Kiwi intelligent ripening control system integrated with multi-sensor feedback

Abstract

The application discloses a kind of integrated multi-sensor feedback kiwi intelligent ripening control system, including multi-sensor array, data processing center, ripening strategy generation unit and environmental regulation execution device.Multiple point space sampling is implemented to the fruit in the warehouse by the distributed nodes of multi-sensor array.Data processing center converts multi-dimensional state information into fruit skin color change rate, fruit hardness curve and water loss rate, calls improved fuzzy comprehensive evaluation algorithm for fusion analysis, and when endogenous ethylene generation rate is lower than effective detection threshold, it is adaptively reduced to three-dimensional evaluation space, and the current comprehensive maturity index is generated.Ripening strategy generation unit queries maturity-regulation parameter mapping table, and outputs dynamic ripening regulation parameter set, which is adjusted by environmental regulation execution device to adjust the working state of ethylene releaser, heating device and humidifying device.

Description

Technical Field

[0001] This invention belongs to the field of intelligent ripening technology for fruits and vegetables, specifically an intelligent ripening control system for kiwifruit that integrates multi-sensor feedback. Background Technology

[0002] Kiwifruit is a typical climacteric fruit, requiring artificial ripening after harvest to reach a suitable level for fresh consumption. Currently, large-scale ripening operations mainly rely on timed and quantitative application of ethylene combined with controlled temperature and humidity. Operators set ripening times and environmental parameters based on experience, and periodically sample fruit to check for changes in firmness or appearance to judge the ripening progress. This method relies on manual judgment, lacks precision in control, and is prone to batch-to-batch quality fluctuations.

[0003] Existing technologies include ripening monitoring schemes based on single or limited sensors. For example, some rely solely on environmental temperature and humidity monitoring for feedback control, while others use image acquisition to assess fruit peel color changes and estimate ripeness. These schemes face two significant drawbacks. First, the microenvironment of stacked fruit in ripening storage exhibits significant spatial heterogeneity. Temperature, humidity, and ethylene concentration vary considerably within the stack, at corners, and near ventilation openings. Monitoring data based on single or limited sampling points cannot accurately reflect the average ripeness of the entire storage, leading to biased control decisions and difficulty in ensuring uniform ripening. Second, in multi-indicator fusion evaluations, the structure and dimensions of the evaluation model are typically fixed. However, in the early stages of ripening, the release of endogenous ethylene from the fruit is extremely low, often preventing sensors from obtaining effective readings. Including this unreliable low-value data in the fusion calculation introduces noise interference, reducing the accuracy of ripeness determination. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes an intelligent ripening control system for kiwifruit integrating multi-sensor feedback, comprising: A multi-sensor array is used to continuously collect multi-dimensional state information of kiwifruit samples in the ripening storage. The multi-sensor array contains distributed sensor nodes to perform multi-point spatial sampling of the stacked fruits in the storage and generate information representing the average state of the overall fruits through statistical fusion. The multi-dimensional state information includes fruit surface color distribution images, fruit mechanical response data, endogenous ethylene production rate, fruit weight change data, and environmental temperature and humidity parameters. The data processing center receives and parses the multi-dimensional state information, converts the fruit surface color distribution image into a peel color change rate, converts the fruit mechanical response data into a fruit firmness curve, converts the fruit weight change data into a water loss rate, and calls an improved fuzzy comprehensive evaluation algorithm to perform fusion analysis and state determination on the peel color change rate, the fruit firmness curve, the endogenous ethylene generation rate, and the water loss rate to generate the current comprehensive maturity index of the kiwifruit. The improved fuzzy comprehensive evaluation algorithm is based on a preset three-dimensional membership function that includes at least color-firmness-humidity, and when the endogenous ethylene generation rate is lower than the effective detection threshold, it adaptively reduces the dimensionality to a three-dimensional evaluation space for calculation. The ripening strategy generation unit is used to query a preset maturity-control parameter mapping table based on the current comprehensive maturity index to generate a dynamic ripening control parameter set including the target ethylene application concentration, target ambient temperature, and target ambient humidity. An environmental control actuator is used to adjust the working status of the ethylene releaser, heating device, and humidification device in the ripening environment according to the dynamic ripening control parameter set.

[0005] Furthermore, the working principle of the improved fuzzy comprehensive evaluation algorithm includes: Construct a three-dimensional evaluation space that includes at least color, hardness, and humidity, and dynamically expand it into a four-dimensional evaluation space of color, hardness, ethylene, and humidity based on the effectiveness of the endogenous ethylene generation rate. Within the current valid evaluation space, define fuzzy subsets for each dimension, with each fuzzy subset containing three linguistic variables: "low", "medium", and "high". Each of the linguistic variables is configured with a membership function based on the shape adjustment of the Sigmoid function. The membership function based on the shape adjustment of the Sigmoid function introduces the inflection point position parameter and the curve steepness parameter, so that the inflection point position is dynamically determined according to the optimal ripening experimental data of the kiwifruit variety, and the curve steepness is adaptively adjusted according to the sensor measurement error range. The specific values ​​of each effective dimension indicator obtained by real-time collection and conversion are input into the membership function based on the Sigmoid function shape adjustment corresponding to each dimension to obtain the membership value of each input value relative to the three linguistic variables of "low", "medium" and "high", forming an initial fuzzy evaluation matrix. An improved weighted generalization operator is applied to aggregate the initial fuzzy evaluation matrix. The improved weighted generalization operator introduces a correlation compensation factor between indicators. When the current evaluation space is four-dimensional, it models the nonlinear coupling relationship between color and hardness, and between ethylene and humidity. When the current evaluation space is three-dimensional, it only models the nonlinear coupling relationship between color and hardness to correct the bias of the traditional weighted average. The results of the aggregate calculation are processed to clarify the output and output a scalar value between zero and one hundred. The scalar value is the current comprehensive maturity index.

[0006] Furthermore, the method for determining the inter-index correlation compensation factor in the improved weighted synthesis operator includes: During the initial calibration phase of the system, multiple rounds of ripening experiments were conducted on a batch of standard kiwifruit samples. The data of the changes in peel color, fruit firmness curve, endogenous ethylene generation rate and water loss rate in each round of experiments were fully recorded. The data of the entire process were processed by principal component analysis to extract the main influencing components, and the first covariance between the peel color change rate and the fruit firmness curve, as well as the second covariance between the endogenous ethylene generation rate and the water loss rate were analyzed. Normalize the first covariance and the second covariance respectively to obtain the first correlation factor characterizing the strength of the color-hardness correlation and the second correlation factor characterizing the strength of the ethylene-humidity correlation; In the aggregate calculation of the improved weighted generalization operator, when calculating the joint contribution of color and hardness indicators, the first correlation factor is introduced to correct the weighted product of the two indicators; when calculating the joint contribution of ethylene and humidity indicators, the second correlation factor is introduced to correct the weighted product of the two indicators. The correlation compensation factor is not a fixed value, but is dynamically selected from a preset correlation factor curve table according to the different maturity stages of the current comprehensive maturity index.

[0007] Further, converting the fruit surface color distribution image into a fruit peel color change rate includes: The color distribution image of the fruit surface is preprocessed to extract the effective area image of the kiwi fruit surface and remove background interference; The effective region image is converted from the red-green-blue color space to the hue-saturation-brightness color space, and the hue component values ​​of all pixels are extracted. Calculate the statistical histogram of the hue component values, identify the hue value corresponding to the main peak in the histogram, and use the hue value corresponding to the main peak as the representative color value of the fruit surface; Obtain the initial fruit surface color distribution image collected when the kiwifruit is put into storage, and calculate the initial representative color value according to the same process; Calculate the absolute difference between the current representative color value and the initial representative color value, and divide the absolute difference by the total monitoring time to obtain the color change per unit time, i.e., the fruit peel color change rate.

[0008] Further, the fruit mechanical response data is converted into a fruit firmness curve, including: By using a contact-type micro force sensor and displacement sensor, the indenter is pressed into the kiwi fruit at a constant speed, and the reaction force and indentation depth are recorded in real time during the pressing process to obtain a force-depth curve. The force value corresponding to the biological yield point is extracted from the force-depth curve as the fruit hardness characteristic value; The fruit firmness characteristic values ​​are recorded continuously in chronological order to form the fruit firmness curve.

[0009] Furthermore, the endogenous ethylene generation rate is obtained in the following manner: At least one sealable sampling chamber is set up in the ripening chamber. Sample fruits are sealed in the sampling chamber at regular intervals. The growth rate of ethylene concentration in the sealed chamber over time is measured using a high-sensitivity ethylene sensor as the endogenous ethylene generation rate. When the growth rate is lower than the preset effective detection limit, the endogenous ethylene generation rate is marked as invalid, triggering the adaptive dimensionality reduction of the fuzzy comprehensive evaluation algorithm.

[0010] Furthermore, based on the current comprehensive maturity index, a preset maturity-regulation parameter mapping table is queried to generate a dynamic ripening regulation parameter set containing the target ethylene application concentration, target ambient temperature, and target ambient humidity, including: The maturity-regulation parameter mapping table is a three-dimensional lookup table, with its three input dimensions being the maturity index range, the regulation parameter value of the previous regulation cycle, and the trend of environmental parameter changes. The current comprehensive maturity index, the actual control parameter values ​​implemented in the previous control cycle, and the recent changes in the environmental temperature and humidity parameters are used as the joint query key; Perform a matching search in the maturity-regulation parameter mapping table and output the set of regulation parameter reference values ​​that best match the joint query key; The reference value of the control parameter is compared with the preset optimal ripening path curve of the kiwi fruit variety. If the deviation is greater than the threshold, the reference value of the control parameter is fine-tuned according to the optimal ripening path curve to generate the final target ethylene application concentration, target ambient temperature and target ambient humidity. The dynamic ripening regulation parameter set also includes adjustment rate limits for each parameter, which are determined based on the difference between the current comprehensive maturity index and the target maturity index.

