Multi-feature fusion Hami melon quality nondestructive testing method and system

Abstract

The invention provides a multi-feature fusion Hami melon quality nondestructive testing method and system, and the method comprises the steps: obtaining quality characterization parameters of different regions in a Hami melon sampling period and environment coupling parameters in the sampling period, and dividing quality discrimination standards; carrying out comprehensive processing on the quality characterization parameters and the environment coupling parameters, and obtaining an input set of quality evaluation models of different regions; the quality characterization parameters and the environment coupling parameters are fused, cantaloupe quality evaluation models of different areas are constructed respectively, and credible deviation boundaries of cantaloupe quality evaluation model results of different areas are calculated; and splicing the quality characterization parameters and the environment coupling parameters of the different regions, constructing a comprehensive quality prediction model of the Hami melons, and calculating a credible deviation limit of a comprehensive quality evaluation model result. According to the invention, through a multi-mode feature fusion detection technology, regional and overall comprehensive evaluation of the quality of the Hami melons is realized, and the accuracy and efficiency of nondestructive detection of the quality of the Hami melons are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of nondestructive detection of agricultural product quality, and in particular to a multi-feature fusion Hami melon quality nondestructive detection method and system. Background Art

[0002] Traditional Hami melon quality testing mainly relies on sensory, physical and chemical methods, which are not only inefficient, but also introduce additional damage and cannot fully evaluate the overall quality of Hami melon. At present, non-destructive testing technology has become a key means of intelligent quality evaluation. The use of physical, chemical or biological methods to test the quality of agricultural products has been widely used in quality grading, quality control, intelligent traceability, etc., which can achieve rapid and accurate quality evaluation and provide consumers with safer and higher-quality products.

[0003] At present, the non-destructive detection of Hami melon quality is mainly based on spectroscopy, electromagnetic waves, sound waves, machine vision and other technologies to obtain the shape, size, color, texture, moisture, sugar content, hardness and other quality indicators of Hami melon, and then extract the characteristics of Hami melon and model it. However, the existing feature extraction of Hami melon quality is relatively simple, lacking multi-feature fusion, and the grading accuracy of Hami melon and the credible deviation limit of model results need to be further improved. With the development of artificial intelligence and machine learning technology, multimodal data fusion intelligent modeling technology is urgently needed to improve the accuracy and efficiency of Hami melon quality detection, thereby improving the quality and safety of Hami melon. Therefore, it is very necessary to design a Hami melon quality non-destructive detection method and system with multi-feature fusion. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the object of the present invention is to provide a multi-feature fusion Hami melon quality non-destructive detection method and system.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] The present invention provides a multi-feature fusion Hami melon quality non-destructive detection method, comprising:

[0007] Hami melons are divided into regions, quality characterization parameters of Hami melons in different regions during the sampling period and environmental coupling parameters during the sampling period are obtained, and quality judgment standards are divided;

[0008] Comprehensively process the quality characterization parameters and environmental coupling parameters of Hami melons in different regions to obtain the input set of the Hami melon quality assessment model in different regions;

[0009] The quality characterization parameters and environmental coupling parameters of Hami melon were integrated to build Hami melon quality assessment models in different regions, and the credible deviation limits of the Hami melon quality assessment model results in different regions were calculated.

[0010] The quality characterization parameters and environmental coupling parameters of Hami melon in different regions were spliced ​​together, and a comprehensive quality prediction model of Hami melon was constructed based on them, and the credible deviation limit of the results of the comprehensive quality evaluation model of Hami melon was calculated;

[0011] The model was graded and corrected by determining the credible deviation limits of the Hami melon quality assessment model results in different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results.

[0012] Preferably, Hami melons are divided into regions to obtain quality characterization parameters of Hami melons in different regions within a sampling period and environmental coupling parameters within the sampling period, specifically:

[0013] From left to right, 1 / 4 of the cantaloupe is the first area, 1 / 4-3 / 4 is the second area, and the remaining 1 / 4 is the third area;

[0014] In each area, data from 6 points are collected at the same time step, and the mean is calculated, which is regarded as the data of one sampling time step in one area;

[0015] The quality characterization parameters are divided into time series data and image data, including appearance texture, sugar content, acidity, moisture content, pulp hardness, electrical conductivity, dielectric constant, sound wave data, spectral data, and visual images;

[0016] The environmental coupling parameters are sampled in one group in each area, including temperature and humidity, gas, vibration, and noise.

