Lyocell fiber production continuous feeding control method and system based on image recognition

Abstract

The invention relates to the field of intelligent control based on image recognition, and discloses a Lyocell fiber production continuous feeding control method and system based on image recognition, and in the method, a feeding and conveying device continuously feeds a pulp raw material into a swelling machine at a preset feeding speed without shutdown; moreover, a camera arranged in the swelling machine obtains a to-be-detected liquid level image in the swelling machine containing the circular liquid level boundary in the non-stop continuous feeding state, then the current liquid level height in the swelling machine is obtained according to the to-be-detected liquid level image, the current liquid level height is compared with the target liquid level height, and a liquid level difference value is obtained; and the liquid level difference value serves as a control factor, the opening degree of a solvent conveying pump at a solvent inlet of the swelling machine is controlled based on a PID control algorithm, and the current liquid level height in the swelling machine is kept within a set range. Waste of production time is avoided, the large-yield requirement is met, and the overall production efficiency is improved; in addition, it is also ensured that the produced pulp porridge has good uniformity.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control based on image recognition, and more specifically to a method and system for controlling continuous feeding of lyocell fiber production based on image recognition. Background Art

[0002] In the traditional lyocell production process, the pulp raw material is fed and swelled in an intermittent feeding and swelling process. This method requires downtime between each batch, which greatly wastes production time. Nowadays, the swelling and dissolution processes are constantly improving, and the production capacity and performance of unit equipment have been greatly improved. The existing intermittent feeding and swelling mode is difficult to meet the needs of large production, and cannot fully adapt to the progress and development of the current swelling and dissolution processes.

[0003] In addition, due to the characteristics of lyocell fiber, it has extremely high requirements for pulp uniformity. Only when the pulp has sufficient uniformity can high-quality glue be prepared. However, the current intermittent feeding is affected by the differences in batch operation, which will cause each batch of materials to swell differently in the swelling machine, and thus make the prepared pulp less uniform in key indicators such as temperature and viscosity. Therefore, it is urgent to find a new way to feed pulp raw materials to solve the above technical problems. Summary of the invention

[0004] In order to solve the problems and shortcomings of the prior art, the present invention proposes a continuous feeding control method and system for lyocell fiber production based on image recognition. The present invention continuously feeds the pulp raw material for lyocell production into a swelling machine for swelling through a feeding and conveying device without stopping, which can significantly improve production efficiency, meet the large output requirements brought about by equipment capacity and performance improvements, and better adapt to the current progress in dissolution technology; and, based on precise liquid level detection and control technology, it also ensures that the pulp porridge produced by continuous feeding has good uniformity and consistency.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows: On one hand, the present invention discloses a method for controlling continuous feeding of lyocell fiber production based on image recognition, and the method is specifically as follows: The feeding and conveying device continuously feeds the pulp raw material into the swelling machine at a preset feeding speed; The camera arranged in the swelling machine obtains the liquid level to be detected image containing the fuzzy liquid level boundary in the swelling machine under the continuous feeding state; Smoothing the liquid level image to be detected, enhancing the area of ​​the image that meets the circular boundary characteristics according to the continuity of the image gradient direction, obtaining a liquid level edge map of the fuzzy liquid level boundary image, and estimating the current liquid level height according to the liquid level edge map; Compare the current liquid level with the target liquid level to obtain the liquid level difference; The liquid level difference is used as a control factor, and the opening of the solvent delivery pump at the solvent inlet of the swelling machine is controlled based on the PID control algorithm to keep the current liquid level height in the swelling machine within the set range.

[0006] Preferably, in the present invention, the liquid level image to be detected is smoothed, and according to the continuity of the image gradient direction, the area of ​​the image that meets the circular boundary characteristics is enhanced to obtain a liquid level edge map of the fuzzy liquid level boundary image, and the current liquid level height is estimated according to the liquid level edge map, which specifically includes the following steps: Step A, performing grayscale conversion on the image to be detected of the liquid level to obtain a preprocessed grayscale image, and performing Gaussian smoothing on the preprocessed image by using a Gaussian kernel function to obtain a smoothed image; Step B, calculating the gradient magnitude and gradient direction of each pixel in the smoothed image; Step C, using a sliding window to analyze the smoothed image, determine the continuity of the gradient direction of each pixel in the sliding window, enhance the area of ​​the image that meets the circular boundary characteristics, and weaken the area that does not meet the circular boundary characteristics; Step D, using the Canny algorithm to obtain a rough edge map of the smoothed image, then using Hough transform to detect the circular boundary of the rough edge map, obtaining circular parameters, and enhancing the rough edge map based on the circular parameters; Step E: Continue to use the Canny algorithm to process the enhanced rough edge map to generate a single-scale edge map; Step F: Use a new Gaussian kernel function to smooth the grayscale converted liquid level image to be detected. Then, the above steps B to E are performed and repeated N-1 times, where N is an integer greater than 1, so as to obtain N different single-scale edge maps; the N different single-scale edge maps are processed by multi-scale processing fusion to obtain a final liquid level edge map; Step G, performing threshold detection on the final liquid level edge map to obtain a binary edge map, then applying morphological operations to optimize the binary edge map, and finally performing Hough transform on the optimized binary edge map to detect circular boundaries and obtain the detection radius of the final liquid level edge map after Hough transform; Step H: Calculate the current liquid level height in the swelling machine according to the detection radius of the final liquid level edge map.

[0007] Preferably, the step A specifically includes the following contents: Step A1: Convert the collected liquid level image to a grayscale image I g ( x ,y ), calculated as follows: Formula (1); in, x and y Respectively represent the coordinates of each pixel point in the collected liquid level image to be detected; R , G , B Respectively represent the intensity values ​​of the three color channels of each pixel; Step A2: Use adaptive histogram to adjust the grayscale image I g ( x , y ) is equalized to obtain a contrast-enhanced preprocessed image I e ( x , y ); Step A3: Use Gaussian kernel function to I e ( x , y ) to obtain a smoothed image I σ ( x , y ), calculated as follows: Formula (2); Formula (3); in, G ( x , y , s ) represents the Gaussian kernel function; s represents the standard deviation of the Gaussian kernel; s 2 Represents the variance of the Gaussian distribution.

