A multi-dimensional analysis method and system for the maturity status of loquats based on non-destructive testing

Abstract

The present invention provides a multi-dimensional loquat maturity analysis method and system based on non-destructive testing, which is applied to the field of non-destructive testing technology. The method includes: collecting image information of the target loquat product; obtaining a loquat hyperspectral image based on a hyperspectral detection device; constructing a multi-dimensional loquat maturity prediction model, which includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch; obtaining a first maturity prediction result and a second maturity prediction result, and inputting them into the maturity calculation branch for weighted calculation to obtain a maturity prediction result; collecting the number of days of full bloom of the product, and combining the maturity prediction result to input into a ripening analysis model to obtain a ripening analysis result for ripening the target loquat product. The technical problems in the prior art that the detection accuracy of the loquat maturity detection method is relatively low, and the subsequent fruit ripening treatment has poor pertinence, resulting in poor treatment effect of the loquat fruit ripening treatment are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of nondestructive testing, and particularly relates to a multi-dimensional loquat maturity analysis method and system based on nondestructive testing. Background Art

[0002] Nondestructive testing refers to the inspection of an object to be inspected without damaging the object to be detected. In the fruit industry, nondestructive testing is often used for the maturity detection of fruit products. However, in the prior art, the maturity detection of fruit products mostly uses the distinction of fruit appearance images to achieve maturity detection. However, the use of only image distinction for the maturity detection of loquats has poor detection effects, resulting in low detection accuracy and poor pertinence for subsequent fruit ripening treatment, and further leading to poor fruit treatment effects.

[0003] Therefore, in the prior art, the detection method for the maturity of loquats has low detection accuracy and poor pertinence for subsequent fruit ripening treatment, resulting in poor treatment effects for loquat fruit ripening treatment. Summary of the Invention

[0004] The present application provides a multi-dimensional loquat maturity analysis method and system based on nondestructive testing, which is used to solve the technical problems in the prior art that the detection method for the maturity of loquats has low detection accuracy and poor pertinence for subsequent fruit ripening treatment, resulting in poor treatment effects for loquat fruit ripening treatment.

[0005] In view of the above problems, the present application provides a multi-dimensional loquat maturity analysis method and system based on nondestructive testing.

[0006] In the first aspect of the present application, a multi-dimensional loquat maturity analysis method based on nondestructive testing is provided. The method includes: collecting image information of a target loquat product; performing hyperspectral image collection on the target loquat product based on a hyperspectral detection device to obtain a hyperspectral image; constructing a multi-dimensional loquat maturity prediction model, where the multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch; inputting the image information into the image prediction module to obtain a first maturity prediction result; inputting the hyperspectral image into the hyperspectral prediction module to obtain a second maturity prediction result; inputting the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain a maturity prediction result; collecting the number of days of full bloom of the target loquat product, and inputting it together with the maturity prediction result into a ripening analysis model to obtain a ripening analysis result, and using the ripening analysis result to ripen the target loquat product.

[0007] The second aspect of the present application provides a multi-dimensional loquat maturity analysis system based on non-destructive testing. The system includes: an image information acquisition module for acquiring image information of a target loquat product; a hyperspectral image acquisition module for acquiring hyperspectral images of the target loquat product based on a hyperspectral detection device; a maturity prediction model acquisition module for constructing a multi-dimensional loquat maturity prediction model, where the multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch; a first maturity prediction result acquisition module for inputting the image information into the image prediction module to obtain a first maturity prediction result; a second maturity prediction result acquisition module for inputting the hyperspectral image into the hyperspectral prediction module to obtain a second maturity prediction result; a maturity prediction result acquisition module for inputting the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain a maturity prediction result; a ripening analysis result acquisition module for collecting the number of days of full bloom of the target loquat product, inputting the maturity prediction result into a ripening analysis model in combination to obtain a ripening analysis result, and using the ripening analysis result to ripen the target loquat product.

