Method for detecting broken slivers of feeding slivers of drawing frame based on visual identification

Abstract

The invention relates to a method for detecting broken strips of a feeding strip of a drawing frame based on visual identification, which comprises the following steps of: acquiring image data containing the feeding strip under various conditions, and obtaining a final feeding strip learning initial identification model based on the image data; feeding strip motion frames are collected and converted into gray images, and the number and positions of feeding strips are obtained; judging whether the obtained number and position of the feeding strips are consistent with the set number and position of the feeding strips or not, and if yes, preprocessing subsequent continuous frames to obtain a gray identification threshold value; obtaining a difference image of the current frame image and the previous preset number of frame images, and determining a current strip feeding state according to the difference image; and according to the gray level identification threshold value and the difference image, identifying the current feeding strip and confirming the motion state of the feeding strip, and according to the motion state of the feeding strip, processing according to a preset rule. According to the invention, an interval section detection means is adopted, so that the staggering of adjacent feeding strips in movement can be detected, movement is contained, and more stable and reliable operation is provided.

Description

Technical Field

[0001] The invention relates to the field of textile detection technology, and in particular to a detection method for broken or missing strips of a feed strip of a drawing frame based on visual recognition. Background Art

[0002] The existing means of detecting broken or missing strips in the feed strips of the textile industry mainly include: through-beam photoelectric broken strip detection, reflective infrared photoelectric multi-path detection, etc. Due to the detection principle of the photoelectric sensor, certain specific feed strip materials or scenes will affect the light transmittance, which is prone to misjudgment. At the same time, due to the low sensitivity of the sensor, it is easily interfered by other objects in a complex environment, resulting in misjudgment, and it is least sensitive to dark feed strips, so that the photoelectric sensor cannot correctly detect the reflection or scattering of light, resulting in limited application. In actual application, the strip often gets stuck in the photoelectric detection area, resulting in failure to work normally, because it is a point detection, and it is difficult to judge the actual situation of all the movement of the feed strip. Therefore, there is an urgent need for a detection method for broken or missing strips in the feed strips of the draw frame based on visual recognition, so as to better solve the related series of problems in actual applications, improve the accuracy and stability of detection, and provide better guarantee for long-term continuous operation. Summary of the invention

[0003] In view of the above problems, the present invention is proposed to provide a method for detecting broken or missing strips in the feed strips of a drawing frame based on visual recognition, which overcomes the above problems or at least partially solves the above problems.

[0004] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions: The embodiment of the present invention discloses a method for detecting broken or missing strips of a drawing frame feed based on visual recognition, comprising: S100. Build a feeding strip image data acquisition environment, collect image data containing feeding strips under various conditions, input the image data into the feeding strip learning initial recognition model for training, and obtain the final feeding strip learning initial recognition model; S200. Continuously collect the motion frames of the feed bar and convert them into grayscale images, input the grayscale images into the final feed bar to learn the initial recognition model, and obtain the number and position of the feed bar; S300. Determine whether the obtained number and position of the feed strips are consistent with the set number and position of the feed strips. If they are consistent, Gaussian blur, aperture effect compensation and learning threshold weighted average calculation are performed on subsequent continuous frames to finally obtain the grayscale recognition threshold; S400. Performing a differential operation on the current frame image and the previous preset number of frame images to obtain a differential image, and comparing the differential image with a specified motion threshold to determine the current feeding bar state; S500. According to the grayscale recognition threshold and the differential image, identify the current feed bar and confirm the motion state of the feed bar, and process according to the preset rules based on the motion state of the feed bar.

[0005] Furthermore, in S100, a feed bar image data acquisition environment is built to collect image data containing the feed bar under various conditions. The specific method includes: setting the feed bar image data acquisition environment, including multiple preset angles, lighting conditions and scenes, the photoresistor senses the intensity of the ambient light, and adjusts the brightness of the fill light LED according to the set threshold. The camera is initialized. The camera uses the CMOS image sensor OV5640. When working, the clock controls the photosensitive array to convert the image signal into an analog signal, which is then converted into a digital signal through an A / D converter and is read as a grayscale image through register configuration. The camera uses the IIC bus to drive the two-wire SCCB interface. The AI ​​smart chip configures the register address of the OV5640 camera through the SCCB protocol and controls the start and end of initialization.