[0008] Preferably, the step B specifically includes the following contents: Step B1: Use Sobel operator to calculate the smoothed image I σ ( x , y ), the horizontal gradient of each pixel G x and vertical gradient G y , calculated as follows: Formula (4); Formula (5); in, xand y Respectively represent the coordinates of each pixel; Step B2: According to the horizontal gradient of the pixel point G x and vertical gradient G y , calculate the gradient direction and initial gradient amplitude of the pixel point, the calculation method is as follows: Formula (6); Formula (7); in, G ( x , y ) represents the initial gradient amplitude of the pixel; i ( x , y ) represents the gradient direction of the pixel.

[0009] Preferably, step C comprises the following steps: Step C1: Use a sliding window to smooth the image I σ ( x , y ) to analyze and calculate the gradient direction difference between each pixel in the sliding window and its adjacent pixels. The calculation method is as follows: Formula (8); Formula (9); in, x and y Respectively represent the coordinates of the pixel points; Δ i row ( x , y ) represents the row direction difference; Δ i col ( x , y ) represents the column direction difference; i ( x , y ) represents the gradient direction of the current pixel; i ( x+ 1, y )and i ( x , y+ 1) Indicates the gradient direction of adjacent pixels; Step C2: Calculate the average value of the gradient direction difference in the sliding window. The calculation method is as follows: Formula (10); in, Represents the average value of the gradient direction difference within the sliding window; N is the total number of gradient direction differences within the sliding window; Step C3, adjusting the gradient amplitude of each pixel in the sliding window according to the continuity of the gradient direction; like , indicating that the gradient direction of the pixels in the sliding window is continuous, then the gradient amplitude of each pixel in the enhanced sliding window is calculated as follows: Formula (11); like , indicating that the gradient direction of the pixel point in the sliding window jumps, which is the interference area, then the gradient amplitude of each pixel point in the sliding window is suppressed, and the calculation method is as follows: Formula (12); like , then the gradient amplitude of each pixel in the sliding window remains unchanged; in, G ( x , y ) represents the initial gradient amplitude of the pixel; Represents the gradient amplitude of the pixel after adjustment according to the continuity of the gradient direction; k 1 and k 2 represent the enhancement coefficient and suppression coefficient respectively.

[0010] Preferably, the step D comprises the following steps: Step D1: Smooth the image I σ ( x , y ) Use the Canny algorithm to generate a rough edge map E pre ( x , y ); Step D2: Rough edge map E pre ( x , y ) Use Hough transform to detect circular boundaries and obtain a rough edge map E pre ( x , y ) of the circle, including the coordinates of the center of the circle ( x c , y c ) and the initial radius r pre ; Step D3: Calculate a rough edge map Epre ( x , y ) and the center of the circle. If the distance is less than the set value, the gradient amplitude of the pixel is further enhanced. The calculation method is as follows: Formula (13); Formula (14); in, Represents the gradient amplitude of the pixel after the second enhancement; k 3 represents the enhancement factor; d ( x , y ) represents the distance from the pixel to the center of the circle.

[0011] Preferably, the step E comprises the following steps: Step E1: along the gradient direction i ( x , y ), compare the gradient amplitude of each pixel after the second enhancement The gradient amplitude of two adjacent pixels in its neighborhood retains the local maximum and suppresses the non-maximum value; Step E2, performing double threshold detection on the gradient amplitude of the pixel after the second enhancement, and classifying the gradient amplitude after non-maximum suppression into strong edge, weak edge or non-edge; Step E3: Connect the strong edges and the adjacent weak edges to generate a single-scale edge map.

[0012] Preferably, in step F, a plurality of different single-scale edge maps are fused through multi-scale processing to obtain a final liquid level edge map, and the calculation method is as follows: Formula (15); in, is the final liquid level edge map; w 1. w 2 and w 3 is the weight coefficient; , , It is a single-scale edge map generated after processing based on different Gaussian kernel functions.

[0013] Preferably, in step G, the morphological operation includes dilation and erosion.

[0014] Preferably, in the step H, the current liquid level height in the swelling machine is calculated according to the detection radius of the final liquid level edge map, and the calculation method is as follows: Formula (16); in,h is the current liquid level in the swelling machine; r is the detection radius of the final liquid level edge map; h max It is the reference value for calibrating the maximum liquid level in the swelling machine; h min It is the reference value for calibrating the minimum liquid level height in the swelling machine; r max The image radius corresponding to the reference value of the maximum liquid level calibration in the swelling machine; r min The image radius corresponding to the reference value of the minimum liquid level calibration in the swelling machine.

[0015] Preferably, the liquid level difference is used as a control factor to control the opening of the solvent delivery pump at the solvent inlet of the swelling machine based on a PID control algorithm, comprising the following steps: According to the liquid level difference, the control amount of the valve opening is calculated based on the following calculation expression: Formula (17); in, Adjust the increment for the solvent delivery pump opening. represents the proportional term, is the integral term, is the differential term; e ( t ) is the liquid level difference; K p represents proportional gain; K i represents the integral gain; K d represents the differential gain; Combination And the current opening of the solvent delivery pump, calculate the target opening of the solvent delivery pump, the calculation method is as follows: Formula (18); in, Indicates the current opening degree of the solvent delivery pump; Indicates the target opening of the solvent delivery pump; If the liquid level is lower than the set value, increase the opening of the solvent delivery pump to the target opening; if the liquid level is higher than the set value, reduce the opening of the solvent delivery pump to the target opening.

[0016] Preferably, the feeding and conveying device is a screw conveyor or a conveyor belt.

[0017] Preferably, the camera is a CCD industrial camera.

[0018] Based on the same inventive concept, the present invention also proposes a continuous feeding control system for Lyocell fiber production based on image recognition, the system is used to implement the above-mentioned continuous feeding control method, and is characterized by comprising: A feeding and conveying device, used for continuously feeding pulp raw materials into the swelling machine at a preset feeding speed; A camera is used to obtain a liquid level image to be detected containing a fuzzy liquid level boundary in the swelling machine under a continuous feeding state; A camera controller, used for controlling the camera to collect the image of the liquid level to be detected in the swelling machine; A liquid level detection module, used to obtain the current liquid level height in the swelling machine according to the liquid level to be detected image transmitted by the camera; The liquid level control module is used to compare the current liquid level height with the target liquid level height to obtain the liquid level difference, and use the liquid level difference as a control factor to control the opening of the solvent delivery pump at the solvent feed inlet of the swelling machine based on the PID control algorithm to keep the liquid level difference within a set range.

