System and method for dynamically regulating and controlling cut stem content of stem based on multi-dimensional visual identification

Abstract

The invention provides a system and a method for dynamically regulating and controlling the cut tobacco content of a stem based on multi-dimensional visual identification, and the system comprises a high-speed visual identification unit which is provided with a high-frame-rate CCD camera and a stroboscopic light source less than or equal to 100 microseconds, eliminates motion blur, and freezes a material track. The dynamic morphological analysis module is used for dividing light-color tobacco shreds (S < = 30%) and dark-color tobacco stems (S > = 50%) in an HSV space, and extracting contour length-width ratios (stem slivers > = 3, stem shreds < = 1.5 and area lt); meanwhile, the light cut stems are judged on the basis that the speed variance is larger than 0.2 m < 2 > / s < 2 >, and the density influence is corrected through a speed-quality model sigma < v2 > = k / m + b; and the problem of misjudgment of fragments with the length-width ratio of 1.5-3.0 is solved by adopting Hu moment template matching. And the closed-loop control unit takes the silk content deviation delta C as input, outputs the displacement of the air cylinder through a PID (Proportion Integration Differentiation) controller, and drives an air door to linearly adjust the air pressure according to K = K0-alpha * delta C (alpha = 0.5-2.0). According to the self-learning mechanism, when the moisture content change of the raw materials is larger than 2%, the color threshold value is automatically adjusted to be + / -5 degrees, and the speed coefficient k is automatically adjusted to be +

Description

Technical Field

[0001] The present invention relates to the field of cigarette manufacturing, and in particular to a system and method for dynamically controlling the shredded content of stems based on multi-dimensional visual recognition. Background Art

[0002] In the tobacco shred process, stem and stick separation is a key step affecting cigarette quality. Traditional air separation systems mainly separate stem and stick tobacco through static screening or fixed air pressure, which has significant defects:

[0003] 1. Insufficient recognition accuracy: Relying on single color or shape features, such as RGB threshold segmentation, it is unable to accurately distinguish between light-colored tobacco and dark-colored tobacco stems of similar colors. In particular, the misidentification rate for transitional fragments with an aspect ratio of 1.5 to 3.0 exceeds 15%.

[0004] 2. Dynamic response lag: Based on offline sampling to detect the silk content, the adjustment cycle is as long as minutes, resulting in a mismatch between wind pressure and real-time working conditions, and the silk content fluctuation range is as high as ±3%;

[0005] 3. Weak anti-interference ability: When the moisture content of the raw material changes by more than 2%, the color characteristics of the tobacco will drift and the H channel will shift by ±5°, causing the traditional threshold method to fail. Lightweight stems will produce abnormal motion trajectories due to airflow disturbances, which cannot be captured by the static model.

[0006] 4. Poor control robustness: The air door opening has a nonlinear relationship with the silk content. When the PID parameters are fixed, the difference in raw material density causes an overshoot of >2%. Summary of the Invention

[0007] The purpose of the present invention is to provide a system and method for dynamically controlling the silk content of stem sticks based on multi-dimensional visual recognition, integrating the three-modal features of color / shape / motion. The system achieves a silk content recognition accuracy of 98.5%±0.3% and a missed detection rate of less than 0.8%. Through millisecond-level dynamic response, the silk content control accuracy reaches ±0.5% of the set value. The system adapts to changes in raw material physical properties and improves the fluctuation suppression efficiency by 40% under complex working conditions.

[0008] The embodiment of the present invention is achieved as follows:

[0009] A system for dynamically controlling the silk content of stem sticks based on multi-dimensional visual recognition, comprising:

[0010] A high-speed visual recognition unit, located above the stem tag collection box, includes a high-frame-rate CCD camera and a stroboscopic light source to continuously capture images of the material as it falls.

[0011] The dynamic morphological analysis module, connected to the CCD camera, performs the following collaborative discrimination:

[0012] Color space segmentation: Separate light-colored tobacco with a saturation of ≤30% from dark-colored tobacco stems with a saturation of ≥50% in the HSV color space;

[0013] Shape feature extraction: Calculate the aspect ratio of the material outline, define the aspect ratio ≥ 3 as a stem, and the aspect ratio ≤ 1.5 and the area < 5mm 2 For the stem shreds;

[0014] Motion trajectory analysis: Calculate the material falling speed variance based on continuous frame images, speed variance>0.2m 2 / s 2 The fragments are judged as stems;

[0015] The closed-loop control unit drives the linear cylinder to adjust the air door opening through the PID controller according to the silk content proportional signal output by the dynamic morphology analysis module, so that the air separation pressure and the silk content deviation are negatively regulated;

[0016] The air pressure actuator includes a continuously adjustable air door driven by a linear cylinder and a position feedback sensor. It dynamically adjusts the air door opening according to the adjustment signal, so that the air separation pressure and the silk content deviation have a negative correlation and linear response.

