A material spreading method and apparatus

Abstract

The present application relates to the technical field of image processing, in particular to a material flattening method and device, the flattening method obtains material images of different angles, carries out image enhancement, feature extraction and feature matching, accurately identifies the material in the multiple angle images, and obtains the material area in the multiple angle images; based on the material area information of multiple angles, a material three-dimensional model is established, and the material area to be flattened is determined; and according to the distance between the material area to be flattened and the inner wall of the mold box, the order of the material area to be flattened to perform flattening processing is determined, and the material area to be flattened is sequentially flattened according to the order; the flattening method has good flattening effect, high flattening efficiency and stable flattening quality.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a material leveling method and apparatus. Background Technology

[0002] When the homogeneous board production line pours the mixed granular cement mixture into the mold box, the surface is generally very uneven due to the characteristics of the material. If it is not leveled, it will affect the subsequent cover plate and molding operations, thus affecting the quality of the finished block.

[0003] Previously, materials needed to be manually leveled, a method that was labor-intensive, inefficient, and produced inconsistent leveling quality. Machine vision-based material leveling automates the process, eliminating the need for manual intervention, thus improving production line efficiency and reducing labor costs. However, machine vision-based leveling is significantly affected by image quality and material position information. Low image quality or inaccurate material position information can lead to incorrect scraper movement trajectories, resulting in poor leveling quality and inconsistent leveling results. Summary of the Invention

[0004] The purpose of this invention is to provide a material leveling method and apparatus that provides good leveling quality and stable leveling effect.

[0005] The technical solution of this invention is as follows:

[0006] A method for spreading materials includes the following operations:

[0007] S1. Obtain images of the material on the mold box to be processed from multiple perspectives to obtain a multi-view material image set;

[0008] In the multi-view material image set, the current view image is enhanced and feature extracted, and then feature matching is performed to obtain the current view classification map; all view classification maps form a multi-view material classification map set; based on the material regions in the multi-view material classification map set and the corresponding position data of the material regions in the multi-view material image set, a three-dimensional model of the material is obtained.

[0009] S2. In the three-dimensional model of the material, the protrusions with a height difference greater than the first threshold are taken as the three-dimensional regions to be flattened. The material regions at the corresponding positions of the three-dimensional regions to be flattened on the mold box to be processed are the material regions to be flattened. All the material regions to be flattened form a set of material regions to be flattened.

[0010] S3. Obtain the concentrated material areas to be flattened, and the distance of each material area to be flattened from the inner wall of the mold box to be processed. Flatten each material area to be flattened in order of increasing distance. During the flattening process, if the height of the material in the current material area to be flattened does not exceed the height of the mold box, then the material in the current material area to be flattened has been flattened, and the next material area to be flattened is flattened.

[0011] The image enhancement operation in S1 specifically involves: processing the current viewpoint image with Gaussian blur to obtain a smoothed image of the current viewpoint; subtracting the corresponding pixels of the current viewpoint image from the smoothed image of the current viewpoint to obtain a detail image of the current viewpoint; superimposing the detail image of the current viewpoint image with the current viewpoint image to obtain an enhanced image of the current viewpoint; and using the enhanced image of the current viewpoint to perform feature extraction operations.

[0012] The feature extraction operation in S1 is as follows: the current view image is enhanced by image enhancement processing to obtain the current view enhanced image, and the gradient is calculated to obtain the horizontal and vertical gradients of each pixel in the current view enhanced image; based on the horizontal and vertical gradients of each pixel, the gradient magnitude corresponding to each pixel is obtained; pixels in the current view enhanced image with gradient magnitudes greater than a second threshold are taken as edge points, and all edge points are subjected to curve fitting processing to obtain a current view feature map containing edge contours; the current view feature map is used to perform feature matching processing.

[0013] The feature matching process in S1 is as follows: the current viewpoint feature map obtained after image enhancement and feature extraction of the current viewpoint image is processed by marking detection boxes according to the edge contours to obtain a feature map to be matched containing multiple detection boxes; the similarity between the feature map of the region corresponding to the current detection box in the feature map to be matched and each standard feature map in the standard database is obtained, and the classification label of the standard feature map corresponding to the maximum similarity is used as the classification result of the feature map of the region corresponding to the current detection box; the classification results of the feature maps of the regions corresponding to all detection boxes form the current viewpoint classification map.

