A bimodal intelligent cotton foreign fiber sorting system and method

Abstract

The application discloses a bimodal fusion intelligent cotton foreign fiber sorting system and method, and belongs to the field of foreign fiber sorting technology.The sorting system comprises a shell, an image acquisition module, an image processing module, a conveying module and a rejection module, and specifically comprises an LED light source, a polarization camera, an RGB camera, an industrial computer, a camera support, a cotton flow channel, a spray valve control board and a high-speed spray valve.Cotton flow enters the cotton flow channel through a cotton inlet pipeline, image data is transmitted into the industrial computer through the image acquisition module, a detection result is quickly obtained through algorithm processing of the image processing module, and the high-speed spray valve is driven by the spray valve control board according to the detection result to remove foreign fibers.The application can realize real-time detection and removal of foreign fibers in cotton, and greatly improves the detection speed and accuracy.

Description

Technical Field

[0001] This invention relates to the fields of cotton processing, foreign fiber identification and removal, and specifically to a dual-modal intelligent cotton foreign fiber sorting system and method. Background Technology

[0002] Cotton is a crucial raw material in my country's textile industry. During cotton harvesting, it's inevitable that foreign fibers, such as polypropylene filaments, mulch film, irrigation tape, and even human hair, will be mixed in. These foreign fibers, when mixed with cotton, are broken into countless tiny fiber defects during processing. These defects are not only difficult to remove during textile processing but also easily broken or combed into even shorter, finer fibers during the impurity removal stage, forming numerous fibrous micro-defects. During spinning, these micro-defects can easily cause yarn breakage, reducing production efficiency; during weaving, they can affect the surface quality of the fabric; and during dyeing, they can lead to uneven dyeing of the fibers and fabric, affecting the appearance of the finished product.

[0003] Traditional methods for detecting foreign fibers in cotton mostly rely on manual visual inspection and simple physical sorting, which suffers from slow detection speed, low efficiency, and poor accuracy, and is also difficult to handle the complex types of foreign fibers and the diversity of cotton quality. In recent years, with the development of image processing and deep learning technologies, foreign fiber detection methods based on image recognition and intelligent algorithms have gradually become a research hotspot. However, existing technologies still face certain challenges in terms of the accuracy and real-time performance of foreign fiber identification. Therefore, a cotton foreign fiber detection and sorting system that combines image processing and deep learning technologies is needed to improve detection accuracy and processing efficiency.

[0004] Chinese patent document CN115201219A discloses a cotton foreign fiber sorting machine, aiming to solve the problems of low efficiency, high labor intensity, and unsatisfactory results of traditional manual foreign fiber sorting through automation technology. The sorting machine includes a main body, a conveyor belt mechanism, a dust blowing mechanism, a dust suction mechanism, a winding mechanism, and a controller. The conveyor belt is used to transport cotton at a uniform speed. The dust blowing and dust suction mechanisms work in conjunction with a CCD camera to automatically blow up and suck up foreign objects, ensuring accurate collection and preventing scattering. The winding mechanism organizes the cotton and stores it in a collection box after sorting. However, this solution may become a bottleneck when handling large quantities of cotton, and it may not be flexible enough in handling different types of foreign objects (such as lightweight plastic film and heavier drip irrigation tape). Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a cotton heterogeneous fiber sorting system and sorting method based on RGB and polarization image information fusion.

[0006] Technical Solution: The technical solution adopted by this invention to solve the problem is as follows: a dual-modal fusion intelligent cotton foreign fiber sorting system, including a housing, an image acquisition module, an image processing module, a conveying module, and a rejection module; the image acquisition module includes: a first RGB camera fixed to a first camera bracket, a second RGB camera fixed to a second camera bracket, a first polarization camera fixed to a third camera bracket, and a second polarization camera fixed to a fourth camera bracket; the conveying module includes a cotton flow channel with several LED light sources alternately installed on both sides; the image information processing module includes an industrial control computer installed inside the housing; the rejection module includes a spray valve control board located on the side of the cotton flow channel, which is wired to the industrial control computer via electrical signals and connected to an air tank and a high-speed spray valve via air pipes respectively.

[0007] Preferably, the housing is made of opaque aluminum alloy and the interior is coated with high-absorption black paint. The LED light source includes a white light source and an ultraviolet light source, which are fixed on both sides of the cotton flow channel. The light source is installed vertically on the same side of each camera's field of view, with the light source angle rotated 45 degrees clockwise or counterclockwise according to the direction of the cotton flow transport pipe.