[0017] As a preferred technical solution, the above-mentioned motion trajectory analysis specifically includes:

[0018] Extract the moving material in each frame of image by background difference method;

[0019] Calculate the displacement Δd of the material at the same position between adjacent frames, and obtain the instantaneous velocity v = Δd / Δt according to the shooting interval Δt;

[0020] Calculate the speed standard deviation σ_v within 10 frames, and determine that the stem is light when σ_v>0.45m / s; and / or,

[0021] Introducing velocity variance analysis into quality factor correction:

[0022] Establish velocity variance-mass mapping model: σ_v 2 =k / m+b, where k and b are airflow characteristic coefficients;

[0023] The fragment mass m is estimated by the image area and image color space data. When the measured σ_v 2 When the value is >1.5×(k / m+b), it is judged as cut stem.

[0024] As a preferred technical solution, the above color space segmentation adopts an adaptive threshold:

[0025] For the H channel, set the tobacco identification intervals to [0°, 30°] (light yellow) and [150°, 180°] (light brown);

[0026] Set the tobacco stem recognition interval [50%, 100%] for the S channel;

[0027] The color and shape features are integrated, and pixels are marked as stems when they meet the color criteria and the aspect ratio is ≤1.5.

[0028] As a preferred technical solution, the flash frequency of the stroboscopic light source is synchronized with the falling speed of the material, and the flash duration is ≤100μs, eliminating motion blur to accurately capture the trajectory.

[0029] As a preferred technical solution, the mapping relationship between the above-mentioned damper opening adjustment and the silk content deviation is:

[0030] K=K0-α·ΔC

[0031] Where K is the air door opening, K0 is the reference opening, ΔC is the deviation between the measured value of silk content and the set value of 7%, α is the wind pressure sensitivity coefficient, α = 0.5 ~ 2.0.

[0032] As a preferred technical solution, the mathematical relationship between the output signal of the PID controller and the silk content deviation is:

[0033] u(t)=K_p·e(t)+K_i·∫e(t)dt+K_d·de(t) / dt

[0034] Where u(t) is the cylinder adjustment value, e(t) is the difference between the actual silk content and the preset threshold, and K_p, K_i, and K_d are adjustable gain parameters.

[0035] As a preferred technical solution, the above shape feature extraction adopts a multi-level screening strategy:

[0036] Primary screening: Exclude dust based on area threshold, area < 0.1mm 2 For dust;

[0037] Secondary screening: Calculate the minimum circumscribed rectangle of the outline, and those with an aspect ratio ≥ 3.5 and a degree of fit > 85% are considered stemmed.

[0038] Level 3 screening: Hu moment matching is performed on fragments with an aspect ratio of 1.5 to 3.0, and those with a similarity of >90% with the preset stem template are classified as stem.

[0039] As a preferred technical solution, the above-mentioned wind pressure actuator includes an airflow stabilizing device:

[0040] A honeycomb guide plate is provided at the damper outlet, and its through hole diameter is ≤3mm;

[0041] The distance between the guide plate and the damper is 0.2 times the diameter of the air duct to ensure that the coefficient of variation of air flow uniformity is less than 5%.

[0042] As a preferred technical solution, the above-mentioned dynamic morphological analysis module includes a self-learning unit:

[0043] Record historical sorting data, including silk content, air door opening, and raw material moisture content;

[0044] When the moisture content of the raw material changes by more than 2%, the color segmentation threshold ΔH = ±5° and the speed variance coefficient k ±0.1 are automatically adjusted.

[0045] An embodiment of the present invention further provides a method for dynamically controlling the silk content of skewer strands based on multi-dimensional visual recognition, based on any of the aforementioned systems for dynamically controlling the silk content of skewer strands, comprising the following steps:

[0046] S1. Use a visual system to capture images of the material in the stem collection box in real time and calculate the silk content based on a dynamic morphological analysis module;

[0047] S2. Compare the measured value of silk content with the preset standard value and generate a deviation signal;

[0048] S3. Input the deviation signal into the PID controller, which outputs the displacement control value of the linear cylinder;

[0049] S4. Drive the damper opening to continuously change, so that the air separation pressure reversely compensates for the deviation in silk content;

[0050] S5. Verify the damper status through position feedback until the silk content returns to the set threshold range.

[0051] The beneficial effects of the embodiments of the present invention are:

[0052] 1. The dynamic morphological analysis module integrates the trimodal features of color, shape, and motion, achieving a stem recognition accuracy of 98.5% ± 0.3%. It uses Hu moment template matching to address misjudgment in transitional regions (aspect ratios of 1.5 to 3.0), enhancing anti-interference capabilities and reducing the missed detection rate to <0.8%. It also employs a dynamic compensation mechanism and a speed-quality correction model to eliminate the effects of raw material density differences, ensuring a stable stem content control accuracy of ± 0.5% of the set value.

