A method and system for intelligent sorting of debris on a conveyor belt based on AI vision guidance

Abstract

This invention relates to the field of machine vision technology, specifically to an AI-guided intelligent sorting method and system for conveyor belt debris. The method first deblurs the image to be detected to eliminate motion blur and enhances the image to amplify the differences between debris and coal. Then, it achieves accurate debris identification through target detection, improving the accuracy of debris identification from the source. Next, it performs image segmentation based on the debris identification results to obtain the geometric and posture parameters of the debris, providing accurate target shape information for grasping. Finally, it combines the conveyor belt speed, debris detection position, maximum and minimum debris width, and the angle of the forward direction to generate a time-synchronized and posture-adapted motion path, which is then converted into a control signal to control the robot gripper to pick up the debris. This effectively solves the problems of debris identification deviation, grasping misalignment, and response lag in complex scenarios. When applied to the automated sorting of debris on conveyor belts in coal washing plants, it can improve sorting accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, specifically to an AI-guided intelligent sorting method and system for conveyor belt debris. Background Technology

[0002] Currently, automatic sorting systems for debris in coal washing plants are mainly deployed on vibrating screens and conveyor belts. Compared with sorting on vibrating screens, sorting debris on conveyor belts has multiple advantages: the conveyor belt runs more smoothly and the speed is controllable, making it easier to deploy intelligent identification and automated sorting systems, and significantly improving sorting efficiency and accuracy.

[0003] Currently, automated sorting equipment on conveyor belts in coal washing plants still faces numerous technical challenges in actual operation. In the identification stage, existing vision systems struggle to accurately distinguish coal blocks from debris such as wood, plastic, wire mesh, and iron, especially under conditions of high coal dust, changing lighting, or high-speed conveyor belt operation, where the misjudgment rate increases significantly. In the grasping stage, limitations in the precision of the robotic arm and the adaptability of the end effector make it difficult to accurately grasp irregular, thin, or misaligned debris, leading to frequent missed or incorrect grasps and impacting sorting efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-guided intelligent sorting method and system for conveyor belt debris.

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

[0006] A method for intelligent sorting of debris on a conveyor belt based on AI vision guidance includes the following operations:

[0007] S1. Acquire an image of the target detection area on the conveyor belt as the image to be detected; the image to be detected is processed by the trained target detection network to obtain the debris recognition detection result; the operation process in the trained target detection network is as follows: the image to be detected is deblurred to obtain the deblurred image to be detected; the deblurred image to be detected is enhanced to obtain the enhanced image to be detected; the enhanced image to be detected is processed by the target detection based on the detection box to obtain the debris recognition detection result.

[0008] S2. Based on the debris identification and detection results, perform image segmentation processing on the image to be detected, obtain the debris segmentation map, determine the maximum length, maximum length direction, maximum width, and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle.

[0009] S3. Based on the conveyor belt speed, the location of the object detection, the maximum width of the object, the minimum width of the object, and the angle of the forward direction, obtain the gripper's gripping waiting time and gripping posture information; based on the gripper's gripping waiting time, gripping posture information, and the robot gripper's initial pose information, obtain the robot gripper's motion path, convert the robot gripper's motion path into a motion control signal, and control the robot gripper to grab the object.

[0010] In S1, the deblurring process is as follows: calculate the blur length based on the conveyor belt speed, camera exposure time and resolution; obtain the blur kernel based on the blur kernel length and the horizontal linear motion blur model; perform noise suppression processing on the image to be detected to obtain the denoised image to be detected; and perform fuzz-iteration-based convolution processing on the denoised image to be detected based on the blur kernel to obtain the deblurred image to be detected.

[0011] The operation of fuzzy iteration-based convolution processing is implemented through the following formula:

[0012] ,

[0013] For the first k+1 The position in the deblurred image to be detected after the next iteration is ( x , y The estimated pixel value at position ) For the first k The position in the deblurred image to be detected after the next iteration is ( x , y The estimated pixel value at position ) The location in the denoised image to be detected is ( x , y The pixel value at ) The location in the image to be detected is ( x , y The fuzzy kernel value at ) This is for convolution processing.

