Green Chinese onion processing and cleaning method integrating visual identification and remote monitoring

Abstract

The invention relates to the technical field of agricultural product processing, and discloses a green Chinese onion processing and cleaning method integrating visual identification and remote monitoring, which comprises the following steps of: directly mapping optical characteristics of organic stains into logic level signals through synchronous acquisition of visible light and 950nm near-infrared band reflection signals and processing of a voltage comparator; the high-pressure or low-pressure cleaning mechanism is driven to execute differential flushing based on logic judgment, essential recognition and targeted cleaning of stain components are achieved, cleaning energy is adjusted in a self-adaptive mode through reflected signal fluctuation characteristics, physical damage is avoided while the decontamination effect is guaranteed, and the cleaning effect is improved. And the vortex cavitation nozzle is used for converting water flow pressure energy into micro cleaning force, so that the cleaning depth is obviously improved.

Description

Technical Field

[0001] The present invention relates to a method for processing and cleaning green onions by integrating visual recognition and remote monitoring, belonging to the technical field of agricultural product processing. Background Art

[0002] In the field of automated agricultural product processing, the cleaning technology based on optical recognition has long relied on visible light image analysis to identify surface stains. The existing mainstream solutions usually use RGB cameras to capture the color and texture features of objects, and judge stains through algorithm comparison with a preset model. However, there is a fundamental bottleneck in this technical path: color features cannot represent the essence of the substance composition, resulting in highly similar optical features of organic stains such as water-containing sticky soil and the outer layer of the green onion epidermis, such as the dry outer layer of green onions, in the visible light band, forming an irreconcilable recognition contradiction.

[0003] In a typical green onion processing scenario, the above contradiction directly leads to systematic failure: when mixed stains pass through the detection area, it is difficult for the system to distinguish foreign objects with similar optical features from the main body, and its decision-making logic is forced to rely on empirical thresholds, which results in the blind action of high-pressure water flow on the epidermis area causing mechanical damage, or low-pressure flushing being unable to peel off deep organic attachments, forming a vicious cycle of ineffective cleaning and excessive damage; to compensate for the recognition defects, the existing technologies generally adopt strategies such as increasing the flushing duration or global high pressure, which not only cause water resource waste to be above the industry average, but also lack a perception mechanism for the mechanical response of organisms and cannot avoid the hidden corruption risk caused by damage to the microscopic tissue structure.

[0004] Although the industry has tried to introduce multispectral imaging or high-precision sensors to improve the recognition rate, these solutions need to rely on complex image processing algorithms and central controllers, which not only significantly increase the hardware cost and maintenance difficulty, but also the millisecond-level decision-making delay leads to a lag in the response of the actuator, making it unable to adapt to the rhythm of high-speed production lines. The deeper core contradiction lies in that the existing technologies separate the substance composition recognition and execution control into two independent subsystems and fail to establish a direct, real-time, and reliable physical-level mapping mechanism from optical features to mechanical actions. Therefore, how to break through the essential limitations of color recognition through a physical-level signal conversion mechanism and synchronously achieve the essential recognition of stain components, the real-time perception of mechanical response, and the precise control of energy release during the conveying process has become the technical problem to be solved by the present invention. Summary of the Invention

[0005] The present invention provides a method for processing and cleaning green onions by integrating visual recognition and remote monitoring, and its main purpose is to solve the problems of inaccurate cleaning strategies, resource waste, and damage to agricultural products caused by the inability of existing optical recognition technologies to distinguish the essence of substance composition.

[0006] To achieve the above object, a scallion processing and cleaning method integrating visual recognition and remote monitoring provided by the present invention includes the following steps: Step a, on the conveying path of scallions, obtain the reflection intensity signal of the scallion surface in the visible light band; at the same time, obtain the reflection intensity signal of the scallion surface in the near-infrared band with a central wavelength of 950 nanometers. Step b, input the reflection intensity signal in the near-infrared band into the first voltage comparator. When the reflection intensity signal in the near-infrared band is lower than the organic matter light absorption threshold determined by pre-calibration, the first voltage comparator outputs a first logic high-level signal. Step c, input the reflection intensity signal in the visible light band into the second voltage comparator. When the reflection intensity signal in the visible light band is lower than the scallion presence threshold determined by pre-calibration, the second voltage comparator outputs a second logic high-level signal. Step d, based on the AND logic judgment of the first logic high-level signal and the second logic high-level signal, drive the high-pressure cleaning mechanism to strongly rinse the scallions. The strong rinsing is performed when the first logic high-level signal is true and the second logic high-level signal is true. Step e, when the first logic high-level signal is false and the second logic high-level signal is true, drive the low-pressure cleaning mechanism to perform a conventional rinse on the scallions.

[0007] Preferably, the reflection intensity signal in the visible light band is obtained by the first photosensitive sensing unit; the reflection intensity signal in the near-infrared band with a central wavelength of 950 nanometers is obtained by the second photosensitive sensing unit, and the second photosensitive sensing unit is only sensitive to the near-infrared light with a central wavelength of 950 nanometers.

[0008] Preferably, both the high-pressure cleaning mechanism and the low-pressure cleaning mechanism are independent water circuits controlled by solenoid valves, and the first logic high-level signal and the second logic high-level signal directly control the opening and closing of the solenoid valves through a logic gate circuit.

[0009] Preferably, the step of strong rinsing further includes: while the high-pressure cleaning mechanism starts rinsing, obtain the reflection intensity signal of the second photosensitive sensing unit; extract the AC fluctuation component of the reflection intensity signal through a high-pass filter; input the AC fluctuation component into the third voltage comparator. When the AC fluctuation component exceeds the vibration amplitude threshold, based on the AC fluctuation component, adjust the energy release mode of the high-pressure cleaning mechanism; the energy release mode is to adjust the single cleaning action of the high-pressure cleaning mechanism into a gradient cleaning sequence composed of at least two pulses with increasing pressure or increasing duration.