[0011] Furthermore, the reference value of the control parameter is compared with the preset optimal ripening path curve for a kiwifruit variety. If the deviation is greater than a threshold, the reference value of the control parameter is fine-tuned according to the optimal ripening path curve, including: The optimal ripening path curve defines the theoretically optimal parameter trajectory for achieving the target maturity level, with the maturity index as the horizontal axis and ethylene concentration, temperature, and humidity as the vertical axis. Calculate the absolute deviations of the ethylene concentration, temperature, and humidity in the reference values ​​of the control parameters from the theoretical values ​​of the points corresponding to the same maturity index on the optimal ripening path curve; If the absolute deviation of any of the parameters, such as ethylene concentration, temperature, and humidity, exceeds the preset independent deviation threshold of the corresponding parameter, it is determined that fine-tuning of the control parameter is required. The fine-tuning strategy is as follows: the reference value of the control parameter is adjusted towards the theoretical value, and the adjustment range is the product of the absolute deviation and a decay factor less than one, and the single adjustment range does not exceed the preset maximum single-step adjustment limit. The fine-tuned parameter values ​​are generated as the final target ethylene concentration, target ambient temperature, or target ambient humidity.

[0012] Furthermore, the application of the improved weighted synthesis operator to perform aggregation calculations on the initial fuzzy evaluation matrix includes: Obtain the initial fuzzy evaluation matrix, which contains the membership values ​​of each evaluation index under the current effective evaluation dimension corresponding to the three linguistic variables of "low", "medium" and "high". The basic weight vector of each effective evaluation index is determined. The basic weight vector is pre-set by the analytic hierarchy process based on the prior influence of each index on the maturity of kiwifruit. When the evaluation space is reduced from four dimensions to three dimensions, the weight of the ethylene dimension is allocated to the other three dimensions according to the preset rules. When the endogenous ethylene generation rate is effective, obtain the first correlation factor characterizing the correlation strength between the color index and the hardness index, and the second correlation factor characterizing the correlation strength between the ethylene index and the humidity index. The current maturity stage of the kiwifruit is determined based on the current comprehensive maturity index, and the dynamic weight adjustment coefficient of the required correlation factors in the current evaluation space is read from the preset correlation factor-maturity relationship table according to the current maturity stage. Based on the dynamic weight adjustment coefficient, the basic weights of each group of indicators with coupling relationship, as well as the basic weights of the ethylene index and the humidity index, are weighted and adjusted respectively to obtain the dynamic effective weights of each effective evaluation index. Based on the coupling relationship between indicators, an aggregation calculation model is constructed. For color and hardness indicators that have a coupling relationship, their aggregation contribution is calculated by the dynamic effective weight, the membership value, and the first correlation factor. For ethylene and humidity indicators that have a coupling relationship only in the four-dimensional evaluation space, their aggregation contribution is calculated by the dynamic effective weight, the membership value, and the second correlation factor. For indicator pairs without a coupling relationship, their aggregation contribution is calculated by the product of their respective dynamic effective weights and membership values. The aggregated contributions of all indicators are summed to obtain the comprehensive fuzzy evaluation set after aggregate calculation.

[0013] Furthermore, the system also includes a process of verifying the reliability of readings from the multi-sensor array and performing data fusion: A confidence weighting coefficient based on the stability of historical readings is configured for each of the fruit surface color sensor, fruit body elasticity sensor, ethylene gas sensor, and temperature and humidity sensor in the multi-sensor array. The rate of change of each sensor reading is monitored in real time. If the instantaneous rate of change of a sensor reading exceeds its normal fluctuation range, its confidence weight coefficient is temporarily reduced. When performing data processing and fusion calculations, the data provided by each sensor is multiplied by its current confidence weight coefficient before being used in the calculation; When the confidence weight coefficient of a certain sensor is continuously lower than the fault threshold, a sensor fault alarm is triggered, and an estimation algorithm based on data from other sensors is attempted to fill in the missing data of the corresponding sensor.

[0014] Compared with the prior art, the beneficial effects of the present invention are: Distributed sensor nodes are used to conduct multi-point spatial sampling of stacked fruits within the ripening storage facility. Statistical fusion is then used to generate multi-dimensional state information representing the overall average state of the fruits. This approach expands the monitoring scope from local points to the spatially distributed fruit population within the storage facility. By utilizing multi-point acquisition and statistical processing, random biases from individual sampling locations are eliminated. This ensures that the data input into subsequent evaluation stages systematically reflects the average ripening level of the entire batch of fruits, thus solving the problem of distortion in the representation of the global state by local sampling due to spatial heterogeneity. This allows ripening control decisions to be based on global state information rather than local, one-sided sampling results.

[0015] When the endogenous ethylene generation rate is below the effective detection threshold, the improved fuzzy comprehensive evaluation algorithm adaptively reduces the dimensionality to a three-dimensional evaluation space of color, hardness, and humidity for calculation, eliminating noise interference introduced by unreliable low-value readings from the ethylene sensor. This adaptive adjustment of the evaluation space dimensions allows maturity determination in the early stages of ripening, when the endogenous ethylene signal is not yet significant, to still rely on the fusion information of three reliable dimensions: color, hardness, and humidity. This avoids the bias caused by unreliable low-value data in the calculation of the comprehensive maturity index. Once the endogenous ethylene generation rate rises above the effective detection threshold, the four-dimensional evaluation space is restored, achieving dynamic adaptation of the evaluation model structure to data quality throughout the entire ripening cycle. Attached Figure Description

[0016] Figure 1 This is a timing diagram of an intelligent ripening control system for kiwifruit integrating multi-sensor feedback, as described in this invention. Figure 2 Flowchart of the work on the improved fuzzy comprehensive evaluation algorithm; Figure 3 A flowchart illustrating the process of determining the compensation factor for the correlation between indicators. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1A multi-sensor feedback-integrated intelligent ripening control system for kiwifruit includes a multi-sensor array, a data processing center, a ripening strategy generation unit, and an environmental control execution device. The multi-sensor array continuously collects multi-dimensional state information of kiwifruit samples, including fruit surface color distribution images, fruit elastic deformation data, ethylene gas concentration, and environmental temperature and humidity parameters. The data processing center receives and analyzes the multi-dimensional state information, converting the fruit surface color distribution image into a peel color change rate, the fruit elastic deformation data into a fruit firmness curve, the ethylene gas concentration into an endogenous ethylene generation rate, and the environmental temperature and humidity parameters into a moisture loss rate. It also uses an improved fuzzy comprehensive evaluation algorithm to assess the peel color change rate. The system integrates and analyzes the fruit firmness curve, endogenous ethylene production rate, and water loss rate to determine the current comprehensive maturity index of kiwifruit. An improved fuzzy comprehensive evaluation algorithm is implemented based on a preset three-dimensional membership function that includes at least color, firmness, and humidity. When the endogenous ethylene production rate is below the effective detection threshold, it adaptively reduces the dimensionality to a three-dimensional evaluation space for computation. A ripening strategy generation unit queries a preset maturity-control parameter mapping table based on the current comprehensive maturity index to generate a dynamic ripening control parameter set containing the target ethylene application concentration, target ambient temperature, and target ambient humidity. An environmental control execution device adjusts the operating status of the ethylene releaser, heating device, and humidification device within the ripening environment according to the dynamic ripening control parameter set.

[0019] In one embodiment of the present invention, during the application of the improved fuzzy comprehensive evaluation algorithm, refer to... Figure 2In specific implementation, taking the ripening monitoring process of the kiwifruit variety "Hayward" as an example, a multi-sensor array collected a set of multi-dimensional status information 12 hours after the ripening process was started. The data processing center converted the fruit surface color distribution image into the peel color change rate, the fruit elastic deformation data into the fruit firmness curve, the released ethylene gas concentration into the endogenous ethylene generation rate, and the environmental temperature and humidity parameters into the water loss rate. At this time, the peel color change rate was 0.08 color units per hour, the instantaneous elastic coefficient of the fruit firmness curve at the current moment was 0.85 Newtons per millimeter, the endogenous ethylene generation rate was 0.02 microliters per kilogram per hour, and the water loss rate was 0.015 grams per kilogram per hour. Because the endogenous ethylene generation rate was lower than the preset value, The effective detection threshold is 0.05 μL / kg / hour. The improved fuzzy comprehensive evaluation algorithm adaptively reduces the evaluation space from a four-dimensional space of color-hardness-ethylene-humidity to a three-dimensional space of color-hardness-humidity. At 48 hours after the ripening process begins, the multi-sensor array collects and converts the data again, obtaining a peel color change rate of 0.18 hue units per hour, an instantaneous elastic coefficient of 0.52 N / mm for the fruit hardness curve at the current moment, an endogenous ethylene generation rate of 0.15 μL / kg / hour, and a moisture loss rate of 0.025 g / kg / hour. The endogenous ethylene generation rate recovers to above the effective detection threshold. The improved fuzzy comprehensive evaluation algorithm expands the evaluation space to a four-dimensional space of color-hardness-ethylene-humidity, thereby forming a data comparison in the time dimension.

