Ginger processing production line control method

Abstract

The invention relates to the technical field of control, in particular to a ginger processing production line control method. Analyzing by combining historical ginger peeling records and preset skin removal indexes to obtain a plurality of expected residual skin proportions of the plurality of peeling stages; initial peeling parameters are set based on the ginger attribute characteristics of the to-be-peeled ginger, and peeling equipment is controlled to execute peeling operation in a first peeling stage according to the initial peeling parameters; performing peeling state identification on the first peeled ginger image to obtain a first predicted residual peel proportion; performing peeling parameter optimization according to the ginger attribute characteristics and the first predicted residual peel proportion, and outputting a second optimal peeling parameter; and according to a second optimal peeling parameter, performing peeling operation of a second peeling stage, performing iterative feedback control, and stopping the fresh ginger peeling operation until the fresh ginger peeling period is completed or the current predicted residual skin proportion is smaller than or equal to a preset skin removal index. The ginger peeling precision, efficiency and product quality stability are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control, in particular to a ginger processing production line control method. BACKGROUND

[0002] Traditional ginger peeling processing relies on manual experience or fixed parameter control of peeling equipment, such as fixed brush head speed and spray flow. Before processing, the peeling difficulty is roughly judged by manual according to the appearance of ginger, and the uniform parameter is set unchanged throughout the process, and the peeling effect is mainly confirmed by manual sampling inspection. The fixed parameter leads to poor adaptability, and lacks precise feedback mechanism, and manual parameter setting and sampling inspection need to consume a lot of time, and the judgment standard is subjective, which is not suitable for the demand of large-scale production line. There are problems of low precision, efficiency and product quality stability of ginger peeling. SUMMARY

[0003] The present application provides a ginger processing production line control method to solve the problem of low precision, efficiency and product quality stability of ginger peeling in the prior art.

[0004] The technical scheme for solving the above technical problems of the present application is as follows: In a first aspect, the present application provides a ginger processing production line control method, comprising: collecting ginger attribute features of ginger to be peeled, wherein the ginger attribute features at least include ginger age, dryness, block size and skin color; filtering historical ginger peeling records with the ginger attribute features as constraints to obtain a plurality of qualified ginger peeling data meeting peeling indexes, wherein the peeling indexes include skin removal indexes and peeling depth indexes; performing residual skin proportion decline rate analysis according to the plurality of qualified ginger peeling data, generating a plurality of residual skin proportion decline rate curves, and performing curve fitting to obtain a standard residual skin proportion decline rate curve; dividing the standard residual skin proportion decline rate curve according to the plurality of peeling stages with the preset skin removal index as the curve end point, to obtain a plurality of expected residual skin proportions of the plurality of peeling stages.

[0005] Wherein, the initial peeling parameters are set based on the ginger attribute features of the ginger to be peeled, including: filtering historical ginger peeling records with the ginger attribute features as product constraints and the attribute features of the peeling equipment as equipment constraints to obtain a plurality of historical initial peeling parameters meeting the peeling indexes, wherein the peeling parameters include brush head movement speed and spray flow; setting the mode of the plurality of historical initial peeling parameters as the initial peeling parameters.

[0006] The first peeling ginger image is subjected to peeling state recognition to obtain a first predicted residual peel ratio, including: extracting ginger contour features by using an edge detection algorithm, segmenting the first peeling ginger image to obtain a plurality of first peeling ginger sub-images; pre-training a residual peel ratio prediction plug-in, wherein the residual peel ratio prediction plug-in includes Q residual peel ratio prediction units, where Q is an integer greater than or equal to 5 and less than or equal to 30; based on the first peeling stage matching, a first unit selection ratio is obtained, multiplied by Q and rounded to obtain P, wherein the unit selection ratio gradually increases with the increase of the peeling stage; P residual peel ratio prediction units are randomly selected in the Q residual peel ratio prediction units, and the plurality of first peeling ginger sub-images are respectively predicted to output a plurality of predicted residual peel ratio sets, and the first predicted residual peel ratio is obtained after two times of mean value calculation.

[0007] The pre-trained residual peel ratio prediction plug-in includes: based on historical ginger peeling records, the ginger attribute features are used as constraints to collect a sample ginger image set, and different sample ginger images are labeled with residual peel ratios to obtain a sample residual peel ratio set; the sample ginger image set and the sample residual peel ratio set are used as training data, divided into Q parts, selected Q times with replacement, a first training set is constructed, and Q training sets are obtained by iterative selection Q times; the Q training sets are used to train a convolutional neural network to convergence respectively to obtain Q residual peel ratio prediction units, and the residual peel ratio prediction plug-in is obtained by combination.

[0008] Optionally, the second expected residual peel ratio of the second peeling stage is used as the expected workload, and the peeling parameter optimization is performed according to the ginger attribute features and the first predicted residual peel ratio to output the second optimal peeling parameter, including: obtaining the second peeling time of the second peeling stage; obtaining the peeling parameter space of the peeling device, wherein the peeling parameter space includes a brush head motion rate threshold and a spraying flow threshold; any parameter is randomly selected within the brush head motion rate threshold and the spraying flow threshold to obtain a first initial peeling parameter, and the parameter is selected multiple times without replacement to obtain a plurality of initial peeling parameters, which are combined with the ginger attribute features, the first predicted residual peel ratio and the second peeling time to obtain a plurality of peeling control schemes; a residual peel ratio predictor is used to perform peeling prediction on the plurality of peeling control schemes to output a plurality of predicted residual peel ratios; the second optimal peeling parameter is output by performing peeling parameter optimization based on the second expected residual peel ratio and the plurality of predicted residual peel ratios.

[0009] The construction process of the residual peel ratio predictor comprises: taking the ginger attribute features as product constraints, taking the attribute features of the peeling device as device constraints, collecting a plurality of sample peeling control schemes and a plurality of sample historical residual peel ratios based on historical ginger peeling records, wherein each sample peeling control scheme comprises sample ginger attribute features, a sample residual peel ratio, sample peeling parameters and a sample peeling duration; and training a deep learning model based on the plurality of sample peeling control schemes and the plurality of sample historical residual peel ratios until the deep learning model converges, to obtain the residual peel ratio predictor.