[0017] Preferably, the quality judgment criteria are divided into:

[0018] Obtaining standard reference thresholds for Hami melon quality characterization parameters;

[0019] Obtain the correlation between the standard reference data of Hami melon quality characterization parameters i :

[0020]

[0021] Where, X i and Y i They represent the standard reference data of different variables of Hami melon quality characterization parameters. and represents the average value of Hami melon standard reference data, i represents the quality parameter variable, r i ∈[-1,1];

[0022] The weights of Hami melon quality characterization parameters are obtained based on the degree of association:

[0023]

[0024] In the formula, θ i is the weight of Hami melon quality parameters, θ i ∈[0,1],|r i | represents the correlation degree of quality parameter variable i, σ i is the standard deviation of the preset quality parameter standard data;

[0025] Calculate the evaluation index of the comprehensive quality judgment standard of Hami melon, and obtain the quality grade of Hami melon according to the quality evaluation index, where the evaluation index of the comprehensive quality judgment standard is:

[0026]

[0027] In the formula, Q i Indicates the overall quality grade of Hami melon, Q i ∈(0,1],θ i Indicates the weight of Hami melon quality characterization parameters, τ i Indicates the reference quality index for each quality characterization parameter.

[0028] Preferably, the quality characterization parameters and environmental coupling parameters of different regions of Hami melon are comprehensively processed, specifically:

[0029] According to the fluctuation of the quality characterization parameters and environmental coupling parameters of each region before and after the data, the discrete coefficient CV of the quality characterization parameters and environmental coupling parameters of each region is obtained. i , calculated as:

[0030]

[0031] In the formula, x i represents the data of parameter i, x i-1 , x i+1 Respectively represent the parameters x i Forward and backward data, represents the mean value of the parameter, σ i represents the standard deviation of parameter i;

[0032] The quality characterization parameters and environmental coupling parameters of Hami melon in different regions were comprehensively processed based on the discrete coefficient.

[0033] Preferably, the quality characterization parameters and environmental coupling parameters of Hami melon are integrated to respectively construct Hami melon quality assessment models for different regions, and the credible deviation limits of the Hami melon quality assessment model results for different regions are calculated, specifically including:

[0034] Construct Hami melon quality assessment models for the first, second, and third regions;

[0035] Training the Hami melon quality assessment model for the first region, the second region, and the third region based on the input set;

[0036] Based on the trained Hami melon quality assessment model, the quality of the first area, second area and third area of ​​Hami melon was identified to obtain the quality discrimination index. The quality grades of the first area, second area and third area of ​​Hami melon were obtained according to the quality discrimination index, and the credible deviation limits of the quality assessment model were calculated respectively.

[0037] Preferably, the quality characterization parameters and environmental coupling parameters of Hami melons in different regions are spliced, and a comprehensive quality prediction model of Hami melons is constructed based on the spliced ​​quality characterization parameters and environmental coupling parameters of Hami melons in different regions, and the credible deviation limit of the results of the Hami melon comprehensive quality evaluation model is calculated, specifically including:

[0038] The dot matrix data of the first area, the second area, and the third area are spliced, and the mean of the dot matrix data is used as the comprehensive quality data input set of Hami melon;

[0039] Construct a comprehensive quality prediction model for Hami melon;

[0040] Training a comprehensive quality prediction model based on a comprehensive quality data input set;

[0041] The quality of Hami melon is identified based on the trained Hami melon comprehensive quality prediction model to obtain the quality discrimination index. The comprehensive quality grade and credible deviation limit of Hami melon are obtained according to the quality discrimination index.

[0042] Preferably, the Hami melon quality assessment model of the first region, the second region and the third region is constructed in the same manner as the Hami melon comprehensive quality prediction model, which is:

[0043] Construct two branch networks, one for processing time series data and the other for processing image data;

[0044] The feature vectors of the two branch networks are extracted respectively, and the feature cascade fusion is performed through the attention mechanism. The feature extraction of time series data and image data is as follows:

[0045]

[0046] In the formula, Represents the features extracted from time series data, represents the kth feature map of the l-1th layer in the network structure, They represent the convolution kernel and bias term connected to the k-th input feature map and the m-th feature map of the l-th layer in the temporal branch network, respectively. represents the features extracted from image data, They represent the convolution kernel and bias term connected to the k-th input feature map and the m-th feature map of the l-th layer in the image branch network, respectively. ReLU represents the activation function. d,g Represents feature output, and concat is a concatenation operation;

