A method for identifying tea varieties based on canopy reflectance spectra

Abstract

The present invention proposes a method for identifying tea varieties based on canopy reflectance spectra, which belongs to the field of detection and analysis technology. It includes: S1. Based on tea germplasm resources, collect near-infrared reflectance spectrum data of tea canopies or leaves, perform data processing, and construct an infrared reflectance spectrum database of different tea varieties; S2. Segment the spectrum database data, one part is used for identification model development, and the other part is used for model diagnosis; S3. Perform machine learning model training on tea variety identification model; S4. For different generated models, establish and verify the correction model, perform correlation analysis, and then evaluate, and finally determine the most effective model. The overall accuracy of the identification model of the present invention is high, and the accuracy of identifying varieties of closely related tea trees is slightly lower, and the application prospect is good.

Description

Technical Field

[0001] The invention relates to the technical field of detection and analysis, and in particular to a method for identifying tea tree varieties based on canopy reflectance spectra. Background Art

[0002] Tea tree nutrient management is the basis for its high quality, high yield and stable yield, and nutrient diagnosis is the theoretical and technical basis of nutrient management. There are many methods for plant nutrient diagnosis, including soil diagnosis, leaf diagnosis, field test, tracer atom, microbial determination and so on. Due to the large root system of tea tree, the tree needs and stores many nutrients, and the site conditions of the garden are complex. The use of soil nutrient diagnosis and field test methods often has problems such as poor diagnostic accuracy and low efficiency, and the results are only regarded as reference values ​​in production. The leaf is the most sensitive organ to nutrition, and its dynamic changes in nutrition can provide real-time feedback on the abundance and deficiency of tree and soil nutrients. Therefore, the leaf nutrient diagnosis method is a relatively ideal nutrient diagnosis method. However, the nutrients in tea leaves have a dynamic change law in time and space. In the measured samples, the collection period and processing method are different, and the analysis and diagnosis results vary greatly. In particular, tea trees use new shoots as their harvesting organs, and they need to be picked continuously from spring to autumn every year, which consumes a lot of nutrients. When measuring the changes in leaf nutrient elements, we must fully consider the changes in a large number of elements caused by picking fresh leaves. When collecting samples, we must consider the changing dynamics of the elements in order to determine the abundance and deficiency of the elements and formulate reasonable types and quantities of fertilizers accordingly.

[0003] The nutrient absorption and conversion rules of tea trees are different in different seasons. Autumn to winter is the nutrient storage period, and the main nutrients are stored in the roots, leaves and dormant buds; spring is the germination period of tea trees, and nutrients are transferred from the roots and leaves to the buds to improve the germination of the buds. With the continuous picking of new shoots, mineral nutrients will be continuously taken away, and nutrients must be replenished in time to meet the needs of the germination of new shoots of tea trees; summer is the vigorous growth season of tea trees, with less nutrient storage, and the germination of buds and the growth of leaves consume a lot of nutrients; at different times, the storage state and operation form of different elements are different, so the nutritional diagnosis of tea leaves is naturally significantly different from that of other crops.

[0004] Tea tree nutrition diagnosis has been carried out for many years, but there is still a lack of fast, convenient and practical diagnostic methods. At present, there is still a blank in my country regarding tea leaf nutrition diagnosis standards and their applications. The present invention obtains a large amount of data on the nutrient content of tea leaves through tea garden investigation, analysis and determination, and fixed-point experiments. Through multiple regression analysis and principal component analysis, the nutritional status of the main varieties of tea gardens is clarified, the standard values ​​of the nutrient elements of representative varieties of tea trees and the nutritional diagnosis technical standards are established, and a tea tree nutrition diagnosis technical system with leaf nutrition analysis as the core is proposed, which is of great significance to improving the fertilizer and water management level of tea gardens.

[0005] China is the country with the richest tea tree planting resources in the world, with more than 2,000 tea tree planting resources. The authenticity identification of tea tree varieties not only affects the protection and utilization of tea tree resources, but also affects the quality of tea product processing and the reputation of famous tea. Therefore, tea tree variety identification is an important part of tea tree planting resource research, but due to the difficulties in collecting and identifying tea tree tissue organ images, it is often impossible to accurately identify tea tree varieties.