[0014] In S1, the image enhancement processing operation is as follows: based on the gray value range of the coal block and the gray value range of the debris, the gray value image of the image to be detected and deblurred is subjected to piecewise linear gray-level stretching to obtain a gray-level stretched image; the gray-level stretched image is divided into several small blocks, the gray-level histogram of each small block is calculated, and histogram equalization is performed according to their respective contrast thresholds to obtain an equalized image; the equalized image is subjected to Laplacian sharpening to obtain the enhanced image to be detected.

[0015] In S3, the gripping posture information includes the gripper opening and the gripper rotation angle; the gripper rotation angle is the angle of the forward direction; the gripper opening is the sum of the maximum width of the debris and the reserved safety gap, or the sum of the minimum width of the debris and the reserved safety gap.

[0016] When the sum of the maximum width of the debris and the reserved safety gap is not greater than the gripper opening threshold, the gripper opening is the sum of the maximum width of the debris and the reserved safety gap, and the gripper gripping position is at the maximum width of the debris; when the sum of the maximum width of the debris and the reserved safety gap is greater than the gripper opening threshold, the gripper opening is the sum of the minimum width of the debris and the reserved safety gap, and the gripper gripping position is at the minimum width of the debris.

[0017] In S3, the gripper waiting time is the difference between the time it takes for the object to move from the object detection position to the position directly below the robot gripper and the response delay of the gripping device.

[0018] An AI-guided intelligent conveyor belt debris sorting system, used to implement the aforementioned AI-guided intelligent conveyor belt debris sorting method, includes:

[0019] Conveyor belts are used to transport coal and other materials.

[0020] The image acquisition unit is used to acquire images of the target detection area on the conveyor belt and send the images to the computing unit;

[0021] The computing unit is used to process the image using a training target detection network to obtain the debris recognition and detection results; and to perform image segmentation processing on the image to be detected based on the debris recognition and detection results, obtain a debris segmentation map, determine the maximum length, maximum length direction, maximum width, and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle; and to obtain the gripper's gripping waiting time and gripping posture information based on the conveyor belt speed, debris detection position, maximum width of the debris, minimum width of the debris, and forward direction angle; to obtain the robot gripper's motion path based on the gripper's gripping waiting time, gripping posture information, and robot gripper's initial pose information; and to convert the robot gripper's motion path into a motion control signal and transmit it to the control unit.

[0022] The control unit controls the robot to pick up objects based on the gripping signals transmitted from the computing unit.

[0023] The robot has grippers at its end for picking up miscellaneous items and moving them to a storage box.

[0024] Storage box for miscellaneous items;

[0025] The reset image acquisition unit has a vision camera mounted on top to acquire images of the gripper after each robot reset and send the images of the gripper to the computing unit. The computing unit determines whether the gripper is in its initial pose state based on the images of the gripper transmitted from the reset image acquisition unit. If not, it issues a gripper abnormality alarm.

[0026] The image acquisition unit is an L-shaped column. An industrial camera and a spotlight are installed at the end of the horizontal arm of the L-shaped column. The shooting direction of the industrial camera and the illumination direction of the spotlight are directly facing the target detection area on the conveyor belt.

[0027] The beneficial effects of this invention are as follows:

[0028] This invention provides an AI-guided intelligent sorting method for debris on conveyor belts. First, the image to be detected is deblurred to eliminate motion blur and image enhancement to amplify the differences between debris and coal. Then, target detection is used to achieve accurate debris identification, improving the accuracy of debris identification from the source. Next, image segmentation is performed based on the debris identification results to obtain the geometric and posture parameters of the debris, providing precise target shape information for grasping. Finally, by combining the conveyor belt speed, debris detection position, maximum and minimum debris width, and the angle of the forward direction, a time-synchronized and posture-adapted motion path is generated and converted into control signals to control the robot gripper to pick up the debris. This achieves full-process optimization from accurate identification to accurate grasping and timely response, effectively solving problems of debris identification deviation, grasping misalignment, and response lag in complex scenarios. Applied to automated debris sorting on conveyor belts in coal washing plants, it can improve sorting accuracy and efficiency. Attached Figure Description

[0029] The solutions and advantages of this application will become clear to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0030] In the attached diagram:

[0031] Figure 1 This is a schematic diagram of the system structure in this embodiment;

[0032] Figure 2 This is a schematic diagram of the gripper structure in this embodiment;

[0033] 1. Industrial camera; 2. Image acquisition unit; 3. Robot; 4. Gripper; 5. Vision camera; 6. Conveyor belt; 7. Miscellaneous storage box. Detailed Implementation

[0034] Exemplary embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings.