[0010] The peeling parameter optimization based on the second expected residual peel ratio and the plurality of predicted residual peel ratios to output the second optimal peeling parameter comprises: taking the second expected residual peel ratio as a benchmark, respectively calculating deviations of the plurality of predicted residual peel ratios to obtain a plurality of ratio deviation values; arranging the plurality of initial peeling parameters in ascending order of the ratio deviation values, and setting the initial peeling parameters as initial solutions to obtain an initial solution sequence; setting the first solution of the initial solution sequence as an optimal solution, setting the residual initial solutions as difference solutions, and taking the optimal solution as an adjustment direction, adjusting a plurality of difference solutions of the initial solution sequence according to an optimization adjustment step, and reordering to obtain an updated initial solution sequence, wherein 5% of the difference solutions of the updated initial solution sequence are eliminated each time the optimization is performed, and initial peeling parameters are randomly selected without replacement in the peeling parameter space for replacement; and continuing the peeling parameter iterative optimization until a preset second optimization number is met, and setting the optimal solution of the current updated initial solution sequence as the second optimal peeling parameter, wherein the optimization number gradually increases as the peeling stage increases.

[0011] By implementing the present application, the ginger peeling cycle can be divided into a plurality of peeling stages, a plurality of expected residual peel ratios of the plurality of peeling stages can be obtained by combining historical ginger peeling records and preset epidermis removal indicators, a clear phased target can be provided for the peeling process to avoid blind operation and ensure the orderly advancement of the peeling process, and the expected value is determined based on historical data and ginger attribute features, so that the target is more suitable for the actual situation and the peeling accuracy is improved. By implementing the present application, the initial peeling parameters can be set based on the ginger attribute features of the ginger to be peeled, the peeling device is controlled to perform the peeling operation of the first peeling stage according to the initial peeling parameters, the initial parameters are determined based on the mode of historical qualified data, the ginger attribute and device constraints are considered, the rationality and stability of the first stage operation are improved, a basis is provided for parameter optimization in subsequent stages, and the cost of blind trial and error is reduced. By implementing the present application, the first peeling image of raw ginger can be acquired after the first peeling stage ends, and the peeling state of the first peeling image can be recognized to obtain the first predicted residual peel ratio. The current peeling state can be accurately obtained by image recognition and multi-prediction unit fusion, and data basis can be provided for parameter adjustment in the next stage. The peeling effect is dynamically fed back to avoid errors caused by relying on experience. By implementing the present application, the second expected residual peel ratio of the second peeling stage can be used as the expected workload. The peeling parameters are optimized according to the property characteristics of the raw ginger and the first predicted residual peel ratio. The second optimal peeling parameter is output. The parameters are optimized based on the actual peeling state and the target difference to ensure that the next stage operation is more accurate to approach the expected target. The effect of different parameters is simulated by the predictor to reduce the time and resource waste of actual debugging. By implementing the present application, the second peeling stage can be performed according to the second optimal peeling parameter, and iterative feedback control can be performed until the raw ginger peeling cycle is completed or the current predicted residual peel ratio is less than or equal to the preset epidermis removal index. The raw ginger peeling operation is terminated to form a closed-loop feedback control. The parameters are adjusted in real time to adapt to the dynamic changes of the raw ginger peeling difficulty to ensure that the final peeling effect meets the index. Over-peeling or insufficient peeling is avoided to improve product quality and processing efficiency.

[0012] In summary, by implementing the present application, the peeling effect can be ensured to meet the epidermis removal index and avoid over-processing. The accuracy, efficiency and product quality stability of raw ginger peeling are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 A flowchart of a raw ginger processing line control method provided by the present application is shown. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0015] In the description of the present application, the terms "first" and "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0016] In the description of the present application, the term "for example" is used to mean "serving as an instance, illustration, or example, of something". Any embodiment described as "for example" in the present application should not be construed as necessarily preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, for the purposes of explanation, specific details are set forth in order to provide a thorough understanding of the application. It will be apparent to one skilled in the art that the present application can be practiced without using these specific details. In other instances, well-known structures and processes are not elaborated upon in order to avoid obscuring the description of the present application. Thus, the present application is not intended to be limited by the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0017] Embodiment one, as shown, the present application provides a processing line control method for fresh ginger, comprising: Figure 1 S100: dividing the peeling cycle of fresh ginger into multiple peeling stages, and analyzing the multiple expected residual peel ratios of the multiple peeling stages in combination with historical fresh ginger peeling records and preset peel removal indicators; S200: setting initial peeling parameters based on the attribute characteristics of the fresh ginger to be peeled, and controlling the peeling equipment to perform peeling work in the first peeling stage according to the initial peeling parameters; S300: after the first peeling stage ends, collecting a first peeled fresh ginger image, and performing peeling state recognition on the first peeled fresh ginger image to obtain a first predicted residual peel ratio; S400: taking the second expected residual peel ratio of the second peeling stage as the expected workload, and performing peeling parameter optimization according to the attribute characteristics of the fresh ginger and the first predicted residual peel ratio, and outputting second optimal peeling parameters; S500: performing peeling work in the second peeling stage according to the second optimal peeling parameters, and performing iterative feedback control until the peeling cycle of the fresh ginger is completed or the current predicted residual peel ratio is less than or equal to the preset peel removal indicator, and terminating the peeling work of the fresh ginger.

[0018] In step S100 of the present application, the peeling cycle of fresh ginger is divided into multiple peeling stages, the purpose of which is to realize fine and controllable management of the peeling process, and through stage-by-stage target decomposition, the peeling parameters can be dynamically adjusted according to the real-time peeling state of fresh ginger, so as to adapt to the changes in peeling difficulty of different fresh ginger, and finally ensure that the peeling effect meets the standard accurately. For example, if the peeling cycle is 60 seconds, it is divided into three peeling stages of 30 seconds, 20 seconds, and 10 seconds according to "rough peeling - medium peeling - fine peeling".