[0008] This invention also provides a dual-modal fusion intelligent cotton foreign fiber sorting method, comprising the following steps:

[0009] S1: Read the RGB image and polarization image into the image processing module on the industrial computer;

[0010] S2: Reconstruct and fuse the RGB image and the polarization image using an autoencoder to generate a dual-modal fused image;

[0011] S3: Use deep learning detection algorithms to detect foreign fibers in dual-modal fused images, and simultaneously use vivid color detection algorithms to detect foreign fibers in RGB images;

[0012] S4: The spray valve control board merges the detection results of the dual-modal fusion image and the RGB image, and controls the high-speed spray valve to open when the foreign fiber reaches the rejection position, thereby rejecting the foreign fiber.

[0013] Preferably, the step in S2 of reconstructing and fusing the RGB image and the polarization image using an autoencoder to generate a dual-modal fused image includes: selecting a variational autoencoder model containing a multi-branch attention mechanism to process the input RGB image. and polarization image Encode into latent vectors respectively and The model training process simultaneously considers reconstruction accuracy and potential spatial distribution constraints using the loss function shown in the following formula, thereby obtaining a fused image that retains key features of visible light and polarization information while suppressing noise:

[0014]

[0015] In the formula, The output image after fusion and reconstruction. Represents reconstruction error ( (for sample number) Relative entropy is used to measure the latent distribution obtained by the model encoding. With prior distribution The difference is used to make the potential vector distribution more stable and improve the model's generalization ability. The weighting coefficients of the loss function effectively enhance the robustness of dual-modal feature fusion and the ability to capture details of heterogeneous fibers by balancing reconstruction accuracy and distribution constraints.

[0016] Preferably, in step S3, the detection of foreign fibers in the dual-modal fused image using a deep learning detection algorithm, and the simultaneous detection of foreign fibers in the RGB image using a vibrant color detection algorithm, includes: designing a guided deep instance segmentation network using the global characteristics of the dual-modal fused image and the prior features of the vibrant color detection algorithm. The network first uses the vibrant color detection algorithm to generate a foreign fiber saliency map. A feature extraction backbone network embedding a deep learning detection algorithm is used, and a saliency weight adjustment mechanism is added to each feature map layer. The feature weights are calculated using the following formula:

[0017]

[0018] In the formula, For the target layer feature weights, To fuse feature maps, This indicates a pixel-weighted operation. This is a multi-resolution interpolation of the saliency map.

[0019] Preferably, in step S4, the spray valve control board merges the detection results of the dual-modal fused image and the RGB image, and controls the high-speed spray valve to open when the foreign fiber reaches the rejection position, thereby rejecting the foreign fiber. Specifically, this includes the following steps:

[0020] S41: Use multi-frame image capture and target similarity analysis to predict the movement speed of defect targets, including continuous shooting of all objects in the cotton flow channel by a high-speed camera, recording the image frame number and timestamp, and ensuring that each impurity target is photographed at least twice;

[0021] S42: Perform feature matching on impurity targets in two consecutive images, and calculate their similarity using the shape contour, color information, and feature point distribution of the impurity targets. The similarity calculation formula is as follows:

[0022]

[0023] in, and Representing the target's first and second moments in frames 1 and 2 respectively. Each impurity target characteristic, For feature similarity measurement function, The total number of feature points; when the similarity If the threshold is exceeded, the imperfection target in both frames is determined to be the same imperfection target. Subsequently, its motion velocity vector is calculated using the positional changes of the imperfection target in the two frames. .

[0024] S43: After determining that the targets are the same defective object, further considering the target's offset and rotation characteristics during movement, the accuracy of removal is improved by optimizing the spray valve control signal. The target's center offset is calculated by analyzing the target's position and contour features in two consecutive frames of images. and the change in shape rotation angle Based on offset and rotation information, the target's motion path is corrected, and the simplified corrected velocity vector is:

[0025]

[0026] In the formula, The corrected velocity vector is used to dynamically adjust the trigger time and angle of the spray valve action. Δt represents the time interval between two adjacent frames.