[0006] Furthermore, in S100, the image data is input into the feed bar learning initial recognition model for training to obtain the final feed bar learning initial recognition model; the specific method includes: labeling the image data, using LabelImg to label the boundary box of the feed bar in each image, and assigning corresponding category labels; dividing the labeled data set into a training set, a validation set and a test set, and using the training set to train the YOLOv2 model; during the training process, monitoring the changes in the model's loss function and evaluation indicators, using the test set to test the final optimized model, and after the test is successful, obtaining the final feed bar learning initial recognition model.

[0007] Furthermore, in S200, the motion frames of the feed bars are continuously collected and converted into grayscale images, and the grayscale images are input into the final feed bar learning initial recognition model to obtain the number and position of the feed bars; the specific method includes: inputting the grayscale image into the final feed bar learning initial recognition model, the model calculates the image, performs YOLO2 operation on the calculation result, and draws and identifies the feed bar feature Blob to obtain the number and position of the feed bars.

[0008] Further, in S300, Gaussian blur, aperture effect compensation and learning threshold weighted average calculation are performed on the subsequent continuous frames, specifically including: using the dynamic frame in the feature blob area of ​​the identified feed strip, the dynamic frame is arranged in the feature blob area, and Gaussian blur filtering is repeatedly performed to eliminate noise and light pollution details to ensure regional uniformity. The Gaussian blur calculation formula is: ,in,( ) is the pixel coordinate in the image, ( , ) is the center coordinate of the Gaussian function, set to the center of the blob area, It is the standard deviation, which controls the blur degree of the Gaussian function. For each pixel in the blob area, the pixel values ​​around it are weighted averaged with the Gaussian function to obtain the blurred pixel value. The aperture effect compensation is added, and the weighted average of the learning value is used to calculate the grayscale recognition threshold number of the feed bar recognition area.

[0009] Further, in S400, when the differential image is larger than the specified motion threshold, the current feed bar state is a moving state, and when the differential image is larger than the specified motion threshold, the current feed bar state is a stopped state.

[0010] Furthermore, the grayscale recognition threshold of the feed bar recognition area identifies the current feed bar and confirms that the feed bar is in motion, uses corrosion operation to perform morphological filtering on the differential binary image of the moving feed bar, identifies the moving feed bar and completes the tracing of the feed bar in the visual recognition interval, and determines whether the tracing feed bar remains within the set area and its integrity. If it is in complete motion, it works normally, otherwise a broken bar signal will be given.

[0011] Furthermore, the grayscale recognition threshold of the feed bar recognition area identifies the current feed bar and confirms that the feed bar is in motion. When the distance between two adjacent feed bar targets is greater than or close to the set threshold, that is, when they are separated and intersected, a feed bar interlacing signal is determined and given; and for inclusion, if the distance between the two targets is less than the threshold, the feed bars overlap and move, and the inclusion signal is determined and given.

[0012] Furthermore, the grayscale recognition threshold of the feed bar recognition area recognizes the current feed bar and confirms that the feed bar is in motion. When the ambient light changes with the external sunrise and sunset, climate change, and indoor lighting factors, timed boundary environment learning is added. First, the environmental state variable is defined. Represents the environmental state vector at time t, which contains multiple environmental parameters, such as light intensity and shadow coverage, that is, ,in, represents the value of the i-th environmental parameter at time t; weighted average is used to update the understanding of the environment, assuming represents the learning function result at time t, is the weight of the i-th environmental parameter, then , based on the learning results of the past period of time, the prediction is Indicates time The predicted environmental state, is the predicted weight coefficient, , , …, , is the learning result of the past n time points, then .

[0013] The embodiment of the present invention further discloses an electronic device, including: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for detecting broken or missing strips in the feed strip of the drawing frame.