[0010] Preferably, the adjustment condition of the energy release mode is that the difference between the AC fluctuation component and the AC fluctuation component of the reflection intensity signal of the first photosensitive sensing unit exceeds the differential fluctuation threshold determined by pre-calibration.

[0011] Preferably, an eddy cavitation nozzle is provided at the end of the high-pressure cleaning mechanism. The eddy cavitation nozzle has an internal channel structure that reduces the local pressure of the flowing high-pressure water flow below the saturated vapor pressure of water, thereby forming cavitation micro-jet flows on the surface of the green onions.

[0012] Preferably, the internal channel structure of the eddy cavitation nozzle includes a spiral tapered channel that increases the water flow velocity, thereby causing a decrease in local pressure, satisfying Bernoulli's principle, i.e.: , where, is the local pressure of the water flow, is the local flow velocity of the water flow, is the density of the water flow, is the energy constant in Bernoulli's principle.

[0013] Preferably, the water pressure of the high-pressure cleaning mechanism is higher than 1.5 MPa.

[0014] Preferably, the following steps are further included: after a predetermined time delay after each grading cleaning operation is completed, the reflection intensity signal in the near-infrared band with a central wavelength of 950 nm of the second photosensitive sensing unit is acquired again; and the acquired reflection intensity signal is input into the fourth voltage comparator. When the signal is lower than the water film residue detection threshold determined by calibrating the epidermis of healthy green onions, the fourth voltage comparator outputs a rejection instruction, and the rejection instruction is used to indicate that the green onions have quality defects and drives the rejection device to remove the green onions from the production line.

[0015] Preferably, the range of the predetermined time delay is from 30 milliseconds to 80 milliseconds.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. By independently collecting the reflection signals in the visible light and 950 nm near-infrared bands, the optical characteristics are directly converted into logic level signals through a voltage comparator. This mechanism bypasses the traditional image processing process and converts the characteristic absorption of organic stains in the near-infrared into an electrical instruction that can drive the actuator from a physical level. This direct mapping based on the optical essence of substances makes the system different from the traditional color recognition mode and realizes the essential distinction between organic stains and agricultural product bodies on the conveying path.

[0017] 2. When the high-pressure cleaning is triggered, the AC fluctuation component of the near-infrared reflection signal is synchronously extracted. This fluctuation feature forms an implicit association with the mechanical response of the green onion structure under the impact of water flow: when the fluctuation amplitude exceeds the preset threshold, it indicates that there is a risk of fragility in the target structure. The system then automatically converts a single cleaning operation into a gradient pressure sequence, and through a progressive energy release mode of pre-wetting to loosen the stains first and then main impact, while maintaining the decontamination efficiency, it avoids irreversible damage to the microscopic cell structure.

[0018] 3. When high-pressure water flows through a nozzle with a spirally tapered channel, a local low-pressure area is generated according to the Bernoulli principle, and a group of cavitation bubbles is spontaneously formed. When the cavitation bubbles collapse on the surface of the agricultural products, the released microjets overcome the van der Waals force at the microscopic scale and peel off the micron-sized particles embedded in the epidermal folds. This process does not rely on additional energy or control units. It only uses the flow channel geometry design to promote the conversion of the pressure energy of the macroscopic water flow into the microscopic interface cleaning force, thereby achieving a substantial improvement in the cleaning depth.

[0019] 4. At a fixed delay after each cleaning action, the thickness difference of the residual water film is detected by the reflection intensity of the 950nm band. Corrupted tissue maintains a thicker water film due to its enhanced hydrophilicity, causing the near-infrared signal to remain below the healthy tissue threshold. This mechanism reuses the physical traces generated by the cleaning process and the existing optical sensing unit. Without adding new hardware, the simple cleaning action is expanded into an early screening link for quality defects, realizing the functional coupling of the processing link and quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a comparison diagram of the near-infrared reflection signal voltage attenuation curve over time of the present invention; Figure 2 This is a timing diagram of the delayed detection and rejection process after cleaning of the present invention; Figure 3 This is a graph showing changes in water pressure over time under different cleaning modes of the present invention; Figure 4 This is a flow chart of vibration amplitude determination for adaptive cleaning mode switching according to the present invention.

[0021] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0022] In order to make the objectives, technical solutions and advantages of the present invention clearer, the specific implementation methods of the present invention will be described in detail below. However, it should be understood that the specific embodiments described here are intended to explain the present invention rather than to limit the present invention.

[0023] A method for processing and cleaning scallions that integrates visual recognition and remote monitoring provided by an embodiment of the present invention systematically integrates multiple stages, such as optical sensing based on the essential material components, real-time logical control at the physical level, adaptive energy regulation based on mechanical response, microscopic cleaning based on fluid dynamics optimization, and invasive detection of quality defects based on physical traces after cleaning, to jointly build a logical closed-loop and highly automated processing and quality control process; the specific engineering parameters of the hardware system of the present invention are set as follows. The spiral tapered channel of the vortex cavitation nozzle adopts a structure with an inlet inner diameter of 5.0 mm, an outlet inner diameter of 1.5 mm, an effective channel length of 25 mm, and an internal spiral lead of 15 mm; the cut-off frequency is determined by the following calibration procedure: collect the signals of the second photosensitive sensing unit and perform spectral analysis in a static state without water flow impact, identify the background noise frequency range mainly concentrated below 30 Hz, then collect the signals again under the impact of the rated water pressure and analyze their spectra, identify the energy main peak frequency range corresponding to the mechanical vibration of the scallion body, usually located between 50 Hz and 200 Hz, and finally set the cut-off frequency to 40 Hz, which is between the noise frequency band and the signal main peak frequency band; the working water pressure of the high-pressure cleaning mechanism is set in the range of 1.4 MPa to 1.8 MPa, and the single-pulse high-pressure flushing reference pressure and the main impact pressure of the gradient cleaning sequence in the embodiment are both uniformly 1.5 MPa.