[0008] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0009] The method provided in the embodiment of the present application acquires image information of a target loquat product. Based on a hyperspectral detection device, a loquat hyperspectral image is obtained. A multi-dimensional loquat maturity prediction model is constructed, and the multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch. The image information is input into the image prediction module to obtain a first maturity prediction result. The hyperspectral image is input into the hyperspectral prediction module to obtain a second maturity prediction result. The first maturity prediction result and the second maturity prediction result are input into the maturity calculation branch for weighted calculation to obtain a maturity prediction result. The number of days of full bloom of the target loquat product is collected, input into a ripening analysis model in combination with the maturity prediction result to obtain a ripening analysis result, and the ripening analysis result is used to ripen the target loquat product. By acquiring fruit images and fruit hyperspectral images, the maturity status of loquats is obtained from multiple dimensions to obtain more accurate maturity data and make targeted treatment methods, improving the accuracy of loquat maturity recognition, further improving the pertinence and applicability of the treatment of immature loquats, and improving the fruit treatment effect. It solves the technical problems in the prior art that the detection accuracy of the loquat maturity detection method is relatively low, and the pertinence of the subsequent fruit ripening treatment is poor, resulting in poor treatment effect of the loquat fruit ripening treatment.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specifically illustrates the specific implementation manners of this application. Description of the Drawings

[0011] Figure 1 It is a schematic flowchart of a multi-dimensional loquat maturity analysis method based on non-destructive testing provided by this application;

[0012] Figure 2 It is a schematic flowchart of the process of obtaining a multi-dimensional loquat maturity prediction model in a multi-dimensional loquat maturity analysis method based on non-destructive testing provided by this application;

[0013] Figure 3 It is a schematic flowchart of the process of obtaining a ripening analysis result in a multi-dimensional loquat maturity analysis method based on non-destructive testing provided by this application;

[0014] Figure 4 It is a schematic structural diagram of a multi-dimensional loquat maturity analysis system based on non-destructive testing provided by this application.

[0015] Description of the reference numerals: Image information acquisition module 11, hyperspectral image acquisition module 12, maturity prediction model acquisition module 13, first maturity prediction result acquisition module 14, second maturity prediction result acquisition module 15, maturity prediction result lake acquisition module 16, ripening analysis result acquisition module 17. Specific Embodiments

[0016] This application provides a multi-dimensional loquat maturity analysis method and system based on non-destructive testing, which is used to solve the technical problems in the prior art that the detection accuracy of the loquat maturity detection method is relatively low, and the subsequent fruit ripening treatment is poorly targeted, resulting in poor treatment effects of the loquat fruit ripening treatment.

[0017] Next, the technical solutions in this application will be clearly and completely described with reference to the drawings. The described embodiments are only part of the content that can be achieved by this application, rather than all of the content of this application.

[0018] Embodiment 1

[0019] As Figure 1 shown, this application provides a multi-dimensional loquat maturity analysis method based on non-destructive testing, and the method includes:

[0020] Step 100: Collect the image information of the target loquat product;

[0021] Step 200: Based on the hyperspectral detection device, collect hyperspectral images of the target loquat product to obtain hyperspectral images;

[0022] Step 300: Construct a multi-dimensional loquat maturity prediction model. The multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch;

[0023] Specifically, non-destructive testing refers to the inspection of the object to be inspected without damaging the object to be detected. In this embodiment, the maturity of the loquat product is detected by collecting images. Collect the image information of the target loquat product to be detected, and based on the hyperspectral detection device, collect hyperspectral images of the target loquat product to obtain hyperspectral images. Construct a multi-dimensional loquat maturity prediction model. The multi-dimensional loquat maturity prediction model includes an image prediction module and a hyperspectral prediction module. The image prediction module is used to determine the maturity of the loquat according to the collected image information of the target loquat product, the hyperspectral prediction module is used to determine the maturity of the loquat according to the hyperspectral image, and the maturity calculation branch is used to perform weighted calculation based on the output results of the two to obtain the final maturity calculation result.

[0024] As Figure 2 shown, step 300 of the method provided in the embodiment of the present application further includes:

[0025] Step 310: Based on the convolutional neural network, construct the image prediction module;

[0026] Step 320: Construct the hyperspectral prediction module;

[0027] Step 330: According to the image prediction module and the hyperspectral prediction module, construct the maturity calculation branch;

[0028] Step 340: According to the constructed image prediction module, hyperspectral prediction module, and maturity calculation branch, obtain the constructed multi-dimensional loquat maturity prediction model.

[0029] Specifically, based on the convolutional neural network, an image prediction module is constructed. By obtaining the image prediction module, the model can obtain the corresponding maturity level according to specific sample image information. Subsequently, a hyperspectral prediction module is constructed, where the hyperspectral prediction module is used to obtain the corresponding maturity level according to the wavelength peak of the hyperspectral image. Subsequently, according to the image prediction module and the hyperspectral prediction module, a maturity calculation branch is performed, where the maturity calculation branch is used to perform weighted calculation on the first maturity prediction result and the second maturity prediction result predicted by the image prediction module and the hyperspectral prediction module to obtain the final maturity calculation result. Through the image prediction module, the hyperspectral prediction module and the maturity calculation branch, a constructed multi-dimensional loquat maturity prediction model is obtained, and the construction of the multi-dimensional loquat maturity prediction model is completed.