[0014] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least: The invention discloses a method for detecting broken or missing strips of a feed strip of a drawing frame based on visual recognition, comprising: building a feed strip image data acquisition environment, acquiring image data containing the feed strip under various conditions, inputting the image data into a feed strip learning initial recognition model for training, and obtaining a final feed strip learning initial recognition model; continuously acquiring feed strip motion frames and converting them into grayscale images, inputting the grayscale images into the final feed strip learning initial recognition model, and obtaining the number and position of the feed strips; judging whether the obtained number and position of the feed strips are consistent with the set number and position of the feed strips, and if they are consistent, performing Gaussian blurring, aperture effect compensation, and learning threshold weighted average evaluation calculation on subsequent continuous frames, and finally obtaining a grayscale recognition threshold; performing differential operation on a current frame image and a preset number of frame images, and obtaining a differential image, and comparing the differential image with a specified motion threshold to determine a current feed strip state; according to the grayscale recognition threshold and the differential image, identifying the current feed strip and confirming the motion state of the feed strip, and performing processing according to preset rules according to the motion state of the feed strip. The present invention adopts interval detection means based on visual recognition of the broken and missing feed strips of the drawing frame. The interval motion detection is more effective than the traditional photoelectric point detection that causes the normal operation to fail due to the stuck points. In addition, for feed strips of different colors, visual recognition does not have this disadvantage at all. It not only judges the motion state and the stop state during detection, but also can detect the staggered and included motion of adjacent feed strips in motion, providing more stable and reliable operation.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a method for detecting broken or missing strips in a draw frame feed based on visual recognition in Example 1 of the present invention; Figure 2 This is a schematic diagram of the structure of an electronic device in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0017] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0018] In order to solve the problems existing in the prior art, an embodiment of the present invention provides a method for detecting broken or missing strips of a drawing frame feeder based on visual recognition.

[0019] Example 1 This embodiment discloses a method for detecting broken or missing strips in the feed of a drawing frame based on visual recognition. Figure 1 ,include: S100. Build a feeding strip image data acquisition environment, collect image data containing feeding strips under various conditions, input the image data into the feeding strip learning initial recognition model for training, and obtain the final feeding strip learning initial recognition model; In S100 of this implementation, a feed bar image data acquisition environment is built to collect image data containing the feed bar under various conditions. The specific method includes: setting the feed bar image data acquisition environment, specifically including a plurality of preset angles, lighting conditions and scenes, the photoresistor senses the intensity of the ambient light, adjusts the brightness of the fill light LED according to the set threshold, and initializes the camera. The camera uses the CMOS image sensor OV5640. When working, the clock controls the photosensitive array to convert the image signal into an analog signal, and then converts it into a digital signal through the A / D converter, and specifies to read it as a grayscale image through the register configuration; the camera uses the IIC bus to drive the two-wire SCCB interface, and the AI ​​smart chip configures the register address of the OV5640 camera through the SCCB protocol, and controls the start and end of the initialization. After the camera initialization configuration is completed, the AI ​​smart chip initializes the neural network processor KPU and loads the feed bar network model. By building a feed bar image data acquisition environment, image data containing the feed bar under various conditions (different angles, lighting conditions and scenes) are collected.

[0020] In S100 of this embodiment, the image data is input into the feed bar learning initial recognition model for training to obtain the final feed bar learning initial recognition model; the specific method includes: labeling the image data, using LabelImg to label the bounding box for the feed bar in each image, and assigning the corresponding category label; dividing the labeled data set into a training set, a validation set and a test set, and using the training set to train the YOLOv2 model; during the training process, monitoring the changes in the loss function and evaluation index of the model, using the test set to test the final optimized model, and after the test is successful, the final feed bar learning initial recognition model is obtained. Specifically, a suitable model architecture is selected according to the complexity of the problem and the characteristics of the data. For image recognition tasks, commonly used model architectures include convolutional neural networks (CNN), ResNet, Inception, etc. This implementation will further optimize the model according to the model evaluation results, such as using data enhancement technology to increase the diversity of data, or using regularization methods to prevent overfitting. You can also try to use ensemble learning methods, such as bagging, boosting, etc., to improve the stability and accuracy of the model.