[0035] This embodiment provides an AI-guided intelligent sorting system for conveyor belt debris. See [link to documentation]. Figure 1 This method is suitable for automated sorting of debris on conveyor belts in coal washing plants. It implements an AI-guided intelligent sorting method for conveyor belt debris, including:

[0036] Conveyor belt 6 is used to transport coal and debris;

[0037] Image acquisition unit 2 is used to acquire images of the target detection area on the conveyor belt and send the images to the computing unit; wherein, the image acquisition unit is L-shaped column, and an industrial camera 1 and a spotlight are installed at the end of the horizontal arm of the L-shaped column. The shooting direction of the industrial camera and the illuminating direction of the spotlight are facing the target detection area on the conveyor belt.

[0038] The computing unit is used to process the image using a training target detection network to obtain the debris recognition and detection results; and to perform image segmentation processing on the image to be detected based on the debris recognition and detection results, obtain a debris segmentation map, determine the maximum length, maximum length direction, maximum width, and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle; and to obtain the gripper's gripping waiting time and gripping posture information based on the conveyor belt speed, debris detection position, maximum width of the debris, minimum width of the debris, and forward direction angle; to obtain the robot gripper's motion path based on the gripper's gripping waiting time, gripping posture information, and robot gripper's initial pose information; to convert the robot gripper's motion path into a motion control signal and transmit it to the control unit; and to determine whether gripper 4 is in the initial pose state based on the image of gripper 4 transmitted by the reset image acquisition unit after each robot reset; if not, to issue a gripper abnormality alarm.

[0039] The control unit is used to control the robot to pick up objects based on the gripping signal transmitted from the computing unit; and to control the vision camera 5 to acquire images of the gripper 4 after each robot reset.

[0040] Robot 3, with gripper 4 at its end (see...) Figure 2 The gripper 4 is used to pick up and place miscellaneous items into the miscellaneous item collection box 7. A grid-based stacking strategy is used to place the items sequentially, improving the space utilization of the collection box. After each placement, the gripper 4 returns to its initial position, maintaining its initial orientation.

[0041] The image acquisition unit is reset. A vision camera 5 is installed on the upper part of one side of the conveyor belt to acquire images of the gripper 4 after each robot reset and send the images of the gripper 4 to the computing unit.

[0042] Storage box for miscellaneous items.

[0043] The image acquisition unit 2 and robot 3 mentioned above have been processed by hand-eye calibration.

[0044] This embodiment also provides an AI-guided intelligent sorting method for conveyor belt debris, including the following operations:

[0045] S1. Acquire an image of the target detection area on the conveyor belt as the image to be detected; the image to be detected is processed by the trained target detection network to obtain the debris recognition and detection result;

[0046] S2. Based on the debris identification and detection results, perform image segmentation processing on the image to be detected, obtain the debris segmentation map, determine the maximum length, maximum length direction, maximum width, and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle.

[0047] S3. Based on the conveyor belt speed, the location of the object detection, the maximum width of the object, the minimum width of the object, and the angle of the forward direction, obtain the gripper's gripping waiting time and gripping posture information; based on the gripper's gripping waiting time, gripping posture information, and the robot gripper's initial pose information, obtain the robot gripper's motion path, convert the robot gripper's motion path into a motion control signal, and control the robot gripper to grab the object.

[0048] The specific steps and details are as follows.

[0049] S1. Obtain an image of the target detection area on the conveyor belt as the image to be detected; the image to be detected is processed by the trained target detection network to obtain the debris recognition detection result.

[0050] By acquiring images of the target detection area on the conveyor belt as the images to be detected, and training a target detection network, the system first removes motion blur caused by the high-speed operation of the conveyor belt, then uses image enhancement to increase the distinction between debris and coal, and finally performs target detection based on the detection boxes to obtain the debris recognition results. This approach can specifically solve the image quality problems caused by high-speed motion blur and coal dust interference in conveyor belt scenarios, effectively eliminate the obscuring of the debris outline by blur, enhance the visual difference between debris and coal, and then accurately locate and classify debris through the detection boxes, ultimately significantly improving the accuracy of debris recognition, reducing false detections and missed detections caused by image problems, and ensuring stable and accurate debris recognition in complex conveyor belt environments.