[0014] The specific steps for obtaining the three-dimensional model of the material in S1 are as follows: 1. Obtain the area where the material is located from the multi-view material classification image set, which is classified as the material area; 2. Randomly select a point on the mold box as the origin of the coordinate system, and obtain the corresponding coordinate data of the material area relative to the origin of the coordinate system in the multi-view material image set, which is used as the position data; 3. Based on the origin of the coordinate system and the position data, establish a three-dimensional model to obtain the three-dimensional model of the material.

[0015] Following the operation in S1, the process further includes filtering the three-dimensional model of the material to obtain a filtered three-dimensional model of the material, which is then used to perform the operation in S2.

[0016] In step S3, if multiple material areas to be flattened are at the same distance from the inner wall of the mold box, then the multiple material areas to be flattened are flattened sequentially in order of increasing distance from the previous material area to be flattened.

[0017] A material leveling device, comprising:

[0018] The image acquisition unit is used to acquire images of the material on the mold box to be processed from multiple perspectives, and obtain a multi-view material image set.

[0019] The image processing unit is used to collect the multi-view material images from the image acquisition unit, and after image enhancement and feature extraction, perform feature matching processing to obtain a current view classification map; all view classification maps form a multi-view material classification map set; it is used to obtain a three-dimensional material model based on the material regions in the multi-view material classification map set and the corresponding position data of the material regions in the multi-view material image set; it is used to identify protrusions in the three-dimensional material model with a height difference greater than a first threshold as three-dimensional regions to be flattened, and the material regions at the corresponding positions on the mold box to be processed as material regions to be flattened; all material regions to be flattened form a set of material regions to be flattened; it is used to obtain the distance from each material region to be flattened to the inner wall of the mold box to be processed in the set of material regions to be flattened;

[0020] The motion control unit is used to flatten each material area to be flattened in sequence according to the order of distance from the material area to be flattened to the inner wall of the mold box in the image processing unit from small to large; during the flattening process, if the height of the material in the current material area to be flattened does not exceed the height of the mold box, then the material in the current material area to be flattened has been flattened, and the next material area to be flattened is flattened.

[0021] The device further includes: a feedback adjustment unit, used to instruct the image acquisition unit to acquire a multi-view material image set of the material on the current mold box to be processed after flattening; instruct the image processing unit to determine, based on the multi-view material image set of the material on the current mold box to be processed after flattening, whether there are protrusions on the material with a height difference greater than a second threshold, and if so, update the material area to be flattened; and instruct the motion control unit to flatten the updated material area to be flattened.

[0022] The flattening operation is achieved by a scraper at the end of a robotic arm that is positionally mapped to the camera in the image acquisition unit.

[0023] The beneficial effects of this invention are as follows:

[0024] This invention provides a material leveling method that, by acquiring material images from different perspectives, performing image enhancement, feature extraction, and feature matching, accurately identifies the material in multiple perspective images, obtaining material regions in multiple perspective images; based on the material region information from multiple perspectives, a three-dimensional model of the material is established to determine the material region to be leveled; and according to the distance between the material region to be leveled and the inner wall of the mold box, the order in which the material region to be leveled is performed is determined, and the material regions to be leveled are leveled sequentially according to the order; this leveling method has good leveling effect, high leveling efficiency, and stable leveling quality. Detailed Implementation

[0025] This embodiment provides a material leveling method, including the following operations:

[0026] S1. Acquire images of the material on the mold box from multiple perspectives to obtain a multi-view material image set; in the multi-view material image set, the current perspective image is enhanced and feature extracted, and then feature matching is performed to obtain a current perspective classification map; all perspective classification maps form a multi-view material classification map set; based on the material regions in the multi-view material classification map set and the corresponding position data of the material regions in the multi-view material image set, a three-dimensional model of the material is obtained;

[0027] S2. In the three-dimensional model of the material, the protrusions with a height difference greater than the first threshold are taken as the three-dimensional regions to be flattened. The material regions at the corresponding positions of the three-dimensional regions to be flattened on the mold box to be processed are the material regions to be flattened. All the material regions to be flattened form a set of material regions to be flattened.

[0028] S3. Obtain the concentrated material areas to be flattened, and the distance of each material area to be flattened from the inner wall of the mold box to be processed. Flatten each material area to be flattened in order of increasing distance. During the flattening process, if the height of the material in the current material area to be flattened does not exceed the height of the mold box, then the material in the current material area to be flattened has been flattened, and the next material area to be flattened is flattened.