[0021] S200. Continuously collect the motion frames of the feed bar and convert them into grayscale images, input the grayscale images into the final feed bar to learn the initial recognition model, and obtain the number and position of the feed bar; In S200 of the present embodiment, the motion frames of the feed bars are continuously collected and converted into grayscale images, and the grayscale images are input into the final feed bar learning initial recognition model to obtain the number and position of the feed bars; the specific method includes: inputting the grayscale image into the final feed bar learning initial recognition model, the model calculates the image, performs YOLO2 operation on the calculation result, and draws and identifies the feed bar feature Blob to obtain the number and position of the feed bars.

[0022] Among them, the YOLO algorithm is a fast, efficient and accurate target detection algorithm. It is based on a convolutional neural network (CNN) and divides the input image into an S×S grid. Each grid unit predicts a fixed number of bounding boxes and category probabilities. Through a single forward pass operation, YOLOv2 can simultaneously predict the location and category of multiple objects in the image. YOLOv2 can process input images in real time and output the location and category information of multiple objects. This is due to its efficient convolutional neural network structure and single forward pass operation. YOLOv2 improves detection accuracy while maintaining real-time performance. This makes YOLOv2 more accurate in target detection tasks. The network structure of YOLOv2 has strong versatility and scalability. It can be applied to a variety of different target detection tasks, and can adapt to different application scenarios by adjusting the network structure and parameters.

[0023] S300. Determine whether the obtained number and position of the feed strips are consistent with the set number and position of the feed strips. If they are consistent, Gaussian blur, aperture effect compensation and learning threshold weighted average calculation are performed on subsequent continuous frames to finally obtain the grayscale recognition threshold; In S300 of this embodiment, Gaussian blur, aperture effect compensation and learning threshold weighted average evaluation calculation are performed on subsequent continuous frames, specifically including: using the dynamic frame in the feature blob area of ​​the identified feed strip, the dynamic frame is arranged in the feature blob area, and Gaussian blur filtering is repeatedly performed to eliminate noise and light pollution details to ensure regional uniformity. The Gaussian blur calculation formula is: ,in,( ) is the pixel coordinate in the image, ( , ) is the center coordinate of the Gaussian function, set to the center of the blob area, It is the standard deviation, which controls the blur degree of the Gaussian function. For each pixel in the blob area, the pixel values ​​around it are weighted averaged with the Gaussian function to obtain the blurred pixel value. The aperture effect compensation is added, and the weighted average of the learning value is used to calculate the grayscale recognition threshold number of the feed bar recognition area.

[0024] Specifically, Gaussian Blur is an image processing technique that uses normal distribution (also known as Gaussian distribution) to calculate the weighted average of each pixel in the image to achieve a blurring effect. The Gaussian Blur calculation steps include: Determine the size and standard deviation of the Gaussian kernel: Choose an appropriate Gaussian kernel size and standard deviation based on the desired blur level.

[0025] Calculate Gaussian kernel: Calculate the value of each element in the Gaussian kernel based on the two-dimensional Gaussian function.

[0026] Apply Gaussian kernel: Convolve the Gaussian kernel with the original image to get the blurred image. The convolution operation is to align the center of the Gaussian kernel with a pixel in the original image, then multiply each element in the Gaussian kernel with the pixel value at the corresponding position, and add all the products to get the blurred value of the pixel. Through the above steps, the Gaussian blur effect can be achieved. In image processing, Gaussian blur is often used in scenes such as removing noise and smoothing images.

[0027] In this embodiment, the aperture effect (or "exposure effect") may cause uneven image brightness due to factors such as lighting changes and lens aperture adjustment. The purpose of compensating for this effect is to keep the brightness of the image sequence consistent, thereby improving the visual experience. The specific method of compensating for the aperture effect includes: Detect aperture effect: Analyze the brightness changes of consecutive frames to detect whether there is an aperture effect.

[0028] Calculate compensation value: Calculate the compensation value for each frame based on the detected brightness changes.