[0019] ​In step S100 of the embodiment, the expected residual peel proportions of the plurality of peeling stages are obtained by analyzing historical ginger peeling records and preset peel removal indicators, including: Collecting ginger attribute features of the ginger to be peeled, wherein the ginger attribute features at least include ginger age, dryness, block size, and peel color; With the ginger attribute features as constraints, historical ginger peeling records are screened to obtain a plurality of qualified ginger peeling data satisfying peeling indicators, wherein the peeling indicators include peel removal indicators and peeling depth indicators; According to the plurality of qualified ginger peeling data, residual peel proportion decline rate analysis is performed to generate a plurality of residual peel proportion decline rate curves, and curve fitting is performed to obtain a standard residual peel proportion decline rate curve; With the preset peel removal indicators as the curve end points, the standard residual peel proportion decline rate curve is divided according to the plurality of peeling stages to obtain a plurality of expected residual peel proportions of the plurality of peeling stages.

[0020] The purpose of step S100 of the embodiment is to provide scientific and accurate stage targets for segmented control of ginger peeling, to ensure that the residual peel proportion decline process of each peeling stage not only meets the ginger attribute features, but also ultimately meets the preset peel removal indicators. By combining historical data with current ginger characteristics, the expected values of each stage are more targeted and achievable, laying a foundation for subsequent parameter optimization and closed-loop control.

[0021] First, the ginger attribute features of the ginger to be peeled need to be collected, that is, the ginger age, dryness, block size, and peel color of the ginger to be peeled are collected. Among them, the ginger age can be determined by planting records or appearance characteristics; the dryness can be detected by a moisture measuring instrument, and the water content of the ginger is used as a measure; the block size is measured by a measuring tool; and the peel color is observed and recorded by a color recognition device or manually.

[0022] For example, the ginger attribute features of a batch of ginger to be peeled can be ginger age of 2 years, measured water content of 65%, block size of 5-8 cm, and yellowish peel.

[0023] Then, the historical ginger peeling records are screened with the ginger attribute features as constraints to obtain a number of qualified ginger peeling data meeting the peeling indicators. That is, the collected target ginger attribute features are used as constraint conditions to screen a large number of historical ginger peeling records. The screening is based on the peeling indicators, which include the skin removal indicator and the peeling depth indicator. The skin removal indicator specifies the final peeling degree to be achieved, and the peeling depth indicator ensures that the ginger flesh is not damaged excessively during peeling. For example, the skin removal indicator can be that the area ratio of the final residual skin to the total surface area of the ginger, i.e., the final residual skin ratio, which can be required to be ≤5%; and the peeling depth indicator can be the depth of the skin removed during peeling, which can be required to be ≤0.3 mm. Only the historical data meeting these indicators are selected to form a qualified ginger peeling data set.

[0024] Next, residual skin ratio decline rate analysis is needed based on the number of qualified ginger peeling data to generate a number of residual skin ratio decline rate curves, and curve fitting is performed to obtain a standard residual skin ratio decline rate curve. That is, the qualified ginger peeling data selected are analyzed to calculate the residual skin ratio decline rate corresponding to each data, and a number of residual skin ratio decline rate curves are generated. Since these curves may have certain discreteness, a mathematical method of curve fitting is used to fit these curves into a standard residual skin ratio decline rate curve to more accurately reflect the trend of the change in the residual skin ratio as the peeling time advances under ideal conditions.

[0025] For example, the “residual skin ratio decline rate”, i.e., the proportion of residual skin reduction per unit time, of 100 data is analyzed to generate 100 discrete decline rate curves. Then, these curves are combined into a standard curve by least squares curve fitting to reflect the ideal peeling trend of this type of ginger.

[0026] For example, the standard curve shows that the initial residual skin ratio is 50%, the expected residual skin ratio is 30% at the end of rough peeling at 20 seconds, the expected residual skin ratio is 15% at the end of medium peeling at 40 seconds, and the expected residual skin ratio is 5% at the end of fine peeling at 60 seconds.

[0027] Finally, the standard residual skin ratio decline rate curve is divided into a plurality of peeling stages according to the preset skin removal indicator as the end point of the curve to obtain a plurality of expected residual skin ratios of the plurality of peeling stages. That is, the preset skin removal indicator is used as the end point of the standard residual skin ratio decline rate curve. According to the plurality of peeling stages that have been divided, the residual skin ratio values of these points are read at the end points corresponding to each stage on the curve, thereby obtaining the expected residual skin ratios of the plurality of peeling stages. These ratio values will serve as control targets for each stage in the subsequent peeling process to guide the adjustment and optimization of the peeling parameters.

[0028] For example, with the end point of final residual skin ≤5%, the standard residual skin percentage decline rate curve is divided into 3 stages, it can be obtained that the expected residual skin percentage at the end of the first peeling stage is 30%; the expected residual skin percentage at the end of the second peeling stage is 15%; and the expected residual skin percentage at the end of the third peeling stage is 5%.

[0029] In step S200 of the embodiment, the initial peeling parameters are set based on the ginger property characteristics of the ginger to be peeled, including: The historical ginger peeling records are filtered with the ginger property characteristics as product constraints and the property characteristics of the peeling equipment as equipment constraints to obtain a plurality of historical initial peeling parameters meeting the peeling indexes, wherein the peeling parameters include the brush head movement rate and the spraying flow rate. The mode of the plurality of historical initial peeling parameters is set as the initial peeling parameters.

[0030] The purpose of step S200 of the embodiment is to provide scientific and adaptive initial parameter benchmarks for the peeling process, to ensure that the operation of the first peeling stage can preliminarily approach the expected peeling effect under the premise of considering the ginger property characteristics and the equipment performance, to lay a reasonable foundation for parameter optimization in the subsequent stages, and to reduce the cost of blind trial and error.

[0031] Firstly, the historical ginger peeling records are filtered with the ginger property characteristics as product constraints and the property characteristics of the peeling equipment as equipment constraints to obtain a plurality of historical initial peeling parameters meeting the peeling indexes. The equipment constraints are the inherent property characteristics of the peeling equipment, such as the maximum speed of the brush head and the maximum flow rate of the spraying system, to ensure that the parameters are within the operating range of the equipment and avoid exceeding the equipment load.

[0032] For example, a set of property characteristics of the ginger to be peeled can be “ginger age of 1 year, water content of 70%, block size of 3-5 cm, and light yellow skin”; and a set of property characteristics of the peeling equipment can be that the maximum speed of the brush head of the peeling equipment is 120 r / min and the maximum flow rate of the spraying system is 5 L / min.