[0019] On the other hand, the present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein when the processor executes the computer program, the above-mentioned continuous feeding control method is implemented.

[0020] On the other hand, the present invention further proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed in a computer processor, the above-mentioned continuous feeding control method is implemented.

[0021] Beneficial effects of the present invention: 1. The present invention aims at feeding pulp raw materials for lyocell fiber production, which changes the intermittent feeding and swelling in the traditional production process, which requires shutdown operation between batches. The feeding and conveying device continuously feeds the pulp raw materials into the swelling machine without stopping, thus avoiding the waste of production time, meeting the demand for large production, and being able to fully adapt to the current status of unit equipment capacity and performance improvement after the progress of swelling and dissolution processes, thereby improving the overall production efficiency. In addition, it also avoids the swelling difference of each batch of materials caused by the operation difference between batches in the traditional intermittent feeding, thereby resulting in the problem of poor uniformity of the produced pulp porridge.

[0022] 2. The present invention combines precise liquid level detection and control technology to control the current liquid level in the swelling machine to always be within the set target height range, ensuring that the pulp raw material and the solvent fully react and swell in the swelling machine under the continuous feeding mode, which is helpful to prepare high-quality glue and meet the high requirements of lyocell fiber production for pulp porridge uniformity. Therefore, it is further ensured that the new continuous feeding process proposed by the present invention for pulp raw materials can ensure that the produced pulp porridge has good quality and production efficiency.

[0023] 3. The feeding method of the present invention can optimize the operation of the equipment, better combine with the efficient dissolution performance of the current dissolution machine, adapt to the progress of the dissolution process, reduce energy waste, and have significant economic benefits; 4. The parameters of the automatic feeding process of the present invention can be monitored and adjusted in real time, which is convenient for optimizing the process and further improving production efficiency and product quality.

[0024] 5. The present invention ensures the correctness of the material sequence when the pulp is continuously fed to the rear end through photoelectric sensors and alternately working transfer conveying devices at the front end of the pulp raw material feeding, ensures the continuity of the feeding process, avoids production interruptions caused by improper feeding, avoids incomplete chemical reactions or side reactions caused by incorrect feeding sequence, reduces the generation of defective products, and improves the consistency of product quality.

[0025] 5. The automatic feeding method of the present invention maximizes the use of production equipment, improves production efficiency, and adapts to the needs of large-scale industrial production.

[0026] 6. In the present invention, at the front end of the pulp raw material feeding conveying, the pulp raw material is transported from the side line bin to the feeding port by the AGV trolley. The application of the AGV trolley realizes the automation of raw material transportation, reduces the time and error of manual operation, and can quickly and accurately transport the pulp raw material for Lyocell production from the side line bin to the feeding port of the feeding line, thereby realizing uninterrupted raw material supply and further improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The foregoing and following detailed description of the present invention will become more apparent when read in conjunction with the following drawings, in which: Figure 1 This is a schematic diagram of feeding the swelling machine of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions for achieving the purpose of the present invention will be further described below through specific embodiments. It should be noted that the technical solutions claimed for protection in the present invention include but are not limited to the following embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0029] Example 1 As a preferred embodiment of the present invention, this embodiment proposes a continuous feeding control method for lyocell fiber production based on image recognition. Figure 1 This is a schematic diagram of the feeding of the swelling machine of the present invention, combined with the attached Figure 1 , the feeding control method is as follows: The pulp raw material that has undergone front-end processing steps such as unpacking, paper removal, loosening, and metal detection is delivered to the feeding and conveying device at the feed port of the swelling machine, and the feeding and conveying device continuously feeds the pulp raw material into the swelling machine at a preset feeding speed without stopping; and a camera arranged in the swelling machine obtains an image of the liquid level to be detected in the swelling machine containing a fuzzy liquid level boundary under a non-stop continuous feeding state, and then the control system obtains the current liquid level height in the swelling machine based on the image of the liquid level to be detected in combination with image detection and recognition technology, compares the current liquid level height with the target liquid level height, and obtains the liquid level difference; the control system uses the liquid level difference as a control factor, and controls the valve opening of the solvent inlet of the swelling machine based on the PID control algorithm to keep the current liquid level height in the swelling machine within a set range.

[0030] Initially, the feeding speed of the feeding and conveying device and the discharging speed of the discharging port of the swelling machine are set. The feeding and conveying device continuously feeds the pulp raw material into the swelling machine at the preset feeding speed without stopping. On this basis, the delivery speed of the solvent delivery pump at the solvent inlet of the swelling machine is adjusted by calculation, and this speed needs to match the feeding and conveying speed of the pulp raw material. In this way, the raw material and the solvent fed into the swelling machine are correctly proportioned and can react fully, and finally it can be ensured that the viscosity and spinnability of the produced pulp porridge meet the requirements, and can provide high-quality raw materials for subsequent glue production.

[0031] The feeding speed of the feeding and conveying device and the discharging speed of the swelling machine are known set values, and the delivery speed of the solvent is also matched with the feeding and conveying speed of the pulp raw material. Then, during the reaction process, the liquid level in the swelling machine is basically in a constant position, with small fluctuations up and down. At this time, the liquid level can be used as the set target liquid level height. Therefore, the current liquid level height in the swelling machine can be used as a basis for judging whether the raw material and the solvent can react fully under continuous feeding to obtain a porridge with better quality. In the later long-term production and operation process, if the speed of the solvent delivery pump fluctuates greatly, it may cause the current liquid level height in the swelling machine to exceed the set target liquid level height, causing the amount of solvent in the current swelling machine to not match the amount of pulp raw material. At this time, if the delivery speed of the solvent delivery pump is not adjusted in time, it may cause the problem of insufficient reaction between the raw material and the solvent, and the quality of the produced porridge is poor.

[0032] Therefore, by monitoring the current liquid level in the swelling machine and controlling it within the set target liquid level range, it can be ensured that the porridge produced under continuous feeding conditions meets the requirements.

