter according to the RGB three-channel component value characteristics; The filtering conditions are used to judge each pixel one by one to obtain the percentage of cement pixels in the entire image, which is the feed grade A of the unit image.

Images