[0030] Step 310 of the method provided in the embodiment of the present application further includes:

[0031] Step 311: Collect and obtain image information of multiple loquat products at different maturity stages to obtain multiple sample image information;

[0032] Step 312: Identify the multiple sample image information according to different multiple maturity levels to obtain multiple sample first maturity level information;

[0033] Step 313: Based on the convolutional neural network, construct the image prediction module;

[0034] Step 314: Use the multiple sample image information and the multiple sample first maturity level information to perform supervised training, verification and testing on the image prediction module to obtain the image prediction module with an accuracy rate meeting the preset requirements.

[0035] Specifically, image information of multiple loquat products at different maturity stages is collected and obtained to acquire multiple sample image information. Subsequently, the obtained sample image information is marked with maturity levels to obtain multiple different maturity levels, and the multiple sample image information is marked to obtain a marking result and multiple sample first maturity level information. Subsequently, the sample image information is used as input data, and the multiple sample first maturity level information is used as marking data to construct training data. The training data is extracted with replacement, and verification data and test data with the same amount as the training data are respectively extracted. Further, based on a convolutional neural network, the training data is input into an untrained convolutional neural network model, and the convolutional neural network model is supervised and trained to obtain a trained model. Then, the verification data and test data are respectively input into the trained model for model verification and testing. When the final output result of the model meets a certain accuracy threshold, the trained model is obtained to get an image prediction module. By constructing the image prediction module, the model can obtain the corresponding maturity level according to specific sample image information, facilitating the subsequent acquisition of a more accurate maturity prediction result.

[0036] Step 320 of the method provided in the embodiment of the present application further includes:

[0037] Step 321: Based on a hyperspectral detection device, use the hyperspectral detection device to collect hyperspectral images of multiple loquat products at different maturity stages to obtain multiple sample hyperspectral images;

[0038] Step 322: Collect the wavelength peaks of the hyperspectral images of the multiple sample hyperspectral images to obtain the wavelength peaks of the multiple sample hyperspectral images;

[0039] Step 323: Sample and detect the pectin content and cellulose content of the multiple loquat products at different maturity stages to obtain multiple sample pectin contents and multiple sample cellulose contents;

[0040] Step 324: Mark according to different multiple maturity levels based on the magnitudes of the multiple sample pectin contents and multiple sample cellulose contents to obtain multiple sample second maturity level information;

[0041] Step 325: Use the wavelength peaks of the multiple sample hyperspectral images, the multiple sample pectin contents, the multiple sample cellulose contents, and the multiple sample second maturity level information as construction data to construct the hyperspectral prediction module.

[0042] Specifically, based on the hyperspectral detection device, the hyperspectral detection device is used to collect hyperspectral images of multiple loquat products at different maturity stages, obtaining multiple sample hyperspectral images, where each sample hyperspectral image corresponds to a corresponding maturity stage. Subsequently, the wavelength peaks of the multiple sample hyperspectral images are collected to obtain the wavelength peaks of the multiple sample hyperspectral images. And the pectin content and cellulose content of multiple loquat products at different maturity stages are sampled and detected to obtain multiple sample pectin contents and multiple sample cellulose contents. Further, according to different multiple maturity levels, the magnitudes of the multiple sample pectin contents and multiple sample cellulose contents are marked, that is, the magnitude levels of the multiple sample pectin contents and multiple sample cellulose contents are marked. Since the data of the multiple sample pectin contents and multiple sample cellulose contents obtained cannot intuitively reflect the numerical differences, it is necessary to mark the detection values of the multiple sample pectin contents and multiple sample cellulose contents as more intuitive data levels, obtaining the second maturity level information of multiple samples. The second maturity level information includes the magnitude levels of the multiple sample pectin contents and multiple sample cellulose contents, as well as the corresponding maturity levels. Finally, based on the wavelength peaks of the multiple sample hyperspectral images, the multiple sample pectin contents, the multiple sample cellulose contents, and the second maturity level information of the multiple samples as construction data, the hyperspectral prediction module is constructed. Among them, the hyperspectral prediction module is constructed through the hyperspectral prediction sub-model and the corresponding mapping relationship. The hyperspectral prediction module inputs the wavelength peaks of the hyperspectral image, obtains the corresponding pectin content and cellulose content through the hyperspectral prediction sub-model, and further maps to obtain the corresponding maturity level information and obtain the corresponding maturity level.