[0033] Based on the automatic feeding method proposed above, the production capacity of lyocell fiber can be significantly improved to meet the current large-scale production needs; the operation of the feeding and transportation device can be optimized, better combined with the efficient dissolution performance of the current dissolving machine, adapted to the progress of the dissolving process, and energy waste can be reduced, which has significant economic benefits; and better quality porridge can be obtained, providing excellent raw materials for the subsequent preparation of glue solution.

[0034] However, when the swelling machine is running, the material state and stirring form of the internal porridge make it extremely difficult to detect the liquid level. When the existing liquid level detection technology is used to detect the current liquid level height in the swelling machine, the detection error is large and the accuracy is low, which makes it impossible to control the current liquid level height within the target height range.

[0035] Due to the high-speed rotation of the agitator, the input of pulp and additives, and the uneven flow of the liquid, the liquid level is usually not a plane, but presents a complex dynamic shape, such as a wave, a vortex, or an inclined shape. The rotation of the agitator will form a vortex in the center of the liquid surface, causing the center of the liquid surface to be concave and the surrounding area to be higher; at the same time, the liquid surface will also produce irregular waves and local fluctuations due to stirring and material input, and may even be further complicated by foam accumulation. This non-planar liquid level morphology makes it difficult for traditional liquid level measurement methods to accurately reflect the actual liquid level. For example, traditional float sensors are prone to getting stuck or failing in high-viscosity fluids, and cannot accurately measure the fluctuating liquid level; foam and liquid level fluctuations will cause ultrasonic sensor measurement errors and require frequent cleaning. The high-definition camera can be installed in the swelling machine to collect the current liquid level image inside the swelling machine, and quickly and accurately identify the current liquid level height in the swelling machine through image processing and target detection algorithms. Furthermore, after obtaining the current liquid level height through image recognition technology, the valve opening of the solvent feed port of the swelling machine is controlled by the PID control algorithm to keep the current liquid level height in the swelling machine within the set range.

[0036] Since the swelling machine is used to mix pulp raw materials with NMMO solvent to prepare a high-concentration solution in the Lyocell production process, its liquid level detection has the following complex characteristics: (1) High viscosity characteristics: The viscosity of NMMO solution is much higher than that of water. The liquid surface boundary presents a smooth transition rather than a sharp edge, and the gradient is weak.

[0037] (2) Dynamic complexity: The high-speed rotation of the agitator causes the liquid surface to form vortices, waves or tilts, with blurred and irregular boundaries. Interference factors: The solution may contain undissolved fiber clumps, bubbles or splashes, which increase noise interference.

[0038] (3) Lighting and reflection: The lighting inside the swelling machine is uneven, and reflections or shadows affect the image quality.

[0039] (4) Real-time requirements: Liquid level control requires high-frequency detection (such as 10 Hz), and the algorithm must take into account both accuracy and efficiency.

[0040] Therefore, these characteristics make it easy for traditional image processing and target detection algorithms (such as traditional canny operators) to detect the liquid surface boundary in the swelling machine, resulting in missed detection, false detection or boundary breakage, which ultimately leads to poor accuracy in liquid level recognition, which in turn affects the PID control algorithm's control of the liquid level in the swelling machine, making it impossible to ensure that the current liquid level in the swelling machine is always within the set range, which in turn has a significant impact on product quality.

[0041] Therefore, the embodiment of the present invention proposes the following detection method for detecting the current liquid level in the swelling machine, and the specific steps are as follows: Step A: A camera arranged in the swelling machine captures a top view of the inside of the swelling machine tank body, and obtains a liquid level image to be detected which is formed by the pulp raw material and the solvent and contains a fuzzy liquid level boundary when the feeding and conveying device is continuously feeding without stopping the machine. In the liquid level image to be detected, the boundary between the liquid surface and the inner wall of the swelling machine presents a circular boundary, the grayscale of the liquid surface area is relatively low (such as 50), and the grayscale of the tank wall area is relatively high (such as 200), and there may be interference such as vortexes and fiber clusters.

[0042] For the collected liquid level image to be detected, the liquid level image to be detected is first preprocessed, including grayscale conversion and Gaussian smoothing, and then a smoothed liquid level image to be detected is obtained; wherein, Step A1: The liquid level image to be detected collected by the camera is a color image I ( x , y ), so first convert the color image I ( x , y ) to process the grayscale image I g ( x , y ), calculated as follows: Formula (1); in, x and y Respectively represent the coordinates of each pixel point in the collected liquid level image to be detected; R , G , B Respectively represent the intensity values ​​of the three color channels of each pixel; Step A2: Use adaptive histogram to adjust the grayscale image I g ( x , y) to equalize and enhance the contrast between the liquid surface and the inner wall, and obtain a contrast-enhanced preprocessed image I e ( x , y ); Step A3: Use Gaussian kernel function to I e ( x , y ) to obtain a smoothed image I σ ( x , y ), calculated as follows: Formula (2); Formula (3); in, G ( x , y , s ) represents the Gaussian kernel function; s represents the standard deviation of the Gaussian kernel; s 2 Represents the variance of the Gaussian distribution.

[0043] Step B: Based on smoothed image I σ ( x , y ), calculate the gradient magnitude and gradient direction of each pixel in the image; where, Step B1: Use Sobel operator to calculate the smoothed image I σ ( x , y ), the horizontal gradient of each pixel G x and vertical gradient G y , calculated as follows: Formula (4); Formula (5); Step B2: According to the horizontal gradient of the pixel point G x and vertical gradient G y , calculate the gradient direction and initial gradient amplitude of the pixel point, the calculation method is as follows: Formula (6); Formula (7); in, G (x , y ) represents the initial gradient amplitude of the pixel; i ( x , y ) represents the gradient direction of the pixel.

[0044] Step C: Circular pattern analysis based on sliding window (first stage of directional enhancement) Smooth the image using a sliding window I σ ( x , y ) to obtain the continuity of the gradient direction of each pixel in the sliding window, and then enhance the area of ​​the image that meets the circular boundary characteristics, and weaken the area that does not meet the circular boundary characteristics; the specific operation method is as follows: Step C1: Based on the following equations (8) and (9), the gradient direction difference between each pixel point in the sliding window and its adjacent pixel points is calculated as follows: Formula (8); Formula (9); Among them, Δ i row ( x , y ) represents the row direction difference; Δ i col ( x , y ) represents the column direction difference; i ( x , y ) represents the gradient direction of the current pixel; i ( x+ 1, y ) represents the gradient direction of the pixels in the same column of the next row; i ( x , y+ 1) Indicates the gradient direction of the pixels in the same column on the right; sets the sliding window size to 32x32 pixels and the stride to 16.