[0033] Then, a plurality of historical initial peeling parameters meeting the peeling indexes, i.e., meeting the skin removal index and the peeling depth index, are selected from the historical ginger peeling records, i.e., the historical brush head movement rate and spraying flow rate used for the same type of ginger and meeting the peeling effect standards.

[0034] For example, in the filtered historical records, there are 20 initial parameters meeting the above conditions, among which the brush head speed “70 r / min” appears 12 times, which is the mode of the brush head speed, and the spraying flow rate “3 L / min” appears 10 times, which is the mode of the spraying flow rate.

[0035] Finally, the initial peeling parameters are set to "brush head speed 70 r / min, spray flow 3 L / min".

[0036] In step S300 of the embodiment, the first peeled ginger image is subjected to peeling state recognition to obtain a first predicted residual peel ratio, including: The ginger contour features are extracted using an edge detection algorithm, the first peeled ginger image is segmented, and a plurality of first peeled ginger sub-images are obtained. A pre-trained peel ratio prediction plug-in is used, wherein the peel ratio prediction plug-in includes Q peel ratio prediction units, and Q is an integer greater than or equal to 5 and less than or equal to 30. Based on the first peeling stage matching, a first unit selection ratio is obtained, multiplied by Q to obtain P, wherein the unit selection ratio gradually increases with the increase of the peeling stage. P peel ratio prediction units are randomly selected from the Q peel ratio prediction units, and the plurality of first peeled ginger sub-images are predicted respectively to output a plurality of predicted peel ratio sets, and a first predicted residual peel ratio is obtained after twice mean value calculation.

[0037] In the embodiment, the first peeled ginger image is segmented to accurately extract the image information of the ginger after the first peeling stage, and to provide high-quality analysis units for subsequent peel ratio recognition. By segmenting the ginger image to obtain a sub-image, the peeling state of a single ginger individual can be focused on, avoiding the influence of other interference factors in the image on the recognition accuracy, and providing independent analysis objects for multi-unit prediction to improve the accuracy of the peel ratio prediction. The first peeled ginger image is obtained by photographing the ginger after the first peeling stage using an industrial camera, a camera, or other image acquisition devices.

[0038] Specifically, the first peeled ginger image can be processed using a Canny algorithm, a Sobel algorithm, or other edge detection algorithms. The edge detection algorithm identifies the regions with sudden changes in gray value in the image, i.e., the boundaries between ginger and background and between ginger and ginger, to outline the complete contours of each ginger, thereby distinguishing the ginger main body from the image background and the boundaries between different ginger individuals.

[0039] For example, in an image containing multiple pieces of ginger, the edge detection algorithm can identify the edge lines of each piece of ginger, and clearly define the shape and position of each piece of ginger.

[0040] Then, based on the extracted ginger contour features, the first peeled ginger image is segmented. That is, according to the contour boundaries of each ginger, the original image is cut into a plurality of independent sub-images, each sub-image only contains the complete image information of one piece of ginger, excluding the background and other ginger interference.

[0041] For example, if there are 5 pieces of fresh ginger in the original image, 5 independent sub-images can be obtained by segmentation, each of which only contains the peeled state of one piece of fresh ginger, such as the distribution of residual peel, the surface characteristics of ginger body, etc.

[0042] In step S300 of the embodiment of the present application, the pre-trained residual peel ratio prediction plug-in is pre-trained, which comprises: Based on the historical fresh ginger peeling records, the sample ginger image set is collected with the ginger attribute features as constraints, and the residual peel ratio of different sample ginger images is labeled to obtain the sample residual peel ratio set. The sample ginger image set and the sample residual peel ratio set are used as training data, divided into Q parts, and selected Q times with replacement to construct a first training set, and selected Q times iteratively to obtain Q training sets. The convolutional neural network is trained to convergence using the Q training sets, and Q residual peel ratio prediction units are obtained, which are combined to obtain a residual peel ratio prediction plug-in.

[0043] In step S300 of the embodiment of the present application, the purpose of training the residual peel ratio prediction plug-in is to build a high-precision and strong-robustness residual peel ratio prediction tool. By pre-training the plug-in composed of multiple prediction units, the adaptability and prediction accuracy of different fresh ginger peeling states are improved. The combination of multiple prediction units can effectively reduce the bias and overfitting risk of a single model, and provide reliable algorithm support for subsequent peeling state recognition.

[0044] Firstly, based on the historical fresh ginger peeling records, the sample ginger image set is collected with the ginger attribute features as constraints, and the residual peel ratio of different sample ginger images is labeled to obtain the sample residual peel ratio set. That is, the representative sample ginger images are selected as constraints based on the ginger attribute features such as ginger age, moisture content, block size, and skin color to form a sample ginger image set, which covers fresh ginger images at different peeling stages and different residual peel states.

[0045] Then, each image in the sample ginger image set is manually or automatically labeled to determine the corresponding residual peel ratio of each image to form a sample residual peel ratio set. For example, for fresh ginger with ginger age of 2 years and moisture content of 60%, images at different stages such as 10 seconds, 20 seconds, and 30 seconds are collected, and the corresponding residual peel ratios such as 30%, 20%, and 10% are labeled.

[0046] Further, the sample ginger image set and the corresponding sample peel ratio set are divided into Q parts as a whole training data, Q is an integer between 5 and 30, and each part is used to train Q peel ratio prediction units. Specifically, bootstrap sampling or other sampling methods with replacement can be used to repeatedly select Q times from the Q parts of data, and a training set is constructed each time, allowing the same data to be selected multiple times. Finally, Q independent training sets that may overlap are obtained.

[0047] For example, if Q = 10, the original data is divided into 10 parts, and 10 training sets are generated by sampling with replacement, each containing part of the original data, ensuring that each training set has both commonalities and differences. Specifically, the total sample size is not less than 5000 ginger sub-image, each image is labeled with a peel ratio of 0-100%. According to the Q training set division: 5000 samples are divided into Q parts, for example, when Q = 10, each part contains 500 samples, and Q training sets are generated by sampling with replacement, each containing 500 samples, allowing repeated selection.

[0048] Next, Q convolutional neural networks (CNNs) need to be trained for Q training sets, i.e., Q peel ratio prediction units, which are collectively used as a peel ratio prediction plug-in.