[0024] In the production line deployment scenario, green onions enter the detection and cleaning area along with the conveying mechanism. The core technical obstacle here is that traditional visible light imaging cannot physically distinguish between the water-containing organic stains attached to the surface of green onions and the dried or wrinkled epidermis of the green onions themselves. The high optical similarity presented by both in the visible light band is the root cause of the blind spot in the existing technology for identification and the inaccuracy of strategies. To overcome this challenge, the procedure adopted in the present invention is set to collect optical reflection signals of two different bands in parallel on the conveying path. Specifically, a first photosensitive sensing unit is used to obtain the reflection intensity signal of the green onion surface in the visible light band, and simultaneously, a second photosensitive sensing unit that is highly sensitive only to near-infrared light with a central wavelength of 950 nanometers is used to obtain the reflection intensity signal of the same area in this specific near-infrared band. The physical basis of this design is that organic stains rich in organic matter have a strong characteristic absorption effect on the near-infrared light with a central wavelength of 950 nanometers, while the healthy plant tissue of green onions shows high reflectivity. This inherent difference in the spectral characteristics of the material composition provides an irrefutable basis for judgment for the system that goes beyond apparent color and texture, thus achieving an essential identification of organic pollutants at the physical level. Given the stringent requirement of high-speed automated production lines for real-time decision-making, the inherent millisecond-level processing delay in the traditional solution that relies on a central processing unit for image comparison and analysis constitutes an insurmountable performance bottleneck. The present invention circumvents this by means of a direct electrical logic processing procedure, which directly inputs the two obtained reflection intensity signals into independent voltage comparators for physical-level instant processing. Among them, the reflection intensity signal in the near-infrared band is fed into the first voltage comparator, and the comparison reference inside it is an organic matter light absorption threshold determined by pre-calibration. The calibration process of this threshold is designed as follows: in the standard production line lighting environment, the second photosensitive sensing unit is used to repeatedly measure various typical organic stain samples, and the statistical average value of their stable output voltage signals is taken as this threshold. When the real-time measured near-infrared reflection intensity signal is lower than this threshold due to absorption by the stain, it indicates that the target organic matter is detected, and the first voltage comparator immediately outputs a first logic high-level signal. At the same time, the reflection intensity signal in the visible light band is fed into the second voltage comparator, and the internal comparison reference is a green onion presence threshold. The calibration process of this threshold is to measure the background reflection intensity during the no-load operation of the conveyor belt and record the reflection intensity when standard-sized green onions pass through, and take a stable intermediate value between the two as this threshold. Therefore, when the real-time signal is lower than this threshold, it indicates that a green onion body is passing through the detection area, and the second voltage comparator outputs a second logic high-level signal.

[0025] Furthermore, these two logic level signals are directly fed into a hardware logic gate circuit for AND logic judgment, and the output level of the judgment result is directly used to control two independent waterways controlled by solenoid valves, which respectively constitute a high-pressure cleaning mechanism and a low-pressure cleaning mechanism; when both the first logic high-level signal and the second logic high-level signal are true, that is, when the system physically determines that there is a scallion in the current detection area and there is organic stain attached to it, the output of the AND logic judgment circuit will drive the solenoid valve of the high-pressure cleaning mechanism to open instantaneously, and perform a strong flushing on this area; correspondingly, when the first logic high-level signal is false and the second logic high-level signal is true, that is, it is determined that the scallion exists but no organic stain is found, the solenoid valve of the low-pressure cleaning mechanism is driven to open, and a conventional low-pressure flushing is performed; this architecture constructs a physical-level mapping path without software delay between the optical characteristics and the mechanical actuator, ensuring a high degree of synchronization in time and space between the triggering of the cleaning action and the precise positioning of the stain. However, the strong flushing process itself also entails the potential risk of causing physical damage to agricultural products. Especially for some scallion individuals with relatively fragile textures, the unchanging high-pressure water flow is very likely to cause microscopic structural damage to the epidermis and even internal tissues. To avoid such risks, the present invention also incorporates an adaptive energy adjustment mechanism in the strong flushing step; at the moment when the high-pressure cleaning mechanism is controlled to start flushing, the system will synchronously enable a high-pass filter to continuously extract its AC fluctuation component from the output signal of the second photosensitive sensing unit in real time. The amplitude of this AC fluctuation component is directly related to the amplitude of the mechanical vibration response generated by the scallion body under the impact of the water flow; this AC fluctuation component is then input into the third voltage comparator and continuously compared with a preset vibration amplitude threshold. The calibration procedure of this threshold is designed as an offline experimental process: select multiple batches of scallion samples, apply continuously increasing water pressure impacts from low to high to them, and use a high-power microscopic device to synchronously observe the integrity of the epidermal cell structure. The amplitude of the reflected signal fluctuation corresponding to the critical impact force at which irreversible microscopic damage begins to occur is rigorously set as the vibration amplitude threshold; during actual operation, once the AC fluctuation component monitored in real time exceeds this threshold, that is, it indicates that the current impact energy may cause damage to the target scallion body, the system will, based on this judgment, immediately adjust the original single high-pressure cleaning action into a gradient cleaning sequence composed of at least two pulses with increasing pressure or increasing duration. For example, by first performing a low-pressure pre-wetting pulse to loosen the stain and then applying a main impact pressure pulse in a progressive energy release mode, the intelligent avoidance of physical damage is achieved while ensuring the decontamination efficiency. The vibration amplitude threshold and the reference duration The setting is completed through a set of combined offline calibration procedures. First, on a test bench with the same lighting and electrical environment as the production line, the most fragile scallion samples are selected. A series of water pressure pulses starting from 0.5 MPa with a step size of 0.05 MPa are applied using a pressure controller driven by a stepper motor. When each pulse acts, the AC fluctuation component of the second photosensitive sensing unit is synchronously collected. After the pulse ends, immediately observe the impacted area of the sample through a 100-fold microscope. Defining the appearance of more than 5 cell wall ruptures or permanent deformations within a 1-square millimeter field of view as microdamage, and record the statistical average value of the peak corresponding to the previous pressure level before this critical pressure. Then, after multiplying this average value by a safety factor of 0.9, it is determined as ; Subsequently, select healthy scallion samples attached with standard viscous organic stains. Apply single-pulse rinses with a duration starting from 20 ms and increasing in steps of 10 ms under standard high pressure. Measure the stain residue rate after rinsing by the weighing method, and determine the shortest rinsing duration corresponding to the first time the residue rate is lower than 1% as the main impact pulse duration in the gradient cleaning sequence, that is, the reference duration. .