[0053] 2. Use a stroboscopic light source with an exposure time of ≤100μs to freeze the motion trajectory and combine it with the background difference method to capture the actual falling behavior; quantify the characteristics of lightweight stems: the velocity standard deviation threshold σ_v>0.45m / s accurately reflects the airflow disturbance effect;

[0054] 3. The air pressure actuator uses a linear cylinder and a magnetostrictive sensor with a resolution of 0.01mm to achieve a 50ms response to the damper opening. PID control works in conjunction with the air pressure mapping model (K = K0 - α·ΔC) to shorten the fluctuation suppression time to within 3 seconds, achieving millisecond-level closed-loop regulation.

[0055] 4. Moisture content linkage adjustment: When the moisture content of the raw materials changes by more than 2%, the color threshold (ΔH = ± 5°) and speed coefficient (k ± 0.1) are automatically corrected to adapt to the physical properties of different batches of raw materials; the wind pressure sensitivity coefficient α is adjustable (0.5 ~ 2.0), supporting the "one category, one policy" sorting strategy and adapting to complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 This is a schematic block diagram of a system for dynamically controlling the silk content of stalks according to an embodiment of the present invention;

[0058] Figure 2 is a schematic block diagram of a dynamic morphological analysis module according to an embodiment of the present invention;

[0059] Figure 3 This is a flow chart of a method for dynamically controlling the silk content of skewer sticks according to an embodiment of the present invention;

[0060] Figure 4 Schematic diagram of the silk content calculation process according to an embodiment of the present invention;

[0061] Figure 5 This is an intelligent hierarchical control chart according to an embodiment of the present invention.

[0062] Icons: high-speed visual recognition unit 110; CCD camera 111; stroboscopic light source 112; dynamic morphological analysis module 120; color space segmentation 121; shape feature extraction 122; motion trajectory analysis 123; self-learning unit 124; closed-loop control unit 130; PID controller 131; deviation calculator 132; wind pressure actuator 140; cylinder 141; adjustable damper 142; humidity sensor 143; magnetostrictive sensor 144. DETAILED DESCRIPTION

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0064] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0065] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0066] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0067] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0068] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0069] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0070] Example 1

[0071] See Figure 1-2 This embodiment provides a dynamic control system for the silk content of stalks based on multi-dimensional visual recognition, which includes a high-speed visual recognition unit 110, a dynamic morphological analysis module 120, a closed-loop control unit 130 and an air pressure actuator 140.

[0072] High-speed visual recognition unit 110, located above the stem tag collection box, includes a high-frame-rate CCD camera 111 and a stroboscopic light source 112, which continuously capture images of the falling material. The flash frequency of stroboscopic light source 112 is synchronized with the falling material's speed, with a flash duration of ≤100μs. Combined with background subtraction, it captures the actual falling behavior and eliminates motion blur to accurately capture the trajectory.

[0073] The dynamic morphological analysis module 120 is connected to the CCD camera 111 and performs the following collaborative discrimination:

[0074] Color Space Segmentation 121: Separate light tobacco with saturation ≤ 30% from dark tobacco stems with saturation ≥ 50% in HSV color space;

[0075] Shape feature extraction 122: Calculate the aspect ratio of the material contour, define the aspect ratio ≥ 3 as a stem, and the aspect ratio ≤ 1.5 and the area < 5mm 2 For the stem shreds;

[0076] Motion trajectory analysis 123: Calculate the standard deviation and / or velocity variance of the falling material based on continuous frame images, with velocity variance > 0.2m 2 / s 2 The fragments are judged as stems.

[0077] The dynamic morphological analysis module 120 includes a self-learning unit 124:

[0078] Record historical sorting data, including silk content, air door opening, and raw material moisture content;

[0079] When the moisture content of the raw material changes by more than 2%, the color segmentation threshold ΔH = ±5° and the speed variance coefficient k ±0.1 are automatically adjusted.

[0080] The dynamic morphological analysis module 120 integrates the three modal features of color, shape, and motion form, achieving a stem silk recognition accuracy of 98.5%±0.3%; through Hu moment template matching, it solves the problem of misjudgment in morphological transition areas (aspect ratio of 1.5 to 3.0), enhances anti-interference capabilities, and reduces the missed detection rate to <0.8%; adopts a dynamic compensation mechanism and a speed-quality correction model to eliminate the influence of raw material density differences, and the silk content control accuracy is stabilized at the set value ±0.5%.