[0043] The method step 325 provided in the embodiment of the present application further includes:

[0044] Step 325-1: Randomly select a sample hyperspectral image wavelength peak within the multiple sample hyperspectral image wavelength peaks to construct the first division node of the hyperspectral prediction module. The first division node performs a one-class division on the input hyperspectral image wavelength peak;

[0045] Step 325-2: Randomly select another sample hyperspectral image wavelength peak within the multiple sample hyperspectral image wavelength peaks to construct the second division node of the hyperspectral prediction module. The second division node performs a two-class division on the division result of the first division node;

[0046] Step 325-3: Continue to construct the multi-level division nodes of the hyperspectral prediction module;

[0047] Step 325-4: Based on the multi-level division nodes, obtain multiple division results;

[0048] Step 325-5: Mark the multiple division results according to the pectin content of the multiple samples and the cellulose content of the multiple samples to obtain a hyperspectral prediction sub-model;

[0049] Step 325-6: Construct a mapping relationship among the pectin content of the multiple samples, the cellulose content of the multiple samples, and the second maturity level information of the multiple samples;

[0050] Step 325-7: Obtain the constructed hyperspectral prediction module according to the hyperspectral prediction sub-model and the mapping relationship.

[0051] Specifically, randomly select a sample hyperspectral image wavelength peak within the wavelength peaks of the multiple sample hyperspectral images, and use this sample hyperspectral image wavelength peak as the first division node. Subsequently, according to the first division node, divide the input hyperspectral image wavelength peaks into division regions greater than, equal to, and less than this wavelength peak to complete a classification division. Subsequently, randomly select another sample hyperspectral image wavelength peak within the wavelength peaks of the multiple sample hyperspectral images to construct the second division node of the hyperspectral prediction module. The second division node performs a binary classification division on the division result of the first division node, and continues to obtain division regions greater than, equal to, and less than this wavelength peak. Continue to construct multiple-level division nodes of the hyperspectral prediction module, and obtain multiple division results according to the multiple multiple-level division nodes until the division of all data is completed, the division of the original data is completed, and the construction of the tree structure is completed. Further, mark the multiple division results according to the pectin content of the multiple samples and the cellulose content of the multiple samples, that is, mark the constructed tree structure, complete the marking of each node, and obtain a hyperspectral prediction sub-model. The hyperspectral prediction sub-model is used to obtain the pectin content and cellulose content of the sample corresponding to the peak according to the sample hyperspectral image wavelength peak. Finally, construct a mapping relationship among the pectin content of the multiple samples, the cellulose content of the multiple samples, and the second maturity level information of the multiple samples. According to the hyperspectral prediction sub-model and the mapping relationship, obtain the constructed hyperspectral prediction module. The hyperspectral prediction module inputs the hyperspectral image wavelength peak, obtains the corresponding pectin content and cellulose content through the hyperspectral prediction sub-model, and further maps to obtain the corresponding maturity level information and obtain the corresponding maturity level.

[0052] The method step 320 provided by the embodiment of the present application further includes:

[0053] Step 326: Obtain the accuracy rates of the image prediction module and the hyperspectral prediction module to obtain a first accuracy rate and a second accuracy rate;

[0054] Step 327: Perform weight allocation according to the magnitudes of the first accuracy rate and the second accuracy rate to obtain a weight allocation result;

[0055] Step 328: Construct a weighted calculation rule according to the weight assignment result to obtain the maturity calculation branch, where the weighted calculation rule is used to perform weighted calculation on the first maturity prediction result and the second maturity prediction result obtained by prediction of the image prediction module and the hyperspectral prediction module.

[0056] Specifically, obtain the accuracies of the image prediction module and the hyperspectral prediction module to obtain the first accuracy and the second accuracy. When obtaining the accuracies, input the test sample data with known maturity into the image prediction module and the hyperspectral prediction module respectively, and obtain the accuracies of the output results respectively. According to the magnitudes of the first accuracy and the second accuracy, perform weight assignment to obtain the weight assignment result, where the higher the accuracy, the greater the assigned weight. When actually setting the weights, it can be based on the accuracies and set according to the actual situation. According to the weight assignment result, construct a weighted calculation rule to obtain the maturity calculation branch. The weighted calculation rule is that the results output by the image prediction module and the hyperspectral prediction module are respectively subjected to weight calculation, and the calculation results are added up to obtain the added-up maturity calculation result. This process is the weighted calculation rule. The weighted calculation rule is used to perform weighted calculation on the first maturity prediction result and the second maturity prediction result obtained by prediction of the image prediction module and the hyperspectral prediction module to obtain the maturity calculation result. By performing weighted calculation on the model output result, the obtained maturity result is made more accurate.