[0045] Step C2: Calculate the average value of the gradient direction difference in the sliding window. The calculation method is as follows: Formula (10); in, Represents the average value of the gradient direction difference within the sliding window; N is the total number of gradient direction differences within the sliding window; Step C3, adjusting the gradient amplitude of each pixel in the sliding window according to the continuity of the gradient direction; like , indicating that the gradient direction of the pixel points in the sliding window is continuous, which conforms to the circular boundary characteristics, and enhances the gradient amplitude of each pixel point in the sliding window. The calculation method is as follows: Formula (11); like , indicating that the gradient direction of the pixel in the sliding window jumps, which is an interference area, suppressing the gradient amplitude of each pixel in the sliding window. The calculation method is as follows: Formula (12); like , then the gradient amplitude of each pixel in the sliding window remains unchanged; in, Represents the gradient amplitude of the pixel after adjustment according to the continuity of the gradient direction; k 1 and k 2 represent the enhancement coefficient and the suppression coefficient respectively; in the case of enhancement, k The value of 1 can generally be 1.5. For suppression, k The value of 2 can generally be 0.5.

[0046] Step D: Circle positioning and re-enhancement based on Hough transform (second stage of directional enhancement) The rough edge map of the smoothed liquid level image to be detected is obtained using the traditional Canny algorithm, and then the circular boundary of the rough edge map is detected using the Hough transform to obtain the circular parameters, and based on the circular parameters, the rough edge map is enhanced on the basis of step C; the specific operation method is as follows: Step D1: Smooth the image I σ ( x , y ) Use the traditional Canny algorithm (low threshold 50, high threshold 150) to generate a rough edge map E pre ( x , y ); Step D2: Rough edge map E pre ( x , y ) Use Hough transform to detect circular boundaries and obtain a rough edge map E pre ( x , y ) of the circle, which includes the coordinates of the center of the circle ( x c , y c ) and the initial radius rpre ; Step D3: Calculate a rough edge map E pre ( x , y ), the distance between each pixel and the center of the circle, if the distance is less than the set value, the gradient amplitude of the pixel is further enhanced, and the calculation method is as follows: Formula (13); Formula (14); in, Represents the gradient amplitude of the pixel after the second enhancement; k 3 represents the enhancement factor, which can usually be 1.5; d ( x , y ) represents the distance from the pixel to the center of the circle.

[0047] Step E: Continue to use the traditional Canny algorithm to enhance the rough edge map E pre ( x , y ) to generate a single-scale edge map; Step E1: along the gradient direction i ( x , y ),Compare E pre ( x , y ) is the gradient amplitude of each pixel after the second enhancement The gradient amplitude of two adjacent pixels in its neighborhood only retains the local maximum value and suppresses the non-maximum value; Step E2, then using a high threshold and a low threshold to perform double threshold detection on the gradient amplitude of the pixel after the second enhancement, and classifying the gradient amplitude after non-maximum suppression into a strong edge, a weak edge or a non-edge; Step E3: Connect the weak edge pixels with the strong edge. If there is a strong edge in the neighborhood of the weak edge, it is retained as an edge. Otherwise, it is removed. Finally, a single-scale edge map is generated.

[0048] Step F: Use different Gaussian kernel functions to process the preprocessed image. I e ( x , y ) for smoothing, Then repeat the above steps B to E for N-1 times, where N is an integer greater than 1, to obtain N different single-scale edge maps; (different Gaussian kernel standard deviations are used for smoothing, and one Gaussian kernel standard deviation corresponds to a single-scale edge map); then fuse multiple different single-scale edge maps through multi-scale processing to obtain the final liquid level edge map, which is calculated as follows: Formula (15); in, It is the final edge map after fusion processing; w 1. w 2 and w 3 is the weight coefficient; , , It is a single-scale edge generated after processing based on different Gaussian kernel functions.

[0049] In the embodiment described in the present invention, the weight coefficient w 1 can be taken as 0.3, w 2 can be taken as 0.4, w 3 can be taken as 0.3. In general, the middle scale ( s = 3) is weighted slightly higher because it is often used to smooth out detail and noise; small scale ( s =1) retain details, large scale ( s =5) Emphasize the overall structure.

[0050] Step G: The final liquid level edge map Perform threshold detection (for example, 128) to obtain a binary edge map, then apply morphological operations to optimize the binary edge map, and finally perform Hough transform on the optimized binary edge map to detect the circular boundary and obtain the detection radius of the final liquid level edge map after Hough transform. r .

[0051] In the described embodiments of the present invention, the morphological operations typically include dilation and erosion.

[0052] Step H: According to the obtained detection radius r , the current liquid level in the swelling machine is calculated by the following formula (16): h : Formula (16); in, h is the current liquid level in the swelling machine; r is the detection radius of the final liquid level edge map; h max It is the reference value for calibrating the maximum liquid level in the swelling machine; hmin It is the reference value for calibrating the minimum liquid level height in the swelling machine; r max The image radius corresponding to the reference value of the maximum liquid level calibration in the swelling machine; r min The image radius corresponding to the reference value of the minimum liquid level calibration in the swelling machine.

[0053] In the embodiment described in the present invention, the core of liquid level conversion is to establish the detection radius in the image. r With the current liquid level h The mapping relationship between the swelling machine tank and the surface area is shown in Figure 1. Given that the cross-section of the swelling machine tank is circular, the liquid surface boundary appears as a circular outline in the top view image, and its radius changes with the liquid level. According to the pre-calibrated liquid level-radius relationship, for example, when the radius is 200 pixels, the corresponding liquid level is 0 mm (tank bottom), and when the radius is 400 pixels, the corresponding liquid level is 500 mm (tank top). The above liquid level calculation formula can be derived by linear interpolation.