[0047] The feature fusion of time series and image data is performed through the attention mechanism, as follows:

[0048] F out =Attention(F d,g )+F d,g

[0049] The output layer of the Hami melon quality index obtained based on the fusion features is:

[0050] Y out =softmax((w y F out +b y ))

[0051] Based on the model output probability, the credible deviation limit of the model output is calculated as:

[0052] θ=[αloss-|e 上 |,αloss+|e 下 |]

[0053] In the formula, Attention is the attention mechanism operation, F out Represents the fusion feature output, where w y is the weight of the output layer, b y is the bias term of the output layer, Y out is the quality index output result. loss represents the cross entropy loss function of the model, e 上 and e 下 Represents the preset maximum and minimum errors, θ∈[0,1], α represents the error correction factor α∈(0,1].

[0054] Preferably, the model is graded and corrected by determining the credible deviation limits of the Hami melon quality assessment model results in different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results, specifically:

[0055] For Hami melon quality assessment models in different regions and Hami melon comprehensive quality assessment models, if the credible deviation limit of the model output is ≤5%, the error correction factor α=1;

[0056] If the credible deviation limit of the model output is >5%, the model is corrected by the error correction factor, α∈(0,1);

[0057] Among them, the comprehensive quality judgment result of Hami melon is consistent with the quality grades of the first area, the second area, and the third area, or is consistent with the judgment result of the lowest quality grade of the first area, the second area, and the third area.

[0058] The present invention also provides a multi-feature fusion Hami melon quality non-destructive detection system, comprising:

[0059] The acquisition unit is used to obtain the quality characterization parameters of Hami melon in different areas during the sampling period and the environmental coupling parameters during the sampling period:

[0060] A data comprehensive processing unit, used for comprehensive processing of quality characterization parameters and environmental coupling parameters of Hami melons in different regions;

[0061] The first model building unit is used to build Hami melon quality assessment models in different regions;

[0062] The second model building unit is used for the comprehensive quality evaluation model of Hami melon.

[0063] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0064] The present invention provides a Hami melon quality nondestructive detection method and system with multi-feature fusion, the method comprising obtaining quality characterization parameters of different regions in a Hami melon sampling period and environmental coupling parameters in the sampling period, dividing quality discrimination standards; comprehensively processing the quality characterization parameters and environmental coupling parameters, obtaining input sets of quality assessment models of different regions; fusing the quality characterization parameters and environmental coupling parameters, respectively constructing Hami melon quality assessment models of different regions, and calculating the credible deviation limits of Hami melon quality assessment model results of different regions; splicing the quality characterization parameters and environmental coupling parameters of different regions, constructing a comprehensive quality prediction model of Hami melon, and calculating the credible deviation limits of the comprehensive quality assessment model results, and hierarchically correcting the model by discriminating the credible deviation limits of the Hami melon quality assessment model results of different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results. The system comprises: an acquisition unit, a data comprehensive processing unit, a first model construction unit, and a second model construction unit. The present invention realizes regional and overall comprehensive evaluation of Hami melon quality through multi-mode feature fusion detection technology, which helps to improve the accuracy and efficiency of non-destructive detection of Hami melon quality, thereby improving the quality and safety of Hami melon. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0066] Figure 1 A schematic diagram of a flow chart of a Hami melon quality nondestructive detection method using multi-feature fusion provided by an embodiment of the present invention;

[0067] Figure 2 A schematic diagram of the Hami melon region division in the embodiment provided by the present invention;

[0068] Figure 3 A schematic diagram of the structure of a Hami melon quality nondestructive detection system with multi-feature fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] The purpose of the present invention is to provide a multi-feature fusion Hami melon quality non-destructive detection method, which realizes regional and overall comprehensive evaluation of Hami melon quality through multi-mode feature fusion detection technology, helps to improve the accuracy and efficiency of Hami melon quality non-destructive detection, and thus improves the quality and safety of Hami melon.