[0057] Step 400: Input the image information into the image prediction module to obtain the first maturity prediction result;

[0058] Step 500: Input the hyperspectral image into the hyperspectral prediction module to obtain the second maturity prediction result;

[0059] Step 600: Input the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain the maturity prediction result;

[0060] Step 700: Collect the number of days of full bloom of the target loquat product, input it into the ripening analysis model in combination with the maturity prediction result to obtain the ripening analysis result, and use the ripening analysis result to ripen the target loquat product.

[0061] Specifically, input the image information into the image prediction module to obtain the first maturity prediction result. Input the hyperspectral image into the hyperspectral prediction module to obtain the second maturity prediction result. Finally, input the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain the maturity prediction result, and obtain the output maturity of the multi-dimensional loquat maturity prediction model. Further, collect the number of days of full bloom of the target loquat product, where the number of days of full bloom is the number of days from the full bloom period to the picking time of the corresponding loquat product. Combine the maturity prediction result and input it into the ripening analysis model to obtain the ripening analysis result, where the ripening analysis result includes the corresponding treatment method for the maturity of the sampled loquat, such as performing ripening treatment with different ethylene concentrations, or other common ripening methods. Use the ripening analysis result to ripen the target loquat product. Obtain the loquat maturity status through multiple dimensions to obtain more accurate maturity data and make targeted treatment methods, improve the accuracy of loquat maturity identification, further improve the pertinence and applicability of immature loquat treatment, and improve the fruit treatment effect.

[0062] As Figure 3 shown, step 700 of the method provided by the embodiment of the present application further includes:

[0063] Step 710: Collect and obtain the number of days of full bloom and the maturity of multiple loquat products at different maturity stages to obtain multiple sample days of full bloom and multiple sample maturities;

[0064] Step 720: Set different ripening parameters for the multiple loquat products at different maturity stages to obtain multiple sample ripening parameters;

[0065] Step 730: Based on the BP neural network, construct the ripening analysis model, where the input data of the ripening analysis model is the number of days of full bloom and the maturity, and the output parameter is the ripening analysis result;

[0066] Step 740: Identify the multiple sample days of full bloom, multiple sample maturities, and multiple sample ripening parameters to obtain a constructed data set;

[0067] Step 750: Based on the k-fold cross-validation method, use the constructed data set to perform iterative supervised training and verification on the ripening analysis model to obtain the ripening analysis model with an accuracy rate meeting the preset requirements.

[0068] Specifically, multiple flowering days and multiple maturity levels of multiple loquat products at different maturity stages are collected to obtain multiple sample flowering days and multiple sample maturity levels. The flowering days refer to the number of days from the full flowering stage to the picking time of the corresponding loquat product. Subsequently, different ripening parameters are set for the multiple loquat products at different maturity stages. For example, the loquat products at different maturity stages are subjected to ripening treatments with different ethylene concentrations or other common ripening methods to obtain multiple sample ripening parameters. Based on the BP neural network, the ripening analysis model is constructed. Among them, the input data of the ripening analysis model are the flowering days and the maturity level, and the output parameter is the ripening analysis result. Further, the multiple sample flowering days, multiple sample maturity levels, and multiple sample ripening parameters are labeled to obtain a constructed data set. When constructing the ripening analysis model, the constructed data set is input into an untrained BP neural network model, and the model is trained. And based on the k-fold cross-validation method, which is a prior art and will not be elaborated here, the constructed data set is used to perform iterative supervised training and verification on the ripening analysis model to obtain the ripening analysis model with an accuracy meeting the preset requirements.