[0054] Furthermore, in the above liquid level calculation formula, the calibration reference value is obtained through calibration experiments, and it can be understood that in image processing and computer vision, radius and pixel can be equivalent, mainly because the physical size in the image (such as radius) can be represented and measured by pixel value. That is, in the image, the radius is a geometric concept, which represents the distance from the center point to the edge. This distance can be represented by pixel value. Therefore, the specific calibration process can be referred to as follows: (1) Place the swelling machine tank in an empty liquid state (minimum liquid level height calibration reference value), take a top view image, measure the radius of the liquid surface boundary, and record the pixel value corresponding to the radius at this time. The pixel value is the image radius corresponding to the minimum liquid level height calibration reference value; (2) Fill the tank to the maximum liquid level (corresponding to the maximum liquid level height calibration reference value), take a top-down image, measure the radius of the liquid surface boundary, and record the pixel value corresponding to the radius at this time. This pixel value is the image radius corresponding to the maximum liquid level height calibration reference value; (3) Then, based on the parameters obtained in steps (1) and (2), the conversion factor in the formula can be calculated: .

[0055] In the embodiment described in the present invention, it can be understood that in order to achieve uniform feeding of pulp raw materials into the swelling machine, the feeding and transporting device is usually selected as a screw conveyor, and the amount of conveyed material can be adjusted by adjusting the speed or pitch of the screw conveyor.

[0056] Furthermore, considering the complexity of the working environment inside the swelling machine, the camera installed in the swelling machine usually selects a CCD industrial camera (for example, a resolution of 1080p and a frame rate of 10Hz), which can still provide high-quality images in complex environments and help achieve accurate liquid level detection. In addition, the CCD industrial camera is usually installed on the top of the swelling machine, and shoots a top view of the inside of the swelling machine tank vertically downward. The captured image contains a circular liquid surface boundary (the boundary between the liquid surface and the tank wall).

[0057] After obtaining the current liquid level in the swelling machine in the above manner, the current liquid level is compared with the target liquid level to obtain a liquid level difference. The liquid level difference is used as a control factor to control the valve opening of the solvent feed port of the swelling machine based on a PID control algorithm, thereby keeping the liquid level in the swelling machine always within a set range. Specifically, the following steps are included: According to the liquid level difference, the control amount of the valve opening is calculated based on the following calculation expression: Formula (17); in, Adjust the increment for the solvent delivery pump opening. represents the proportional term, is the integral term, is the differential term; e ( t ) is the liquid level difference; K p represents proportional gain; K i represents the integral gain; K d represents the differential gain; Combination And the current opening of the solvent delivery pump, calculate the target opening of the solvent delivery pump, the calculation method is as follows: Formula (18); in, Indicates the current opening degree of the solvent delivery pump; Indicates the target opening of the solvent delivery pump; If the liquid level is lower than the set value, the solvent delivery speed is slow, resulting in a small amount of solvent in the swelling machine. At this time, the solvent delivery speed needs to be increased, that is, by increasing the opening of the solvent delivery pump at the solvent inlet of the swelling machine to the calculated target opening, thereby increasing the amount of solvent added; if the liquid level is higher than the set value, the solvent delivery speed is fast, resulting in an increase in the amount of solvent in the swelling machine. At this time, the solvent delivery speed needs to be reduced, that is, by reducing the opening of the solvent delivery pump at the solvent inlet of the swelling machine to the target opening, thereby reducing the amount of solvent added.

[0058] Further, If the limit processing is performed If it exceeds the preset range (for example, 0% to 100%, corresponding to fully closed and fully open states respectively), it will be limited to the boundary value, that is, when <0% is set to 0%, when >100% is set to 100%; finally, It is converted into electrical signals or mechanical instructions, and the valve opening of the pump is accurately adjusted through a driving device (such as a servo motor or solenoid valve), thereby dynamically controlling the flow of the swelling machine and achieving highly stable adjustment of the liquid level.

[0059] Example 2 On the basis of Example 1, a pulp raw material receiving and preliminary processing step is further provided at the front end of the feeding and conveying device, and the specific process is as follows: Step P1: Import the production plan into the material preparation information management module. The material preparation information management module receives the production plan and then generates batching instructions. After receiving the batching instructions, the AGV transports the required type and quantity of pulp raw material bags from the side warehouse to the bag cutting area; Step P2: the bale shearing mechanism in the bale shearing area performs a bale-opening process on the pulp raw material bale, and splits the pulp raw material bale into a plurality of pulp raw material small bales; Step P3: After the large bag is opened, different pulp raw material packets are transported to the corresponding preparation area A, preparation area B and preparation area C; the first photoelectric gate, the second photoelectric gate and the third photoelectric gate are respectively provided at the exits of the preparation area A, preparation area B and preparation area C. When the pulp raw material packet passes through the exit of each preparation area, the corresponding photoelectric gate emits a pulse signal; Step P4: the pulp raw material packets in each preparation area are transported from the material outlet of the preparation area to the packet cutting wire area according to the set order and quantity. Two groups of transfer conveying devices are arranged in the packet cutting wire area. One side of the two groups of transfer conveying devices is respectively provided with a packet cutting mechanism; the two groups of transfer conveying devices alternately complete the work of receiving the pulp raw material packets output from the preparation area, and alternately transport the received and packet-cut pulp raw materials to the debarking machine; the outlets of the two groups of transfer conveying devices are respectively provided with a fourth photoelectric gate and a fifth photoelectric gate. When the pulp raw materials with packet cuts pass through the outlet of the transfer conveying device, the corresponding photoelectric gate emits a pulse signal; Step P5: After the pulp raw material is removed in the paper remover, it is transported to the opening and loosening machine, which opens the fibers in the pulp raw material into small fiber bundles or single fibers; the pulp raw material after opening is further transported to the metal detector for metal detection, and the outlet of the metal detector is provided with a sixth photoelectric gate; Step P6: The pulp raw material after metal detection is transported to the feeding and conveying device.