[0029] S1. Acquire images of the material on the mold box from multiple perspectives to obtain a multi-view material image set; In the multi-view material image set, the current perspective image is enhanced and feature extracted, and then feature matching is performed to obtain the current perspective classification map; All perspective classification maps form a multi-view material classification map set; Based on the material regions in the multi-view material classification map set and the corresponding position data of the material regions in the multi-view material image set, a three-dimensional model of the material is obtained.

[0030] By acquiring material images from different perspectives, performing image enhancement, feature extraction, and feature matching, the material in the images from multiple perspectives is identified, and material regions in the images from multiple perspectives are obtained. Based on the material region information from multiple perspectives, a three-dimensional model of the material is established, which is used to control the end effector of the robotic arm to perform material leveling based on machine vision.

[0031] First, front, back, left, right and top views of the material on the mold box to be processed are obtained to obtain images of the material from multiple perspectives. The images of the material from multiple perspectives form a multi-view material image set, which is used to build a three-dimensional model of the material.

[0032] Then, image enhancement is performed on the images from each viewpoint to improve the clarity and detail, facilitating accurate material identification. The image enhancement process involves: applying a Gaussian blur to the current viewpoint image to obtain a smoothed image; subtracting the corresponding pixels from the smoothed image to obtain a detail image; and then overlaying the detail image with the smoothed image to obtain the enhanced image. This enhanced image is then used for feature extraction. All the enhanced images obtained after image enhancement are combined to form a multi-view enhancement map set, which is used for feature extraction.

[0033] Next, feature extraction is performed on the view enhancement map of each image enhancement process. By focusing on the local structure and texture features of the view enhancement map, the edge contours of different targets in the image are obtained, which facilitates the determination of the edge contours of the material area, helps to obtain accurate material location data, and improves the accuracy of the material 3D model.

[0034] The feature extraction process can be as follows: The current viewpoint image, after image enhancement processing, yields an enhanced current viewpoint image. Gradient calculation is then performed to obtain the horizontal and vertical gradients for each pixel in the enhanced current viewpoint image. Based on these gradients, the gradient magnitude for each pixel is calculated. Pixels with gradient magnitudes greater than a second threshold in the enhanced current viewpoint image are designated as edge points. All edge points are then subjected to curve fitting to obtain a current viewpoint feature map containing edge contours. This current viewpoint feature map is used to perform feature matching. After feature extraction from all the enhanced viewpoint images, the resulting multi-viewpoint feature maps form a multi-viewpoint feature map set, which is used to perform feature matching, improving the accuracy of classification in feature matching.

[0035] To improve the continuity of edge contours, the current viewpoint feature map is smoothed to obtain a smoothed feature map, which is then used for feature matching. The smoothing process involves: filtering the current viewpoint feature map at different scales to obtain multiple filtered feature maps; fusing each of these filtered feature maps with the current viewpoint feature map, and then overlaying the resulting image to obtain a first filtered fused image; downsampling this first fused image and fusing it with the current viewpoint feature map to obtain a second filtered fused image; and finally, convolutional processing of this second fused image yields a smoothed feature map with enhanced details and edge contours. All the smoothed feature maps obtained from these edge contour smoothing processes form a multi-view smoothed feature map set, which is used for feature matching to further improve classification accuracy.

[0036] Subsequently, feature matching processing is performed on each view feature map or view smoothing feature map containing edge contours to identify the area where the material is located, which is beneficial for extracting material location information in the future.

[0037] Taking a specific viewpoint feature map or a smoothed viewpoint feature map as an example, specifically the current viewpoint feature map or the current viewpoint smoothed feature map, the feature matching process is as follows: The current viewpoint feature map or the smoothed viewpoint feature map, obtained after image enhancement and feature extraction from the current viewpoint image, is labeled with detection boxes based on edge contours to obtain a feature map to be matched containing multiple detection boxes. The similarity between the feature map of the region corresponding to the current detection box in the feature map to be matched and each standard feature map in the standard database is obtained. The classification label of the standard feature map corresponding to the maximum similarity is used as the classification result of the feature map of the region corresponding to the current detection box. The classification results of all feature maps corresponding to the detection boxes form the current viewpoint classification map.