[0049] Each peel ratio prediction unit contains four parts: an input layer, a convolutional layer, a pooling layer, and a fully connected layer.

[0050] The input layer is used to receive the segmented ginger sub-image, and the image size needs to be uniform, such as 224x224 pixels, in 3-channel RGB format.

[0051] The convolutional layer contains three convolutional operations, the first layer has 32 3x3 convolutional kernels with a step size of 1 and uses a ReLU activation function; the second layer has 64 3x3 convolutional kernels with a step size of 1 and uses a ReLU activation function; the third layer has 128 3x3 convolutional kernels with a step size of 1 and uses a ReLU activation function.

[0052] Each convolutional layer is connected to a 2x2 max pooling layer with a step size of 2 to reduce the feature dimension.

[0053] The fully connected layer contains two fully connected operations, the first layer has 512 neurons and uses a ReLU activation function; the second layer has one neuron and uses a Sigmoid activation function, with an output range of 0-1, corresponding to a peel ratio of 0-100%.

[0054] In the training of the residual skin ratio prediction unit, the initial learning rate is set to 0.001, and an adaptive learning rate adjustment strategy such as the Adam optimizer is used. The batch size is 32 images per input for training. L2 regularization is added to the fully connected layer with a coefficient of 0.0001 to avoid overfitting. Each prediction unit is initially trained for 50 rounds. In the next 5 rounds of training, the average absolute error of the predicted residual skin ratio and the labeled value, i.e., the average of the absolute values of the difference between the predicted value and the true value, is ≤2%, and the error reduction amplitude is <0.1%. When the model converges, a residual skin ratio prediction unit is obtained.

[0055] By the above method, the convolutional neural network is trained to convergence using the Q training sets, and Q residual skin ratio prediction units are obtained, which are combined to obtain a residual skin ratio prediction plug-in.

[0056] Further, based on the first peeling stage matching, the first unit selection ratio is obtained, multiplied by Q to obtain P, and P residual skin ratio prediction units are randomly selected from the Q residual skin ratio prediction units to predict the plurality of first peeled ginger sprout images.

[0057] The purpose of this step is to balance the prediction efficiency and accuracy in the peeling process by dynamically adjusting the number of prediction units and the ensemble learning strategy. The logic is that as the peeling stage progresses, the residual skin ratio gradually decreases and the distribution becomes more uneven. By gradually increasing the unit selection ratio, more prediction units are used to improve the recognition accuracy in complex scenarios.

[0058] First, the number of prediction units P needs to be determined. Specifically, the mapping relationship between the unit selection ratio and the peeling stage can be predefined, such as a first peeling stage unit selection ratio of 20%, a second peeling stage unit selection ratio of 40%, a third peeling stage unit selection ratio of 60%, etc., with the ratio increasing with the peeling stage.

[0059] For the first peeling stage, the corresponding first unit selection ratio, such as 20%, is matched, multiplied by the total number of prediction units Q (Q ∈ [5, 30]) and rounded to obtain P. For example, if Q = 10 and the first unit selection ratio is 20%, then P = 10 × 20% = 2.

[0060] Next, P residual skin ratio prediction units are randomly selected from the Q pre-trained residual skin ratio prediction units, and the plurality of ginger sprout images obtained by segmenting the first peeling stage, such as 5 ginger sprout images, are input into the P units. Each unit outputs a residual skin ratio prediction value for each sub-image, forming P sets of prediction results, such as 2 residual skin ratio prediction units each outputting 5 prediction values, which form the plurality of predicted residual skin ratio sets.

[0061] Then, the average of all sub-image prediction values of each residual skin ratio prediction unit is calculated. For example, if the residual skin ratio prediction values of residual skin ratio prediction unit 1 for 5 sub-images are 30%, 32%, 29%, 31%, and 30%, the average is 30.4%.

[0062] Further, the average of P residual skin ratio prediction units is calculated again to obtain the final first predicted residual skin ratio. For example: The average of residual skin ratio prediction unit 1 is 30.4%, the average of residual skin ratio prediction unit 2 is 31.4%, and the final first predicted residual skin ratio is (30.4%+31.4%) / 2=30.9%.

[0063] In step S400 of the embodiment, the second expected residual skin ratio of the second peeling stage is used as the expected workload, and the peeling parameters are optimized according to the ginger attribute characteristics and the first predicted residual skin ratio, and the second optimal peeling parameters are output, including: The second peeling time of the second peeling stage is obtained. The peeling parameter space of the peeling device is obtained, wherein the peeling parameter space includes a brush head movement rate threshold and a spraying flow threshold. Any parameter is randomly selected within the brush head movement rate threshold and the spraying flow threshold for combination to obtain a first initial peeling parameter, and the selection is continued multiple times without replacement to obtain multiple initial peeling parameters, which are combined with the ginger attribute characteristics, the first predicted residual skin ratio, and the second peeling time to obtain multiple peeling control schemes. The residual skin ratio predictor is used to perform peeling prediction on the multiple peeling control schemes respectively, and multiple predicted residual skin ratios are output. The peeling parameters are optimized based on the second expected residual skin ratio and the multiple predicted residual skin ratios, and the second optimal peeling parameters are output.

[0064] In the embodiment, the purpose of obtaining the multiple peeling control schemes in the above steps is to screen a batch of feasible initial peeling parameter schemes for the second peeling stage as the basis for subsequent parameter optimization. By combining the peeling time of the second peeling stage, the device parameter limit, and the actual state of the current ginger, diversified control schemes are generated to ensure that the most suitable peeling parameters can be found in a reasonable range in the subsequent optimization process, so as to accurately achieve the expected residual skin ratio target of the second peeling stage.

[0065] Specifically, the second peeling time of the second peeling stage is first obtained, i.e., according to the pre-divided peeling stages, such as a total period of 60 seconds, divided into 3 stages, the second peeling stage is 20 seconds, and the fixed time of the second peeling stage is determined as the second peeling time, which is used as the time constraint for parameter optimization, i.e., all schemes need to complete the peeling operation within the time.