[0081] The color space segmentation 121 uses an adaptive threshold:

[0082] For the H channel, set the tobacco identification intervals to [0°, 30°] (light yellow) and [150°, 180°] (light brown);

[0083] Set the tobacco stem recognition interval [50%, 100%] for the S channel;

[0084] Color and shape features are integrated to label pixels as stems when both the color criteria and an aspect ratio ≤ 1.5 are met. Dynamically adjusting the threshold to accommodate lighting changes and batch variations ensures robustness.

[0085] However, in actual operation, the saturation (S) of cut tobacco decreases significantly when the moisture content is high, and it is easy to be mistaken for cut stems. However, the color of cut stems is less affected by moisture. Therefore, this embodiment also integrates color, shape characteristics and moisture data verification. According to the different water absorption characteristics of cut tobacco and cut stems, the moisture value can be obtained through the near-infrared (NIR) or humidity sensor 143 at the air door position to establish a moisture-color association model. For example, high moisture leads to a decrease in the S value, and the S channel threshold needs to be dynamically adjusted.

[0086] The typical moisture content of cut tobacco is 12%-15%. If it is judged to be in a high moisture state, the S channel value is reduced;

[0087] The typical moisture content of shredded stems is 18%-22%. When the moisture content is high, the S channel value remains stable.

[0088] The specific operations are as follows:

[0089] Map the moisture point cloud to image pixel coordinates through affine transformation;

[0090] Collect samples with different moisture contents (12-25%), measure the S channel attenuation curve, fit the linear coefficient, and obtain the calibrated moisture compensation coefficient;

[0091] Spatial interpolation is performed on the moisture data to generate a moisture mask with the same resolution as the image. Threshold correction is calculated in parallel using GPUs, and >1000 frames per second are processed. This deeply integrates the physical property of moisture with visual features, significantly improving classification robustness under complex working conditions.

[0092] This embodiment adopts moisture content linkage adjustment: when the moisture content of the raw materials changes by more than 2%, the color threshold (ΔH = ±5°) and speed coefficient (k ± 0.1) are automatically corrected to adapt to the physical properties of different batches of raw materials; the wind pressure sensitivity coefficient α is adjustable (0.5 ~ 2.0), supporting the "one category, one policy" sorting strategy and adapting to complex working conditions.

[0093] Shape feature extraction 122 adopts a multi-level screening strategy:

[0094] Primary screening: Exclude dust based on area threshold, area < 0.1mm 2 For dust;

[0095] Secondary screening: Calculate the minimum circumscribed rectangle of the outline, and those with an aspect ratio ≥ 3.5 and a degree of fit > 85% are considered stemmed.

[0096] Level 3 screening: Hu moment matching is performed on fragments with an aspect ratio of 1.5 to 3.0, and those with a similarity of >90% with the preset stem template are classified as stem.

[0097] Motion trajectory analysis 123 specifically includes:

[0098] Extract the moving material in each frame of image by background difference method;

[0099] Calculate the displacement Δd of the material at the same position between adjacent frames, and obtain the instantaneous velocity v = Δd / Δt according to the shooting interval Δt;

[0100] Calculate the speed standard deviation σ_v within 10 frames, and determine that the stem is light when σ_v>0.45m / s; and / or,

[0101] Introducing velocity variance analysis into quality factor correction:

[0102] Establish velocity variance-mass mapping model: σ_v 2 =k / m+b, where k and b are airflow characteristic coefficients;

[0103] The fragment mass m is estimated by the image area and image color space data. When the measured σ_v 2 When the value is >1.5×(k / m+b), it is judged as cut stem.

[0104] The mass estimation model is used to estimate the fragment mass m through image area and image color space data:

[0105] Area: number of pixels × calibration factor (0.02mm 2 / pixel);

[0106] Color density: Dark tobacco stems (saturation ≥ 50%) density = 0.55 g / cm 3 Light-colored tobacco (saturation ≤ 30%) density = 0.18 g / cm 3 ;

[0107] Real-time calibration: When the raw material batch changes, the self-learning unit 124 dynamically updates the k,b values ​​to adapt to the tobacco characteristics of different origins.

[0108] By calculating the standard deviation of the material's velocity, rapid classification learning can be achieved. To further improve classification accuracy, this embodiment also introduces velocity variance analysis as a quality factor correction, which can distinguish true random motion (stem shreds) from pseudo-random motion (wet stems slightly affected by airflow). For example, the stem breaking theory σ_v 2 =0.10, but the measured σ_v 2 =0.13 (<1.5×0.10), then the stems are excluded; the actual measured σ_v of the stems 2=0.32 (>1.5×0.10), which can be captured correctly.