[0026] To further enhance the peeling ability of microscopically attached particles, at the spray end of the high-pressure cleaning mechanism, a vortex cavitation nozzle designed based on the principle of hydrodynamics is specially configured. Its core structure is a spiral tapered channel inside; when high-pressure water flow with a pressure stably maintained above 1.5 MPa flows through this specially designed channel, according to the fluid energy conservation relationship described by Bernoulli's principle, that is , where is the local water pressure, is the local water flow velocity, [[ID=1B]] is the water density, is the energy constant, and the sudden increase in the water flow velocity will lead to a decrease in its local pressure Correspondingly, it suddenly drops below the saturated vapor pressure of water; this purely physical process causes tiny cavitation bubbles to spontaneously and densely form in the water flow. These cavitation bubbles collapse instantly when they impact the surface of the scallions and release cavitation microjets with highly concentrated energy within microseconds. This microjet can generate instantaneous impacts of several atmospheres and strong shear effects at the microscopic scale, sufficient to overcome the van der Waals forces that bind micron-sized dust particles deep in the epidermal folds, thus achieving a deep cleaning effect that is difficult to achieve with conventional water jet technology. Finally, the present invention also cleverly reuses the physical traces generated by the cleaning action itself and seamlessly integrates the online quality control function into the end of the processing flow; specifically, after each grading cleaning action is completed and after a predetermined time delay between thirty milliseconds and eighty milliseconds, the system will drive the second photosensitive sensing unit again to obtain the reflection intensity signal of the scallion surface at this moment for the near-infrared band with a central wavelength of 950 nanometers; the setting of this predetermined time delay is based on the physical property differences between the hydrophobicity of the healthy scallion epidermis and the hydrophilic tissues due to damage or incipient spoilage. During this delay period, the residual water film on the surface of healthy tissues will quickly drain or significantly thin, while damaged or spoiled tissues will maintain a relatively thicker water film due to their increased hydrophilicity. The thicker water film will absorb more near-infrared light, resulting in a significant decrease in the reflection signal intensity; the re-obtained reflection intensity signal is input into the fourth voltage comparator and compared with a pre-calibrated water film residue detection threshold. The calibration of this threshold is achieved by statistically analyzing the near-infrared reflection intensities measured from a large number of clean and healthy scallion samples after cleaning and passing through the above standard delay, thereby establishing a signal baseline representing the health status; when the real-time signal intensity is lower than this threshold, it indicates that abnormal water film residues are detected, and the fourth voltage comparator will output an ejection instruction to drive the ejection device at the end of the production line to precisely separate the scallions determined to have latent quality defects from the mainstream products; for the dynamic update mechanism of the stain optical feature library, its background analysis program specifically integrates a density-based DBSCAN clustering algorithm. The neighborhood radius parameter Eps of this algorithm is set to twice the standard deviation of all near-infrared reflection intensity signal values in the existing feature library, and the minimum number of points parameter MinPts is fixed at 100; when the system is running, all near-infrared signal values triggering high-pressure cleaning events are input into this algorithm model in real time. When the cumulative number of noise points determined by the algorithm to not belong to any known cluster exceeds 500 within a continuous one-hour monitoring period, the system determines that a new data cluster with a significant statistical difference from the existing feature library has been detected, automatically triggers an alarm on the remote monitoring terminal, and simultaneously highlights the timestamps and production line position information corresponding to the noise points of this batch to guide physical sampling and subsequent feature library update operations.

[0027] At the same time, when the high-pressure water flow impacts and effectively removes the organic stains on the surface of the green onion, due to the strong absorption characteristics of the stain debris to the 950nm near-infrared light, the AC fluctuation component of the second photosensor unit It will produce violent and random fluctuations, and the visible light AC fluctuation component of the first photosensitive sensor unit It is not sensitive to this, resulting in the difference signal between the two On the contrary, when the impact energy mainly acts on the onion itself and causes damage risk, the signal fluctuations of the two channels are derived from the macroscopic physical vibration of the onion, and their waveforms are highly correlated, and the difference is maintained at a low level; therefore, the differential fluctuation threshold The calibration procedure is set as follows: first, standard high-pressure flushing is applied to group B samples, and the difference signal is recorded and calculated simultaneously. Peak collection , and then take the statistical mean of the set , and set the threshold to In actual operation, only when and Only when these two conditions are met at the same time will the system finally determine that the current impact poses a risk of structural damage and switch to the gradient cleaning sequence, thereby effectively distinguishing the violent signal generated by stain removal from the onion body overload vibration.