[0038] Specifically, based on the edge contours and edges of the current viewpoint feature map or the current viewpoint smoothed feature map, detection boxes are marked on the current viewpoint feature map or the current viewpoint smoothed feature map. The number of detection boxes is equal to the number of regions enclosed by the edge contours and edges of the current viewpoint feature map or the current viewpoint smoothed feature map, resulting in a feature map to be matched containing multiple detection boxes. The similarity between the feature map of the region corresponding to each detection box in the feature map to be matched and each standard feature map in the standard database is obtained. Taking the current detection box as an example, the similarity between the feature map of the region corresponding to the current detection box and each standard feature map in the standard database is obtained. The classification label of the standard feature map corresponding to the maximum similarity is used as the classification result of the feature map of the region corresponding to the current detection box. The classification results of all the feature maps of the region corresponding to all detection boxes form a current viewpoint classification map containing classification results and corresponding regions. This allows for the distinction between material regions and non-material regions, facilitating the subsequent extraction of the position coordinate information of the material region. All viewpoint classification maps form a multi-view material classification map set.

[0039] Finally, based on the location data of the material in the material classification map of each perspective in the multi-view material classification map, a three-dimensional data model containing the material is established.

[0040] The specific steps to obtain the 3D model of the material are as follows: acquire the material classification image set from multiple perspectives, and obtain the material region by classifying the region where the material is located; randomly select a point on the mold box as the origin of the coordinate system to obtain the material region, and obtain the corresponding coordinate data of the material region relative to the origin of the coordinate system in the multiple perspective material image set as the position data; based on the origin of the coordinate system and the position data, establish a 3D model to obtain the 3D model of the material.

[0041] Specifically, the area where the material is located in the multi-view material classification image set is taken as the material area; a point on the mold box is randomly selected as the coordinate origin, for example, the right inflection point of the inner wall above the mold box is taken as the coordinate origin, and the coordinate data of the material area relative to the coordinate origin in the multi-view material image set is obtained as the material position data; based on the coordinate origin, the image scale, and the material position data, a three-dimensional data model is established to obtain the three-dimensional model of the material.

[0042] To remove noise data (mainly some free coordinate points near the contour in the material's 3D model), the material's 3D model is filtered. The resulting filtered material 3D model is used to perform the operation in S2.

[0043] S2. In the material 3D model, the protrusions with a height difference greater than the first threshold are taken as the 3D regions to be flattened. The material regions at the corresponding positions on the mold box to be processed are the material regions to be flattened. All the material regions to be flattened form a set of material regions to be flattened.

[0044] After obtaining the 3D model of the material or the filtered 3D model of the material, the protrusions in the 3D model of the material or the filtered 3D model whose height difference exceeds a threshold (areas protruding in the Z-axis direction relative to the surrounding area on the surface of the 3D model of the material or the filtered 3D model, usually appearing as raised shapes) are taken as the 3D areas to be flattened. According to the mapping relationship between the material areas to be flattened and the 3D areas to be flattened, the material areas corresponding to the 3D areas to be flattened are shrunk on the mold box to be processed, and these are taken as the material to be flattened. All the material areas to be flattened are counted to form a set of material areas to be flattened. In the actual flattening process, the movement path and flattening action of the end scraper of the flattening robot can be controlled by establishing a position mapping relationship between the material areas to be flattened and the end scraper of the flattening robot, so as to achieve the flattening of the material areas to be flattened.

[0045] S3. Obtain the concentrated material areas to be flattened. The distance of each material area to be flattened from the inner wall of the mold box to be processed is calculated. Flatten each material area to be flattened in order of increasing distance. If the height of the material in the current material area to be flattened does not exceed the height of the mold box, then the material in the current material area to be flattened has been flattened. Flatten the next material area to be flattened.

[0046] The order in which the material to be flattened is determined by the distance between the area to be flattened and the inner wall of the mold box. The material to be flattened is then flattened in sequence to improve the efficiency of the flattening process.

[0047] During the leveling process, if the height of a certain area of ​​material to be leveled does not exceed the height of the mold box, it is considered that the area of ​​material to be leveled has been leveled, and the end scraper will then begin to level the next area of ​​material to be leveled.

[0048] The operation to obtain the distance between the current material area to be leveled and the inner wall of the mold box is as follows: Based on the planar dimensions of the mold box to be processed, the actual leveling working range is obtained; based on the actual leveling working range and the mapping relationship between the material area to be leveled and the three-dimensional area to be leveled, the simulated leveling working range corresponding to the actual leveling working range on the material three-dimensional model or the filtered material three-dimensional model is obtained; the closest distance between the current three-dimensional area to be leveled and the boundary of the simulated leveling working range is obtained, and the simulated distance is obtained; based on the simulated distance and the mapping relationship between the material area to be leveled and the three-dimensional area to be leveled, the distance between the current material area to be leveled and the inner wall of the mold box is obtained.

