CN-122004051-A - Mulberry leaf grading harvesting method based on hyperspectrum and machine learning

Abstract

The invention relates to the technical field of mulberry leaf grading harvesting, in particular to a mulberry leaf grading harvesting method based on hyperspectrum and machine learning. The hyperspectral and machine learning-based mulberry leaf grading harvesting method has the advantages that a hyperspectral reflectance-based leaf physiological state classification model is established for distinguishing different types of mulberry leaves, healthy leaves are harvested, a chlorophyll content prediction model is developed for distinguishing the tender degree of the leaves, and the grading harvesting purpose is achieved, the system compares 3 sample set dividing methods, 5 data preprocessing methods and 5 machine learning algorithms, the sample set dividing adopts random data set dividing and original data or multi-element scattering correction preprocessing, a support vector model is constructed, and R2 of a test set is optimally represented.

Inventors

• LUO YIWEI
• HE NINGJIA
• YANG RUI
• LI LI
• LIU HONGJIANG
• LI WEI

Assignees

• 西南大学

Dates

Publication Date

20260512

Application Date

20251218

Claims (4)

1. 1. A mulberry leaf grading harvesting method based on hyperspectral and machine learning is characterized by comprising the following specific steps: Firstly, determining chlorophyll content, namely randomly selecting a blade, presetting three points a, b and c at the tip of the blade as an interested region, repeatedly measuring each point for 3 times, and taking an average value as a SPAD value of the point; Collecting RGB and hyperspectral imaging data of the blade to be detected, and preprocessing the collected hyperspectral imaging data to reduce baseline drift influence; And thirdly, constructing and evaluating classification models and chlorophyll content regression models, namely processing the pretreatment data by constructing the classification models and the chlorophyll content regression models to obtain physiological states and chlorophyll content of the leaves, and grading the leaves according to the physiological states and the chlorophyll content of the leaves.
2. 2. The method for classifying and harvesting mulberry leaves based on hyperspectral and machine learning as claimed in claim 1, wherein the method is characterized in that an orbital high-flux phenotype platform is selected for harvesting, the orbital high-flux phenotype platform integrates Dualix hyperspectral modules and RGB modules, and the resolution of the RGB modules is 2000 ten thousand pixels.
3. 3. The method for classifying and harvesting mulberry leaves based on hyperspectral and machine learning as claimed in claim 1, wherein the preprocessing comprises a black-and-white correction module and a smoothing module, hyperspectral data passes through the SPECTRAVIEW software black-and-white correction module and the smoothing module, and reflection spectrums of ROI points corresponding to the SPAD instrument are selected in HYPERSCAN software for subsequent analysis.
4. 4. The method for classifying and harvesting mulberry leaves based on hyperspectral and machine learning as claimed in claim 1, wherein the classification model is characterized in that the classification model is composed of four classification of mulberry leaves, namely a health model, a pest model, a shading model and an auxotroph model.

Description

Mulberry leaf grading harvesting method based on hyperspectrum and machine learning Technical Field The invention relates to the technical field of mulberry leaf grading harvesting, in particular to a hyperspectral and machine learning-based mulberry leaf grading harvesting method. Background Mulberry is an important foundation stone of ancient silk roads, and has an irreplaceable position in the mulberry industry and traditional medicines for thousands of years, and the description of Ben Cao gang mu is recorded for the first time. Studies show that mulberry has important pharmacological activity in the aspects of treating obesity, hypertension, diabetes and the like. The most prominent is medicinal mulberry, and the variety is originally produced by Iran and widely planted in Xinjiang areas of China, and has a special genetic structure (22 times body, 2n=308) and rich secondary metabolites. At present, the medicinal mulberry is mainly asexual propagation, and grafting seedling is mainly adopted in Xinjiang areas, however, the method has the defects of low survival rate, low nursery yield and slow garden building. In addition, the medicinal mulberry can quickly obtain a large number of seedlings through tissue culture, but the tissue culture seedlings are easy to be affected by diseases and insects when cultivated outdoors, and the accumulation of secondary metabolites can be obviously affected by environmental factors. Therefore, the innovation of the cultivation mode of the medicinal mulberry is important to realize standardized and accurate production. The prior art has some problems that at present, the classification and classification of the mulberry materials are often judged by human experience in production, however, the health state of the mulberry materials cannot be accurately judged by human experience, especially when drought, nutritional defects or diseases and insect pests are just started and the phenotype is not obvious. In addition, the young degree can be judged only by experience when people pick the fruits. The method has great subjectivity, and is not beneficial to large-scale factory production. Hyperspectral technology captures fine plant features through multiband reflectance spectroscopy, and modern agricultural systems are widely integrating hyperspectral sensors for stress phenotype classification and analysis of compound content. In addition, hyperspectral data can develop a band screening model, and create new indexes with stronger prediction capability, wherein the indexes are important for prediction of various agricultural indexes in modern agriculture management. With the development of the computer field, the prediction model based on machine learning can accurately capture sensitivity response characteristics of chlorophyll dynamics to insect pest stress, shading stress and nutrition deficiency stress, so that a key basis is provided for dynamic monitoring of crop health states. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides a mulberry leaf grading harvesting method based on hyperspectral and machine learning, which has the advantage of rapid analysis. In order to achieve the purpose, the invention provides the technical scheme that the mulberry leaf grading harvesting method based on hyperspectral and machine learning comprises the following specific steps: Firstly, determining chlorophyll content, namely randomly selecting a blade, presetting three points a, b and c at the tip of the blade as an interested region, repeatedly measuring each point for 3 times, and taking an average value as a SPAD value of the point; And step two, spectrum data acquisition and preprocessing, namely acquiring RGB and hyperspectral imaging data of the blade to be detected, preprocessing the acquired hyperspectral imaging data to reduce the influence of baseline drift, systematically applying five preprocessing strategies for improving the prediction accuracy and stability of a detection model, and extracting RAW spectrum (RAW) from hyperspectral images of medicinal mulberry leaves through SPECTRAVIEW spectrum analysis software. MMN and SG pre-treatments showed stronger amplitude changes in the visible range 400-700nm compared to the original spectra. In the near infrared region 750-1000nm, mmn treatment resulted in spectral contraction, while SG treatment produced larger amplitude changes. After MSC and SNV pretreatment, the spectrum range is obviously shrunk in the 500-1000nm area, so that the spectrum change caused by scattering is effectively lightened, and meanwhile, the baseline drift and offset in the spectrum data are corrected. Both MMN, MSC, SNV and BC pretreatment effectively enhance the spectral convergence in the region above 750nm, resulting in overlapping spectral trajectories with reduced variability. In contrast, SG processing amplifies signal resolution in the 750-1000nm critical band, and sample division has a large infl