[0109] This embeds the laws of aerodynamics (Stokes drag model) into visual recognition, breaking through the limitations of traditional image algorithms that rely on apparent features. It filters instantaneous fluctuations in airflow with a 1.5-fold safety factor, avoids false triggering caused by over-sensitivity, and improves anti-interference capabilities. The quality estimation model is linked with moisture content self-correction to cover the physical property changes of tobacco from initial baking to re-baking. The quality factor correction enables the system to maintain high-precision closed-loop control when raw materials fluctuate or operating conditions change, reducing the average annual waste of tobacco companies due to misjudgment by approximately 1.27 million yuan per production line.

[0110] The closed-loop control unit 130 drives the linear cylinder to adjust the air door opening through the PID controller 131 according to the silk content proportional signal output by the dynamic morphology analysis module 120, so that the air separation pressure and the silk content deviation are negatively feedback regulated.

[0111] The mathematical relationship between the output signal of the PID controller 131 and the silk content deviation is:

[0112] u(t)=K_p·e(t)+K_i·∫e(t)dt+K_d·de(t) / dt

[0113] Where u(t) is the cylinder adjustment value, e(t) is the difference between the actual silk content and the preset threshold, and K_p, K_i, and K_d are adjustable gain parameters.

[0114] The closed-loop control unit 130 is the control link of the entire air separation system. Its goal is to dynamically adjust the negative pressure (air separation pressure) of the air separation zone according to the real-time tobacco content status of the tobacco material flow, to ensure that the tobacco content is stable near the optimal threshold set by the process, thereby achieving a stable and efficient air separation effect.

[0115] This unit is constructed based on the principle of negative feedback control. Specifically, it receives real-time signal input from the upstream dynamic morphological analysis module 120, which represents the proportion of silk content in the current material flow. The control unit compares this measured value with the preset silk content process threshold and calculates the deviation value. Then, the PID (proportional-integral-differential) control algorithm is used to calculate the corresponding control action amount based on the deviation. This action amount drives the linear cylinder actuator to accurately adjust the air door opening and change the negative pressure in the air selection area. The adjusted wind pressure directly affects the sorting effect of the material, thereby changing the silk content to form a closed-loop control loop. The ultimate goal is to make the measured value of the silk content close to the set value.

[0116] Its detailed components and workflow include:

[0117] Input signal - measured signal of silk content: It comes from the real-time output of the dynamic morphological analysis module 120. This module continuously analyzes the morphological characteristics of the material flow through high-speed imaging and image processing technology, and calculates the silk content ratio value S at the current moment. (t) This signal is the core feedback quantity of closed-loop control.

[0118] Silk content setting threshold: The target silk content value S is set by the operator or the superior control system according to the process requirements. This is a relatively stable reference value.

[0119] Deviation Calculator 132:

[0120] Core function: Calculate the error e(t) between the real-time measured value of silk content and the set value, e(t) = SS (t) .

[0121] When e(t)>0: it means that the measured silk content is lower than the set threshold (there are too few qualified filaments in the material), and it is necessary to reduce the air separation pressure (reduce the suction force) to allow more qualified filaments to remain.

[0122] e(t)<0: This indicates that the measured silk content is higher than the set threshold (there are too many impurities such as stalks in the material), and it is necessary to increase the air separation pressure (increase the suction force) to remove more impurities.

[0123] e(t)=0: reaches the ideal state, no adjustment is required.

[0124] The PID controller 131 receives the error signal e(t) and calculates a control output signal u(t) based on the proportional (P), integral (I), and differential (D) functions. This signal determines the amplitude and direction of the linear cylinder's adjustment.

[0125] The control algorithm is: u(t) = K_p*e(t) + K_i*∫e(t)dt + K_d*de(t) / dt. The proportional term, K_p*e(t), provides an immediate response proportional to the current deviation, e(t). A larger K_p (proportional gain) provides a stronger correction for the current deviation, but too large a value can cause system oscillation. The integral term, K_i*∫e(t)dt, accumulates historical deviations (∫e(t)dt represents the integral of the deviation over time). K_i (integral gain) eliminates steady-state errors (i.e., small, long-lasting deviations). It continues to operate until the deviation is zero. However, a large K_i value can cause the system to respond more slowly or overshoot. The differential term, K_d*de(t) / dt, reflects the rate of change of the deviation (de(t) / dt is the differential of the deviation, i.e., the rate of change). The role of K_d (differential gain) is to predict future deviation trends, provide damping, suppress rapid changes and oscillation tendencies of the system, and make the control process smoother. It is sensitive to high-frequency noise. Adjustable parameters: K_p, K_i, K_d are key parameters that need to be adjusted according to the specific characteristics of the air separation equipment and the material characteristics. Well-tuned PID parameters can ensure that the system responds quickly, stably and with small overshoot. Output limiting: The controller output u(t) is usually limited to ensure that its value is within the acceptable operating range (U_min to U_max) of the damper actuator (linear cylinder 141) to prevent over-adjustment or damage to the equipment.