[0020] When constructing the aforementioned three-dimensional and four-dimensional evaluation spaces, a fuzzy subset is defined for each effective dimension. Each fuzzy subset contains three linguistic variables: "low," "medium," and "high." A membership function based on Sigmoid function morphological adjustment is configured for each linguistic variable. This Sigmoid function morphological adjustment introduces inflection point position parameters and curve steepness parameters. The inflection point position is dynamically determined based on the optimal ripening experimental data of the kiwifruit variety "Hayward," and the curve steepness is adaptively adjusted based on the sensor measurement error range. In the three-dimensional evaluation space, color, hardness, and humidity indicators each correspond to their respective Sigmoid function morphological adjustment membership functions. The complete mathematical expression of the improved fuzzy comprehensive evaluation algorithm is as follows. First, for each evaluation indicator (color... ,hardness ethylene ,humidity Define a membership function based on the Sigmoid function's shape adjustment. For any index Its membership degree belongs to the linguistic variables "low", "medium", and "high". , , Calculated by the following functions respectively:

[0021]

[0022]

[0023] in, These are the real-time collected values ​​of the indicator. , , The inflection point position parameter of the membership function. , and Parameters used to control the steepness of the curve. Inflection point location parameters. The optimal ripening experimental data for the target kiwifruit variety were determined using the following method: Measured data sequences of various indicators for this variety during the ripening process were collected, the sequences were sorted numerically, and the top 25% quantiles were used as the baseline. The initial value, the median as The initial value, and the last 75th percentile as The initial value is then fine-tuned based on expert experience. The curve steepness parameter is determined according to the sensor measurement error range. Adaptive adjustment, the adjustment rules are as follows: , ,in , This is a constant proportional factor set according to the variety. When the endogenous ethylene production rate is invalid and the evaluation space is reduced from four dimensions to three dimensions, the original weight of the ethylene dimension is retained. The data is reassigned to the color, hardness, and humidity dimensions according to preset rules. The preset rules are: ... proportionally Allocation, among which , , These are allocation coefficients pre-set based on the importance of color, hardness, and humidity to maturity in the absence of ethylene information, satisfying... Then the new basic weights for each dimension after dimensionality reduction. , , The correlation factor calculation is as described in claim 3, and the expression for the coupling term is embodied in the improved weighted synthesis operator. For index pairs with coupling relationships (such as color...) With hardness ), their joint contributions Calculated by the following formula:

[0024] in, , For the dynamic effective weight of the indicator, , This is a vector consisting of the membership values ​​of the corresponding linguistic variables. As a color-hardness correlation factor, This represents element-wise multiplication of vectors. For ethylene... With humidity The indicators are paired, and the calculation methods are the same. The clarification calculation uses the centroid method; let the comprehensive fuzzy evaluation set obtained after aggregation calculation be a vector. These represent the overall membership degree to "low," "medium," and "high," respectively. The current comprehensive maturity index is then clearly output. for:

[0025] in, , , These are predefined typical values ​​corresponding to "low", "medium", and "high" language variables, for example, they can be set to 0, 50, and 100 respectively.

[0026] In the four-dimensional evaluation space, a membership function based on the Sigmoid function morphology adjustment corresponding to the ethylene index is added. Taking the 12-hour data as an example, the peel color change rate of 0.08 is input into the Sigmoid function morphology adjustment membership function of the color index, resulting in a membership value of 0.85 relative to "low", 0.14 relative to "medium", and 0.01 relative to "high". The instantaneous elastic coefficient of the fruit firmness curve of 0.85 is input into the Sigmoid function morphology adjustment membership function of the firmness index, resulting in a membership value of 0.02 relative to "low", 0.18 relative to "medium", and 0.80 relative to "high". The water loss rate of 0.015 is input into the Sigmoid function morphology adjustment membership function of the humidity index, resulting in a membership value of 0.90 relative to "low", 0.10 relative to "medium", and 0.01 relative to "high". At 0.00, the aforementioned membership values ​​constitute the initial fuzzy evaluation matrix in the three-dimensional effective evaluation space of the 12th hour. Taking the data of the 48th hour as an example, the fruit peel color change rate of 0.18 is input into the membership function of the color index based on the Sigmoid function morphology adjustment, resulting in membership values ​​of 0.04, 0.91, and 0.05 for "low," "medium," and "high," respectively. The hardness value of 0.52 is input into the hardness index into the membership function of the hardness index based on the Sigmoid function morphology adjustment, resulting in membership values ​​of zero. The membership function of the ethylene index, with 0.15 as the input of the endogenous ethylene generation rate and 0.15 as the input of the ethylene index, is adjusted based on the Sigmoid function, resulting in membership values ​​of 0.03, 0.89, and 0.08 respectively. The membership function of the humidity index, with 0.025 as the input of the water loss rate, is adjusted based on the Sigmoid function, resulting in membership values ​​of 0.06, 0.90, and 0.04 respectively. These values ​​constitute the initial fuzzy evaluation matrix in the four-dimensional evaluation space at the 48th hour.

[0027] When applying the improved weighted generalization operator to aggregate the initial fuzzy evaluation matrix, the basic weight vectors of each effective evaluation index are first determined. These basic weight vectors are pre-set using the analytic hierarchy process (AHP) based on the prior influence of each index on the ripeness of kiwifruit. In the four-dimensional evaluation space, the basic weights of the color, hardness, ethylene, and humidity indices are set to 0.3, 0.3, 0.2, and 0.2, respectively. When the evaluation space is reduced from four-dimensional to three-dimensional, the basic weight of the ethylene dimension is allocated according to a preset rule, distributing the 0.2 basic weight of the ethylene index evenly to the other three dimensions. This ensures that the basic weights of the color, hardness, and humidity indices in the three-dimensional evaluation space are adjusted to their original basic weights plus 0.067, i.e., [the weights are adjusted accordingly]. Instead of 0.367, 0.367, and 0.266, when obtaining the first correlation factor characterizing the correlation strength between color and hardness indices, and the second correlation factor characterizing the correlation strength between ethylene and humidity indices, the system has already processed the entire process data of the Hayward variety standard sample through principal component analysis during the initial calibration stage, obtaining a normalized first correlation factor of 0.65 and a second correlation factor of 0.58. The improved weighted comprehensive operator also determines the current maturity stage of the kiwifruit based on the current comprehensive maturity index, and reads the dynamic weight adjustment coefficients of the required correlation factors in the current evaluation space from the preset correlation factor-maturity relationship table. At the 12th hour, based on the current comprehensive maturity estimated at the previous moment... The moisture index is at a relatively low maturity stage of approximately 15%. The adjustment coefficient for the color-hardness combination at this time, as shown in the correlation factor-maturity relationship table, is 0.75. At the 12th hour, in the three-dimensional evaluation space, there is no coupling between the ethylene index and the humidity index. Therefore, only the first correlation factor and its dynamic weight adjustment coefficient are used to adjust the basic weights of the coupled color and hardness indices. The resulting dynamic effective weight for the color index is 0.546, calculated as the base weight of 0.367 multiplied by (with parentheses 1), plus the first correlation factor of 0.65 multiplied by the dynamic weight adjustment coefficient of 0.75. The dynamic effective weight for the hardness index is also 0.546. The humidity index has no coupling relationship, and its dynamic effective weight remains unchanged. Holding the base weight of 0.266, at the 48th hour, the current comprehensive maturity index is approximately 55%, indicating a mid-term stage. The dynamic weight adjustment coefficients for color and hardness, read from the correlation factor-maturity relationship table, are 0.88 and 0.82, respectively. After weighting the base weights of the color and hardness indicators, the dynamic effective weight for the color indicator is 0.3 multiplied by (1 in parentheses) plus 0.65 multiplied by 0.88, equaling 0.4716. The dynamic effective weight for the hardness indicator is also 0.4716. After weighting the base weights of the ethylene and humidity indicators, the dynamic effective weight for the ethylene indicator is 0.2 multiplied by (1 in parentheses) plus 0.58 multiplied by 0.82, equaling 0.2951.The dynamic effective weight of the humidity index is 0.2 multiplied by 1 (in parentheses) plus 0.58 multiplied by 0.82, which equals 0.2951.

[0028] In the constructed aggregation calculation model, the aggregation contribution of each indicator is calculated based on the coupling relationship between the indicators. For color and hardness indicators with coupling relationships, their aggregation contributions are calculated by their respective dynamic effective weights, the membership values ​​corresponding to each linguistic variable in the initial fuzzy evaluation matrix, and the first correlation factor. The aggregation contributions of color and hardness indicators to the linguistic variables "low," "medium," and "high" are obtained by weighted summation of their membership values ​​and superimposing the cross-terms affected by the first correlation factor. For ethylene and humidity indicators with coupling relationships only in the four-dimensional evaluation space, the aggregation contribution is calculated using the second correlation factor in the same way. For indicator pairs without coupling relationships, the aggregation contribution is directly obtained by multiplying the dynamic effective weight of the indicator with its membership value. Taking the 12-hour three-dimensional evaluation space as an example, the humidity indicator has no coupling relationship, and its aggregation contributions to "low," "medium," and "high" are 0.266 multiplied by 0.90, 0.266 multiplied by 0.10, and 0.266 multiplied by 0.00, respectively. The aggregated contributions of color and hardness indices are calculated by combining the sequence 0.546, the membership sequence, and the first correlation factor 0.65. Taking the four-dimensional evaluation space at the 48th hour as an example, the aggregated contributions of color and hardness indices are calculated by their respective dynamic effective weights of 0.4716, membership, and the first correlation factor. The aggregated contributions of ethylene and humidity indices are calculated by their dynamic effective weights of 0.2951, membership, and the second correlation factor. The aggregated contributions of all indices are summed according to linguistic variables to obtain the aggregated comprehensive fuzzy evaluation set. The aggregated comprehensive fuzzy evaluation set is then clarified by using the centroid method to transform the fuzzy set into scalar values ​​between zero and one hundred. After clarification, the comprehensive fuzzy evaluation set at the 12th hour outputs a current comprehensive maturity index of sixteen, and at the 48th hour outputs a current comprehensive maturity index of fifty-three. This achieves a continuous numerical representation of the kiwifruit maturity status, and the dynamic adjustment of the spatial dimension and correlation factor allows the maturity index to evolve smoothly with changes in multi-sensor feedback information.