[0027] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0028] (1) This invention combines data acquisition technologies from RGB and polarization cameras, utilizes an autoencoder for image reconstruction and fusion to generate a dual-modal fused image, and combines deep learning with a vibrant color detection algorithm to achieve high-precision detection of foreign fibers in cotton. The polarization image effectively reduces interference from ambient light sources, especially for objects with high reflectivity (such as plastic films and artificial fibers), making the detection system more robust and adaptable under complex lighting conditions.

[0029] (2) This invention employs multi-frame image capture technology, predicts the movement speed through impurity target similarity analysis, and calculates the corrected velocity vector of the target by combining offset and rotation characteristics, thereby realizing dynamic spray valve control. This method effectively solves the problem of rejection deviation caused by the difference in flight speed of cotton and foreign fibers in the channel and the complexity of their movement trajectory, greatly improving the accuracy of rejection action and meeting the needs of large-scale high-speed sorting.

[0030] (3) This invention significantly enhances the ability to process large-scale image data through the collaborative work of industrial control computer and FPGA, while ensuring high stability of the system during long-term operation. With the high-quality light source layout and housing design, noise interference and image deviation are reduced, making it suitable for industrial applications of large-volume cotton processing, and greatly improving production efficiency and sorting reliability. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the structure of the present invention;

[0032] Figure 2 This is a flowchart of the foreign fiber sorting method. Detailed Implementation

[0033] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0034] like Figure 1-2As shown, a cotton foreign fiber sorting system includes a housing 1, an image acquisition module, an image processing module, a transmission module, and a rejection module. Inside the housing 1, on both sides of the cotton flow channel, are installed a first LED light source 51, a second LED light source 52, a third LED light source 53, a fourth LED light source 54, a fifth LED light source 55, a sixth LED light source 56, a seventh LED light source 57, and an eighth LED light source 58. The image acquisition module and the transmission module are located inside the housing 1, and the image processing module is installed on the side of the housing. The image acquisition module includes an RGB camera 31 fixed to a first camera bracket and an RGB camera fixed to a second camera bracket. Camera 32, polarization camera 33 fixed to the third camera bracket, and polarization camera 34 fixed to the fourth camera bracket; the transmission module includes cotton flow channel 4, waste cotton pipe 10, and cotton outlet pipe 11; the image information processing module includes industrial control computer 6 and spray valve control board 7, which is connected to the first RGB camera 31, the second RGB camera 32, the first polarization camera 33, the second polarization camera 34, the first LED light source 51, the second LED light source 52, the third LED light source 53, the fourth LED light source 54, the fifth LED light source 55, the sixth LED light source 56, the seventh LED light source 57, and the eighth LED light source 58. The housing 1 is made of opaque aluminum alloy, with the interior coated with a high-absorption black paint. Inside, the cotton flow channel is flanked by alternating, sequentially arranged LED light sources 51, 52, 53, 54, 55, 56, 57, and 58. The first camera bracket 21 is positioned vertically between the first LED light source 51 and the second LED light source 52; the second camera bracket 22 is positioned vertically between the fourth LED light source 54 and the fifth LED light source 55; the third camera bracket 23 is positioned vertically between the third LED light source and the fourth LED light source 54; and the fourth camera bracket 24 is positioned between the seventh LED light source 57. The first LED light source 51 and the second LED light source 52 are installed at 45 degrees in opposite directions, perpendicular to the cotton flow channel 4, with the first RGB camera 31 as the center of the line scan field of view; the third LED light source 53 and the fourth LED light source 54 are installed at 45 degrees in opposite directions, perpendicular to the cotton flow channel 4, with the second RGB camera 32 as the center of the line scan field of view; the fifth LED light source 55 and the sixth LED light source 56 are installed at 45 degrees in opposite directions, perpendicular to the cotton flow channel 4, with the first polarization camera 33 as the center of the line scan field of view; the seventh LED light source 57 and the eighth LED light source 58 are installed at 45 degrees in opposite directions, perpendicular to the cotton flow channel 4, with the second polarization camera 34 as the center of the line scan field of view.

[0035] The housing 1, the first camera bracket 21, the second camera bracket 22, the third camera bracket 23, and the fourth camera bracket 24 are all independent of the cotton flow channel 4, which avoids the vibration generated by the fan during the transportation of cotton flow from causing vibration interference to the first RGB camera 31, the second RGB camera 32, the first polarization camera 33, and the second polarization camera 34 installed on the camera bracket.