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a multi-feature fusion Hami melon quality non-destructive detection method, comprising:

[0073] Step 100: dividing Hami melon into regions, obtaining quality characterization parameters of Hami melon in different regions within a sampling period and environmental coupling parameters within the sampling period, and dividing quality discrimination standards;

[0074] Step 200: Comprehensively process the quality characterization parameters and environmental coupling parameters of Hami melons in different regions to obtain an input set of Hami melon quality evaluation models in different regions;

[0075] Step 300: fusing the Hami melon quality characterization parameters and the environmental coupling parameters, respectively constructing Hami melon quality assessment models for different regions, and calculating the credible deviation limits of the Hami melon quality assessment model results for different regions;

[0076] Step 400: splicing the quality characterization parameters and environmental coupling parameters of Hami melons in different regions, building a comprehensive quality prediction model for Hami melons based on the quality characterization parameters and environmental coupling parameters, and calculating the credible deviation limit of the Hami melon comprehensive quality evaluation model result;

[0077] Step 500: The model is graded and corrected by determining the credible deviation limits of the Hami melon quality assessment model results in different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results.

[0078] In step 100, Hami melons are divided into regions to obtain quality characterization parameters of Hami melons in different regions within a sampling period and environmental coupling parameters within the sampling period, specifically:

[0079] From left to right, 1 / 4 of the cantaloupe is the first area, 1 / 4-3 / 4 is the second area, and the remaining 1 / 4 is the third area. The specific division diagram is as follows Figure 2 As shown;

[0080] In each area, data from 6 points are collected at the same time step, and the mean is calculated, which is regarded as the data of one sampling time step in one area;

[0081] The method for solving the mean value is as follows: remove the maximum and minimum values ​​from the six point matrix data obtained at the same time step in each area, and then average them, and use the mean result as the data of one sampling time step in one area;

[0082] The quality characterization parameters are divided into time series data and image data, including but not limited to appearance texture, sugar content, acidity, moisture content, pulp hardness, electrical conductivity, dielectric constant, sound wave data, spectral data, visual images, etc.;

[0083] The environmental coupling parameters are sampled in one group in each area, including temperature and humidity, gas, vibration, and noise;

[0084] It should be noted that the present invention provides an embodiment in which Hami melon timing and image data acquisition conditions are set to two groups, one group of temperature is 0-5°C, humidity is 75%-85%, and the other group of temperature is 5-10°C, humidity is 80%-90%.

[0085] In step 100, the quality judgment criteria are divided into:

[0086] Obtaining standard reference thresholds for Hami melon quality characterization parameters;

[0087] Obtain the correlation between the standard reference data of Hami melon quality characterization parameters i :

[0088]

[0089] Where, X i and Y i They represent the standard reference data of different variables of Hami melon quality characterization parameters. and represents the average value of Hami melon standard reference data, i represents the quality parameter variable, r i ∈[-1,1];

[0090] The weights of Hami melon quality characterization parameters are obtained based on the degree of association:

[0091]

[0092] In the formula, θ i is the weight of Hami melon quality parameters, θ i ∈[0,1],|r i | represents the correlation degree of quality parameter variable i, σ i is the standard deviation of the preset quality parameter standard data;

[0093] Calculate the evaluation index of the comprehensive quality judgment standard of Hami melon, and obtain the quality grade of Hami melon according to the quality evaluation index, where the evaluation index of the comprehensive quality judgment standard is:

[0094]

[0095] In the formula, Q i Indicates the overall quality grade of Hami melon, Q i ∈(0,1],θ i Indicates the weight of Hami melon quality characterization parameters, τ i A reference quality index representing each quality characterization parameter;

[0096] Determine the standard reference thresholds of Hami melon quality characterization parameters based on national standards, industry standards, group standards, enterprise standards, etc., and determine the reference quality index of each quality characterization parameter based on the sensory reference thresholds, thereby classifying the quality grades of Hami melons;

[0097] The present invention provides an embodiment in which the quality grade of Hami melon is divided into five grades: A+ grade, A grade, B+ grade, B grade and C grade, specifically:

[0098] If Q i ∈[0.9,1], the quality grade of Hami melon is A+;

[0099] If Q i ∈[0.8,0.9), the quality grade of Hami melon is judged as A;

[0100] If Q i ∈[0.7,0.8), the quality grade of Hami melon is B+;

[0101] If Q i ∈[0.5,0.7), the quality grade of Hami melon is B;

[0102] If Q i ∈(0,0.5), the quality grade of Hami melon is judged to be C.