[0029] Apply Compensation: Apply the calculated compensation value to each frame to adjust its brightness so that the brightness of the image sequence remains consistent.

[0030] In this embodiment, the learning threshold weighted average evaluation is an adaptive weighted average method, which combines the learning algorithm and the threshold control to achieve further smoothing and optimization of the image sequence. This method can adaptively adjust the weight according to the image content, thereby obtaining a better processing effect.

[0031] The learning threshold weighted average evaluation calculation method includes: Initialize learning parameters: Set the initial parameters of the learning algorithm, such as learning rate, threshold, etc.

[0032] Calculate weighted average: For each frame of image, calculate its weighted average based on the learning algorithm and threshold control.

[0033] Learning algorithm: adaptively adjusts weights based on image content.

[0034] Threshold control: Limit the range of weights to avoid the influence of extreme values.

[0035] Update learning parameters: Update the parameters of the learning algorithm according to the processing results to optimize the subsequent processing effect. Applying the above three methods in the processing of continuous frame images can obtain better image quality and visual experience. The specific steps are as follows: Gaussian blur is applied to each frame to obtain a smoothed image sequence.

[0036] Detect aperture effect and calculate compensation value for each frame to adjust image brightness.

[0037] The processed image sequence is further smoothed and optimized using a learning threshold weighted average evaluation method.

[0038] Through the above steps, high-quality processing of continuous frame images can be achieved, the clarity and stability of the images can be improved, thereby enhancing the visual experience.

[0039] S400. Perform a differential operation on the current frame image and a preset number of frame images to obtain a differential image, and compare the differential image with a specified motion threshold to determine the current feed bar state; specifically, when the differential image is larger than the specified motion threshold, the current feed bar state is a moving state, and when the differential image is larger than the specified motion threshold, the current feed bar state is a stopped state.

[0040] S500. According to the grayscale recognition threshold and the differential image, identify the current feed bar and confirm the motion state of the feed bar, and process according to the preset rules based on the motion state of the feed bar.

[0041] In some preferred embodiments, the grayscale recognition threshold of the feed bar recognition area identifies the current feed bar and confirms that the feed bar is in motion, uses corrosion operation to perform morphological filtering on the differential binary image of the moving feed bar, identifies the moving feed bar and completes the tracing of the feed bar in the visual recognition interval, and determines whether the tracing feed bar remains within the set area and its integrity. If it is in complete motion, it works normally, otherwise a broken bar signal will be given.

[0042] In some preferred embodiments, the grayscale recognition threshold of the feed bar identification area identifies the current feed bar and confirms that the feed bar is in motion. When the distance between two adjacent feed bar targets is greater than or close to the set threshold, that is, when they are separated and intersected, a feed bar interlacing signal is determined and given; and for inclusion, if the distance between the two targets is less than the threshold, the feed bars overlap and move, and the inclusion signal is determined and given.

[0043] In some preferred embodiments, the grayscale recognition threshold of the feed bar recognition area recognizes the current feed bar and confirms that the feed bar is in motion. When the ambient light changes with the external sunrise and sunset, climate change, and indoor lighting factors, timed boundary environment learning is added. First, the environmental state variable is defined. Represents the environmental state vector at time t, which contains multiple environmental parameters, such as light intensity and shadow coverage, that is, ,in, represents the value of the i-th environmental parameter at time t; weighted average is used to update the understanding of the environment, assuming represents the learning function result at time t, is the weight of the i-th environmental parameter, then , based on the learning results of the past period of time, the prediction is Indicates time The predicted environmental state, is the predicted weight coefficient, , , …, , is the learning result of the past n time points, then .