[0028] Baseline duration in gradient wash sequence Water film residual detection threshold in quality inspection The setting is completed through a set of joint calibration procedures. The procedure first applies a single pulse high voltage of 1.5 MPa to the samples in group B. The pulse duration starts from 20 milliseconds and increases in steps of 5 milliseconds. The stain residual rate after each rinse is measured by weighing method. The shortest rinse time corresponding to the first residual rate below 1% is determined as the benchmark time. ; Then, 50 samples from group H and group D were taken, and the obtained attenuation curve was used to find the optimal function under the same wetting conditions. Calculate the optimal detection delay time with maximum signal separation , and the water film residual detection threshold Set in At the arithmetic mean of the two groups of sample signal voltages at the moment, that is, This method integrates the originally independent parameter settings into a data-driven optimization process based on measurable physical endpoints.

[0029] Example 1: In this example, in a scallion processing center that continuously conducts high-throughput processing, its production line faces a batch of scallions just harvested from moist soil after rain. This batch of raw materials presents extremely challenging and complex working conditions. Not only are they generally attached with organic matter dirt with high water content and strong viscosity, but there are also significant differences in the size, texture, and maturity of the scallion bodies, among which there are some tender scallions with fragile textures. Traditional cleaning strategies that rely on global high pressure or fixed-duration cleaning will inevitably lead to a technical dilemma in this scenario, that is, strong flushing will damage the microscopic tissue structure of the tender scallions, while gentle flushing cannot effectively remove the viscous dirt, resulting in a negative result of coexistence of incomplete cleaning and physical damage.

[0030] When a scallion with both wet dirt and dry dust attached to its surface enters the detection area, the system simultaneously collects visible light and near-infrared band reflection signals with a central wavelength of 950 nanometers, and conducts physical-level voltage comparison and logical judgment. Given the strong absorption of near-infrared light by the wet organic matter dirt, the signal intensity of the second photosensitive sensing unit will drop sharply, triggering a first logic high-level signal. At the same time, the first photosensitive sensing unit confirms the presence of the scallion and triggers a second logic high-level signal. The AND logic judgment result of the two immediately drives the high-pressure cleaning mechanism to start. Here, the system first resolves the fundamental contradiction between cleaning intensity and precision. It does not blindly perform flushing, but completely hands over the trigger right of the high-pressure water flow to the recognition result of organic stains based on the spectral characteristics analysis of the material composition, thereby ensuring the targeted release of energy. Furthermore, when the high-pressure water flow impacts a relatively fragile tender scallion, another synergistic mechanism of the solution is activated. The mechanical vibration of the scallion body caused by the water flow impact is captured by the second photosensitive sensing unit as the AC fluctuation component of the reflection signal in real time. The magnitude of this AC fluctuation component directly characterizes the stress state of the scallion body. Through the analysis of this fluctuation component by the high-pass filter and the third voltage comparator, once its amplitude exceeds the preset vibration amplitude threshold, the system determines that there is a risk of damage to the current impact energy, and immediately adjusts the subsequent single high-pressure cleaning action to a gradient cleaning sequence consisting of low-pressure pre-wetting and high-pressure main impact. Here, the adaptive energy regulation mechanism based on mechanical response forms a key synergy with the front-end stain recognition mechanism. The former takes the judgment of the latter as the execution premise, and the latter relies on the dynamic regulation of the former to avoid application limitations. The two jointly ensure that when the system faces complex and changeable cleaning objects, its operation mode can achieve a deterministic balance driven by internal logic between the upper limit of ensuring decontamination effect and the bottom line of avoiding physical damage.

[0031] At a deeper level, the solution of the present invention redefines the technical boundary of the cleaning process by reusing the physical traces in the processing flow. In traditional understanding, the residual water film generated after cleaning is a useless thing to be removed, while this solution transforms it into a key information carrier for quality inspection; when a seemingly clean green onion that has undergone deep cleaning by an eddy current cavitation nozzle passes through the re-inspection area at the rear stage, the system obtains the reflection signal in the 950-nanometer near-infrared band again after a predetermined time delay of 30 milliseconds to 80 milliseconds. If there are hidden defects such as internal contusions or early spoilage in the green onion, the hydrophilicity of the damaged tissue will cause local water film thickening, thereby resulting in the near-infrared reflection signal being lower than the water film residual detection threshold calibrated by healthy samples, and finally triggering the rejection instruction; this procedure does not resort to adding independent non-destructive testing equipment, but through secondary interpretation of the physical phenomena left by the cleaning action, the cleaning process itself has derived the function of quality control, which is the embodiment of the design idea of redefining the problem. The originally intractable problem of on-line quality screening is transformed into a low-cost and high-efficiency signal comparison task under this new framework.

[0032] Example 2: To objectively verify the actual effectiveness and engineering feasibility of the core technical features in the solution of the present invention, especially to quantitatively evaluate its comprehensive ability to achieve differential precise cleaning and adaptive damage avoidance under complex working conditions, a verification test of this example is specifically set up and implemented.