[0036] The housing 1 has an opening on the side for easy installation, debugging, and maintenance of the camera. It is made entirely of aluminum, which helps dissipate heat from the internal camera equipment while isolating it from external light and interference signals. The cotton flow channel 4 has a horizontal length of 1.3 meters and a channel width of 20 centimeters. A frequency converter controls the speed of the fan in the cotton flow channel. The airflow generated by the fan provides the movement speed for the cotton entering the channel, thus regulating the cotton flow speed. The completely enclosed sides of the channel effectively reduce internal and external airflow interference and also provide good wear resistance and corrosion resistance.

[0037] The cotton raw material is input from the top of the top cotton flow channel 4. The encoder at the motor shaft sends a square wave signal to the spray valve control board 7. The square wave signal can be used to track the change of motor speed, thereby tracking the movement of the object. The spray valve control board 7 divides the square wave signal and outputs the frequency that triggers the camera to take pictures. The program written using Marix-SDK enables two RGB cameras and two polarization cameras to simultaneously acquire images of the cotton flow. Several LED light sources are installed on both sides of the cotton flow channel 4, with the camera's line scan field of view as the center, and rotated 45 degrees in opposite directions perpendicular to the cotton flow transport direction. This effectively supplements the light source, enabling the cameras to accurately capture images of the cotton raw material.

[0038] The first RGB camera 31, the second RGB camera 32, the first polarization camera 33, and the second polarization camera 34 all use line-scan industrial cameras. These four cameras are connected to the industrial control computer via a CameraLink interface to ensure image data transmission. The industrial control computer is equipped with an image acquisition card and a GPU. The acquisition card is responsible for reading image data, while the GPU performs rapid processing of large amounts of image information, ensuring the efficient operation of the detection algorithm within the industrial control computer. The industrial control computer and the spray valve control board are connected via a network cable to transmit image processing results. The spray valve control board is equipped with an onboard FPGA and an ARM core. The ARM core communicates with the industrial control computer to facilitate the acquisition of image processing results and to control the operation of corresponding modules on the FPGA. The FPGA is responsible for dividing the encoder square wave signal to output a stable square wave frequency to trigger the camera, storing different image processing results row-by-row in a FIFO, merging different image processing results, and then controlling the high-speed spray valve to remove foreign fibers from the cotton.

[0039] like Figure 2The diagram shows an algorithm flowchart for a cotton foreign fiber sorting method. A camera acquisition program is written, and through an industrial control computer and a spray valve control board, an external trigger mode is used to simultaneously control the first RGB camera 31, the second RGB camera 32, the first polarization camera 33, and the second polarization camera 34 to stably acquire image data. The RGB cameras acquire color image data of the cotton. The polarization cameras capture polarization characteristic images of foreign fibers, especially transparent or reflective fibers.

[0040] For RGB data, the OpenCV library is used to preprocess the RGB images, including denoising and color enhancement operations, to optimize image quality. Subsequently, the RGB images are converted to LAB and HSV color spaces respectively, and information from different color channels is extracted to determine if they match the corresponding target colors. A saturation channel threshold is also extracted to determine if the saturation meets the vibrant color standard. Based on preset threshold conditions, a vibrant color detection algorithm is applied to identify brightly colored foreign fibers, generating corresponding masking as the result of the vibrant color detection algorithm and generating a foreign fiber saliency map. .

[0041] RGB and polarization images are reconstructed and fused using an autoencoder. A variational autoencoder model with a multi-branch attention mechanism is used to encode the RGB and polarization images as latent vectors respectively. The loss function shown in the following equation balances reconstruction accuracy and latent spatial distribution constraints during model training, thereby obtaining a fused image that retains key features of visible light and polarization information while suppressing noise:

[0042]

[0043] In the formula, The output image after fusion and reconstruction. Represents reconstruction error ( (for sample number) Relative entropy is used to measure the latent distribution obtained by the model encoding. With prior distribution The difference is used to make the potential vector distribution more stable and improve the model's generalization ability. The weighting coefficients of the loss function enhance the robustness of dual-modal feature fusion and the ability to capture details of heterogeneous fibers by balancing reconstruction accuracy and distribution constraints.

[0044] The label-studio annotation tool was used to annotate the preprocessed cotton images, covering various types of foreign fibers such as polypropylene filaments, mulch film, irrigation tape, and human hair.