[0051] First, an image of the target detection area on the conveyor belt is acquired as the image to be detected. Specifically, an image acquisition unit is placed on one side of the conveyor belt in the coal washing plant. The industrial camera of the image acquisition unit and the spotlight are aligned with the conveyor belt, and the spotlight illuminates the conveyor belt, forming a uniformly bright area. At the same time, the camera's field of view is adjusted so that it covers the illuminated area, and a rectangular area covering the width of the conveyor belt is selected within the camera's field of view as the debris recognition and detection area, i.e., the target detection area. This avoids the impact of uneven lighting on the debris recognition effect. The industrial camera then captures an image of the target detection area to obtain the image to be detected.

[0052] The industrial camera continuously captures images at fixed intervals, acquiring images of debris within different time ranges of the target detection area. The fixed interval is the ratio of the target detection area length to the conveyor belt speed. This ensures that the tail information (e.g., the head of a debris) of the target detection area in the previous image is sequentially connected to the head information (e.g., the tail of a debris) of the target detection area in the current image, preventing any debris from being missed.

[0053] Then, the image to be detected is processed by the trained target detection network to identify the debris on the conveyor belt and obtain the debris detection result.

[0054] The method for obtaining the training target detection network is as follows: images of debris of different types and under different lighting conditions on the conveyor belt are acquired, and the corresponding types are marked to form a training dataset; the training dataset is used to train the target detection network to obtain the training target detection network.

[0055] The operations in training the object detection network are as follows: the image to be detected is deblurred to avoid motion blur caused by the conveyor belt running, resulting in a deblurred image; the deblurred image is then enhanced to improve the distinction between debris and coal on the conveyor belt, resulting in an enhanced image; the enhanced image is then processed by bounding box-based object detection to obtain the debris detection result. The bounding box-based object detection operation includes, but is not limited to, implementation using the Faster R-CNN model.

[0056] The specific steps for the above deblurring process are as follows.

[0057] Step 1: Calculate the blur length based on the conveyor belt speed, the camera exposure time and resolution of the industrial camera; obtain the blur kernel based on the blur kernel length and the horizontal linear motion blur model.

[0058] The fuzzy kernel is calculated using the following formula:

[0059] ,

[0060] ,

[0061] The location in the image to be detected is ( x , y The fuzzy kernel value at ) For fuzzy length, , , These are conveyor belt speed, camera exposure time, and resolution, respectively.

[0062] Step 2: To avoid the high-frequency noise of the image (caused by coal dust) being amplified during the deblurring process and affecting the blurring effect, noise suppression is performed on the image to be detected (this can be achieved by weighting each pixel with its neighboring pixels). This effectively filters out the speckle noise generated by coal dust and preserves the image edge information to the greatest extent, resulting in the denoised image to be detected.

[0063] Step 3: Based on the blur kernel, perform iterative convolution processing on the image to be detected to deblur the image, improve the image sharpness, and obtain the deblurred image to be detected.

[0064] The operation of fuzzy iteration-based convolution processing can be implemented using the following formula:

[0065] ,

[0066] For the first k+1 The position in the deblurred image to be detected after the next iteration is ( x , y The estimated pixel value at position ) For the first k The position in the deblurred image to be detected after the next iteration is ( x , y The estimated pixel value at position ) The location in the denoised image to be detected is ( x , y The pixel value at ) This is for convolution processing.

[0067] The specific steps for the above image enhancement processing are as follows.

[0068] Step 1: Based on the grayscale value range of the coal block and the grayscale value range of the impurities, perform piecewise linear grayscale stretching on the grayscale image of the image to be deblurred to expand the global grayscale difference between the impurities and the coal block, and obtain the grayscale stretched image.

[0069] The piecewise linear grayscale stretching process can be achieved using the following formula.

[0070] When the location in the deblurred image to be detected ( x , y grayscale value at ) When the first stretching threshold (used to prevent dark areas of the coal block from being mismapped to extreme dark areas) is reached, the grayscale of the dark area of ​​the coal block is mapped to the [0,80] range to preserve the internal layering of the coal block. The formula is: , The position in the grayscale stretch image ( x , y The grayscale value at ().