[0029] In one embodiment of the present invention, the method for determining the inter-index correlation compensation factor in the improved weighted synthesis operator is as follows. (See reference...) Figure 3In the initial calibration phase of the system, multiple rounds of ripening experiments were conducted on a batch of standard kiwifruit samples. The changes in peel color, fruit firmness curve, endogenous ethylene production rate, and water loss rate were recorded in detail for each round. Principal component analysis was used to process this data, extracting the main influencing components and analyzing the first covariance between the peel color change rate and the fruit firmness curve, as well as the second covariance between the endogenous ethylene production rate and the water loss rate. The first and second covariances were normalized to obtain a first correlation factor characterizing the color-firmness correlation strength and a second correlation factor characterizing the ethylene-humidity correlation strength. In the aggregate calculation of the improved weighted synthesis operator, when calculating the joint contribution of color and firmness indices, the first correlation factor was introduced to correct the weighted product of the two indices; when calculating the joint contribution of ethylene and humidity indices, the second correlation factor was introduced to correct the weighted product of the two indices. These correlation compensation factors are not fixed values ​​but are dynamically selected from a pre-set correlation factor curve table based on the different maturity stages of the current comprehensive maturity index.

[0030] In practical implementation, determining the inter-indicator correlation compensation factor in the improved weighted comprehensive operator involves the initial system calibration of standard kiwifruit samples. During the initial calibration phase, multiple rounds of ripening experiments were conducted on a batch of "Golden Fruit" variety standard kiwifruit samples. In each round, the changes in peel color, fruit firmness curve, endogenous ethylene production rate, and water loss rate were fully recorded from the start of the experiment until the fruit reached the target maturity. The complete process data includes time-series data of peel color change rate, eigenvalues ​​of the fruit firmness curve, endogenous ethylene production rate, and water loss rate. Principal component analysis (PCA) was used to process the recorded data. PCA calculated the covariance matrix of all indicators in the data and extracted the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvalues ​​were arranged from largest to smallest, and several principal components with a cumulative contribution rate exceeding a preset threshold were selected as the main influencing components. Based on the analysis of the main influencing components, the first covariance between the numerical sequence of the peel color change rate and the characteristic value sequence of the fruit firmness curve was calculated, and the second covariance between the numerical sequence of the endogenous ethylene production rate and the numerical sequence of the water loss rate was also calculated.

[0031] In some embodiments, the calculated first covariance and second covariance are normalized to obtain a first correlation factor characterizing the strength of the color-hardness correlation and a second correlation factor characterizing the strength of the ethylene-humidity correlation. The formula for calculating the first correlation factor is:

[0032] in: Represents the first correlation factor. The first covariance represents the relationship between the rate of change in peel color and the fruit firmness curve. The standard deviation of the numerical series representing the rate of change in fruit peel color. The standard deviation of the eigenvalue sequence representing the fruit firmness curve. The formula for calculating the second correlation factor is:

[0033] in: Represents the second correlation factor. The second covariance represents the relationship between the rate of endogenous ethylene formation and the rate of water loss. The standard deviation of the numerical sequence representing the endogenous ethylene formation rate. This represents the standard deviation of the numerical sequence of water loss rates. Based on the above calculations, the first correlation factor... With the second correlation factor The range of values ​​is normalized to the interval [-1, 1].

[0034] In practical implementation, a correlation factor is introduced for correction in the aggregation calculation of the improved weighted synthesis operator. When calculating the joint contribution of color and hardness indices, a first correlation factor is introduced. The product of the base weights for color and hardness indices is adjusted by adding a term to the aggregate calculation model. Dynamic effective weight of color index Dynamic effective weight of hardness index Related coupling terms. A second correlation factor is introduced when calculating the joint contribution of the ethylene index and the humidity index. The product of the basic weights of the ethylene index and the humidity index is corrected by adding a term to the polymerization calculation model. Dynamic effective weight of ethylene index Dynamic effective weighting of humidity index Related coupling terms. The correlation compensation factor does not use a fixed value; the first correlation factor... With the second correlation factor The specific value is dynamically selected based on the different maturity stages determined by the current comprehensive maturity index. The first correlation factor is stored in the system's pre-set correlation factor curve table. Relationship curve with maturity stage and the second correlation factor Relationship curve with maturity stage Relationship curve and The shape was obtained by fitting the results of analysis of multiple rounds of experimental data during the initial calibration phase.

[0035] Optional, relationship curve and This can be represented as a discrete lookup table. For example, the maturity stage is divided into five stages: "green maturity," "early color change," "mid-color change," "soft maturity," and "fully mature." The correlation factor curve table stores the corresponding first correlation factor recommendation value for each stage. Recommended value of the second correlation factor During the aggregate calculation, the maturity stage of the kiwifruit is determined based on the current comprehensive maturity index, and the corresponding maturity stage is retrieved from the correlation factor curve table. Value and The value is used for the current comprehensive evaluation calculation.

[0036] It is understandable that extracting the main influencing components through principal component analysis and then calculating the covariance can reduce the interference of data noise on the correlation strength analysis, thus improving the calculated first correlation factor. With the second correlation factor It better reflects the essential correlation between indicators. In some embodiments, the first covariance is calculated. With the second covariance The data used is reconstructed data after dimensionality reduction through principal component analysis, rather than the original full-process data. The reconstructed data retains the main characteristics of the original data but filters out some random fluctuations. This can be understood as correlation factors. and Based on the dynamic selection of the ripening stage, compared with the use of fixed correlation factors, the dynamically selected correlation factors can better reflect the changes in the coupling relationship between the internal physiological and biochemical indicators of kiwifruit at different ripening stages. For example, in the "mid-coloring stage", the correlation between color change and decrease in firmness may be the strongest, while the correlation is weaker in the "green ripening stage". The dynamic correlation factors can adapt to this non-linear change.

[0037] In one embodiment of the invention, this system is deployed in a commercial ripening facility housing Hayward kiwifruit. A sampling node in the multi-sensor array triggers a mechanical measurement cycle 48 hours after the ripening process begins. A contact-type micro-force sensor uses a 3mm diameter cylindrical flat-head indenter. The displacement sensor resolution is set to 0.01mm. The indenter is vertically pressed into the equatorial region of the kiwifruit at a constant speed of 0.5mm per second, with an upper limit of 8mm indentation depth. During the pressing process, the force sensor and displacement sensor synchronously collect data, generating a force-depth curve. The curve records the reaction force values ​​corresponding to the increasing indentation depth from 0mm to 8mm. Initially, the reaction force increases approximately linearly with depth, then the slope gradually decreases until a clear inflection point appears. The force-depth curve is processed using a second derivative zero-crossing detection method to locate the depth at which the second derivative crosses zero from a negative value. The force value corresponding to this location is 12.3 Newtons, and this force value is identified as the fruit hardness characteristic value corresponding to the biological yield point.

[0038] Four mechanical measurements had been completed at the same sampling node prior to this measurement. The measurements were taken at 0, 12, 24, and 36 hours after the start of ripening, with corresponding fruit firmness values ​​of 19.6 N, 17.8 N, 15.9 N, and 14.2 N, respectively. These five fruit firmness values ​​were recorded sequentially to form a fruit firmness curve. The curve shows a monotonically decreasing trend from 19.6 N to 12.3 N, with a faster rate of decrease initially followed by a gradual slowdown, thus fully depicting the dynamic process of fruit pulp softening during ripening. After each mechanical measurement, the indenter automatically returned to its initial position, the sensor zero point was automatically calibrated, and the sample fruit remained stationary in place during the measurement interval to avoid external interference affecting subsequent measurement results.

[0039] Inside the ripening chamber, two sealable sampling chambers were set up, each with a volume of 20 liters and stainless steel inner walls. Each chamber was equipped with an automatically opening and closing airtight lid and a high-sensitivity ethylene sensor, with a detection limit of 0.005 μL / L. Forty-eight hours after the ripening process began, the automatic airtight lid of sampling chamber one closed, sealing three sample fruits inside for 10 minutes. The high-sensitivity ethylene sensor recorded an ethylene concentration of 0.12 μL / L at the moment of closure and 0.28 μL / L at the end of closure, calculating an increase of 0.16 μL / L in ethylene concentration over 10 minutes. This translates to an endogenous ethylene generation rate of 0.016 μL / L / min. This endogenous ethylene generation rate is higher than the preset effective detection limit of 0.005 μL / L / min; therefore, the endogenous ethylene generation rate obtained in this measurement is marked as valid. The fuzzy comprehensive evaluation algorithm maintains the four-dimensional evaluation space, incorporating the ethylene dimension into the maturity evaluation.