[0033] The experiment was conducted on the platform of an industrial production line, which is integrated with a variable-speed controlled conveying system. An optical detection module including a first photosensitive sensing unit and a second photosensitive sensing unit is deployed above it. The signal processing and control unit is based on the direct electrical logic procedure of a voltage comparator and a hardware logic gate circuit, and cascades to control a high-pressure cleaning mechanism and a low-pressure cleaning mechanism respectively equipped with eddy cavitation nozzles. To ensure the effectiveness and comparability of the test data, the test samples were preset into three groups. Group A is healthy green onions with dry floating soil attached, Group B is healthy green onions smeared with calibrated highly viscous organic dirt, and Group C is green onions with relatively fragile texture but also smeared with this highly viscous organic dirt, respectively used to simulate three typical working conditions of mild pollution, severe pollution, and severe pollution and vulnerability. Before the experiment starts, the core control parameters are precisely calibrated. Among them, the setting of the vibration amplitude threshold on which the adaptive energy regulation mechanism depends aims to establish a deterministic decision boundary to distinguish normal cleaning impacts from damaging impacts. Its calibration follows the following operating procedures: Select a batch of standard Group C samples, apply a gradient pressure pulse with a step of 0.1 MPa from low to high to them. During each pulse action, synchronously record the peak value of the AC fluctuation component in the near-infrared reflection intensity signal output by the second photosensitive sensing unit through a data acquisition system. And immediately after each pulse ends, use a high-resolution digital microscope to conduct a surface microstructure inspection on the impacted area of the sample. Set the peak value of the AC fluctuation component measured at the previous pressure level corresponding to the critical pressure when the first microscopic damage to the cell wall is observed as the vibration amplitude threshold. For the sample characteristics used in this experiment, this threshold is determined to be 25 mV. During the experiment, the samples of Groups A, B, and C are placed on the conveying system in a random order. The system runs continuously and records its decision-making and execution status, showing a clear differential response phenomenon. When the Group A sample passes, since the system does not detect a near-infrared signal below the organic matter light absorption threshold, it stably executes the regular flushing of the low-pressure cleaning mechanism. When the Group B sample passes, the system uniformly determines it as a strong flushing condition and executes the standard single-pulse high-pressure flushing action. When the Group C sample passes, the system also first starts the high-pressure cleaning mechanism. However, at the initial stage when the high-pressure water flow contacts the green onion body, the data acquisition system generally records that the AC fluctuation component instantaneously exceeds the vibration amplitude threshold of 25 mV, and it is observed that the control system switches the cleaning mode to a gradient cleaning sequence almost at the same time. The specific test data are shown in Table 1.

[0034] Table 1: Data table of differential cleaning and adaptive energy regulation test.

[0035]

[0036] Analysis of the data shown in Table 1 indicates that when samples B-01 and C-01 faced exactly the same severe pollution, the system correctly initiated high-pressure cleaning. However, precisely because the latter generated a measurable physical phenomenon of alternating current fluctuation components far exceeding the threshold under the impact, it triggered the adaptive energy regulation mechanism based on mechanical response, causing the final execution mode to switch from single-pulse high pressure to a gradient cleaning sequence. The direct result of this mode switch is that the cell damage rate of group C samples was successfully controlled at the same order of magnitude as that of group B samples, far lower than that of the control group without the intervention of this mechanism. At the same time, its residual rate after cleaning also remained at a level below 1%, which is direct evidence of the effectiveness of this internal mechanism.

[0037] Example 3: This example combines Figures 1 to 4 , and illustrates the implementation of a green onion processing and cleaning method that integrates visual recognition and remote monitoring. As Figure 1 shown, Figure 1 shows the typical attenuation curves of the near-infrared reflection signal voltage (unit: mV) over time (unit: ms) during the natural evaporation of the surface water film after cleaning for healthy green onions (group H) and damaged green onions (group D), which are used to reflect the differences in near-infrared reflection characteristics of the two groups of samples at different time points. In the figure, healthy green onions (group H) are marked by solid lines, and their near-infrared reflection signal voltage drops rapidly over time, indicating rapid evaporation of the water film. While damaged green onions (group D) are marked by dashed lines, and the decline in their reflection signal is relatively slow, reflecting the enhanced hydrophilicity and thicker water film residue due to tissue damage. The figure also clearly marks the optimal detection time point with a vertical gray dashed line, which is determined by analyzing the maximum distinguishability of the two groups of samples in the near-infrared reflection signal voltage, that is, the maximum signal separation difference. As a decision window for judging the existence of quality defects, it provides a highly reliable time node basis for subsequent automatic rejection using near-infrared optical characteristics. This figure, as an important part of the quality screening mechanism derived from physical traces, reflects the core technical path of the present invention to accurately identify hidden damage to agricultural products through signal attenuation laws without adding new hardware.

[0038] As Figure 2As shown, after the cleaning system completes the cleaning action first, it starts a delay timer and enters a set delay stage of waiting for 30 - 80 ms to ensure the formation of water film evaporation differences. After the water film evaporation differences are formed, a secondary detection is triggered, and the second photosensitive sensing unit re - collects the reflected signal voltage. If it is a detection of healthy green onions, the output reflected intensity signal is the thin water film reflection signal (high). When this signal is judged by the fourth voltage comparator to be higher than the signal water film residue threshold, a non - rejection instruction is output, the rejection device does not act, and through quality inspection, the green onions are confirmed as healthy green onions and moved to the healthy green onion path; if it is a detection of defective green onions, the output reflected intensity is the thick water film reflection signal (low), the signal is lower than the signal water film residue threshold, and the fourth voltage comparator outputs and sends a rejection instruction, and the rejection device immediately executes the rejection action. This green onion is confirmed to have strong hydrophilicity and a thick water film, indicating that there are signs of tissue damage or early spoilage, and it is removed from the main line to the defective green onion path.

[0039] As Figure 3 shown, Figure 3 shows the comparison curves of water pressure (unit: MPa) changing with time (unit: ms) in two cleaning modes of single - pulse high - pressure cleaning and gradient cleaning sequence in the present invention, reflecting the differential characteristics of the cleaning energy control strategy. The solid line represents single - pulse high - pressure cleaning, starting from 0 MPa at 0 ms, the water pressure instantaneously rises to about 1.45 MPa at 100 ms, and maintains a constant high pressure between 100 ms and 250 ms, and quickly drops back to 0 MPa after 250 ms. Overall, it is a single - time, high - intensity, and fixed - duration pulsed water flow. The dotted line represents the gradient cleaning sequence, which realizes the gradual release of water pressure through segmented control. Specifically, it rises to about 0.75 MPa at about 50 ms as a pre - wetting pulse, briefly returns to zero after 150 ms, then rises again to about 1.45 MPa in the main impact section, and returns to zero after 250 ms. This sequence effectively avoids structural damage to the fragile green onion body through pressure - grading progression.