[0071] when (When the second stretching threshold is used to accurately separate the grayscale transition zone between coal blocks and debris, the grayscale span of the transition zone is expanded to enhance the grayscale distinction between coal blocks and debris. The formula is:) )+80.

[0072] when When the object is in a bright area, map its grayscale to the [200, 255] range to highlight its outline and avoid overexposure. The formula is: )+200.

[0073] The above , , , These are the minimum and maximum values ​​within the range of ash values ​​for the coal block, respectively. This represents the minimum value in the range of grayscale values ​​for the debris. , , All are empirical values.

[0074] Step 2: Divide the grayscale stretched image into several small blocks, calculate the grayscale histogram of each small block, and perform histogram equalization processing according to their respective contrast thresholds to optimize local contrast, eliminate the image haze caused by coal dust, and obtain an equalized image.

[0075] Step 3: Perform Laplacian sharpening on the equalized image to highlight the outline features of the debris and enhance the morphological distinction between the debris and the coal block, thus obtaining the enhanced image to be detected.

[0076] S2. Based on the debris identification and detection results, perform image segmentation processing on the image to be detected to obtain a debris segmentation map, determine the maximum length, maximum length direction, maximum width, and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle.

[0077] Based on the debris recognition results, image segmentation is performed to obtain a debris segmentation map. The maximum length, maximum length direction, maximum width, and minimum width of the debris are determined. The angle between the maximum length direction and the forward direction of the conveyor belt is calculated and used as the forward direction angle. This provides accurate geometric and attitude parameters of the debris for subsequent robot gripper path planning, reducing planning time to adapt to the high-speed operation of the conveyor belt and reducing gripping deviation caused by parameter ambiguity.

[0078] First, based on the clutter detection results, image segmentation processing is performed on the image to be detected (including but not limited to using the YOLOv8 model or GAN network) to extract the clutter image and obtain the initial segmented image; then, the clutter in the initial segmented image is fitted with edge contours to obtain the clutter segmentation map.

[0079] Then, determine the maximum length of the debris in the debris segmentation diagram and the direction in which the maximum length is located (the direction of the maximum length); and take the maximum width and minimum width of the debris in the debris segmentation diagram, which are perpendicular to the direction of the maximum length, as the maximum width and minimum width of the debris, respectively, for subsequent determination of the robot's gripper opening, and control the robot's gripper to grasp and pick up the debris from the width direction.

[0080] Finally, the angle between the maximum length of the debris and the forward direction of the conveyor belt is taken as the forward direction angle, which is used to adjust the posture of the robot gripper when picking up the debris.

[0081] S3. Based on the conveyor belt speed, the location of the object detection, the maximum width of the object, the minimum width of the object, and the angle of the forward direction, obtain the gripper's gripping waiting time and gripping posture information; based on the gripper's gripping waiting time, gripping posture information, and the robot gripper's initial pose information, obtain the robot gripper's motion path, convert the robot gripper's motion path into a motion control signal, and control the robot gripper to grab the object.

[0082] Based on parameters such as conveyor belt speed and object detection position, the gripping waiting time (matching the conveyor belt movement to avoid premature or delayed gripping) and gripping posture (adapting to the maximum or minimum width of the object, and combining the angle of the forward direction to ensure accurate gripping direction) are obtained. Then, the motion path is generated by combining the initial position of the gripper and converted into a control signal. This provides accurate parameters for path planning that fit the scene, avoids response delays that may cause missed gripping opportunities, and ultimately achieves accurate path planning, stable gripping without slippage, and timely response to adapt to high-speed conveyor belts, effectively improving gripping accuracy and response efficiency.

[0083] First, based on the conveyor belt speed, the location of the foreign object detection, the maximum width of the foreign object, the minimum width of the foreign object, and the angle of the forward direction, the gripper waiting time and gripping posture information are obtained.

[0084] The conveyor belt speed can be measured using existing auxiliary measuring instruments. Another method for obtaining the conveyor belt speed is as follows: select two images of the same debris detected along the coal flow direction within a rectangular area of ​​interest on the conveyor belt, record the time interval between the two images, and combine the pixel position difference of the debris's feature points in the two images (used to determine the distance the debris has moved with the conveyor belt) to obtain the conveyor belt speed.