[0044] The present embodiment discloses a method for detecting broken or missing feed strips of a drawing frame based on visual recognition, comprising: building a feed strip image data acquisition environment, acquiring image data containing feed strips under various conditions, inputting the image data into a feed strip learning initial recognition model for training, and obtaining a final feed strip learning initial recognition model; continuously acquiring feed strip motion frames and converting them into grayscale images, inputting the grayscale images into the final feed strip learning initial recognition model, and obtaining the number and position of feed strips; judging whether the obtained number and position of feed strips are consistent with the set number and position of feed strips, and if so, performing Gaussian blurring, aperture effect compensation, and learning threshold weighted average evaluation calculation on subsequent continuous frames, and finally obtaining a grayscale recognition threshold; performing a differential operation on a current frame image and a preset number of frame images before, and obtaining a differential image, and comparing the differential image with a specified motion threshold to determine a current feed strip state; identifying the current feed strip and confirming the feed strip motion state according to the grayscale recognition threshold and the differential image, and performing processing according to the feed strip motion state according to preset rules. The present invention adopts interval detection means based on visual recognition of the broken and missing feed strips of the drawing frame. The interval motion detection is more effective than the traditional photoelectric point detection that causes the normal operation to fail due to the stuck points. In addition, for feed strips of different colors, visual recognition does not have this disadvantage at all. It not only judges the motion state and the stop state during detection, but also can detect the staggered and included motion of adjacent feed strips in motion, providing more stable and reliable operation.

[0045] Example 2 Based on the same inventive concept, an embodiment of the present disclosure also provides an electronic device. Figure 2 FIG. 1 is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Figure 2 As shown, an embodiment of the present disclosure provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement any optimization method in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, and are configured to implement information interaction between the processor and the memory.

[0046] Among them, the processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU), etc.; the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH); the I / O interface (read-write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus), etc.

[0047] In some embodiments, the processor 101 , the memory 102 , and the I / O interface 103 are connected to each other via a bus 104 , and further connected to other components of the computing device.

[0048] In some embodiments, the one or more processors 101 include a field programmable gate array.

[0049] According to an embodiment of the present disclosure, a computer-readable medium is further provided, wherein a computer program is stored on the computer-readable medium, wherein when the program is executed by a processor, the steps in any optimization method in the above-mentioned embodiment are implemented.

Claims

1. A method for detecting broken or missing strips of a drawing frame feed based on visual recognition, characterized in that: include: S100. Build a feeding strip image data acquisition environment, collect image data containing feeding strips under various conditions, input the image data into the feeding strip learning initial recognition model for training, and obtain the final feeding strip learning initial recognition model; S200. Continuously collect the motion frames of the feed bar and convert them into grayscale images, input the grayscale images into the final feed bar to learn the initial recognition model, and obtain the number and position of the feed bar; S300. Determine whether the obtained number and position of the feed strips are consistent with the set number and position of the feed strips. If they are consistent, Gaussian blur, aperture effect compensation and learning threshold weighted average calculation are performed on subsequent continuous frames to finally obtain the grayscale recognition threshold; S400. Performing a differential operation on the current frame image and the previous preset number of frame images to obtain a differential image, and comparing the differential image with a specified motion threshold to determine the current feeding bar state; S500. According to the grayscale recognition threshold and the differential image, identify the current feed bar and confirm the motion state of the feed bar, and process according to the preset rules based on the motion state of the feed bar.

2. A method for detecting broken or missing strips of a draw frame feed based on visual recognition as claimed in claim 1, characterized in that: In S100, a feed bar image data acquisition environment is built to collect image data containing the feed bar under various conditions. The specific method includes: setting the feed bar image data acquisition environment, including multiple preset angles, lighting conditions and scenes, the photoresistor senses the intensity of the ambient light, adjusts the brightness of the fill light LED according to the set threshold, and initializes the camera. The camera uses the CMOS image sensor OV5640. When working, the clock controls the photosensitive array to convert the image signal into an analog signal, which is then converted into a digital signal through an A / D converter and is specified to be read as a grayscale image through register configuration; the camera uses the IIC bus to drive the two-wire SCCB interface inside, and the AI ​​smart chip configures the register address of the OV5640 camera through the SCCB protocol, and controls the start and end of initialization.

3. A method for detecting broken or missing strips of a draw frame feed based on visual recognition as claimed in claim 1, characterized in that: In S100, the image data is input into the feed strip learning initial recognition model for training to obtain the final feed strip learning initial recognition model; the specific method includes: labeling the image data, using LabelImg to label the boundary box of the feed strip in each image, and assigning corresponding category labels; dividing the labeled data set into a training set, a validation set and a test set, and using the training set to train the YOLOv2 model; during the training process, monitoring the changes in the model's loss function and evaluation indicators, using the test set to test the final optimized model, and after the test is successful, obtaining the final feed strip learning initial recognition model.