[0126] The actuator (linear cylinder 141 and damper) is used to convert the electrical signal u(t) output by the PID controller 131 into a physical displacement change of the damper opening, including:

[0127] Electric-pneumatic converter / servo valve: receives a control signal u(t), such as a voltage or current signal, and proportionally adjusts the flow and direction of compressed air.

[0128] The linear cylinder 141 is used to generate a linear motion corresponding to u(t) by its piston rod under the drive of compressed air.

[0129] The damper / baffle is used to be directly connected to the piston rod of the cylinder 141. The linear motion of the cylinder 141 drives the damper to rotate or translate, thereby changing the flow cross-sectional area of ​​the ventilation duct, that is, the damper opening.

[0130] Position feedback: In this embodiment, a displacement sensor, such as an LVDT or a magnetostrictive sensor 144, is installed on the cylinder 141 or the damper. This embodiment uses the magnetostrictive sensor 144 to form a secondary closed loop of the damper position, ensuring that the cylinder 141 moves accurately to the position required by the u(t) instruction, thereby improving control accuracy and anti-interference capability.

[0131] Controlled object and negative feedback effect:

[0132] Controlled object: Negative pressure in the air separation area (air separation pressure). The damper opening is the main regulating variable.

[0133] The negative feedback regulation mechanism is:

[0134] S_measured(t) increases (e(t) becomes negative) -> PID output u(t) increases -> cylinder 141 drives the damper opening to increase -> air duct resistance decreases -> air separation negative pressure decreases -> suction force weakens -> qualified filaments sucked away decrease -> S_measured(t) tends to decrease (correcting the deviation).

[0135] S_measured(t) decreases (e(t) becomes positive) -> PID output u(t) decreases -> cylinder 141 drives the damper opening to decrease -> air duct resistance increases -> air separation negative pressure increases -> suction force increases -> more impurities are sucked away -> S_measured(t) tends to increase (correcting the deviation).

[0136] This reverse relationship between the change in air separation pressure and the deviation in silk content is the core embodiment of negative feedback, which results in the silk content S_measured(t) automatically stabilizing near the set value S_setpoint.

[0137] The closed-loop control unit 130 in this embodiment has the following advantages: it can respond in real time to fluctuations in the silk content of the material caused by uneven incoming materials, automatically adjust the wind pressure, maintain a stable sorting effect, and have dynamic adaptability; by setting S_setpoint, the silk content level of the final product can be accurately controlled; by fine-tuning the K_p, K_i, K_d parameters, the best balance can be achieved between response speed, stability and robustness; it can reduce manual intervention, realize continuous and stable automated production, and achieve automation and continuity.

[0138] The wind pressure actuator 140 includes a continuously adjustable damper 142 driven by a linear cylinder 141 and a position feedback sensor. The damper opening is dynamically adjusted according to the adjustment signal so that the wind selection pressure and the silk content deviation have a negative correlation linear response.

[0139] The mapping relationship between the air door opening adjustment and the silk content deviation is: K=K0-α·ΔC, where K is the air door opening, K0 is the reference opening, ΔC is the deviation between the actual silk content and the set value of 7%, and α is the wind pressure sensitivity coefficient, α=0.5~2.0.

[0140] The wind pressure actuator 140 also includes an additional airflow stabilization device. The airflow stabilization device in this embodiment is a honeycomb guide plate provided at the damper outlet, which is made of corrosion-resistant, high-strength materials, such as anodized aluminum alloy and stainless steel, and has a porous honeycomb structure as a whole. Its core feature is that it is composed of a large number of closely arranged, slender parallel straight-through pipes (honeycomb holes), and its through-hole diameter is ≤3mm. This small aperture design is the key, which can force the irregular, large-scale turbulent vortexes from the damper to be divided into a large number of tiny streams. At the same time, the airflow is constrained to flow strictly in the axial direction of the honeycomb duct (that is, perpendicular to the plane of the guide plate and parallel to the mainstream direction of the air duct), which greatly suppresses the transverse (radial) velocity component and swirl. The hole wall produces a viscous shear effect on the airflow, effectively dissipating the turbulent kinetic energy and significantly reducing the turbulence of the downstream airflow.

[0141] The distance between the guide plate and the damper is 0.2 times the duct diameter. When the damper is adjusted, especially at a small opening, a high-speed jet may form at its outlet. The 0.2D_h spacing provides this jet with a brief space for diffusion and momentum exchange, initially widening and homogenizing its velocity profile. This prevents the high-speed jet from directly impacting the honeycomb panel inlet end face, causing localized high pressure loss, noise, or structural vibration. Ensure that the coefficient of variation of airflow uniformity is less than 5%.