[0060] In the embodiment described in the present invention, the production of lyocell fibers has very strict requirements on the order of material delivery, and the wrong order of feeding will affect the quality of the product. Assuming that the feeding order set by the system is A1-A2-A3-B1-B2-C1-C2-C3, then at the front-end transportation and processing end, it is necessary to first transport 3 packages of material A from the preparation area A to the feeding conveyor, and then transport 2 packages of material B from the preparation area B, and finally 3 packages of material C from the preparation area C. There are generally no major problems in the process of transporting materials from the preparation area, and the point where the order is prone to errors is the area where the wire is cut for small packages. Two sets of transfer conveying devices are set in the area where the wire is cut for small packages. Although the two sets of transfer conveying devices can improve the processing efficiency of materials, it is easy for the speed of cutting the wire for small packages to be different, which leads to the wrong order of the final material delivery. Therefore, the embodiment of the present invention specifically sets the two groups of transfer and transportation devices to work alternately, and respectively sets a fourth photoelectric gate (hereinafter referred to as light 4) and a fifth photoelectric gate (hereinafter referred to as light 5) on one side of the two transfer and transportation devices. Since the feeding sequence is initially set to A1-A2-A3-B1-B2-C1-C2-C3, the system will think that the record of light 4-light 5-light 4-light 5-light 4-light 5-light 4-light 5 corresponds to A1-A2-A3-B1-B2-C1-C2-C3. However, in actual operation, if the transfer and transportation devices do not work alternately, since the small package cutting mechanisms corresponding to the two transfer and transportation devices have different processing speeds for cutting wire, it is possible that the first three packages of materials are cut by the small package cutting mechanism where light 4 is located, and are transported by the transfer and transportation device corresponding to light 4, and the fourth package of materials is cut by the small package cutting mechanism where light 5 is located, and is transported by the corresponding transfer and transportation device. In this case, the system will record the pulse signals corresponding to these three packages as A1, A3 and B2, but they are actually transported in the order of A1, A2 and A3, which causes the system to not correspond to the actual situation, resulting in information errors, which ultimately affects the generation.

[0061] It can be understood that the two groups of transfer conveying devices are usually conveyor belts arranged in parallel; and the transfer conveying devices are slidably switched so that the output ends of the two groups of transfer conveying devices are alternately connected with the feed port of the debarking machine.

[0062] Furthermore, a connecting conveyor line is provided between each material preparation area and the transfer conveyor device, and the connecting conveyor line connects each material preparation area and the input end of the transfer conveyor device. The connecting conveyor line is generally a conveyor belt.

[0063] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for controlling continuous feeding of lyocell fiber production based on image recognition, characterized in that: The feeding and conveying device continuously feeds the pulp raw material into the swelling machine at a preset feeding speed; The camera arranged in the swelling machine obtains the liquid level to be detected image containing the fuzzy liquid level boundary in the swelling machine under the continuous feeding state; Smoothing the liquid level image to be detected, enhancing the area of ​​the image that meets the circular boundary characteristics according to the continuity of the image gradient direction, obtaining a liquid level edge map of the fuzzy liquid level boundary image, and estimating the current liquid level height according to the liquid level edge map; Compare the current liquid level with the target liquid level to obtain the liquid level difference; The liquid level difference is used as a control factor, and the opening of the solvent delivery pump at the solvent inlet of the swelling machine is controlled based on the PID control algorithm to keep the current liquid level height in the swelling machine within the set range.

2. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 1, characterized in that: The liquid level image to be detected is smoothed, and according to the continuity of the image gradient direction, the area of ​​the image that meets the circular boundary characteristics is enhanced to obtain a liquid level edge map of the fuzzy liquid level boundary image, and the current liquid level height is estimated according to the liquid level edge map, which specifically includes the following steps: Step A, performing grayscale conversion on the image to be detected of the liquid level to obtain a preprocessed grayscale image, and performing Gaussian smoothing on the preprocessed image by using a Gaussian kernel function to obtain a smoothed image; Step B, calculating the gradient magnitude and gradient direction of each pixel in the smoothed image; Step C, using a sliding window to analyze the smoothed image, determine the continuity of the gradient direction of each pixel in the sliding window, enhance the area of ​​the image that meets the circular boundary characteristics, and weaken the area that does not meet the circular boundary characteristics; Step D, using the Canny algorithm to obtain a rough edge map of the smoothed image, then using Hough transform to detect the circular boundary of the rough edge map, obtaining circular parameters, and enhancing the rough edge map based on the circular parameters; Step E: Continue to use the Canny algorithm to process the enhanced rough edge map to generate a single-scale edge map; Step F: Use a new Gaussian kernel function to smooth the grayscale converted liquid level image to be detected. Then, the above steps B to E are performed and repeated N-1 times, where N is an integer greater than 1, so as to obtain N different single-scale edge maps; the N different single-scale edge maps are processed by multi-scale processing fusion to obtain a final liquid level edge map; Step G, performing threshold detection on the final liquid level edge map to obtain a binary edge map, then applying morphological operations to optimize the binary edge map, and finally performing Hough transform on the optimized binary edge map to detect circular boundaries and obtain the detection radius of the final liquid level edge map after Hough transform; Step H: Calculate the current liquid level height in the swelling machine according to the detection radius of the final liquid level edge map.

3. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 2, characterized in that: The step A specifically includes the following contents: Step A1: Convert the collected liquid level image to a grayscale image I g ( x , y ), calculated as follows: Formula (1); in, x and y Respectively represent the coordinates of each pixel point in the collected liquid level image to be detected; R , G , B Respectively represent the intensity values ​​of the three color channels of each pixel; Step A2: Use adaptive histogram to adjust the grayscale image I g ( x , y ) is equalized to obtain a contrast-enhanced preprocessed image I e ( x , y ); Step A3: Use Gaussian kernel function to I e ( x , y ) to obtain a smoothed image I σ ( x , y ), calculated as follows: Formula (2); Formula (3); in, G ( x , y , σ ) represents the Gaussian kernel function; σ represents the standard deviation of the Gaussian kernel; σ 2 Represents the variance of the Gaussian distribution.

4. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 3, characterized in that: The step B specifically includes the following contents: Step B1: Use Sobel operator to calculate the smoothed image I σ ( x , y ), the horizontal gradient of each pixel G x and vertical gradient G y , calculated as follows: Formula (4); Formula (5); in, x and y Respectively represent the coordinates of each pixel; Step B2: According to the horizontal gradient of the pixel point G x and vertical gradient G y , calculate the gradient direction and initial gradient amplitude of the pixel point, the calculation method is as follows: Formula (6); Formula (7); in, G ( x , y ) represents the initial gradient amplitude of the pixel; θ ( x , y ) represents the gradient direction of the pixel.

5. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 4, characterized in that: The step C comprises the following steps: Step C1: Use sliding window to smooth the image I σ ( x , y ) to analyze and calculate the gradient direction difference between each pixel in the sliding window and its adjacent pixels. The calculation method is as follows: Formula (8); Formula (9); in, x and y Respectively represent the coordinates of the pixel points; Δ θ row ( x , y ) represents the row direction difference; Δ θ col ( x , y ) represents the column direction difference; θ ( x , y ) represents the gradient direction of the current pixel; θ ( x+ 1, y )and θ ( x , y+ 1) Indicates the gradient direction of adjacent pixels; Step C2: Calculate the average value of the gradient direction difference in the sliding window. The calculation method is as follows: Formula (10); in, Represents the average value of the gradient direction difference within the sliding window; N is the total number of gradient direction differences within the sliding window; Step C3, adjusting the gradient amplitude of each pixel in the sliding window according to the continuity of the gradient direction; like , the gradient direction of the pixels in the sliding window is continuous, and the gradient amplitude of each pixel in the sliding window is enhanced. The calculation method is as follows: Formula (11); like , the gradient direction of the pixel points in the sliding window jumps, suppressing the gradient amplitude of each pixel point in the sliding window. The calculation method is as follows: Formula (12); like , then the gradient amplitude of each pixel in the sliding window remains unchanged; in, G ( x , y ) represents the initial gradient amplitude of the pixel; Represents the gradient amplitude of the pixel after adjustment according to the continuity of the gradient direction; k 1 and k 2 represent the enhancement coefficient and suppression coefficient respectively.

6. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 5, characterized in that: Said k The value of 1 is 1.5, k The value of 2 is 0.

5.

7. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 5, characterized in that: The step D comprises the following steps: Step D1: Smooth the image I σ ( x , y ) Use the Canny algorithm to generate a rough edge map E pre ( x , y ); Step D2: Rough edge map E pre ( x , y ) Use Hough transform to detect circular boundaries and obtain a rough edge map E pre ( x , y ) of the circle, including the coordinates of the center of the circle ( x c , y c ) and the initial radius r pre ; Step D3: Calculate a rough edge map E pre ( x , y ) and the center of the circle. If the distance is less than the set value, the gradient amplitude of the pixel is further enhanced. The calculation method is as follows: Formula (13); Formula (14); in, Represents the gradient amplitude of the pixel after the second enhancement; k 3 represents the enhancement factor; d ( x , y ) represents the distance from the pixel to the center of the circle.

8. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 7, characterized in that: The step E comprises the following steps: Step E1: along the gradient direction θ ( x , y ), compare the gradient amplitude of each pixel after the second enhancement The gradient amplitude of two adjacent pixels in its neighborhood retains the local maximum and suppresses the non-maximum value; Step E2, performing double threshold detection on the gradient amplitude of the pixel after the second enhancement, and classifying the gradient amplitude after non-maximum suppression into strong edge, weak edge or non-edge; Step E3: Connect the strong edges and the adjacent weak edges to generate a single-scale edge map.

9. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 8, characterized in that: In step F, a plurality of different single-scale edge maps are fused through multi-scale processing to obtain a final liquid level edge map, which is calculated as follows: Formula (15); in, is the final liquid level edge map; w 1. w 2 and w 3 is the weight coefficient; , , It is a single-scale edge map generated after processing based on different Gaussian kernel functions.

10. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 2, characterized in that: In step G, the morphological operation includes dilation and erosion.

11. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 2, characterized in that: In the step H, the current liquid level height in the swelling machine is calculated according to the detection radius of the final liquid level edge map, and the calculation method is as follows: Formula (16); in, h is the current liquid level in the swelling machine; r is the detection radius of the final liquid level edge map; h max It is the reference value for calibrating the maximum liquid level in the swelling machine; h min It is the reference value for calibrating the minimum liquid level height in the swelling machine; r max The image radius corresponding to the reference value of the maximum liquid level calibration in the swelling machine; r min The image radius corresponding to the reference value of the minimum liquid level calibration in the swelling machine.

12. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 1, characterized in that: The method uses the liquid level difference as a control factor and controls the opening of the solvent delivery pump at the solvent inlet of the swelling machine based on a PID control algorithm, comprising the following steps: According to the liquid level difference, the control amount of the valve opening is calculated based on the following calculation expression: Formula (17); in, u ( t ) is the solvent delivery pump opening adjustment increment, represents the proportional term, is the integral term, is the differential term; e ( t ) is the liquid level difference; K p represents proportional gain; K i represents the integral gain; K d represents the differential gain; Combination u ( t ) and the current opening of the solvent delivery pump, the target opening of the solvent delivery pump is calculated as follows: Formula (18); in, Indicates the current opening degree of the solvent delivery pump; Indicates the target opening of the solvent delivery pump; If the liquid level is lower than the set value, increase the opening of the solvent delivery pump to the target opening; if the liquid level is higher than the set value, reduce the opening of the solvent delivery pump to the target opening.

13. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 1, characterized in that: The feeding and conveying device is a screw conveyor or a conveyor belt.

14. The method for controlling continuous feeding of lyocell fiber production based on image recognition according to claim 1, characterized in that: The camera is a CCD industrial camera.

15. A continuous feeding control system for Lyocell fiber production based on image recognition, characterized in that: The system is used to implement the continuous feeding control method according to any one of claims 1 to 14, comprising: A feeding and conveying device, used for continuously feeding pulp raw materials into the swelling machine at a preset feeding speed; A camera is used to obtain a liquid level image to be detected containing a fuzzy liquid level boundary in the swelling machine under a continuous feeding state; A camera controller, used for controlling the camera to collect the image of the liquid level to be detected in the swelling machine; A liquid level detection module, used to obtain the current liquid level height in the swelling machine according to the liquid level to be detected image transmitted by the camera; The liquid level control module is used to compare the current liquid level height with the target liquid level height to obtain the liquid level difference, and use the liquid level difference as a control factor to control the opening of the solvent delivery pump at the solvent feed inlet of the swelling machine based on the PID control algorithm to keep the liquid level difference within a set range.

Images