[0006] With the continuous advancement of technologies such as deep learning, machine vision, and image recognition, the use of key technologies such as generative adversarial networks can be used to process images of tea tree tissue organs, and then accurately identify and classify tea tree varieties, which can solve the above problems well. Summary of the invention

[0007] The purpose of the present invention is to propose a method for identifying tea tree varieties based on canopy reflectance spectra. The overall accuracy of the identification model of the present invention is 84%. The identification accuracy of tea tree varieties with leaf color mutations such as Anji White Tea, Golden Bud, and Purple Azalea can reach 100%. The identification accuracy of varieties of closely related tea trees is slightly lower, and the application prospect is good.

[0008] The technical solution of the present invention is achieved in this way:

[0009] The present invention provides a method for identifying tea tree varieties based on canopy reflectance spectrum, comprising the following steps:

[0010] S1. Based on tea germplasm resources, collect near-infrared reflectance spectrum data of tea canopy or leaves, process the data, and build an infrared reflectance spectrum database of different tea varieties;

[0011] S2. Split the spectral database data, one part is used for recognition model development, and the other part is used for model diagnosis;

[0012] S3. Perform machine learning model training on tea variety recognition model;

[0013] S4. For the different models generated, establish and verify the correction model, perform correlation analysis, and then evaluate and finally determine the most effective model.

[0014] As a further improvement of the present invention, the method for collecting near-infrared reflectance spectrum data of the tea tree canopy or leaves in step S1 is as follows: take the sample powder and put it into a sample cup, mix it evenly, flatten it, and use an integrating sphere diffuse reflectance accessory to collect it to obtain near-infrared reflectance spectrum data.

[0015] As a further improvement of the present invention, the scanning range collected by the integrating sphere diffuse reflection accessory is 4000-10000cm -1 , scanning times 60-70 times, resolution 5-10cm-1 , ambient temperature 18-25℃, InGaAs detector.

[0016] As a further improvement of the present invention, the scanning number is 64 times and the resolution is 8 cm -1 , ambient temperature 20℃.

[0017] As a further improvement of the present invention, the data processing in step S1 uses MATLAB 7.0 software combined with TQAnalyst 7.2 data processing software.

[0018] As a further improvement of the present invention, the machine learning method in step S3 includes linear discriminant, partial least squares method, neural network, decision tree, random forest and support vector machine.

[0019] As a further improvement of the present invention, the method for establishing and verifying the calibration model in step S4 is to use TQAnalyst 7.2 data processing software to establish a calibration model between the near-infrared spectrum characteristics of tea trees and the content of the components to be measured, and then analyze the verification set samples.

[0020] As a further improvement of the present invention, the correlation analysis method in step S4 is to use SPSS software to perform pairwise correlation analysis on the four index components of chlorophyll, water content, ash content and alcohol-soluble extract of tea plants.

[0021] As a further improvement of the present invention, the present invention specifically comprises the following steps:

[0022] S1. Based on tea germplasm resources, collect near-infrared reflectance spectrum data of tea canopy or leaves, process the data, and build an infrared reflectance spectrum database of different tea varieties;

[0023] The collection method is as follows: put the sample powder into the sample cup, mix it evenly, flatten it, and use the integrating sphere diffuse reflectance accessory to collect the near-infrared reflectance spectrum data. The scanning range of the integrating sphere diffuse reflectance accessory is 4000-10000cm -1 , scan times 64 times, resolution 8cm -1 , ambient temperature 20°C, InGaAs detector;

[0024] S2. Split the spectral database data, one part is used for recognition model development, and the other part is used for model diagnosis;

[0025] S3. Perform machine learning model training on tea variety identification model, wherein the machine learning method includes linear discriminant, partial least squares method, neural network, decision tree, random forest and support vector machine;

[0026] S4. For the different models generated, TQ Analyst 7.2 data processing software was used to establish a calibration model between the near-infrared spectral characteristics of tea trees and the content of the components to be measured, and then the validation set samples were analyzed. SPSS software was used to perform pairwise correlation analysis on the four indicator components of tea trees, including chlorophyll, moisture, ash and alcohol-soluble extracts, and then an evaluation was performed to finally determine the best model.

[0027] The present invention has the following beneficial effects: the overall accuracy of the recognition model of the present invention is 84%, the recognition accuracy of tea tree varieties with leaf color mutations such as Anji White Tea, Golden Bud, and Purple Azalea can reach 100%, and the recognition accuracy of varieties with closely related tea trees is slightly lower, and the application prospect is good. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] 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 or the description of the prior art 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.