[0049] To improve computational efficiency, when obtaining the distance between the current material area to be leveled and the inner wall of the mold box, the points on the current material area to be leveled are used as the current tracking targets, and the points in other areas are used as the current non-tracking targets. The current non-tracking targets are masked (not displayed) to obtain the current simplified 3D model of the material, which is used to perform the operation of obtaining the simulated distance.

[0050] The material to be leveled is leveled in order of increasing distance from the previous material to the inner wall of the mold box. If multiple material areas are equidistant from the inner wall of the mold box, they are leveled in order of increasing distance from the previous material area.

[0051] If the height of the material in the current area to be leveled does not exceed the height of the mold box, then the material in the current area to be leveled has been leveled, and the next area to be leveled will be leveled.

[0052] This embodiment also provides a material leveling device, including:

[0053] The image acquisition unit is used to acquire images of the material on the mold box to be processed from multiple perspectives, and obtain a multi-view material image set.

[0054] The image processing unit is used to collect multi-view material images from the image acquisition unit, and after image enhancement and feature extraction, perform feature matching processing to obtain a current view classification map. All view classification maps form a multi-view material classification map set. Based on the material regions in the multi-view material classification map set and the corresponding position data of the material regions in the multi-view material image set, a three-dimensional model of the material is obtained. Protrusions in the three-dimensional material model with a height difference greater than a first threshold are used as three-dimensional regions to be flattened, and the material regions at the corresponding positions on the mold box to be processed are the material regions to be flattened. All material regions to be flattened form a set of material regions to be flattened. The unit is used to obtain the distance of each material region to be flattened from the inner wall of the mold box to be processed in the set of material regions to be flattened.

[0055] The motion control unit is used to flatten each material area to be flattened in sequence according to the order of distance from the material area to be flattened to the inner wall of the mold box in the image processing unit from small to large. During the flattening process, if the height of the material in the current material area to be flattened does not exceed the height of the mold box, then the material in the current material area to be flattened has been flattened, and the next material area to be flattened is flattened.

[0056] The device also includes: a feedback adjustment unit, used to instruct the image acquisition unit to acquire a multi-view image set of the material on the current mold box to be processed after flattening; to instruct the image processing unit to determine, based on the multi-view image set of the material on the current mold box to be processed after flattening, whether there are protrusions on the material with a height difference greater than a second threshold; if so, to update the material area to be flattened; and to instruct the motion control unit to flatten the updated material area to be flattened. This feedback adjustment unit, based on the actual operating effect of the motion control unit, performs re-identification through the image processing unit and adjusts and optimizes the motion path and actions of the motion control unit to improve the material flattening effect.

[0057] The leveling operation described above is achieved through a scraper at the end of a robotic arm that is positionally mapped to the camera in the image acquisition unit. The camera is positionally mapped to the material on the mold to be processed. A light source is located above the camera for supplemental lighting.

[0058] This embodiment provides a material leveling method that, by acquiring material images from different perspectives, performing image enhancement, feature extraction, and feature matching, accurately identifies the material in multiple perspective images, obtaining material regions in multiple perspective images; based on the material region information from multiple perspectives, a three-dimensional model of the material is established to determine the material region to be leveled; and according to the distance between the material region to be leveled and the inner wall of the mold box, the order in which the material region to be leveled is performed is determined, and the material regions to be leveled are leveled sequentially according to the order; this leveling method has good leveling effect, high leveling efficiency, and stable leveling quality.

Claims

1. A method for spreading materials, characterized in that, This includes the following operations: S1. Obtain images of the material on the mold box to be processed from multiple perspectives to obtain a multi-view material image set; In the multi-view material image set, the current view image is enhanced and feature extracted, and then feature matching is performed to obtain the current view classification image. All perspective classification diagrams form a multi-perspective material classification atlas; Based on the material regions in the multi-view material classification map set and the location data of the material regions in the multi-view material image set, a three-dimensional model of the material is obtained. S2. In the three-dimensional model of the material, the protrusions with a height difference greater than the first threshold are taken as the three-dimensional regions to be flattened. The material regions at the corresponding positions of the three-dimensional regions to be flattened on the mold box to be processed are the material regions to be flattened. All the material regions to be flattened form a set of material regions to be flattened. S3. Obtain the concentrated area of ​​material to be flattened, the distance of each area of ​​material to be flattened from the inner wall of the mold box to be processed, and flatten each area of ​​material to be flattened in order of increasing distance. During the leveling process, if the height of the material in the current area to be leveled does not exceed the height of the mold box, then the material in the current area to be leveled has been leveled, and the next area to be leveled is then leveled.