[0040] 72 hours after the ripening process began, sampling chamber 1 underwent another closed measurement, with the closure lasting 10 minutes. The initial ethylene concentration was 0.31 μL / L, and the concentration at the end of the closure was 0.33 μL / L, resulting in a calculated increase of 0.02 μL / L, corresponding to an endogenous ethylene production rate of 0.002 μL / L / min. This value is below the preset effective detection limit of 0.005 μL / L / min, indicating that the release of endogenous ethylene from the fruit is extremely weak at this point, and the noise content in the sensor readings is significantly increased. The system marked this measured endogenous ethylene production rate as invalid. This invalidation triggered the fuzzy comprehensive evaluation algorithm to perform an adaptive dimensionality reduction operation, reducing the evaluation space from four dimensions to three. In subsequent fusion calculations, the ethylene dimension was excluded, and the maturity status was determined solely based on three dimensions: the rate of change in peel color, the fruit firmness curve, and the water loss rate.

[0041] The formula used to calculate the endogenous ethylene formation rate is as follows:

[0042] in This indicates the rate of endogenous ethylene production, expressed in microliters per liter per minute. This indicates the ethylene concentration in the sealed chamber at the end of the sealing process, expressed in microliters per liter. This indicates the initial ethylene concentration in the sealed chamber, expressed in microliters per liter. Indicates the duration of the closure, in minutes. This represents the proportion of the total volume of the sampled fruit within the sealed chamber to the total volume of the sealed chamber. In the measurement example above, taken 48 hours after the ripening process was initiated, The value is 0.28 microliters per liter. The value is 0.12 microliters per liter. The value is 10 minutes. The value is 0.15, which is then substituted into the formula to obtain the result. The result is 0.016 microliters per liter per minute, consistent with the simplified difference division result, but a scaling factor is introduced into the formula. To correct for the effect of sample fruit volume differences on the dilution effect of ethylene concentration. When the calculated endogenous ethylene generation rate is lower than 0.005 μL / L / min, the system automatically marks the corresponding rate as invalid and triggers dimensionality reduction. When the calculated value is higher than or equal to 0.005 μL / L / min, normal four-dimensional calculation is maintained.

[0043] In another time window, 96 hours after the ripening process was initiated, sampling chamber number two underwent closed-loop measurement for 10 minutes, and the results were obtained. It is 0.45 microliters per liter. It is 0.62 microliters per liter. The value was set to 0.18, and the calculated endogenous ethylene generation rate was 0.031 μL / min. This value is significantly higher than the effective detection limit, so the measurement result is marked as valid, and the system continues to operate in the four-dimensional evaluation space. The closed measurement sequence and the mechanical measurement sequence of the sampling chamber are independently scheduled without conflict. The two sampling chambers work alternately to cover the sample fruits in different areas of the storage. After the closed measurement is completed, the airtight cover automatically opens and ventilation is activated to ensure that the initial ethylene concentration in the sealed chamber returns to the environmental baseline level before the next closed measurement.

[0044] In one embodiment of the present invention, the process of generating a dynamic ripening control parameter set, including the target ethylene application concentration, target ambient temperature, and target ambient humidity, by querying a preset maturity-control parameter mapping table based on the current comprehensive maturity index is as follows. The maturity-control parameter mapping table is a three-dimensional lookup table, with its three input dimensions being the maturity index range, the control parameter value of the previous control cycle, and the trend of environmental parameter changes. The specific data support for the maturity-control parameter mapping table comes from statistical summarization and machine learning model training of historical ripening experimental data for a specific variety (such as "Hayward"). The table is constructed as follows: First, large-scale ripening experiments are conducted under various initial conditions. The complete records are kept for each experimental batch throughout the ripening process, detailing the actual combination of control parameters (ethylene concentration, temperature, humidity) used to achieve good ripening results under different comprehensive maturity index ranges (e.g., 0-20, 20-40, 40-60, 60-80, 80-100), based on parameters from the previous control cycle (e.g., ethylene concentration in the previous cycle was in the range of 10-15 ppm, 15-20 ppm), and environmental parameter trends (e.g., "temperature rise and humidity fall," "temperature and humidity stable," "temperature fall and humidity rise"). Second, a random forest model is trained using this historical data, taking the comprehensive maturity index range, the previous control parameter state, and the environmental change trend as input, and the recommended control parameters as output. Finally, the input space of the model is discretized, and its prediction results are used to create a three-dimensional lookup table, which is the maturity-control parameter mapping table. For example, for the "Hayward" variety, the table might contain the following specific record: When the comprehensive maturity index is in the 40-50 range, the ethylene concentration in the previous cycle was 20 ppm, and the environment shows a "temperature rise and humidity fall" trend, the reference values ​​for the output control parameters are: ethylene concentration 22 ppm, temperature 17.5 degrees Celsius, and humidity 82%. The optimal ripening path curve is also generated based on historical experimental data, defining the path from initial maturity to target maturity (e.g., ...). The theoretically most time- and energy-efficient trajectory of ethylene concentration, temperature, and humidity parameters during the ripening process is determined by curve fitting. This curve can be a piecewise function or a high-dimensional lookup table. For example, some data points can be expressed as follows: when the overall maturity index is 30, the corresponding theoretical optimal parameters are (ethylene 18 ppm, temperature 19 degrees Celsius, humidity 85%); when the overall maturity index is 45, the corresponding theoretical optimal parameters are (ethylene 20.5 ppm, temperature 18.2 degrees Celsius, humidity 83%); and when the overall maturity index is 60, the corresponding theoretical optimal parameters are (ethylene 25 ppm, temperature 16.5 degrees Celsius, humidity 80%). These theoretical values ​​are obtained by analyzing the parameter paths taken by the batches with the fastest ripening speed and best quality uniformity in historical data.

[0045] The current comprehensive maturity index, the actual control parameter values ​​implemented in the previous control cycle, and the recent changes in environmental temperature and humidity parameters are used as the joint query key. A matching search is performed in the maturity-control parameter mapping table, outputting the set of control parameter reference values ​​that best match the joint query key. The control parameter reference values ​​are compared with the preset optimal ripening path curve for kiwifruit varieties. If the deviation exceeds a threshold, the control parameter reference values ​​are fine-tuned based on the optimal ripening path curve to generate the final target ethylene application concentration, target environmental temperature, and target environmental humidity. The dynamic ripening control parameter set also includes adjustment rate limits for each parameter, determined based on the difference between the current comprehensive maturity index and the target maturity index.

[0046] The process of comparing and fine-tuning the reference values ​​of the control parameters with the preset optimal ripening path curve for kiwifruit varieties includes: the optimal ripening path curve defines the theoretical optimal parameter trajectory for achieving the target maturity level, with the maturity index as the abscissa and ethylene concentration, temperature, and humidity as the ordinate. The absolute deviations of the ethylene concentration, temperature, and humidity in the control parameter reference values ​​from the theoretical values ​​at the corresponding points on the optimal ripening path curve for the same maturity index are calculated. If the absolute deviation of any of the ethylene concentration, temperature, or humidity parameters exceeds the preset independent deviation threshold for the corresponding parameter, fine-tuning of the control parameters is required. The fine-tuning strategy is to adjust the control parameter reference values ​​towards the theoretical values, with the adjustment magnitude being the product of the absolute deviation and a decay factor less than one, and the single adjustment magnitude not exceeding the preset maximum single-step adjustment limit. The fine-tuned parameter values ​​are generated as the final target ethylene application concentration, target ambient temperature, or target ambient humidity.

[0047] In practice, the process of querying the preset maturity-regulation parameter mapping table based on the current comprehensive maturity index and generating a dynamic ripening regulation parameter set involves a specific regulation cycle. The maturity-regulation parameter mapping table is a three-dimensional lookup table, with three input dimensions: the maturity index range, the regulation parameter values ​​from the previous regulation cycle, and the environmental parameter change trend. When generating the dynamic ripening regulation parameter set, the current comprehensive maturity index, the actual regulation parameter values ​​executed in the previous regulation cycle, and the recent changes in environmental temperature and humidity parameters are used as the joint query key. The current comprehensive maturity index is 45, the actual regulation parameter values ​​executed in the previous regulation cycle were ethylene concentration of 20 ppm, temperature of 18 degrees Celsius, and humidity of 85%, and the recent changes in environmental temperature and humidity parameters show "temperature rising, humidity falling."

[0048] A matching search is performed in the maturity-regulation parameter mapping table. Based on the nearest neighbor matching principle, the search outputs the set of regulation parameter reference values ​​that best match the joint query key. These reference values ​​are ethylene concentration 22 ppm, temperature 17.5 degrees Celsius, and humidity 82%. The regulation parameter reference values ​​are then compared with the preset optimal ripening path curve for the kiwifruit variety "Jinyan." This optimal ripening path curve, with maturity index on the x-axis and ethylene concentration, temperature, and humidity on the y-axis, defines the theoretically optimal parameter trajectory for achieving the target maturity level. If the deviation exceeds a threshold, the regulation parameter reference values ​​are fine-tuned based on the optimal ripening path curve to generate the final target ethylene concentration, target ambient temperature, and target ambient humidity. The dynamic ripening control parameter set also includes adjustment rate limits for each parameter. The adjustment rate limits are determined based on the difference between the current comprehensive maturity index and the target maturity index. The current comprehensive maturity index is 45, the target maturity index is 70, and the difference is 25. Based on the difference-rate limit mapping relationship, the adjustment rate limit for ethylene concentration is determined to be 2 ppm per hour, the adjustment rate limit for temperature is 0.5 degrees Celsius per hour, and the adjustment rate limit for humidity is 2% per hour.