[0029] Figure 1 It is a schematic diagram of the process of the tea tree variety identification method based on canopy reflectance spectrum of the present invention;

[0030] Figure 2 This is an aerial view of the field planting area of ​​different varieties of tea trees;

[0031] Figure 3 This is the prediction effect of the support vector machine model (SVM) on tea tree varieties. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. 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.

[0033] Example 1

[0034] Reference Figure 1 The tea tree variety identification method based on canopy reflectance spectrum specifically includes the following steps:

[0035] S1. Based on tea germplasm resources, collect near-infrared reflectance spectrum data of tea canopy or leaves, process the data, and build an infrared reflectance spectrum database of different tea varieties;

[0036] The collection method is as follows: put the sample powder into the sample cup, mix it evenly, flatten it, and use the integrating sphere diffuse reflectance accessory to collect the near-infrared reflectance spectrum data. The scanning range of the integrating sphere diffuse reflectance accessory is 4000-10000cm -1 , scan times 64 times, resolution 8cm -1 , ambient temperature 20°C, InGaAs detector;

[0037] S2. Split the spectral database data, one part is used for recognition model development, and the other part is used for model diagnosis;

[0038] S3. Perform machine learning model training on tea variety identification model, wherein the machine learning method includes linear discriminant, partial least squares method, neural network, decision tree, random forest and support vector machine;

[0039] S4. For the different models generated, TQ Analyst 7.2 data processing software was used to establish a calibration model between the near-infrared spectral characteristics of tea trees and the content of the components to be measured, and then the validation set samples were analyzed. SPSS software was used to perform pairwise correlation analysis on the four indicator components of tea trees, namely chlorophyll, moisture, ash and alcohol-soluble extracts, and then an evaluation was performed to finally determine the most effective model.

[0040] Example 2

[0041] according to Figure 2 and Figure 3 , multiple canopy leaves of different varieties were collected in the tea planting plots, and their canopy reflectance spectral data were collected; part of the data was used as training data, and machine learning models such as linear discriminant, partial least squares, neural network, decision tree, random forest and support vector machine were constructed for variety identification; data that did not participate in modeling was used as test data, and brought into different models, and the best recognition model was selected as the support vector machine model (SVM) according to the recognition accuracy, and its recognition accuracy for most tea varieties reached 100%.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying tea varieties based on canopy reflectance spectra, characterized in that: The following steps are involved: S1. Based on tea germplasm resources, collect near-infrared reflectance spectrum data of tea canopy or leaves, process the data, and build an infrared reflectance spectrum database of different tea varieties; The collection method is as follows: put the sample powder into the sample cup, mix it evenly, flatten it, and use the integrating sphere diffuse reflectance accessory to collect near-infrared reflectance spectrum data. The scanning range of the integrating sphere diffuse reflectance accessory is 4000-10000cm -1 ; S2. Split the spectral database data, one part is used for recognition model development, and the other part is used for model diagnosis; S3. Perform machine learning model training on tea variety identification model, wherein the machine learning method includes linear discriminant, partial least squares method, neural network, decision tree, random forest and support vector machine; S4. For the different models generated, TQ Analyst 7.2 data processing software was used to establish a calibration model between the near-infrared spectral characteristics of tea trees and the content of the components to be measured, and then the validation set samples were analyzed. SPSS software was used to perform pairwise correlation analysis on the four indicator components of tea trees, including chlorophyll, moisture, ash and alcohol-soluble extracts, and then an evaluation was performed to finally determine the optimal model.

2. The method for identifying tea varieties based on canopy reflectance spectrum according to claim 1, characterized in that: The method for collecting the near-infrared reflectance spectrum data of the tea tree canopy or leaves in step S1 is as follows: take the sample powder and put it into a sample cup, mix it evenly, flatten it, and collect it using an integrating sphere diffuse reflectance accessory to obtain near-infrared reflectance spectrum data.

3. The method for identifying tea varieties based on canopy reflectance spectrum according to claim 2, characterized in that: Scan times: 60-70 times, resolution: 5-10cm -1 , ambient temperature 18-25℃, InGaAs detector.

4. The method for identifying tea varieties based on canopy reflectance spectrum according to claim 3, characterized in that: The scanning number is 64 times, and the resolution is 8cm -1 , ambient temperature 20℃.

5. The method for identifying tea varieties based on canopy reflectance spectrum according to claim 1, characterized in that: The data processing described in step S1 uses MATLAB 7.0 software combined with TQ Analyst 7.2 data processing software.

Images