[0085] The gripper waiting time is the difference between the time it takes for the debris to move from the debris detection position to the position directly below the robot gripper and the gripping device response delay (gripper response delay). The debris detection position is the position of the debris on the conveyor belt when the image to be detected is acquired.

[0086] The gripping posture information includes the gripper opening and the gripper rotation angle.

[0087] The claw rotation angle is the angle of the forward direction. If the angle of the forward direction is counterclockwise, the claw rotation angle is positive; if the angle of the forward direction is clockwise, the claw rotation angle is negative.

[0088] In addition, the gripper opening is the sum of the maximum width of the debris and the reserved safety gap, or the sum of the minimum width of the debris and the reserved safety gap. Specifically, when the sum of the maximum width of the debris and the reserved safety gap is not greater than the gripper opening threshold, the gripper opening is the sum of the maximum width of the debris and the reserved safety gap, and the gripper grips at the maximum width of the debris; when the sum of the maximum width of the debris and the reserved safety gap is greater than the gripper opening threshold, the gripper opening is the sum of the minimum width of the debris and the reserved safety gap, and the gripper grips at the minimum width of the debris.

[0089] Then, based on the gripper's gripping waiting time and gripping posture information, as well as the robot gripper's initial pose information, the robot's motion path is obtained.

[0090] Finally, the robot gripper's movement path is used as a motion control signal to control the robot gripper to pick up the debris.

[0091] During the gripper's grasping process, objects are grasped sequentially. If the gripper fails to grasp an object in the current grasping process, it means that the object in the previous grasping process is the same as the object in the current grasping process and has already been grasped in the previous grasping process. In this case, the gripper immediately returns to its initial position and executes the next grasping task.

[0092] This embodiment provides an AI-guided intelligent sorting method for conveyor belt debris. First, the image to be detected is deblurred to eliminate motion blur and the image is enhanced to amplify the differences between debris and coal. Then, target detection is used to achieve accurate debris identification, improving the accuracy of debris identification from the source. Next, image segmentation is performed based on the debris identification results to obtain the geometric parameters and posture parameters (angle of the forward direction) of the debris, providing accurate target shape basis for grasping. Finally, by combining the conveyor belt speed, debris detection position, maximum and minimum width of the debris, and angle of the forward direction, a time-synchronized and posture-adapted motion path is generated and converted into control signals to control the robot gripper to pick up the debris. The whole process is optimized from accurate identification to accurate grasping to timely response, effectively solving the problems of debris identification deviation, grasping misalignment, and response lag in complex scenarios. When applied to the automated sorting of debris on conveyor belts in coal washing plants, it can improve sorting accuracy and efficiency.

Claims

1. An AI vision-guided conveyor belt foreign matter intelligent sorting method, characterized in that, This includes the following operations: S1. Acquire an image of the target detection area on the conveyor belt as the image to be detected; the image to be detected is processed by the trained target detection network to obtain the debris recognition and detection result; The operation process in training the object detection network is as follows: the image to be detected is deblurred to obtain the deblurred image to be detected; The deblurred image to be detected undergoes image enhancement processing to obtain the enhanced image to be detected. The image enhancement processing operation is as follows: based on the gray value range of the coal block and the gray value range of the impurities, the gray value image of the deblurred image to be detected is subjected to piecewise linear gray-level stretching processing to obtain a gray-level stretched image; the gray-level stretched image is divided into several small blocks, the gray-level histogram of each small block is calculated, and histogram equalization processing is performed according to their respective contrast thresholds to obtain an equalized image; the equalized image is then subjected to Laplacian sharpening processing to obtain the enhanced image to be detected. The enhanced image to be detected is processed by target detection based on detection boxes to obtain the clutter detection result; S2. Based on the debris identification and detection results, perform image segmentation processing on the image to be detected, obtain the debris segmentation map, determine the maximum length, maximum length direction, maximum width and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle; S3. Based on the conveyor belt speed, debris detection location, maximum debris width, minimum debris width, and the angle of the forward direction, obtain the gripper waiting time and gripping posture information; the gripping posture information includes the gripper opening and the gripper rotation angle; the gripper rotation angle is the angle of the forward direction; the gripper opening is the sum of the maximum debris width and the reserved safety gap, or the sum of the minimum debris width and the reserved safety gap; when the sum of the maximum debris width and the reserved safety gap is not greater than the gripper opening threshold, the gripper opening is the sum of the maximum debris width and the reserved safety gap, and the gripper gripping position is at the maximum debris width; when the sum of the maximum debris width and the reserved safety gap is greater than the gripper opening threshold, the gripper opening is the sum of the minimum debris width and the reserved safety gap, and the gripper gripping position is at the minimum debris width. Based on the gripper's gripping waiting time, gripping posture information, and the robot gripper's initial pose information, the robot gripper's motion path is obtained. The robot gripper's motion path is then converted into a motion control signal to control the robot gripper to grasp the object.