[0045] The image containing polarization and RGB fusion is paired with the RGB image. Using the label-studio annotation tool, bounding boxes are created to annotate the foreign fibers in the RGB image. Foreign fiber types include, but are not limited to, polypropylene fibers, mulch film, irrigation tape, and human hair. This yields a TXT annotation file corresponding to the image, containing the foreign fiber category and bounding box coordinates. By pairing the RGB image with the image containing fusion information, a vibrant color detection algorithm is used to generate a foreign fiber saliency map. The YOLO deep learning detection algorithm is embedded into the feature extraction backbone network, and a saliency weight adjustment mechanism is added to each feature map. The feature weights are calculated as shown in the following formula.

[0046] In the formula, For the target layer feature weights, To fuse feature maps, This indicates a pixel-weighted operation. This is a multi-resolution interpolation of the saliency map.

[0047] Multi-frame image capture and target similarity analysis are performed on the cotton flow to predict the movement speed of defective targets. Continuous high-frequency triggering is used for each camera to ensure that each impurity target is captured at least twice. Based on the triggering frequency, the time interval between image captures can be calculated, and the position coordinates of the detected impurities in different images can be calculated to predict the object's trajectory.

[0048] Similarity is calculated using the shape, color information, and feature point distribution of impurity targets. The similarity calculation formula is as follows:

[0049]

[0050] in, and Representing the target's first and second moments in frames 1 and 2 respectively. Each impurity target characteristic, For feature similarity measurement function, The total number of feature points; when the similarity If the threshold is exceeded, the object in both frames is determined to be the same object. The velocity vector of this object is calculated using its positional changes across the two frames. .

[0051] After determining that the targets are the same defective object, the accuracy of removal is further improved by optimizing the spray valve control signal, taking into account the target's offset and rotation characteristics during movement. The target's center offset is calculated by analyzing the target's position and contour features in two consecutive frames of images. and the change in shape rotation angle Based on offset and rotation information, the target's motion path is corrected, and the simplified corrected velocity vector is:

[0052]

[0053] In the formula, The corrected velocity vector is used to dynamically adjust the trigger time and angle of the spray valve action. Δt represents the time interval between two adjacent frames.

[0054] After calculating the trajectory of the impurity target, the processing result is generated as a corresponding mask. Areas with a value of 0 represent normal cotton, while areas with a value of 1 represent foreign fibers that need to be removed. Before sending the result to the spray valve control board, the detection results from both cameras need to be gradually merged. Since the right RGB camera and polarization camera are generally higher than the left, to ensure the integrity of the final data, the processing results from both images need to be merged before being sent to the lower-level machine. The complete detection result from the right camera is fused with a partial detection result from the left camera. Unmerged portions are processed in the next round of fusion to ensure the integrity of the final data. The final image mask result after each round of fusion is sent to the spray valve control board.

[0055] The spray valve control board stores the image detection results in a FIFO queue and outputs them to the spray valve control system to control the opening and closing of the spray valve. Based on the detection results, the time it takes for foreign fibers to reach the spray valve nozzle is calculated. Taking into account the distance between the camera and the nozzle and the speed of the cotton flow, the image delay time is precisely determined. This delay time is then matched with the dequeue frequency of the FIFO queue to ensure that the spray valve can accurately activate when the foreign fiber reaches the rejection position. When the detection result is "1", the spray valve opens to reject the foreign fiber; when the result is "0", the spray valve closes. This achieves efficient detection and accurate rejection of foreign fibers in cotton, improving the automation level and detection accuracy of cotton sorting.

[0056] The above-described specific embodiments are merely preferred embodiments of the present invention and are not intended to limit the implementation of the present invention or the scope of the claims. All equivalent changes and modifications made in accordance with the scope of patent protection of the present invention should be included within the scope of the present invention patent application.