[0040] As Figure 4 shown, the process starts from the start node. First, the high - pressure cleaning mechanism is started, and then the signal of the second photosensitive sensing unit is collected in real - time. The system extracts the AC fluctuation component through a high - pass filter to obtain the signal waveform reflecting the stress state of the green onion body. This process corresponds to the vibration response waveform shown in the figure, and its characteristic quantity is the AC component (AC component). The system compares this with a vibration threshold determined in advance through a vibration threshold calibration process. If the judgment result is > (Vibration threshold), it indicates that there is a risk of damage in the current impact, and the system switches to a gradient cleaning sequence; the gradient cleaning sequence consists of two pulses, where Pulse 1: 50% pressure, Pulse 2: 100% pressure, and the corresponding times are and , forming a progressive energy release mode; conversely, if ≤ (Vibration threshold), it enters the path of maintaining single-pulse high-pressure cleaning (standard pressure), performs single-pulse cleaning, with its continuous pressure: 100%, duration: standard duration, and completes the cleaning operation with standard high pressure; finally, both paths point to the end node.

[0041] Example 4: In a production scenario for processing a specific category of organic green onions with fragile epidermal texture in terms of physical properties, in order to further reduce the microscopic physical damage rate and improve the accuracy of subsequent quality defect elimination based on water film residue, it is necessary to perform more precise, data-driven joint optimization on two key dynamic links in the system, namely the energy structure of the gradient cleaning sequence, and the time delay and determination threshold in quality detection.

[0042] Regarding the optimization of the gradient cleaning sequence, its goal is to establish an energy release model with an accurate correlation between the response intensity and the risk level. For this purpose, the following procedures are adopted in this example: When high-pressure cleaning is triggered and the AC fluctuation component monitored by the system in real time exceeds the preset vibration amplitude threshold , the system executes a double-pulse cleaning sequence. The first pre-wetting pulse of this sequence has its pressure set at fifty percent of the standard high pressure, and its duration is determined by the relational expression , where is a reference duration, and the pressure of the second main impact pulse is the standard high pressure, and the duration is ; According to this relational expression, when the vibration of the green onion body intensifies, that is, the value increases, the duration of the pre-wetting pulse will be correspondingly shortened, thereby adaptively adjusting the total energy input and action time. Furthermore, in order to avoid the setting uncertainty brought about by the mutual dependence between the two parameters of the predetermined time delay and the water film residue detection threshold in the quality detection link, this example introduces an offline joint calibration experimental procedure. The initial state of this procedure is defined as: Prepare two groups of samples, group H is healthy green onions confirmed to have no damage, and group D is green onions with hidden bruises made on the surface by standardized micro-impacts; the execution steps of the procedure are to wet the two groups of samples under exactly the same conditions and immediately place them within the detection range of the second photosensitive sensing unit, from Starting from a certain moment, the near-infrared reflection intensity signal voltage of each sample is continuously collected and recorded at intervals of five milliseconds until the signal reaches stability, thereby obtaining a voltage-time decay curve representing the surface water film evaporation process of each sample.

[0043] The analysis steps for the collected data are as follows. Since the damaged tissues of the D-group samples have stronger hydrophilicity, the water film evaporation rate is slower than that of the H-group, manifested as the attenuation of its curve is more gentle. To find the peak point that distinguishes the signal differences between the two groups of samples, a time-related signal separation function is defined, and its value is the average voltage of the H-group samples at moment minus the average voltage of the D-group samples at moment That is ; Through calculation, find the moment when reaches the maximum value, this is determined as a pre-determined time delay optimized by data ; Once is determined, the water film residue detection threshold can be set at the intermediate value of the voltages of the two groups of samples at this moment, that is ; Through this joint calibration procedure, the setting process of two interrelated parameters is transformed into an optimization problem driven by data.

[0044] Example 5: To ensure that the technical solution of the present invention can maintain the stability of its design performance in different regions, different seasons, and in the face of different batches of raw materials, when the system is first deployed or the processing object is replaced, a set of standardized pre-calibration and baseline establishment procedures can be executed. This procedure first requires the operator to select a set of reference samples from the current batch to be processed. This set of samples must include healthy and clean green onions, green onions with typical stains of this batch, and green onions with the most fragile texture confirmed. After the system runs without load and records the background reflection signal of the conveyor belt, this set of reference samples is passed through the optical detection module in turn to complete the signal acquisition of the reflection intensity in the visible and near-infrared bands in the non-cleaning state; based on the data collected by the system, a set of initial working thresholds for the current batch is automatically calculated and set. Among them, the green onion presence threshold is set at the intermediate value of the visible light reflection intensity signals of the clean green onions and the conveyor belt background, and the organic matter light absorption threshold is established based on the near-infrared absorption characteristics of the typical stain samples. At the same time, by applying a standard test water flow pulse to the sample with the most fragile texture, its vibration response data can be obtained to provide a basis for setting the vibration amplitude threshold.

[0045] To further enhance the system's long-term adaptive ability to environmental and raw material changes and improve the recognition robustness for new or atypical organic stains, the system also integrates a dynamic update and maintenance mechanism for the stain optical feature library. During the regular processing, the system continuously records the reflected intensity signal values in the 950-nanometer near-infrared band corresponding to all events that trigger the high-pressure cleaning mechanism and stores them in the database. During the system maintenance period, a background analysis program will perform statistical clustering analysis on the signal values accumulated in the database. This analysis aims to identify new signal distribution clusters that have significant statistical differences from the data in the existing feature library. If such new clusters are detected, the system will prompt the operator to extract physical samples associated with the spatio-temporal position of the abnormal signal from the production line for inspection. Once a new stain is manually confirmed, the near-infrared reflection feature data of this sample will be officially incorporated into the system's core stain optical feature library, and the organic matter light absorption threshold will also be recalculated and adjusted based on the updated feature library to achieve the long-term stability of the system's adaptability to environmental changes and its recognition accuracy. The specific procedures for the remote monitoring function and the dynamic maintenance of the stain feature library disclosed in the present invention are as follows. The remote monitoring of the system is realized through an industrial Ethernet module integrated into the main controller. This module uses the OPC-UA protocol to push data to the local server or cloud platform. The data presented in real time on its remote terminal interface includes the processing count of the current batch of green onions, the minute trigger frequency of the high- and low-pressure cleaning mechanisms, the activation ratio of the gradient cleaning sequence, and the real-time rejection rate of defective products. At the same time, this interface authorizes users with secondary or higher-level permissions to remotely call the preset threshold parameter groups after changing the processing batch. In addition, the system packs and records the near-infrared reflection intensity signal values of all events that trigger the high-pressure cleaning and the peak values of their corresponding AC fluctuation components. When the background analysis program uses the clustering algorithm to identify new data clusters that have significant statistical differences from the existing stain optical feature library, the remote monitoring interface will issue an alarm to the maintenance personnel and highlight the timestamp and production line position information of the physical samples corresponding to this data cluster to guide them to conduct physical sampling and feature library update.