[0049] In practice, the process of comparing and fine-tuning the reference values ​​of the control parameters with the preset optimal ripening path curve for kiwifruit varieties is as follows. The optimal ripening path curve defines the theoretically optimal parameter trajectory for achieving the target maturity level, with the maturity index as the abscissa and ethylene concentration, temperature, and humidity as the ordinate. The absolute deviations of the ethylene concentration, temperature, and humidity in the reference values ​​of the control parameters from the theoretical values ​​at the corresponding points of the same maturity index on the optimal ripening path curve are calculated. Looking up the optimal ripening path curve, when the maturity index is 45, the corresponding theoretically optimal parameters are an ethylene concentration of 20.5 ppm, a temperature of 18.2 degrees Celsius, and a humidity of 83%. The calculated absolute deviations are 1.5 ppm for ethylene concentration, 0.7 degrees Celsius for temperature, and 1% for humidity.

[0050] If the absolute deviation of any of the parameters—ethylene concentration, temperature, and humidity—exceeds the preset independent deviation threshold for that parameter, fine-tuning of the control parameters is required. The preset independent deviation thresholds are 1 ppm for ethylene concentration, 0.5 degrees Celsius for temperature, and 1.5% for humidity. Comparison results show that the absolute deviation of 1.5 ppm for ethylene concentration is greater than the independent deviation threshold of 1 ppm, the absolute deviation of 0.7 degrees Celsius for temperature is greater than the independent deviation threshold of 0.5 degrees Celsius, and the absolute deviation of 1% for humidity is less than the independent deviation threshold of 1.5%. Therefore, fine-tuning of the reference values ​​for the control parameters of ethylene concentration and temperature is necessary. The fine-tuning strategy involves adjusting the reference values ​​of the control parameters towards the theoretical values. The adjustment magnitude is the product of the absolute deviation and a decay factor less than 1, and the magnitude of a single adjustment does not exceed the preset maximum single-step adjustment limit. The decay factor is preset to 0.6, the maximum single-step adjustment limit for ethylene concentration is 1 ppm, and the maximum single-step adjustment limit for temperature is 0.3 degrees Celsius.

[0051] The ethylene concentration adjustment range was calculated as 1.5 ppm multiplied by 0.6, which equals 0.9 ppm. However, 0.9 ppm exceeds the maximum single-step adjustment limit of 1 ppm, so the adjustment range was set to 0.9 ppm. The temperature adjustment range was calculated as 0.7 degrees Celsius multiplied by 0.6, which equals 0.42 degrees Celsius. 0.42 degrees Celsius exceeds the maximum single-step adjustment limit of 0.3 degrees Celsius, so the adjustment range was set to 0.3 degrees Celsius. The fine-tuned parameter values ​​were generated: the ethylene concentration was fine-tuned to 22 ppm minus 0.9 ppm, which equals 21.1 ppm; the temperature was fine-tuned to 17.5 degrees Celsius minus 0.3 degrees Celsius, which equals 17.2 degrees Celsius; and the humidity did not exceed the threshold, so the original reference value of 82% was maintained. The fine-tuned parameter values ​​were used as the final target ethylene application concentration, target ambient temperature, or target ambient humidity. A simplified maturity-control parameter mapping table is shown in Table 1. Table 1: Maturity-Regulation Parameter Mapping Table

[0052] In some embodiments, the construction of the three-dimensional lookup table is based on historical ripening experiment data and the training of a machine learning model. The fine-tuning calculation of the optimal ripening path curve can be expressed as follows:

[0053] in: This represents the final target parameter value (ethylene concentration, temperature, or humidity) after fine-tuning. This represents the reference value of the regulation parameter obtained from the maturity-regulation parameter mapping table. This represents the theoretically optimal value for the same maturity level, obtained from the optimal ripening path curve. Represents the decay factor. This represents the maximum single-step adjustment limit preset for this parameter. It is a symbolic function. This is the absolute value operation. When hour, , indicates a downward adjustment; when hour, This indicates an upward adjustment. It's understandable that by introducing a maximum single-step adjustment limit, drastic fluctuations in control parameters caused by excessive deviations in a single query are avoided, resulting in smoother control of the ripening environment.

[0054] Optionally, the input dimension of environmental parameter change trends can be further refined into specific quantitative indicators such as "rate of temperature increase" and "rate of humidity decrease," rather than a simple qualitative description. It can be understood that the three-dimensional structure of the maturity-regulation parameter mapping table allows queries to not only depend on the current maturity level but also consider the regulation status and environmental change trends of the previous cycle. This gives regulation memory and adaptability, avoiding control oscillations that may occur when simply looking up the table based on maturity. In some embodiments, the adjustment rate limit is not a fixed value; it is negatively correlated with the difference between the current comprehensive maturity index and the target maturity index. That is, the larger the difference, the larger the allowable adjustment rate limit, in order to accelerate the regulation response in the early stages of maturity and slow down the adjustment to prevent over-adjustment as the target approaches.

[0055] In one embodiment of the present invention, a data processing center deployed in the "Hayward" kiwifruit ripening warehouse obtains the initial fuzzy evaluation matrix at the 60th hour of the ripening process. The initial fuzzy evaluation matrix includes membership values ​​for the color, hardness, ethylene, and humidity dimensions corresponding to the three linguistic variables "low," "medium," and "high," respectively. The membership values ​​for the color dimension are {0.18, 0.61, 0.21}, the hardness dimension is {0.29, 0.52, 0.19}, the ethylene dimension is {0.12, 0.43, 0.45}, and the humidity dimension is {0.38, 0.49, 0.13}. The data processing center calls a pre-defined base weight vector using the analytic hierarchy process (AHP). The base weight vector has a weight of 0.30 for the color dimension, 0.25 for the hardness dimension, 0.20 for the ethylene dimension, and 0.25 for the humidity dimension. Since the endogenous ethylene generation rate measured at the 60th hour of the ripening process was 0.031 μL / L / min, which is higher than the preset effective detection limit of 0.005 μL / L / min, the endogenous ethylene generation rate was marked as valid. The fuzzy comprehensive evaluation algorithm maintained the operation of the four-dimensional evaluation space and did not trigger adaptive dimensionality reduction. The data processing center read the first correlation factor, which characterizes the correlation strength between the color index and the hardness index, from the preset correlation factor storage area. The current value of the first correlation factor is 0.18. At the same time, it reads the second correlation factor, which characterizes the correlation strength between the ethylene index and the humidity index. The current value of the second correlation factor is 0.12. Based on the current comprehensive maturity index of 54, the data processing center consulted the preset correlation factor-maturity relationship table. The current comprehensive maturity index of 54 falls within the maturity range of 50 to 70, corresponding to the middle stage of ripening. From the correlation factor-maturity relationship table, the color-hardness dynamic weight adjustment coefficient is obtained as 0.75, and the ethylene-humidity dynamic weight adjustment coefficient is obtained as 1.15. Based on the dynamic weight adjustment coefficient, the data processing center performs a weighted adjustment on the basic weights of the color and hardness indicators, which are coupled. The adjustment method is as follows: the basic weight of the color indicator is 0.30 multiplied by the color-hardness dynamic weight adjustment coefficient of 0.75 to obtain an intermediate weight of 0.225; the basic weight of the hardness indicator is 0.25 multiplied by the color-hardness dynamic weight adjustment coefficient of 0.75 to obtain an intermediate weight of 0.1875; the basic weight of the ethylene indicator is 0.20 multiplied by the ethylene-humidity dynamic weight adjustment coefficient of 1.15 to obtain an intermediate weight of 0.23; the basic weight of the humidity indicator is 0.25 multiplied by the ethylene-humidity dynamic weight adjustment coefficient of 1.15 to obtain an intermediate weight of 0.2875.The four intermediate weights were then normalized, and the sum of the intermediate weights was 0.225 + 0.1875 + 0.23 + 0.2875 = 0.93. After normalization, the dynamic effective weights of each effective evaluation index were obtained: the dynamic effective weight of color index was 0.242, the dynamic effective weight of hardness index was 0.202, the dynamic effective weight of ethylene index was 0.247, and the dynamic effective weight of humidity index was 0.309. In the four-dimensional evaluation space, index pairs without coupling relationships were considered as independent contributions. However, all four indices were included in the coupling relationship pairings. That is, the color index and hardness index formed the first coupling pair, and the ethylene index and humidity index formed the second coupling pair. There were no indices outside the coupling pairs. Therefore, the aggregate contribution of all indices was calculated using the coupling contribution method.