[0069] In summary, the method provided in the embodiment of the present application acquires the image information of the target loquat product. Based on the hyperspectral detection device, the loquat hyperspectral image is obtained. A multi-dimensional loquat maturity prediction model is constructed, and the model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch. The first maturity prediction result is obtained, and the second maturity prediction result is obtained. And they are input into the maturity calculation branch for weighted calculation to obtain the maturity prediction result. The flowering days of the product are collected, and combined with the maturity prediction result, they are input into the ripening analysis model to obtain the ripening analysis result to ripen the target loquat product. By acquiring the fruit image and the fruit hyperspectral image, the maturity status of the loquat is obtained from multiple dimensions to obtain more accurate maturity data and make targeted treatment methods, improving the accuracy of loquat maturity recognition, further improving the pertinence and applicability of the treatment of immature loquats, and improving the fruit treatment effect. This solves the technical problems in the prior art that the detection accuracy of the loquat maturity detection method is relatively low, and the subsequent fruit ripening treatment has poor pertinence, resulting in poor treatment effect of the loquat fruit ripening treatment.

[0070] Embodiment 2

[0071] Based on the same inventive concept as the multi-dimensional loquat maturity status analysis method based on non-destructive detection in the foregoing embodiment, as Figure 4 shown, the present application provides a multi-dimensional loquat maturity status analysis system based on non-destructive detection, and the system includes:

[0072] An image information acquisition module 11, configured to acquire the image information of the target loquat product;

[0073] The hyperspectral image acquisition module 12 is used to acquire hyperspectral images of the target loquat products based on a hyperspectral detection device, and obtain hyperspectral images;

[0074] The maturity prediction model acquisition module 13 is used to construct a multi-dimensional loquat maturity prediction model, wherein the multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch;

[0075] The first maturity prediction result acquisition module 14 is used to input the image information into the image prediction module to obtain a first maturity prediction result;

[0076] The second maturity prediction result acquisition module 15 is used to input the hyperspectral image into the hyperspectral prediction module to obtain a second maturity prediction result;

[0077] The maturity prediction result acquisition module 16 is used to input the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain a maturity prediction result;

[0078] The ripening analysis result acquisition module 17 is used to collect the number of days from full bloom of the target loquat products, input it into the ripening analysis model in combination with the maturity prediction result, obtain a ripening analysis result, and use the ripening analysis result to ripen the target loquat products.

[0079] Furthermore, the maturity prediction model acquisition module 13 is further used for:

[0080] Construct the image prediction module based on a convolutional neural network;

[0081] Construct the hyperspectral prediction module;

[0082] Construct the maturity calculation branch according to the image prediction module and the hyperspectral prediction module;

[0083] Obtain the constructed multi-dimensional loquat maturity prediction model according to the constructed image prediction module, hyperspectral prediction module, and maturity calculation branch.

[0084] Furthermore, the maturity prediction model acquisition module 13 is further used for:

[0085] Collect and obtain image information of multiple loquat products at different maturity stages to obtain multiple sample image information;

[0086] Identify the multiple sample image information according to different multiple maturity levels to obtain multiple sample first maturity level information;

[0087] Construct the image prediction module based on a convolutional neural network;

[0088] Using the multiple sample image information and the multiple sample first maturity level information, the image prediction module is supervised, trained, verified and tested to obtain the image prediction module with an accuracy rate meeting the preset requirements.

[0089] Furthermore, the maturity prediction model acquisition module 13 is further configured to:

[0090] Based on a hyperspectral detection device, the hyperspectral detection device is used to collect hyperspectral images of multiple loquat products at different maturity stages to obtain multiple sample hyperspectral images;

[0091] Collect the wavelength peaks of the hyperspectral images of the multiple sample hyperspectral images to obtain the wavelength peaks of the multiple sample hyperspectral images;

[0092] Sample and detect the pectin content and cellulose content of the multiple loquat products at different maturity stages to obtain multiple sample pectin contents and multiple sample cellulose contents;

[0093] According to different multiple maturity levels, identify them according to the magnitudes of the multiple sample pectin contents and multiple sample cellulose contents to obtain multiple sample second maturity level information;

[0094] Using the wavelength peaks of the multiple sample hyperspectral images, the multiple sample pectin contents, the multiple sample cellulose contents and the multiple sample second maturity level information as construction data, construct the hyperspectral prediction module.

[0095] Furthermore, the maturity prediction model acquisition module 13 is further configured to:

[0096] Randomly select a sample hyperspectral image wavelength peak within the wavelength peaks of the multiple sample hyperspectral images to construct the first division node of the hyperspectral prediction module, and the first division node performs a one-class classification division on the input hyperspectral image wavelength peak;

[0097] Randomly select a sample hyperspectral image wavelength peak again within the wavelength peaks of the multiple sample hyperspectral images to construct the second division node of the hyperspectral prediction module, and the second division node performs a two-class classification division on the division result of the first division node;

[0098] Continue to construct the multi-level division nodes of the hyperspectral prediction module;

[0099] Based on the multi-level division nodes, obtain multiple division results;

[0100] Mark the multiple division results according to the multiple sample pectin contents and multiple sample cellulose contents to obtain a hyperspectral prediction sub-model;

[0101] Construct the mapping relationships among the pectin contents of the multiple samples, the cellulose contents of the multiple samples, and the second maturity level information of the multiple samples;

[0102] According to the hyperspectral prediction sub-model and the mapping relationships, obtain the constructed hyperspectral prediction module.