[0046] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cleaning method for scallion processing that integrates visual recognition and remote monitoring, characterized in that, Including the following steps: Step a: On the conveying path of green onions, obtain the reflection intensity signal of the green onion surface in the visible light band; meanwhile, obtain the reflection intensity signal of the green onion surface in the near-infrared band with the central wavelength. Step b: Input the reflection intensity signal in the near-infrared band into the first voltage comparator. When the reflection intensity signal in the near-infrared band is lower than the organic matter light absorption threshold determined through pre-calibration, the first voltage comparator outputs a first logic high-level signal. Step c: Input the reflection intensity signal in the visible light band into the second voltage comparator. When the reflection intensity signal in the visible light band is lower than the green onion presence threshold determined through pre-calibration, the second voltage comparator outputs a second logic high-level signal. Step d: Based on the AND logic judgment of the first logic high-level signal and the second logic high-level signal, drive the high-pressure cleaning mechanism to perform a strong rinse on the green onions. The strong rinse is executed when the first logic high-level signal is true and the second logic high-level signal is true. Step e: When the first logic high-level signal is false and the second logic high-level signal is true, drive the low-pressure cleaning mechanism to perform a conventional rinse on the green onions.

2. The method for processing and cleaning green onions by integrating visual recognition and remote monitoring according to claim 1, wherein The reflection intensity signal in the visible light band is obtained by the first photosensitive sensing unit; the reflection intensity signal in the near-infrared band with a central wavelength of 950 nanometers is obtained by the second photosensitive sensing unit, and the second photosensitive sensing unit is only sensitive to the near-infrared light with a central wavelength of 950 nanometers.

3. The method for processing and cleaning green onions by integrating visual recognition and remote monitoring according to claim 1, characterized in that, Both the high-pressure cleaning mechanism and the low-pressure cleaning mechanism are independent water circuits controlled by solenoid valves. The first logic high-level signal and the second logic high-level signal directly control the opening and closing of the solenoid valves through a logic gate circuit.

4. The method for processing and cleaning scallions by integrating visual recognition and remote monitoring according to claim 3, wherein The steps of the strong rinse further include: while the high-pressure cleaning mechanism starts rinsing, obtain the reflection intensity signal of the second photosensitive sensing unit; extract the AC fluctuation component of the reflection intensity signal through a high-pass filter; input the AC fluctuation component into the third voltage comparator. When the AC fluctuation component exceeds the vibration amplitude threshold, based on the AC fluctuation component, adjust the energy release mode of the high-pressure cleaning mechanism; the energy release mode is to adjust the single cleaning action of the high-pressure cleaning mechanism into a gradient cleaning sequence composed of at least two pulses with increasing pressure or increasing duration.

5. The scallion processing and cleaning method integrating visual recognition and remote monitoring according to claim 4, wherein The adjustment condition of the energy release mode is that the difference between the AC fluctuation component and the AC fluctuation component of the reflection intensity signal of the first photosensitive sensing unit exceeds the differential fluctuation threshold determined through pre-calibration.

6. The scallion processing and cleaning method integrating visual recognition and remote monitoring according to claim 5, characterized in that, An eddy cavitation nozzle is provided at the end of the high-pressure cleaning mechanism. The eddy cavitation nozzle has an internal channel structure, and the internal channel structure reduces the local pressure of the flowing high-pressure water flow below the saturated vapor pressure of water, thereby forming cavitation microjets on the surface of the green onions.

7. The scallion processing and cleaning method integrating visual recognition and remote monitoring according to claim 6, characterized in that, The internal channel structure of the vortex cavitation nozzle includes a spiral tapered channel, which increases the water flow velocity, resulting in a decrease in local pressure, satisfying Bernoulli's principle, that is: , where is the local water pressure, is the local water flow velocity, is the water density, is the energy constant in Bernoulli's principle.

8. The method for processing and cleaning green onions by integrating visual recognition and remote monitoring according to claim 7, characterized in that, The water pressure of the high-pressure cleaning mechanism is higher than 1.5 MPa.

9. The scallion processing and cleaning method integrating visual recognition and remote monitoring according to claim 8, characterized in that, Also including the following steps: After a predetermined time delay after each grading and cleaning operation is completed, the reflection intensity signal in the near-infrared band with a central wavelength of 950 nanometers of the second photosensitive sensing unit is acquired again; and the acquired reflection intensity signal is input to the fourth voltage comparator. When the signal is lower than the water film residue detection threshold determined by calibrating the surface of healthy green onions, the fourth voltage comparator outputs a rejection instruction. The rejection instruction is used to indicate that there are quality defects in the green onions and drives the rejection device to remove the green onions from the production line.

10. The scallion processing and cleaning method integrating visual recognition and remote monitoring according to claim 9, characterized in that, The range of the predetermined time delay is from 30 milliseconds to 80 milliseconds.

Images