[0066] Then, the peeling parameter space of the peeling device needs to be obtained, specifically, based on the performance limit of the peeling device, the brush head movement rate threshold and the spraying flow threshold are determined, such as the minimum brush head movement rate allowed by the device is 50 r / min, and the maximum brush head movement rate is 150 r / min; the minimum spraying flow is 2 L / min, and the maximum spraying flow is 8 L / min. The boundary of the parameter space is formed to ensure that all selected parameters are within the operating range of the device, avoiding overload and causing device failure.

[0067] Further, a plurality of peeling control schemes need to be obtained, specifically, each set of initial peeling parameters is associated with ginger attribute characteristics, a first predicted residual peel ratio, and a second peeling duration to form a plurality of complete peeling control schemes, each scheme containing complete information of "initial peeling parameters + ginger attribute characteristics + first predicted residual peel ratio + second peeling duration", providing input data for subsequent evaluation of scheme effect by the residual peel ratio predictor.

[0068] First, within the brush head movement rate threshold and the spraying flow threshold, a set of parameters is randomly selected, such as a brush head rate of 80 r / min and a spraying flow of 5 L / min, as the first initial peeling parameter. Then, using the non-replacement sampling method, continue to randomly select multiple sets of parameters to ensure that the brush head rate and the spraying flow of each set of parameters are within the threshold, and the parameters of each set are not completely repeated.

[0069] Finally, each set of initial peeling parameters is associated with the attribute characteristics of the ginger to be peeled, the first predicted residual peel ratio, and the second peeling duration to form a plurality of complete peeling control schemes, providing input data for subsequent evaluation of scheme effect by the residual peel ratio predictor.

[0070] In step S400 of the embodiment of the present application, the construction process of the residual peel ratio predictor includes: Based on the historical ginger peeling records, a plurality of sample peeling control schemes and a plurality of sample historical residual peel ratios are collected, wherein each sample peeling control scheme includes sample ginger attribute characteristics, sample residual peel ratio, sample peeling parameters, and sample peeling duration; The plurality of sample peeling control schemes and the plurality of sample historical residual peel ratios are used to train the deep learning model to convergence, obtaining the residual peel ratio predictor.

[0071] In the embodiments of the present application, the core purpose of the above steps is to build a tool that can accurately predict the peeling effect, i.e., a residual peel ratio predictor. Through the residual peel ratio predictor, the residual peel ratio change under different peeling parameters can be simulated in advance based on ginger attribute characteristics, peeling equipment attribute characteristics, peeling time, etc., to provide a scientific basis for subsequent peeling parameter optimization, avoid blindness in actual debugging, and improve the efficiency and accuracy of parameter optimization.

[0072] First, a plurality of sample peeling control schemes and a plurality of sample historical residual peel ratios need to be collected as sample data. Specifically, the ginger age, dryness, block size, and peel color are used as product constraints to ensure that the samples match the characteristics of the ginger to be processed. The brush head speed range and the spraying flow range are used as equipment constraints to ensure that the sample parameters are within the equipment operating range. Then, based on the above constraints, a plurality of "sample peeling control schemes" and corresponding "sample historical residual peel ratios" are extracted from historical ginger peeling records. Each sample peeling control scheme includes sample ginger attribute characteristics, sample residual peel ratio, sample peeling parameters, and sample peeling time. For example, the sample peeling parameters can be a brush head speed of 80 r / min and a spraying flow of 3 L / min, and the sample peeling time can be 20 seconds, etc.

[0073] According to the above standard, no less than 10,000 samples are prepared, each sample being a complete "sample peeling control scheme + sample historical residual peel ratio" data pair, which is used to train the residual peel ratio predictor.

[0074] In the construction of the residual peel ratio predictor, a multi-layer perceptron (MLP) in the deep learning model can be used according to the task type. This model is suitable for processing the mapping relationship between multi-dimensional input and continuous value output.

[0075] Specifically, the residual peel ratio predictor is a three-layer structure, including an input layer, a hidden layer, and an output layer. The input layer is used to receive multi-dimensional feature input corresponding to each feature parameter in the sample peeling control scheme. The hidden layer includes three fully connected layers, the first layer includes 128 neurons, uses a ReLU activation function, the second layer includes 64 neurons, uses a ReLU activation function, and the third layer includes 32 neurons, uses a ReLU activation function. The output layer has one neuron, uses a Sigmoid activation function, and the output range is 0-1, corresponding to 0%-100% of the residual peel ratio.

[0076] In the training of the residual skin ratio predictor, the initial learning rate is set to 0.001, and the Adam optimizer is used for adaptive adjustment. The batch size is 64 samples per input for training. L2 regularization is added to each hidden layer with a coefficient of 0.0005 to prevent model overfitting. A dropout layer is added after the first hidden layer with a dropout probability of 0.2 to enhance the model's generalization ability.

[0077] In the continuous 10 rounds of training, the average absolute error of the predicted residual skin ratio and the historical residual skin ratio of the sample, i.e., the average value of the absolute value of the difference between the predicted value and the true value, is ≤3%, and the error change amplitude is <0.5%. When the model converges, the residual skin ratio predictor is obtained.

[0078] Inputting the aforementioned multiple peeling control schemes can predict multiple residual skin ratios, such as 15% and 16%.

[0079] In step S400 of the embodiment, the second expected residual skin ratio and the multiple predicted residual skin ratios are used to optimize the peeling parameters, and the second optimal peeling parameters are output, including: The deviation of each predicted residual skin ratio from the second expected residual skin ratio is calculated to obtain multiple ratio deviation values; The multiple initial peeling parameters are arranged in ascending order of ratio deviation values, and the initial peeling parameters are set as initial solutions to obtain an initial solution sequence; The first solution of the initial solution sequence is set as an optimal solution, and the remaining initial solutions are set as difference solutions. The optimal solution is used as the adjustment direction, and the difference solutions of the initial solution sequence are adjusted according to the optimization adjustment step to obtain an updated initial solution sequence. The last 5% of the difference solutions of the updated initial solution sequence are eliminated each time, and initial peeling parameters are randomly selected without replacement in the peeling parameter space for replacement; The peeling parameter iterative optimization is continued until the preset second optimization number is met, and the optimal solution of the current updated initial solution sequence is set as the second optimal peeling parameter. The optimization number gradually increases with the increase of the peeling stage.