[0103] In step 200, the quality characterization parameters and environmental coupling parameters of different regions of Hami melon are comprehensively processed, specifically:

[0104] According to the fluctuation of the quality characterization parameters and environmental coupling parameters of each region before and after the data, the discrete coefficient CV of the quality characterization parameters and environmental coupling parameters of each region is obtained. i , calculated as:

[0105]

[0106] In the formula, x i represents the data of parameter i, x i-1 , x i+1 Respectively represent the parameters x i Forward and backward data, represents the mean value of the parameter, σ i represents the standard deviation of parameter i;

[0107] Based on the discrete coefficient, the quality characterization parameters and environmental coupling parameters of Hami melon in different regions were comprehensively processed;

[0108] The present invention provides an embodiment, in which the discrete fluctuation range of the preset quality characterization parameter and the environmental coupling parameter reference data is set to [0.1, 0.6]. If the discrete coefficient of the quality characterization parameter and the environmental coupling parameter of each area is greater than 0.6, this point is judged as a discrete point. If the proportion of discrete points is greater than 10%, interpolation correction is required.

[0109] In step 300, the quality characterization parameters of Hami melon and the environmental coupling parameters are integrated to respectively construct Hami melon quality assessment models for different regions, and the credible deviation limits of the Hami melon quality assessment model results for different regions are calculated, specifically including:

[0110] Construct Hami melon quality assessment models for the first, second, and third regions;

[0111] Training the Hami melon quality assessment model for the first region, the second region, and the third region based on the input set;

[0112] Based on the trained Hami melon quality assessment model, the quality of the first area, second area and third area of ​​Hami melon was identified to obtain the quality discrimination index. The quality grades of the first area, second area and third area of ​​Hami melon were obtained according to the quality discrimination index, and the credible deviation limits of the quality assessment model were calculated respectively.

[0113] In step 400, the quality characterization parameters and environmental coupling parameters of Hami melons in different regions are spliced, and a comprehensive quality prediction model of Hami melons is constructed based on the quality characterization parameters and environmental coupling parameters of Hami melons in different regions. The credible deviation limit of the result of the comprehensive quality evaluation model of Hami melons is calculated, which specifically includes:

[0114] The dot matrix data of the first area, the second area, and the third area are spliced, and the mean of the dot matrix data is used as the comprehensive quality data input set of Hami melon;

[0115] Construct a comprehensive quality prediction model for Hami melon;

[0116] Training a comprehensive quality prediction model based on a comprehensive quality data input set;

[0117] The quality of Hami melon is identified based on the trained Hami melon comprehensive quality prediction model to obtain the quality discrimination index. The comprehensive quality grade and credible deviation limit of Hami melon are obtained according to the quality discrimination index.

[0118] The Hami melon quality assessment model for the first region, the second region and the third region is constructed in the same manner as the Hami melon comprehensive quality prediction model, which is:

[0119] Construct two branch networks, one for processing time series data and the other for processing image data;

[0120] The feature vectors of the two branch networks are extracted respectively, and the feature cascade fusion is performed through the attention mechanism. The feature extraction of time series data and image data is as follows:

[0121]

[0122]

[0123] In the formula, Represents the features extracted from time series data, represents the kth feature map of the l-1th layer in the network structure, They represent the convolution kernel and bias term connected to the k-th input feature map and the m-th feature map of the l-th layer in the temporal branch network, respectively. represents the features extracted from image data, They represent the convolution kernel and bias term connected to the k-th input feature map and the m-th feature map of the l-th layer in the image branch network, respectively. ReLU represents the activation function. d,g Represents feature output, and concat is a concatenation operation;

[0124] The feature fusion of time series and image data is performed through the attention mechanism, as follows:

[0125] F out =Attention(F d,g )+F d,g

[0126] The output layer of the Hami melon quality index obtained based on the fusion features is:

[0127] Y out =softmax((w y F out +b y ))

[0128] Based on the model output probability, the credible deviation limit of the model output is calculated as:

[0129] θ=[αloss-|e 上 |,αloss+|e 下 |]

[0130] In the formula, Attention is the attention mechanism operation, F out Represents the fusion feature output, where w y is the weight of the output layer, b y is the bias term of the output layer, Y out is the quality index output result. loss represents the cross entropy loss function of the model, e 上 and e 下 represents the preset maximum and minimum errors, and α represents the error correction factor.