Claims

1. A bimodal intelligent cotton foreign fiber sorting method, characterized in that: The application discloses a double-mode fusion intelligent cotton foreign fiber sorting system, which comprises a shell (1), an image acquisition module, an image processing module, a conveying module and a rejecting module. The image acquisition module comprises a first RGB camera (31) fixed on a first camera support (21), a second RGB camera (32) fixed on a second camera support (22), a first polarization camera (33) fixed on a third camera support (23) and a second polarization camera (34) fixed on a fourth camera support (24); the conveying module comprises a cotton flow channel (4) with a plurality of LED light sources alternately installed on both sides; the image processing module comprises an industrial computer (6) installed in the box; the rejecting module comprises a spray valve control board (7) located on the side of the industrial computer (6) and connected with the industrial computer (6) through an electric signal and a gas pipe, and connected with a gas tank (8) and a high-speed spray valve (9) through the gas pipe; the following steps are implemented: S1: reading the RGB image and the polarization image into the image processing module on the industrial computer; S2: reconstructing and fusing the RGB image and the polarization image using a self-encoder to generate a bimodal fusion image, including: selecting a variational self-encoder model containing a multi-branch attention mechanism, encoding the input RGB image and the polarization image into latent vectors and respectively, and simultaneously considering reconstruction accuracy and latent space distribution constraints in the model training process through a loss function shown in the following formula to obtain a fusion image that retains key features of visible light and polarization information and suppresses noise: ， In the formula, is the output image after reconstruction; represents the reconstruction error (D(x, x')) is the sample number; is the relative entropy, which is used to measure the difference between the latent distribution obtained by model coding and the prior distribution, so as to make the latent vector distribution more stable and improve the model generalization ability; is the difference between the prior distribution ; and is the weighting coefficient of the loss function, which balances the reconstruction accuracy and the distribution constraint, enhances the robustness of the bimodal feature fusion, and improves the ability to capture the details of the different fibers. S3: detecting the foreign fiber in the double-mode fusion image by using a deep learning detection algorithm, and simultaneously detecting the foreign fiber in the RGB image by using a bright color detection algorithm; S4: merging the detection results of the double-mode fusion image and the RGB image by the spray valve control board, and controlling the high-speed spray valve to be opened when the foreign fiber reaches the rejecting position so as to reject the foreign fiber.

2. A bimodal intelligent cotton foreign fiber sorting method according to claim 1, characterized in that: The S3 utilizes a deep learning detection algorithm to detect foreign fibers in the bimodal fusion image, and simultaneously utilizes a bright color detection algorithm to detect foreign fibers in the RGB image, including: utilizing the global characteristics of the bimodal fusion image and the prior characteristics of the bright color detection algorithm, designing a guided deep instance segmentation network, the network first utilizes the bright color detection algorithm to generate a foreign fiber saliency map A feature extraction backbone network embedded with a deep learning detection algorithm is added with a saliency weight adjustment mechanism in each layer of feature maps, and the feature weight is calculated through the following formula: ， In the formula, is a target layer feature weight, is a fused feature map, represents a pixel-wise weighting operation, is a multi-resolution interpolation of the saliency map.

3. A bimodal intelligent cotton foreign fiber sorting method according to claim 1, characterized in that: The S4 nozzle control panel merges the detection results of the bimodal fusion image and the RGB image, and controls the high-speed nozzle to open when the foreign fiber reaches the removal position to remove the foreign fiber, comprising: using multi-frame image capture and target similarity analysis to predict the motion speed of the defect target, including continuously shooting all objects in the cotton flow channel by a high-speed camera, recording the image frame number and time stamp, and ensuring that each impurity target is shot at least twice; then, the impurity targets in the two consecutive images are matched, and the similarity of the impurity targets is calculated by using the shape profile, color information and feature point distribution of the impurity targets Wherein the similarity calculation formula is ， in, and Representing the target's first and second moments in frames 1 and 2 respectively. Each impurity target characteristic, For feature similarity measurement function, The total number of feature points; when the similarity When the threshold is exceeded, the impurity target in the two frames is determined to be the same impurity target. Then, its motion velocity vector is calculated using the positional change of the impurity target in the two frames. .

4. A bimodal intelligent cotton foreign fiber sorting method according to claim 3, characterized in that: The position change of the impurity target in two frames of images is used to calculate the motion speed of the target in S4 After determining that the target is the same defect object, further combining the offset and rotation characteristics of the target in the motion process, the rejection accuracy is improved by optimizing the control signal of the spray valve, the position and contour characteristics of the target in the continuous two frames of images are analyzed, and the center offset of the target is calculated The motion path of the target is corrected based on the offset, and the simplified correction speed vector is: ， In the formula, is the corrected velocity vector, while the trigger time and position of the injection valve operation are dynamically adjusted, and Δt represents the time interval of adjacent two frames of images.

Images