[0080] In the embodiment, the above steps aim to select the optimal parameter that can most accurately achieve the second-stage expected residual skin ratio from multiple candidate peeling parameters. Through deviation calculation, iterative optimization, and dynamic adjustment, the parameter can meet the device constraints and adapt to the dynamic changes of ginger peeling difficulty, ultimately achieving the peeling goal of the second stage and laying a foundation for the closed-loop control of the subsequent stage.

[0081] Firstly, the deviation of each predicted residual skin ratio is calculated based on the second expected residual skin ratio, and a plurality of ratio deviation values are obtained. That is, the deviation value of the predicted residual skin ratio of each peeling control scheme is calculated based on the second expected residual skin ratio of the second stage, wherein the second expected residual skin ratio of the second stage can be 15%, and the predicted residual skin ratio of each peeling control scheme is, for example, 22% for scheme 1, 18% for scheme 2, etc. The second expected residual skin ratio is the target value of the residual skin ratio that should be reached at the end of the second peeling stage.

[0082] Specifically, the formula for calculating the deviation value can be: ratio deviation value = |predicted residual skin ratio-second expected residual skin ratio|, for example: if the second expected residual skin ratio is 15% and the predicted residual skin ratio of scheme 1 is 22%, the deviation value is 7%; and if the predicted residual skin ratio of scheme 5 is 15%, the deviation value is 0%.

[0083] Further, the plurality of initial peeling parameters, i.e. the brush head speed and the spraying flow rate of each scheme, are sorted in ascending order according to the corresponding ratio deviation values to form an initial solution sequence, and the initial peeling parameter with the smallest deviation is placed at the first position. For example, the initial solution sequence can be: [scheme 5 (deviation value 0%), scheme 4 (deviation value 1%), scheme 2 (deviation value 3%), scheme 3 (deviation value 3%), scheme 1 (deviation value 7%)].

[0084] Then, the peeling parameters need to be iteratively optimized, and the first solution of the initial solution sequence, i.e. the initial peeling parameter with the smallest deviation, is set as the optimal solution, and the rest are the difference solutions.

[0085] Then, the optimal solution is used as the direction, and the fixed optimization adjustment step is used, such as adjusting the brush head speed by 5 r / min each time and adjusting the spraying flow rate by 0.5 L / min each time, to fine-tune part of the difference solutions, such as the last 30% of the difference solutions with larger deviation. For example: the peeling parameters of scheme 1 (70 r / min, 3 L / min) are adjusted to the optimal solution (100 r / min, 3.5 L / min) to obtain new peeling parameters (85 r / min, 3.2 L / min).

[0086] The adjusted peeling parameters and the original peeling parameters are combined, and the updated initial solution sequence is obtained by re-sorting the deviation values; the last 5% of the peeling parameters with the largest deviation in the sequence are eliminated, and new parameters are randomly selected from the peeling parameter space without replacement to replace them, and the length of the sequence is kept unchanged.

[0087] The above iterative optimization process is repeated until a preset second optimization number, such as 10 times, is reached. The optimization number increases with the peeling stage, such as 5 times for the first peeling stage, 10 times for the second peeling stage, and 15 times for the third peeling stage, to adapt to the more precise peeling requirements in the later period. Finally, the first solution in the initial solution sequence, i.e., the peeling parameters with the smallest deviation, such as the brush head speed 102 r / min and the spray flow 3.6 L / min, are set as the second optimal peeling parameters.

[0088] In the step S500 of the embodiment of the present application, the peeling operation of the second peeling stage is also performed according to the second optimal peeling parameters, and iterative feedback control is performed until the ginger peeling cycle is completed or the predicted remaining residual skin ratio is less than or equal to the preset epidermis removal index, and the ginger peeling operation is terminated.

[0089] In the embodiment of the present application, the purpose of step S500 is to realize closed-loop intelligent control of the ginger peeling process. By performing operations in each stage according to the optimal peeling parameters and continuously adjusting the feedback, it is ensured that the peeling effect ultimately meets the preset index. Specifically, by completing the second peeling stage operation according to the second optimal peeling parameters, and by adapting to the dynamic changes of the subsequent stages through the iterative feedback mechanism, the entire peeling cycle is efficiently completed under the premise of ensuring the peeling quality.

[0090] Specifically, the second optimal peeling parameters obtained in S400 are needed to control the equipment to complete the second peeling stage operation; after the operation is completed, the ginger image is continuously collected according to the method of step S300, and the current remaining residual skin ratio is identified; if the remaining residual skin ratio ≤ preset epidermis removal index at this time, the operation is directly terminated; if it does not meet the standard and all peeling stages are not completed, the third peeling stage is taken as the target, and the parameter optimization process of S400 is repeated to obtain the third optimal peeling parameters; The third peeling stage operation is performed according to the new parameters, and the remaining residual skin ratio is fed back through image recognition to continue the iterative optimization of the parameters; This cycle continues until all peeling stages are completed, or the remaining residual skin ratio meets the standard after the end of a peeling stage, and the operation is finally stopped.

[0091] The entire process continuously adjusts dynamically to adapt the peeling parameters to the changes in the ginger state, ensuring that the final effect meets the requirements.

[0092] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0093] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code thereon for use by or in connection with an instruction execution system. For the purposes of this description, a computer-usable or computer readable storage medium can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-usable storage medium can be a computer- readable storage medium that can be any media that can be accessed by the computer. By way of example, and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer- readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or other

[0094] The present application is described in reference to the flowchart illustrations and / or block diagrams of the methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0095] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0097] Although preferred embodiments of the application have been described herein, those skilled in the art will readily devise many additional variations of these preferred embodiments that will be within the scope of the present application.

[0098] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the present application and its equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method for controlling a ginger processing line, characterized by, The method comprises: dividing a ginger peeling cycle into multiple peeling stages, combining historical ginger peeling records and preset skin removal indicators to analyze to obtain multiple expected residual skin ratios of the multiple peeling stages; setting initial peeling parameters based on ginger attribute characteristics of ginger to be peeled, and controlling a peeling device to perform peeling work of a first peeling stage according to the initial peeling parameters; after the first peeling stage ends, collecting a first peeled ginger image, and performing peeling state recognition on the first peeled ginger image to obtain a first predicted residual skin ratio; taking a second expected residual skin ratio of a second peeling stage as an expected workload, and performing peeling parameter optimization according to the ginger attribute characteristics and the first predicted residual skin ratio to output second optimal peeling parameters; performing peeling work of the second peeling stage according to the second optimal peeling parameters, and performing iterative feedback control until the ginger peeling cycle is completed or the current predicted residual skin ratio is less than or equal to the preset skin removal indicator, and then terminating the ginger peeling work.