2. The material leveling method according to claim 1, characterized in that, The image enhancement operation in S1 is specifically as follows: The current view image is Gaussian blurred to obtain a smoothed image of the current view; the current view image and the smoothed image of the current view are subtracted at corresponding positions to obtain a detail image of the current view; the detail image of the current view is superimposed on the current view image to obtain an enhanced image of the current view. The current view augmentation map is used to perform feature extraction operations.

3. The material leveling method according to claim 1, characterized in that, The feature extraction operation in S1 is as follows: The current view image is enhanced by image enhancement processing to obtain the current view enhanced image. Gradient calculation is then performed to obtain the horizontal and vertical gradients of each pixel in the current view enhanced image. Based on the horizontal and vertical gradients of each pixel, the gradient magnitude corresponding to each pixel is obtained; In the current view enhancement map, the pixels with gradient magnitude greater than the second threshold are taken as edge points. All edge points are subjected to curve fitting to obtain the current view feature map containing edge contours. The current view feature map is used to perform feature matching processing.

4. The material leveling method according to claim 1, characterized in that, The feature matching process in S1 is as follows: The current view feature map obtained after image enhancement and feature extraction of the current view image is processed by marking detection boxes according to the edge contour to obtain a feature map to be matched containing multiple detection boxes. Obtain the similarity between the feature map of the region corresponding to the current detection box in the feature map to be matched and each standard feature map in the standard database. Take the classification label of the standard feature map corresponding to the maximum similarity as the classification result of the feature map of the region corresponding to the current detection box. The classification results of the feature maps of the regions corresponding to all detection boxes form the current view classification map.

5. The material leveling method according to claim 1, characterized in that, The specific steps for obtaining the three-dimensional model of the material in S1 are as follows: Obtain the multi-view material classification map set, and use the area where the material is located as the material area as the classification result; Randomly select a point on the mold box as the origin of coordinates, obtain the material area, and use the corresponding coordinate data of the material area relative to the origin of coordinates in the multi-view material image set as the position data; Based on the coordinate origin and position data, a three-dimensional model is established to obtain the three-dimensional model of the material.

6. The material leveling method according to claim 1, characterized in that, Following the operation in S1, the process further includes filtering the three-dimensional model of the material to obtain a filtered three-dimensional model of the material, which is then used to perform the operation in S2.

7. The material leveling method according to claim 1, characterized in that, In step S3, if multiple material areas to be flattened are at the same distance from the inner wall of the mold box, then the multiple material areas to be flattened are flattened sequentially in order of increasing distance from the previous material area to be flattened.

8. A material leveling device, characterized in that, include: The image acquisition unit is used to acquire images of the material on the mold box to be processed from multiple perspectives, and obtain a multi-view material image set. The image processing unit is used to collect the multi-view material images from the image acquisition unit, and after image enhancement and feature extraction, perform feature matching processing to obtain the current view classification image. All perspective classification images form a multi-perspective material classification image set; used to obtain a three-dimensional material model based on the material regions in the multi-perspective material classification image set and the corresponding position data of the material regions in the multi-perspective material image set; used to identify protrusions in the material three-dimensional model with a height difference greater than a first threshold as three-dimensional regions to be flattened, and the material regions at the corresponding positions of the three-dimensional regions to be flattened on the mold box to be processed as material regions to be flattened; all material regions to be flattened form a set of material regions to be flattened; used to obtain the distance from each material region to be flattened to the inner wall of the mold box to be processed in the set of material regions to be flattened; The motion control unit is used to flatten each material area to be flattened in sequence according to the order of distance from the material area to be flattened to the inner wall of the mold box in the image processing unit from small to large; during the flattening process, if the height of the material in the current material area to be flattened does not exceed the height of the mold box, then the material in the current material area to be flattened has been flattened, and the next material area to be flattened is flattened.

9. The material leveling device according to claim 8, characterized in that, The device further includes: The feedback adjustment unit is used to instruct the image acquisition unit to acquire a multi-view material image set of the material on the current mold box to be processed after flattening; instruct the image processing unit to determine whether there are protrusions on the material with a height difference greater than the second threshold based on the multi-view material image set of the material on the current mold box to be processed after flattening, and if so, update the material area to be flattened; and instruct the motion control unit to flatten the updated material area to be flattened.

10. The material leveling device according to claim 8, characterized in that, The flattening operation is achieved by a scraper at the end of a robotic arm that has a positional mapping relationship with the camera in the image acquisition unit.