[0142] The wind pressure actuator 140 uses a linear cylinder 141 and a magnetostrictive sensor 144 with a resolution of 0.01mm to achieve a 50ms response to the damper opening; PID control and the wind pressure mapping model (K=K0-α·ΔC) work together to shorten the fluctuation suppression time to within 3 seconds and achieve millisecond-level closed-loop regulation.

[0143] Example 2

[0144] The embodiment of the present invention also provides a method for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition. Based on the dynamic control system for the silk content of stalks described in the first embodiment, see Figure 3-5 , including the following steps:

[0145] S1. The image of the material in the stem tag collection box is captured in real time by the visual system, and the silk content is calculated based on the dynamic morphological analysis module 120, specifically:

[0146] S11: high-speed image acquisition, starting the stroboscopic light source 112 (flash duration ≤ 100 μs), and triggering the CCD camera synchronously with the falling speed of the material (≥ 500 fps);

[0147] S12: Continuously capture a three-channel dynamic image sequence just above the stem tag collection box;

[0148] S13: Dynamic morphological analysis

[0149] S131: Color space segmentation and shape feature extraction;

[0150] S132: Motion trajectory calculation: perform background subtraction on 10 consecutive frames of images, extract the contour of the moving material, calculate the displacement Δd of the fragment (sub-pixel accuracy), and the instantaneous velocity v = Δd / Δt (Δt = 2ms); calculate the velocity standard deviation σ_v, and mark the object as a lightweight candidate when σ_v>0.45m / s;

[0151] S133: Mass correction determination, first estimate the fragment mass: m = Area × ρ (ρ is distributed by color: tobacco ρ = 0.18 g / cm 3 , tobacco stem ρ=0.55g / cm 3 ), then calculate the theoretical velocity variance: σ_v 2 Theoretical = k / m+b (k, b are calibrated by the fan configuration), the final determination of the stem is: when σ_v 2 Measured >1.5×σ_v 2 Theoretically, the aspect ratio must be less than or equal to 1.5.

[0152] S134: Output of silk content, silk content C = (∑ silk stem pixel area / ∑ total material pixel area) × 100%.

[0153] S2. Compare the measured value of silk content with the preset standard value and generate a deviation signal. The specific operation is as follows:

[0154] S21: Comparator operation: e(t) = C_measured - C_setting, where C_setting in this embodiment defaults to 7% and can be remotely configured;

[0155] S22: If |ΔC|≤0.5%, maintain the current damper opening;

[0156] If 0.5%<|ΔC|≤1%, then enter PID fine-tuning mode;

[0157] If |ΔC|>1%, emergency adjustment (α coefficient × 2) is initiated.

[0158] S3. Input the deviation signal into the PID controller 131, output the displacement control value of the linear cylinder 141, and perform a discretized PID operation: u(k) = K_p·e(k)+K_i·T_s·\sum_{i=0}^{k}e(i)+K_d·\frac{e(k)-e(k-1)}{T_s}, with a control period of 200ms. After dynamic optimization by the self-learning unit 124, K_p = 0.8, K_i = 0.05, and K_d = 0.3; output the displacement limit of the cylinder 141: the displacement of the cylinder 141 u(k)∈[U_min,U_max], corresponding to a damper opening of 0-100%.

[0159] S4. Drive the damper opening to continuously change, causing the air separation pressure to reversely compensate for the deviation in silk content. The actuator responds: Linear cylinder 141 receives the u(k) signal and pushes the damper to the target opening K = K_0 - α·ΔC, where K_0 is the reference opening for the current operating condition. For example, when the moisture content is 12%, K_0 = 45%. The wind pressure sensitivity coefficient α defaults to 1.2, which is increased to 2.4 in emergency mode. The airflow is rectified by a honeycomb guide plate (aperture 3mm) to ensure that the turbulence in the air separation box is less than 5%.

[0160] S5. Verify the damper status through position feedback until the silk content returns to the set threshold range.

[0161] Position feedback verification: The magnetostrictive sensor 144 transmits the damper opening K_actual in real time. If |K_target - K_actual|>0.5mm, the closed-loop position compensation of the cylinder 141 is triggered. Wire content regression detection continuously monitors for three control cycles (600ms). When |ΔC| is continuously ≤0.5%, it is determined to be stable. After five consecutive adjustments, if |ΔC|>1%, abnormal safety processing is performed and it is switched to safe mode: K_safe=K_0+0.3×(M_current-M_baseline), where M is the moisture content. The alarm information is uploaded to the MES system, triggering manual intervention.