2. The ginger processing line control method according to claim 1, wherein The method of combining historical ginger peeling records and preset skin removal indicators to analyze to obtain multiple expected residual skin ratios of the multiple peeling stages comprises: collecting ginger attribute characteristics of ginger to be peeled, wherein the ginger attribute characteristics at least include ginger age, dryness, block size and skin color; taking the ginger attribute characteristics as constraints, screening historical ginger peeling records to obtain a plurality of qualified ginger peeling data satisfying peeling indicators, wherein the peeling indicators include skin removal indicators and peeling depth indicators; performing residual skin ratio decline rate analysis according to the plurality of qualified ginger peeling data, generating a plurality of residual skin ratio decline rate curves, and performing curve fitting to obtain a standard residual skin ratio decline rate curve; taking the preset skin removal indicator as a curve endpoint, dividing the standard residual skin ratio decline rate curve according to the multiple peeling stages to obtain multiple expected residual skin ratios of the multiple peeling stages.

3. The method according to claim 2, wherein The method of setting initial peeling parameters based on ginger attribute characteristics of ginger to be peeled comprises: taking the ginger attribute characteristics as product constraints and taking attribute characteristics of a peeling device as device constraints, screening historical ginger peeling records to obtain a plurality of historical initial peeling parameters satisfying the peeling indicators, wherein the peeling parameters include brush head movement rate and spraying flow; taking the mode of the plurality of historical initial peeling parameters as the initial peeling parameters.

4. The processing line control method of ginger as claimed in claim 1, wherein, The method of performing peeling state recognition on the first peeled ginger image to obtain a first predicted residual skin ratio comprises: extracting ginger contour features using an edge detection algorithm, segmenting the first peeled ginger image to obtain a plurality of first peeled ginger sub-images; pre-training a residual skin ratio prediction plug-in, wherein the residual skin ratio prediction plug-in includes Q residual skin ratio prediction units, wherein Q is an integer greater than or equal to 5 and less than or equal to 30; based on the first peeling stage, a first unit selection ratio is matched and obtained, and P is obtained by multiplying Q and taking the integer part, wherein the unit selection ratio gradually increases with the increase of the peeling stage; Randomly select P residual skin ratio prediction units in the Q residual skin ratio prediction units, and respectively predict the plurality of first peeled ginger sub-image to output a plurality of predicted residual skin ratio sets, and obtain the first predicted residual skin ratio after twice mean value calculation.

5. The method of claim 4, wherein the step of controlling the ginger processing line is characterized by, The pre-training residual skin ratio prediction plug-in comprises: Based on the historical ginger peeling record, the ginger attribute features are used as constraints to collect a sample ginger image set and label the residual skin ratio of different sample ginger images to obtain a sample residual skin ratio set; The sample ginger image set and the sample residual skin ratio set are used as training data, divided into Q parts, randomly selected Q times, and a first training set is constructed by iteration for Q times to obtain Q training sets; The convolutional neural network is trained to convergence respectively using the Q training sets to obtain Q residual skin ratio prediction units, and a residual skin ratio prediction plug-in is obtained by combination.

6. The method of claim 1, wherein the method further comprises: The second expected residual skin ratio of the second peeling stage is used as the expected workload, and the ginger attribute features and the first predicted residual skin ratio are used for peeling parameter optimization to output the second optimal peeling parameter, which comprises: Obtaining the second peeling time length of the second peeling stage; Obtaining the peeling parameter space of the peeling device, wherein the peeling parameter space comprises a brush head movement rate threshold and a spraying flow threshold; Randomly selecting any parameter within the brush head movement rate threshold and the spraying flow threshold to obtain a first initial peeling parameter, and continuing to select multiple times without replacement to obtain multiple initial peeling parameters, which are combined with the ginger attribute features, the first predicted residual skin ratio and the second peeling time length to obtain multiple peeling control schemes; Using a residual skin ratio predictor, the multiple peeling control schemes are respectively predicted to output multiple predicted residual skin ratios; Based on the second expected residual skin ratio and the multiple predicted residual skin ratios, the peeling parameter optimization is performed to output the second optimal peeling parameter.

7. The method of claim 6, wherein the step of controlling the ginger processing line is characterized by, The construction process of the residual skin ratio predictor comprises: Based on the historical ginger peeling record, the ginger attribute features are used as product constraints and the attribute features of the peeling device are used as device constraints to collect multiple sample peeling control schemes and multiple sample historical residual skin ratios, wherein each sample peeling control scheme comprises a sample ginger attribute feature, a sample residual skin ratio, a sample peeling parameter and a sample peeling time length; Using the multiple sample peeling control schemes and the multiple sample historical residual skin ratios, a deep learning model is trained to convergence to obtain a residual skin ratio predictor.

8. The processing line control method of ginger as claimed in claim 6, wherein, Based on the second expected residual skin ratio and the multiple predicted residual skin ratios, the peeling parameter optimization is performed to output the second optimal peeling parameter, which comprises: Based on the second expected residual skin ratio, the multiple predicted residual skin ratios are respectively calculated for deviation to obtain multiple ratio deviation values; The multiple initial peeling parameters are arranged in order from small to large according to the ratio deviation values, and the initial peeling parameters are set as initial solutions to obtain an initial solution sequence; The first solution of the initial solution sequence is set as an optimal solution, and the remaining initial solutions are set as poor solutions, and the optimal solution is used as an adjustment direction, and the poor solutions of the initial solution sequence are adjusted according to an optimization adjustment step, and an updated initial solution sequence is obtained by reordering, wherein 5% of the poor solutions of the updated initial solution sequence are eliminated each time, and an initial peeling parameter is randomly selected without replacement in the peeling parameter space to replace it. The peeling parameter iterative optimization is continuously performed until a preset second optimization number is satisfied, and the optimal solution of the current updated initial solution sequence is set as a second optimal peeling parameter, wherein the optimization number gradually increases with the increase of the peeling stage.

Images