[0131] In step 500, the model is graded and corrected by determining the credible deviation limits of the Hami melon quality assessment model results in different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results, specifically:

[0132] For Hami melon quality assessment models in different regions and Hami melon comprehensive quality assessment models, if the credible deviation limit of the model output is ≤5%, the error correction factor α=1;

[0133] If the credible deviation limit of the model output is >5%, the model is corrected by the error correction factor, α∈(0,1);

[0134] Among them, the comprehensive quality judgment result of Hami melon is consistent with the quality grades of the first area, the second area, and the third area, or is consistent with the judgment result of the lowest quality grade of the first area, the second area, and the third area.

[0135] like Figure 3 As shown, the present invention also provides a multi-feature fusion Hami melon quality non-destructive detection system, comprising:

[0136] The acquisition unit is used to obtain the quality characterization parameters of Hami melon in different areas during the sampling period and the environmental coupling parameters during the sampling period:

[0137] A data comprehensive processing unit, used for comprehensive processing of quality characterization parameters and environmental coupling parameters of Hami melons in different regions;

[0138] The first model building unit is used to build Hami melon quality assessment models in different regions;

[0139] The second model building unit is used for the comprehensive quality evaluation model of Hami melon.

[0140] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0141] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A multi-feature fusion nondestructive detection method for Hami melon quality, characterized in that: include: Hami melons are divided into regions, the quality characterization parameters of Hami melons in different regions during the sampling period and the environmental coupling parameters during the sampling period are obtained, and the quality judgment standards are divided; The quality characterization parameters and environmental coupling parameters of Hami melons in different regions are processed comprehensively to obtain the input set of the Hami melon quality assessment model in different regions; The quality characterization parameters and environmental coupling parameters of Hami melon were integrated to build Hami melon quality assessment models in different regions, and the credible deviation limits of the Hami melon quality assessment model results in different regions were calculated. The quality characterization parameters and environmental coupling parameters of Hami melon in different regions were spliced ​​together, and a comprehensive quality prediction model of Hami melon was constructed based on them, and the credible deviation limit of the results of the comprehensive quality evaluation model of Hami melon was calculated; The model was graded and corrected by determining the credible deviation limits of the Hami melon quality assessment model results in different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results.

2. The method according to claim 1, characterized in that: Hami melons were divided into regions to obtain the quality characterization parameters of Hami melons in different regions during the sampling period and the environmental coupling parameters during the sampling period, specifically: From left to right, 1 / 4 of the cantaloupe is the first area, 1 / 4-3 / 4 is the second area, and the remaining 1 / 4 is the third area; In each area, data from 6 points are collected at the same time step, and the mean is calculated, which is regarded as the data of one sampling time step in one area; The quality characterization parameters are divided into time series data and image data, including appearance texture, sugar content, acidity, moisture content, pulp hardness, electrical conductivity, dielectric constant, sound wave data, spectral data, and visual images; The environmental coupling parameters are sampled in one group in each area, including temperature and humidity, gas, vibration, and noise.

3. The method according to claim 1, characterized in that: The quality judgment criteria are divided into: Obtaining standard reference thresholds for Hami melon quality characterization parameters; Obtain the correlation between the standard reference data of Hami melon quality characterization parameters i : Where, X i and Y i They represent the standard reference data of different variables of Hami melon quality characterization parameters. and represents the average value of Hami melon standard reference data, i represents the quality parameter variable, r i ∈[-1,1]; The weights of Hami melon quality characterization parameters are obtained based on the degree of association: In the formula, is the weight of Hami melon quality parameters, |r i | represents the correlation degree of quality parameter variable i, σ i is the standard deviation of the preset quality parameter standard data; Calculate the evaluation index of the comprehensive quality judgment standard of Hami melon, and obtain the quality grade of Hami melon according to the quality evaluation index, where the evaluation index of the comprehensive quality judgment standard is: In the formula, Q i Indicates the overall quality grade of Hami melon, Q i ∈(0,1], represents the weight of Hami melon quality characterization parameters, τ i Indicates the reference quality index for each quality characterization parameter.

4. The method according to claim 1, characterized in that The quality characterization parameters and environmental coupling parameters of Hami melon in different regions were comprehensively processed, specifically: According to the fluctuation of the quality characterization parameters and environmental coupling parameters of each region before and after the data, the discrete coefficient CV of the quality characterization parameters and environmental coupling parameters of each region is obtained. i , calculated as: In the formula, x i represents the data of parameter i, x i-1 , x i+1 Represents the parameter x i Forward and backward data, represents the mean value of the parameter, σ i represents the standard deviation of parameter i; The quality characterization parameters and environmental coupling parameters of Hami melon in different regions were comprehensively processed based on the discrete coefficient.