[0056] The data processing center constructs an aggregation calculation model based on the coupling relationship between indicators. For the linguistic variable "medium", the first coupling contribution between the color index and the hardness index is calculated. The first coupling contribution is calculated by multiplying the dynamic effective weight of color (0.242) by the membership value of color "medium" (0.61), adding the dynamic effective weight of hardness (0.202) by the membership value of hardness "medium" (0.52), and then multiplying the sum by the first correlation factor (0.18). The calculated value is 0.18 multiplied by (0.242 multiplied by 0.61 plus 0.202 multiplied by 0.52) equals 0.18 multiplied by (0.14762 plus 0.10504) equals 0.18 multiplied by 0.25266 equals 0.04548. The second coupling contribution of the ethylene index and the humidity index is calculated. This second coupling contribution is obtained by multiplying the dynamic effective weight of ethylene (0.247) by the membership value of ethylene "medium" (0.43), plus the dynamic effective weight of humidity (0.309) multiplied by the membership value of humidity "medium" (0.49). The sum is then multiplied by the second correlation factor (0.12). The calculated value is 0.12 multiplied by (0.247 multiplied by 0.43 plus 0.309 multiplied by 0.49) equals 0.12 multiplied by (0.10621 plus 0.15141) equals 0.12 multiplied by 0.25762 equals 0.03091. The aggregated contributions of all indices are summed to obtain the membership value of the "medium" linguistic variable in the comprehensive fuzzy evaluation set, which equals the first coupling contribution (0.04548) plus the second coupling contribution (0.03091), equaling 0.07639. The comprehensive membership degree of the linguistic variables "low" and "high" is calculated in the same way to obtain the comprehensive fuzzy evaluation set. After the centroid method is used for declarative processing, a scalar value between zero and one hundred is output as the current comprehensive maturity index.

[0057] At the 96th hour of the ripening process, the measured endogenous ethylene generation rate was 0.002 μL / L / min, below the effective detection limit. Therefore, the endogenous ethylene generation rate was marked as invalid, and the fuzzy comprehensive evaluation algorithm performed adaptive dimensionality reduction, shrinking the evaluation space from four dimensions to three. After dimensionality reduction, the original base weight of 0.20 belonging to the ethylene dimension was allocated to the remaining three dimensions according to preset rules: the new base weight for color was 0.30 + 0.08 = 0.38; the new base weight for hardness was 0.25 + 0.06 = 0.31; and the new base weight for humidity was 0.25 + 0.06 = 0.31. After normalization, the three-dimensional base weights were: color 0.38, hardness 0.31, and humidity 0.31. At this point, only one coupling pair existed: color-hardness. The first correlation factor was updated to 0.22 with the maturity stage, and the dynamic weight adjustment coefficient for color-hardness, read from the correlation factor-maturity relationship table, was 0.70. The base weights of color and hardness indices were adjusted. The median weight for color was 0.38 x 0.70 = 0.266, and the median weight for hardness was 0.31 x 0.70 = 0.217. The humidity index, which has no coupling relationship, maintained its base weight of 0.31. The sum of the three is 0.266 + 0.217 + 0.31 = 0.793. After normalization, the effective three-dimensional dynamic weights are: color 0.335, hardness 0.274, and humidity 0.391. The three-dimensional membership values ​​collected at the 96th hour were: color {0.10, 0.30, 0.60}, hardness {0.15, 0.35, 0.50}, and humidity {0.40, 0.45, 0.15}. For the linguistic variable "medium", the contribution of the color-hardness coupling is the first correlation factor 0.22 multiplied by (the dynamic effective weight of color 0.335 multiplied by the membership degree of color "medium" 0.30 plus the dynamic effective weight of hardness 0.274 multiplied by the membership degree of hardness "medium" 0.35), which equals 0.22 multiplied by (0.100). 5 plus 0.0959 equals 0.04321. The independent contribution of the humidity index is the dynamic effective weight of humidity 0.391 multiplied by the membership degree of humidity "medium" 0.45, which equals 0.17595. The comprehensive membership degree of the linguistic variable "medium" is 0.04321 plus 0.17595, which equals 0.21916. Compared with the predicted value in the four-dimensional space, it shows a significant change. The current comprehensive maturity index obtained after clarification is updated accordingly, driving the ripening strategy generation unit to output a dynamic ripening control parameter set that matches the maturity status.

[0058] During the reliability verification and data fusion process of the multi-sensor array, the data processing center configured an initial reliability weighting coefficient of 0.92 for the fruit surface color sensor, 0.88 for the fruit body mechanics sensor, 0.90 for the ethylene gas sensor, 0.95 for the temperature and humidity sensor, and 0.85 for the weight sensor. At the 72-hour mark of the ripening process, the instantaneous change rates of the fruit body mechanics sensor readings in three consecutive acquisition cycles were 15%, 22%, and 19% of the previous cycle, respectively. Since the upper limit of the historical normal fluctuation range for the fruit body mechanics sensor is 8%, and the instantaneous change rate continuously exceeded this limit, the data processing center determined that the fruit body mechanics sensor might be experiencing instantaneous interference and temporarily reduced its reliability weighting coefficient from 0.88 to 0.35. In the calculation of converting fruit mechanical response data into fruit firmness curves during the 72nd hour, the force-depth curve data collected by the fruit mechanical sensors were multiplied by the current confidence weighting coefficient of 0.35 before being used to extract fruit firmness feature values. This weighted force value was compressed overall, causing the stress value of the biological yield point extracted from the force-depth curve to be corrected from 14.2 Newtons before weighting to 4.97 Newtons. During the fusion calculation, the influence of this low-weight data on the fruit firmness curve was significantly suppressed, avoiding interference from single-sensor disturbances on the accuracy of the current comprehensive maturity index calculation. In the subsequent six collection cycles, the rate of change of fruit mechanical sensor readings recovered to within 5% and remained stable. The confidence weighting coefficient increased by 0.08 periodically according to the preset recovery strategy, eventually recovering to 0.88.

[0059] During the 120th hour of the ripening process, the reliability weight coefficient of the ethylene gas sensor remained below the preset fault threshold of 0.40 for more than 30 minutes, triggering a sensor fault alarm and sending a notification to the maintenance terminal. During operation before the ethylene gas sensor was repaired, the data processing center used an estimation algorithm based on data from other normal sensors. Utilizing data from temperature and humidity sensors, fruit color sensors, and weight sensors, a multiple regression model was used to generate an estimate of the endogenous ethylene production rate, filling in the missing data from the ethylene gas sensor. This allowed the fuzzy comprehensive evaluation algorithm to continue operating in the four-dimensional evaluation space even with the temporary lack of ethylene-dimensional data. After the ethylene gas sensor was replaced and recalibrated, its reliability weight coefficient was reset to 0.90 and it was re-integrated into the fusion calculation. The estimation algorithm for other sensor data was specifically a multiple linear regression model used to estimate missing readings (such as the endogenous ethylene production rate) when a specific sensor (such as the ethylene gas sensor) malfunctioned. The construction and training steps of this estimation model are as follows: First, the input features of the model were determined. During normal system operation, a large amount of synchronous time-series data was collected, including the rate of change in fruit peel color. Fruit firmness characteristic value Water loss rate Ambient temperature Ambient humidity As an input feature, the corresponding endogenous ethylene generation rate As the target output. Secondly, construct a multiple linear regression model, whose mathematical expression is:

[0060] in, It is the ethylene production rate estimated by the model. It is the intercept term. to These are the model coefficients, corresponding to the weights of each input feature. Then, the model is trained using collected historical data, and the optimal coefficients are calculated using the least squares method. to This makes the model estimate Compared with actual measured value The mean square error between them is minimized. After training, the model's coefficients are fixed and embedded into the fault handling module of the data processing center. When the system determines that the ethylene gas sensor is faulty and its confidence weight coefficient remains below the fault threshold, this estimation process is triggered: the data processing center will use other valid sensor data at the current moment ( , , , , The input is fed into a pre-trained multiple linear regression model, which calculates and outputs an estimated rate of endogenous ethylene production based on the formula. This value is used to temporarily fill in the missing ethylene data for use by the fuzzy comprehensive evaluation algorithm until the faulty sensor is repaired or replaced.

[0061] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A kiwifruit intelligent ripening control system integrated with multi-sensor feedback, characterized in that, The system includes: A multi-sensor array is used to continuously collect multi-dimensional state information of kiwifruit samples in the ripening storage. The multi-sensor array contains distributed sensor nodes to perform multi-point spatial sampling of the stacked fruits in the storage and generate information representing the average state of the overall fruits through statistical fusion. The multi-dimensional state information includes fruit surface color distribution images, fruit mechanical response data, endogenous ethylene production rate, fruit weight change data, and environmental temperature and humidity parameters. The data processing center receives and parses the multi-dimensional state information, converts the fruit surface color distribution image into a peel color change rate, converts the fruit mechanical response data into a fruit firmness curve, converts the fruit weight change data into a water loss rate, and calls an improved fuzzy comprehensive evaluation algorithm to perform fusion analysis and state determination on the peel color change rate, the fruit firmness curve, the endogenous ethylene generation rate, and the water loss rate to generate the current comprehensive maturity index of the kiwifruit. The improved fuzzy comprehensive evaluation algorithm is based on a preset three-dimensional membership function that includes at least color-firmness-humidity, and when the endogenous ethylene generation rate is lower than the effective detection threshold, it adaptively reduces the dimensionality to a three-dimensional evaluation space for calculation. The ripening strategy generation unit is used to query a preset maturity-control parameter mapping table based on the current comprehensive maturity index to generate a dynamic ripening control parameter set including the target ethylene application concentration, target ambient temperature, and target ambient humidity. An environmental control actuator is used to adjust the working status of the ethylene releaser, heating device, and humidification device in the ripening environment according to the dynamic ripening control parameter set.