[0162] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0163] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0164] The units described as separate components may or may not be physically separated. As a unit, a person of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0165] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0166] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or grid device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0167] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition, characterized in that: include: A high-speed visual recognition unit, located above the stem tag collection box, includes a high-frame-rate CCD camera and a stroboscopic light source to continuously capture images of the material as it falls. The dynamic morphological analysis module is connected to the CCD camera and performs the following collaborative discrimination: Color space segmentation: Separate light-colored tobacco with a saturation of ≤30% from dark-colored tobacco stems with a saturation of ≥50% in the HSV color space; Shape feature extraction: Calculate the aspect ratio of the material outline, define the aspect ratio ≥ 3 as a stem, and the aspect ratio ≤ 1.5 and the area < 5mm 2 For the stem shreds; Motion trajectory analysis: Calculate the material falling velocity variance based on continuous frame images. Fragments with velocity variance > 0.2m2 / s2 are identified as stems. The closed-loop control unit drives the linear cylinder to adjust the air door opening through the PID controller according to the silk content proportional signal output by the dynamic morphology analysis module, so that the air separation pressure and the silk content deviation are negatively regulated; The air pressure actuator includes a continuously adjustable air door driven by a linear cylinder and a position feedback sensor, which dynamically adjusts the air door opening according to the adjustment signal so that the air selection pressure and the silk content deviation have a negative correlation linear response.

2. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The motion trajectory analysis specifically includes: Extract the moving material in each frame of image by background difference method; Calculate the displacement Δd of the material at the same position between adjacent frames, and obtain the instantaneous velocity v = Δd / Δt according to the shooting interval Δt; Calculate the speed standard deviation σ_v within 10 frames, and determine that the stem is light when σ_v>0.45m / s; and / or, Introducing velocity variance analysis into quality factor correction: Establish velocity variance-mass mapping model: σ_v 2 =k / m+b, where k and b are airflow characteristic coefficients; The fragment mass m is estimated by the image area and image color space data. When the measured σ_v 2 When the value is >1.5×(k / m+b), it is judged as cut stem.

3. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The color space segmentation uses adaptive thresholding: For the H channel, set the tobacco identification intervals to [0°, 30°] (light yellow) and [150°, 180°] (light brown); Set the tobacco stem recognition interval [50%, 100%] for the S channel; The color and shape features are integrated, and pixels are marked as stems when they meet the color criteria and the aspect ratio is ≤1.

5.

4. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The flash frequency of the stroboscopic light source is synchronized with the falling speed of the material, and the flash duration is ≤100μs, eliminating motion blur to accurately capture the trajectory.

5. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The mapping relationship between the damper opening adjustment and the silk content deviation is: K=K0-α·ΔC Where K is the air door opening, K0 is the reference opening, ΔC is the deviation between the measured value of silk content and the set value of 7%, α is the wind pressure sensitivity coefficient, α = 0.5 ~ 2.

0.

6. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The mathematical relationship between the output signal of the PID controller and the silk content deviation is: u(t)=K_p·e(t)+K_i·∫e(t)dt+K_d·de(t) / dt Where u(t) is the cylinder adjustment value, e(t) is the difference between the actual silk content and the preset threshold, and K_p, K_i, and K_d are adjustable gain parameters.

7. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The shape feature extraction adopts a multi-level screening strategy: Primary screening: Exclude dust based on area threshold, area < 0.1mm 2 For dust; Secondary screening: Calculate the minimum circumscribed rectangle of the outline, and those with an aspect ratio ≥ 3.5 and a degree of fit > 85% are considered stemmed. Level 3 screening: Hu moment matching is performed on fragments with an aspect ratio of 1.5 to 3.0, and those with a similarity of >90% with the preset stem template are classified as stem.

8. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1 is characterized in that: The wind pressure actuator includes an airflow stabilizing device: A honeycomb guide plate is provided at the damper outlet, with a through-hole diameter of ≤3mm. The distance between the guide plate and the damper is 0.2 times the diameter of the air duct, ensuring that the coefficient of variation of air flow uniformity is less than 5%.

9. The system for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition according to claim 1, characterized in that: The dynamic morphological analysis module includes a self-learning unit: Record historical sorting data, including silk content, air door opening, and raw material moisture content; When the moisture content of the raw material changes by more than 2%, the color segmentation threshold ΔH = ±5° and the speed variance coefficient k ±0.1 are automatically adjusted.

10. A method for dynamically controlling the silk content of stalks based on multi-dimensional visual recognition, based on the system for dynamically controlling the silk content of stalks according to any one of claims 1 to 9, characterized in that: The steps include: S1. Use a visual system to capture images of the material in the stem collection box in real time and calculate the silk content based on a dynamic morphological analysis module; S2. Compare the measured value of silk content with the preset standard value and generate a deviation signal; S3. Input the deviation signal into the PID controller, which outputs the displacement control value of the linear cylinder; S4. Drive the damper opening to continuously change, so that the air separation pressure reversely compensates for the deviation in silk content; S5. Verify the damper status through position feedback until the silk content returns to the set threshold range.

Images