2. The intelligent sorting method for conveyor belt debris based on AI vision guidance according to claim 1, characterized in that, In S1, the deblurring operation is as follows: The blur length is calculated based on the conveyor belt speed, camera exposure time, and resolution; the blur kernel is obtained based on the blur kernel length and the horizontal linear motion blur model; and the image to be detected is subjected to noise suppression processing to obtain the denoised image to be detected. Based on the fuzzy kernel, the image to be detected and denoised is subjected to fuzzy iteration-based convolution processing to obtain the image to be detected and deblurred.

3. The intelligent sorting method for conveyor belt debris based on AI vision guidance according to claim 2, characterized in that, The operation of fuzzy iteration-based convolution processing is implemented through the following formula: ， For the first k+1 The position in the deblurred image to be detected after the next iteration is ( x , y The pixel value at ) For the first k The position in the deblurred image to be detected after the next iteration is ( x , y The pixel value at ) The location in the denoised image to be detected is ( x , y The pixel value at ) The location in the image to be detected is ( x , y The fuzzy kernel value at ) This is for convolution processing.

4. The intelligent sorting method for conveyor belt debris based on AI vision guidance according to claim 1, characterized in that, The gripper waiting time in S3 is the difference between the time it takes for the object to move from the object detection position to the position directly below the robot gripper and the response delay of the gripping device.

5. An AI-guided intelligent sorting system for conveyor belt debris, characterized in that, The method for intelligent sorting of conveyor belt debris based on AI vision guidance as described in claim 1 is characterized by comprising: Conveyor belts are used to transport coal and other materials. The image acquisition unit is used to acquire images of the target detection area on the conveyor belt and send the images to the computing unit; The computing unit is used to process the image using a training target detection network to obtain the debris recognition and detection results; and to perform image segmentation processing on the image to be detected based on the debris recognition and detection results, obtain a debris segmentation map, determine the maximum length, maximum length direction, maximum width, and minimum width of the debris, and take the angle between the maximum length direction of the debris and the forward direction of the conveyor belt as the forward direction angle; and to obtain the gripper's gripping waiting time and gripping posture information based on the conveyor belt speed, debris detection position, maximum width of the debris, minimum width of the debris, and forward direction angle; to obtain the robot gripper's motion path based on the gripper's gripping waiting time, gripping posture information, and initial pose information of the robot gripper; and to convert the robot gripper's motion path into a motion control signal and transmit it to the control unit. The control unit controls the robot to pick up objects based on the gripping signals transmitted from the computing unit. The robot has grippers at its end for picking up miscellaneous items and moving them to a storage box. A storage box for miscellaneous items.

6. The AI ​​vision-guided intelligent sorting system for conveyor belt debris according to claim 5, wherein the image acquisition unit is an L-shaped column, and an industrial camera and a spotlight are installed at the end of the horizontal arm of the L-shaped column, with the shooting direction of the industrial camera and the irradiation direction of the spotlight facing the target detection area on the conveyor belt.

7. The AI ​​vision-guided intelligent sorting system for conveyor belt debris according to claim 5 further includes: The image acquisition unit is reset, with a vision camera mounted on the upper part to acquire images of the gripper after each robot reset, and then sends the images of the gripper to the computing unit. The calculation unit obtains the image of the gripper transmitted from the reset image unit and determines whether the gripper is in the initial pose state; if not, it issues a gripper abnormality alarm.

Images