5. The method according to claim 1, characterized in that The quality characterization parameters and environmental coupling parameters of Hami melon are integrated to build Hami melon quality assessment models for different regions, and the credible deviation limits of the Hami melon quality assessment model results for different regions are calculated, including: Construct Hami melon quality assessment models for the first, second, and third regions; Training the Hami melon quality assessment model for the first region, the second region, and the third region based on the input set; Based on the trained Hami melon quality assessment model, the quality of the first area, second area and third area of ​​Hami melon was identified to obtain the quality discrimination index. The quality grades of the first area, second area and third area of ​​Hami melon were obtained according to the quality discrimination index, and the credible deviation limits of the quality assessment model were calculated respectively.

6. The method according to claim 5, characterized in that The quality characterization parameters and environmental coupling parameters of Hami melons in different regions were spliced ​​together, and a comprehensive quality prediction model of Hami melons was constructed based on them. The credible deviation limits of the results of the Hami melon comprehensive quality assessment model were calculated, including: The dot matrix data of the first area, the second area, and the third area are spliced, and the mean of the dot matrix data is used as the comprehensive quality data input set of Hami melon; Construct a comprehensive quality prediction model for Hami melon; Training a comprehensive quality prediction model based on a comprehensive quality data input set; The quality of Hami melon is identified based on the trained Hami melon comprehensive quality prediction model to obtain the quality discrimination index. The comprehensive quality grade and credible deviation limit of Hami melon are obtained according to the quality discrimination index.

7. The method according to claim 6, characterized in that The Hami melon quality assessment model for the first region, the second region and the third region is constructed in the same manner as the Hami melon comprehensive quality prediction model, which is: Construct two branch networks, one for processing time series data and the other for processing image data; The feature vectors of the two branch networks are extracted respectively, and the feature cascade fusion is performed through the attention mechanism. The feature extraction of time series data and image data is as follows: In the formula, Represents the features extracted from time series data, represents the kth feature map of the l-1th layer in the network structure, They represent the convolution kernel and bias term connected to the k-th input feature map and the m-th feature map of the l-th layer in the temporal branch network, respectively. represents the features extracted from image data, They represent the convolution kernel and bias term connected to the k-th input feature map and the m-th feature map of the l-th layer in the image branch network, respectively. ReLU represents the activation function. d,g Represents feature output, and concat is a concatenation operation; The feature fusion of time series and image data is performed through the attention mechanism, as follows: F out =Attention(F d,g )+F d,g The output layer of the Hami melon quality index obtained based on the fusion features is: Y out =softmax((w y F out +b y )) Based on the model output probability, the credible deviation limit of the model output is calculated as: θ=[αloss-|e 上 |,αloss+|e 下 |] In the formula, Attention is the attention mechanism operation, F out Represents the fusion feature output, where w y is the weight of the output layer, b y is the bias term of the output layer, Y out is the quality index output result. loss represents the cross entropy loss function of the model, e 上 and e 下 Represents the preset maximum and minimum errors, θ∈[0,1], α represents the error correction factor α∈(0,1].

8. The method according to claim 7, characterized in that The model was graded and corrected by determining the credible deviation limits of the Hami melon quality assessment model results in different regions and the credible deviation limits of the Hami melon comprehensive quality assessment model results, specifically: For Hami melon quality assessment models in different regions and Hami melon comprehensive quality assessment models, if the credible deviation limit of the model output is ≤5%, the error correction factor α=1; If the credible deviation limit of the model output is >5%, the model is corrected by the error correction factor, α∈(0,1); Among them, the comprehensive quality judgment result of Hami melon is consistent with the quality grades of the first area, the second area, and the third area, or is consistent with the judgment result of the lowest quality grade of the first area, the second area, and the third area.

9. A multi-feature fusion Hami melon quality non-destructive detection system, characterized in that: include: The acquisition unit is used to obtain the quality characterization parameters of Hami melon in different areas during the sampling period and the environmental coupling parameters during the sampling period: A data comprehensive processing unit, used for comprehensive processing of quality characterization parameters and environmental coupling parameters of Hami melons in different regions; The first model building unit is used to build Hami melon quality assessment models in different regions; The second model building unit is used for the comprehensive quality evaluation model of Hami melon.

Images