[0103] Further, the maturity prediction model acquisition module 13 is further configured to:

[0104] Obtain the accuracies of the image prediction module and the hyperspectral prediction module, and obtain a first accuracy and a second accuracy;

[0105] Perform weight allocation according to the magnitudes of the first accuracy and the second accuracy, and obtain a weight allocation result;

[0106] According to the weight allocation result, construct a weighted calculation rule to obtain the maturity calculation branch, where the weighted calculation rule is used to perform weighted calculation on the first maturity prediction result and the second maturity prediction result predicted by the image prediction module and the hyperspectral prediction module.

[0107] Further, the ripening analysis result acquisition module 17 is further configured to:

[0108] Collect and obtain the multiple blooming days and the multiple maturity levels of multiple loquat products at different maturity stages, and obtain multiple sample blooming days and multiple sample maturity levels;

[0109] Set different ripening parameters for the multiple loquat products at different maturity stages, and obtain multiple sample ripening parameters;

[0110] Based on the BP neural network, construct the ripening analysis model, where the input data of the ripening analysis model is the blooming days and the maturity level, and the output parameter is the ripening analysis result;

[0111] Identify the multiple sample blooming days, the multiple sample maturity levels, and the multiple sample ripening parameters to obtain a constructed data set;

[0112] Based on the k-fold cross-validation method, use the constructed data set to perform iterative supervised training and verification on the ripening analysis model, and obtain the ripening analysis model with an accuracy meeting the preset requirements.

[0113] The second above-mentioned embodiment is used to execute the method in the first embodiment. Its execution principle and execution basis can both be obtained from the content described in the first embodiment, and will not be elaborated here. Although the present application has been described in combination with specific features and their embodiments, the present application is not limited by the example embodiments described herein. Based on the embodiments of the present application, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application, and the content obtained in this way also belongs to the scope protected by the present application.

Claims

1. A multi-dimensional analysis method for the maturity status of loquats based on non-destructive testing, characterized in that The method includes: Collecting image information of the target loquat product; Based on a hyperspectral detection device, performing hyperspectral image acquisition on the target loquat product to obtain a hyperspectral image; Constructing a multi-dimensional loquat maturity prediction model, wherein the multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch; Inputting the image information into the image prediction module to obtain a first maturity prediction result; Inputting the hyperspectral image into the hyperspectral prediction module to obtain a second maturity prediction result; Inputting the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain a maturity prediction result; Collecting the number of days since full bloom of the target loquat product, inputting it together with the maturity prediction result into a ripening analysis model to obtain a ripening analysis result, and using the ripening analysis result to ripen the target loquat product.

2. The method according to claim 1, wherein The constructing of the multi-dimensional loquat maturity prediction model includes: Based on a convolutional neural network, constructing the image prediction module; Constructing the hyperspectral prediction module; According to the image prediction module and the hyperspectral prediction module, constructing the maturity calculation branch; According to the constructed image prediction module, hyperspectral prediction module, and maturity calculation branch, obtaining the constructed multi-dimensional loquat maturity prediction model.

3. The method according to claim 2, wherein The constructing of the image prediction module based on a convolutional neural network includes: Collecting and obtaining image information of multiple loquat products at different maturity stages to obtain multiple sample image information; Identifying the multiple sample image information according to different multiple maturity levels to obtain multiple sample first maturity level information; Based on a convolutional neural network, constructing the image prediction module; Using the multiple sample image information and the multiple sample first maturity level information to perform supervised training, verification, and testing on the image prediction module to obtain the image prediction module with an accuracy rate meeting the preset requirements.