4. A method for detecting broken or missing strips in a draw frame feed based on visual recognition as claimed in claim 1, characterized in that: In S200, the motion frames of the feed bars are continuously collected and converted into grayscale images, and the grayscale images are input into the final feed bar learning initial recognition model to obtain the number and position of the feed bars; The specific method includes: inputting the grayscale image into the final feed bar learning initial recognition model, the model calculates the image, performs YOLO2 operation on the calculation result, and draws the recognition feed bar feature Blob to obtain the number and position of the feed bars.

5. The method for detecting broken or missing strips of a draw frame feed based on visual recognition according to claim 1, characterized in that: In S300, Gaussian blur, aperture effect compensation and learning threshold weighted average calculation are performed on the subsequent continuous frames, specifically including: using the dynamic frame in the feature blob area of ​​the identified feed strip, the dynamic frame is arranged in the feature blob area, and Gaussian blur filtering is repeatedly performed to eliminate noise and light pollution details to ensure regional uniformity. The Gaussian blur calculation formula is: ,in,( ) is the pixel coordinate in the image, ( , ) is the center coordinate of the Gaussian function, set to the center of the blob area, It is the standard deviation, which controls the blur degree of the Gaussian function. For each pixel in the blob area, the pixel values ​​around it are weighted averaged with the Gaussian function to obtain the blurred pixel value. The aperture effect compensation is added, and the weighted average of the learning value is used to calculate the grayscale recognition threshold number of the feed bar recognition area.

6. A method for detecting broken or missing strips in a draw frame feed based on visual recognition as claimed in claim 1, characterized in that: In S400, when the difference image is larger than the specified motion threshold, the current feeding bar state is a moving state, and when the difference image is larger than the specified motion threshold, the current feeding bar state is a stopped state.

7. A method for detecting broken or missing strips in a draw frame feed based on visual recognition as claimed in claim 6, characterized in that: The grayscale recognition threshold of the feed bar recognition area identifies the current feed bar and confirms that the feed bar is in motion, uses corrosion operation to perform morphological filtering on the differential binary image of the moving feed bar, identifies the moving feed bar and completes the tracing of the feed bar in the visual recognition interval, and determines whether the tracing feed bar remains within the set area and its integrity. If it is in complete motion all the time, it will work normally, otherwise a broken bar signal will be given.

8. The method for detecting broken or missing strips of a draw frame feed based on visual recognition as claimed in claim 6, characterized in that: The grayscale recognition threshold of the feed bar recognition area identifies the current feed bar and confirms that the feed bar is in motion. When the distance between two adjacent feed bar targets is greater than or close to the set threshold, that is, when they are separated and intersected, a feed bar interlacing signal is determined and given; and for inclusion, if the distance between the two targets is less than the threshold, the feed bars overlap and move, and an inclusion signal is determined and given.

9. A method for detecting broken or missing strips in a draw frame feed based on visual recognition as claimed in claim 6, characterized in that: The grayscale recognition threshold of the feed bar recognition area recognizes the current feed bar and confirms that the feed bar is in motion. When the ambient light changes with the external sunrise and sunset, climate change, and indoor lighting factors, timed boundary environment learning is added. First, the environmental state variable is defined. Represents the environmental state vector at time t, which contains multiple environmental parameters, such as light intensity and shadow coverage, that is, ,in, represents the value of the i-th environmental parameter at time t; weighted average is used to update the understanding of the environment, assuming represents the learning function result at time t, is the weight of the i-th environmental parameter, then , based on the learning results of the past period of time, the prediction is Indicates time The predicted environmental state, is the predicted weight coefficient, , , …, , is the learning result of the past n time points, then .

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for detecting broken or missing strips in the feed strip of the drawing frame as described in any one of claims 1 to 9.

Images