2. The integrated multi-sensor feedback kiwifruit intelligent ripening control system according to claim 1, characterized in that, The working principle of the improved fuzzy comprehensive evaluation algorithm includes: Construct a three-dimensional evaluation space that includes at least color, hardness, and humidity, and dynamically expand it into a four-dimensional evaluation space of color, hardness, ethylene, and humidity based on the effectiveness of the endogenous ethylene generation rate. Within the current valid evaluation space, define fuzzy subsets for each dimension, with each fuzzy subset containing three linguistic variables: "low", "medium", and "high". Each of the linguistic variables is configured with a membership function based on the shape adjustment of the Sigmoid function. The membership function based on the shape adjustment of the Sigmoid function introduces the inflection point position parameter and the curve steepness parameter, so that the inflection point position is dynamically determined according to the optimal ripening experimental data of the kiwifruit variety, and the curve steepness is adaptively adjusted according to the sensor measurement error range. The specific values ​​of each effective dimension indicator obtained by real-time collection and conversion are input into the membership function based on the Sigmoid function shape adjustment corresponding to each dimension to obtain the membership value of each input value relative to the three linguistic variables "low", "medium" and "high", forming an initial fuzzy evaluation matrix. An improved weighted generalization operator is applied to aggregate the initial fuzzy evaluation matrix. The improved weighted generalization operator introduces a correlation compensation factor between indicators. When the current evaluation space is four-dimensional, it models the nonlinear coupling relationship between color and hardness, and between ethylene and humidity. When the current evaluation space is three-dimensional, it only models the nonlinear coupling relationship between color and hardness to correct the bias of the traditional weighted average. The results of the aggregate calculation are processed to clarify the output and output a scalar value between zero and one hundred. The scalar value is the current comprehensive maturity index.

3. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 2, characterized in that, The method for determining the inter-indicator correlation compensation factor in the improved weighted synthesis operator includes: During the initial calibration phase of the system, multiple rounds of ripening experiments were conducted on a batch of standard kiwifruit samples. The data of the changes in peel color, fruit firmness curve, endogenous ethylene generation rate and water loss rate in each round of experiments were fully recorded. The data of the entire process were processed by principal component analysis to extract the main influencing components, and the first covariance between the peel color change rate and the fruit firmness curve, as well as the second covariance between the endogenous ethylene generation rate and the water loss rate were analyzed. Normalize the first covariance and the second covariance respectively to obtain the first correlation factor characterizing the strength of the color-hardness correlation and the second correlation factor characterizing the strength of the ethylene-humidity correlation; In the aggregate calculation of the improved weighted generalization operator, when calculating the joint contribution of color and hardness indicators, the first correlation factor is introduced to correct the weighted product of the two indicators; when calculating the joint contribution of ethylene and humidity indicators, the second correlation factor is introduced to correct the weighted product of the two indicators. The correlation compensation factor is not a fixed value, but is dynamically selected from a preset correlation factor curve table according to the different maturity stages of the current comprehensive maturity index.

4. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 1, characterized in that, Converting the fruit surface color distribution image into a fruit peel color change rate includes: The color distribution image of the fruit surface is preprocessed to extract the effective area image of the kiwi fruit surface and remove background interference; The effective region image is converted from the red-green-blue color space to the hue-saturation-brightness color space, and the hue component values ​​of all pixels are extracted. Calculate the statistical histogram of the hue component values, identify the hue value corresponding to the main peak in the histogram, and use the hue value corresponding to the main peak as the representative color value of the fruit surface; Obtain the initial fruit surface color distribution image collected when the kiwifruit is put into storage, and calculate the initial representative color value according to the same process; Calculate the absolute difference between the current representative color value and the initial representative color value, and divide the absolute difference by the total monitoring time to obtain the color change per unit time, i.e., the fruit peel color change rate.

5. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 1, characterized in that, The fruit mechanical response data is converted into a fruit firmness curve, including: By using a contact-type micro force sensor and displacement sensor, the indenter is pressed into the kiwi fruit at a constant speed, and the reaction force and indentation depth are recorded in real time during the pressing process to obtain a force-depth curve. The force value corresponding to the biological yield point is extracted from the force-depth curve as the fruit hardness characteristic value; The fruit firmness characteristic values ​​are recorded continuously in chronological order to form the fruit firmness curve.

6. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 1, characterized in that, The endogenous ethylene generation rate is obtained in the following manner: At least one sealable sampling chamber is set up in the ripening chamber. Sample fruits are sealed in the sampling chamber at regular intervals. The growth rate of ethylene concentration in the sealed chamber over time is measured using a high-sensitivity ethylene sensor as the endogenous ethylene generation rate. When the growth rate is lower than the preset effective detection limit, the endogenous ethylene generation rate is marked as invalid, triggering the adaptive dimensionality reduction of the fuzzy comprehensive evaluation algorithm.

7. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 1, characterized in that, Based on the current comprehensive maturity index, a preset maturity-regulation parameter mapping table is queried to generate a dynamic ripening regulation parameter set containing the target ethylene application concentration, target ambient temperature, and target ambient humidity, including: The maturity-regulation parameter mapping table is a three-dimensional lookup table, with its three input dimensions being the maturity index range, the regulation parameter value of the previous regulation cycle, and the trend of environmental parameter changes. The current comprehensive maturity index, the actual control parameter values ​​implemented in the previous control cycle, and the recent changes in the environmental temperature and humidity parameters are used as the joint query key; Perform a matching search in the maturity-regulation parameter mapping table and output the set of regulation parameter reference values ​​that best match the joint query key; The reference value of the control parameter is compared with the preset optimal ripening path curve of the kiwi fruit variety. If the deviation is greater than the threshold, the reference value of the control parameter is fine-tuned according to the optimal ripening path curve to generate the final target ethylene application concentration, target ambient temperature and target ambient humidity. The dynamic ripening regulation parameter set also includes adjustment rate limits for each parameter, which are determined based on the difference between the current comprehensive maturity index and the target maturity index.

8. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 7, characterized in that, The reference value of the control parameter is compared with the preset optimal ripening path curve for a kiwifruit variety. If the deviation is greater than a threshold, the reference value of the control parameter is fine-tuned according to the optimal ripening path curve, including: The optimal ripening path curve defines the theoretically optimal parameter trajectory for achieving the target maturity level, with the maturity index as the horizontal axis and ethylene concentration, temperature, and humidity as the vertical axis. Calculate the absolute deviations of the ethylene concentration, temperature, and humidity in the reference values ​​of the control parameters from the theoretical values ​​of the points corresponding to the same maturity index on the optimal ripening path curve; If the absolute deviation of any of the parameters, such as ethylene concentration, temperature, and humidity, exceeds the preset independent deviation threshold of the corresponding parameter, it is determined that fine-tuning of the control parameter is required. The fine-tuning strategy is as follows: the reference value of the control parameter is adjusted towards the theoretical value, and the adjustment range is the product of the absolute deviation and a decay factor less than one, and the single adjustment range does not exceed the preset maximum single-step adjustment limit. The fine-tuned parameter values ​​are generated as the final target ethylene concentration, target ambient temperature, or target ambient humidity.

9. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 2, characterized in that, The improved weighted synthesis operator is used to perform aggregation calculations on the initial fuzzy evaluation matrix, including: Obtain the initial fuzzy evaluation matrix, which contains the membership values ​​of each evaluation index under the current effective evaluation dimension corresponding to the three linguistic variables of "low", "medium" and "high". The basic weight vector of each effective evaluation index is determined. The basic weight vector is pre-set by the analytic hierarchy process based on the prior influence of each index on the maturity of kiwifruit. When the evaluation space is reduced from four dimensions to three dimensions, the weight of the ethylene dimension is allocated to the other three dimensions according to the preset rules. When the endogenous ethylene generation rate is effective, obtain the first correlation factor characterizing the correlation strength between the color index and the hardness index, and the second correlation factor characterizing the correlation strength between the ethylene index and the humidity index. The current maturity stage of the kiwifruit is determined based on the current comprehensive maturity index, and the dynamic weight adjustment coefficient of the required correlation factors in the current evaluation space is read from the preset correlation factor-maturity relationship table according to the current maturity stage. Based on the dynamic weight adjustment coefficient, the basic weights of each group of indicators with coupling relationship, as well as the basic weights of the ethylene index and the humidity index, are weighted and adjusted respectively to obtain the dynamic effective weights of each effective evaluation index. Based on the coupling relationship between indicators, an aggregation calculation model is constructed. For color and hardness indicators that have a coupling relationship, their aggregation contribution is calculated by the dynamic effective weight, the membership value, and the first correlation factor. For ethylene and humidity indicators that have a coupling relationship only in the four-dimensional evaluation space, their aggregation contribution is calculated by the dynamic effective weight, the membership value, and the second correlation factor. For indicator pairs without a coupling relationship, their aggregation contribution is calculated by the product of their respective dynamic effective weights and membership values. The aggregated contributions of all indicators are summed to obtain the comprehensive fuzzy evaluation set after aggregate calculation.

10. The kiwifruit intelligent ripening control system integrating multi-sensor feedback according to claim 1, characterized in that, The system also includes a process of verifying the reliability of readings from the multi-sensor array and fusing the data. A confidence weighting coefficient based on the stability of historical readings is configured for each of the fruit surface color sensor, fruit body mechanical sensor, fruit body elasticity sensor, ethylene gas sensor, temperature and humidity sensor, and weight sensor in the multi-sensor array. The rate of change of each sensor reading is monitored in real time. If the instantaneous rate of change of a sensor reading exceeds its normal fluctuation range, its confidence weight coefficient is temporarily reduced. When performing data processing and fusion calculations, the data provided by each sensor is multiplied by its current confidence weight coefficient before being used in the calculation; When the confidence weight coefficient of a certain sensor is continuously lower than the fault threshold, a sensor fault alarm is triggered, and an estimation algorithm based on data from other sensors is attempted to fill in the missing data of the corresponding sensor.

Images