4. The method according to claim 3, wherein Constructing the hyperspectral prediction module includes: Based on a hyperspectral detection device, using the hyperspectral detection device to perform hyperspectral image acquisition on multiple loquat products at different maturity stages to obtain multiple sample hyperspectral images; Performing hyperspectral image wavelength peak acquisition on the multiple sample hyperspectral images to obtain multiple sample hyperspectral image wavelength peaks; Performing sampling and detection on the pectin content and cellulose content of the multiple loquat products at different maturity stages to obtain multiple sample pectin contents and multiple sample cellulose contents; Identifying according to different multiple maturity levels based on the magnitudes of the multiple sample pectin contents and multiple sample cellulose contents to obtain multiple sample second maturity level information; Using the multiple sample hyperspectral image wavelength peaks, multiple sample pectin contents, multiple sample cellulose contents, and the multiple sample second maturity level information as construction data to construct the hyperspectral prediction module.

5. The method according to claim 4, characterized in that Using the multiple sample hyperspectral image wavelength peaks, multiple sample pectin contents, multiple sample cellulose contents, and the multiple sample second maturity level information as construction data to construct the hyperspectral prediction module includes: Randomly select a peak wavelength of a sample hyperspectral image within the peak wavelengths of the multiple sample hyperspectral images, construct a first partitioning node of the hyperspectral prediction module, and the first partitioning node performs a one-class partitioning on the input peak wavelength of the hyperspectral image; Randomly select another peak wavelength of a sample hyperspectral image within the peak wavelengths of the multiple sample hyperspectral images, construct a second partitioning node of the hyperspectral prediction module, and the second partitioning node performs a two-class partitioning on the partitioning result of the first partitioning node; Continue to construct multiple-level partitioning nodes of the hyperspectral prediction module; Based on the multiple-level partitioning nodes, obtain multiple partitioning results; Mark the multiple partitioning results according to the multiple sample pectin contents and multiple sample cellulose contents to obtain a hyperspectral prediction sub-model; Construct a mapping relationship among the multiple sample pectin contents, multiple sample cellulose contents, and the multiple sample second maturity level information; According to the hyperspectral prediction sub-model and the mapping relationship, obtain the constructed hyperspectral prediction module.

6. The method according to claim 4, characterized in that, According to the image prediction module and the hyperspectral prediction module, construct the maturity calculation branch, including: Obtain the accuracies of the image prediction module and the hyperspectral prediction module to obtain a first accuracy and a second accuracy; Perform weight allocation according to the magnitudes of the first accuracy and the second accuracy to obtain a weight allocation result; According to the weight allocation result, construct a weighted calculation rule to obtain the maturity calculation branch, where the weighted calculation rule is used to perform weighted calculation on the first maturity prediction result and the second maturity prediction result predicted by the image prediction module and the hyperspectral prediction module.

7. The method according to claim 1, characterized in that, Collect the number of days of full bloom of the target loquat product, and input the maturity prediction result into the ripening analysis model to obtain a ripening analysis result, including: Collect and obtain the number of days of full bloom and the maturity of multiple loquat products at different maturity stages to obtain multiple sample days of full bloom and multiple sample maturities; Set different ripening parameters for the multiple loquat products at different maturity stages to obtain multiple sample ripening parameters; Based on the BP neural network, construct the ripening analysis model, where the input data of the ripening analysis model is the number of days of full bloom and the maturity, and the output parameter is the ripening analysis result; Identify the multiple sample days of full bloom, multiple sample maturities, and multiple sample ripening parameters to obtain a constructed data set; Based on the k-fold cross-validation method, use the constructed data set to perform iterative supervised training and verification on the ripening analysis model to obtain the ripening analysis model with an accuracy meeting the preset requirements.

8. A multi-dimensional loquat maturity analysis system based on non-destructive testing, characterized in that, The system includes: An image information acquisition module for acquiring image information of the target loquat product; A hyperspectral image acquisition module for acquiring a hyperspectral image of the target loquat product based on a hyperspectral detection device; A maturity prediction model acquisition module for constructing a multi-dimensional loquat maturity prediction model, where the multi-dimensional loquat maturity prediction model includes an image prediction module, a hyperspectral prediction module, and a maturity calculation branch; The first maturity prediction result acquisition module is used to input the image information into the image prediction module to obtain the first maturity prediction result; The second maturity prediction result acquisition module is used to input the hyperspectral image into the hyperspectral prediction module to obtain the second maturity prediction result; The maturity prediction result lake acquisition module is used to input the first maturity prediction result and the second maturity prediction result into the maturity calculation branch for weighted calculation to obtain the maturity prediction result; The ripening analysis result acquisition module is used to collect the number of days from full bloom of the target loquat product, input it into the ripening analysis model in combination with the maturity prediction result to obtain the ripening analysis result, and use the ripening analysis result to ripen the target loquat product.

Images