KR102138808B1 - Prediction System and Prediction Method for Germination Rate of Grain - Google Patents

Abstract

The present invention relates to a germination rate prediction system and a germination rate prediction method for grains, and it is possible to more accurately predict the germination rate of grain samples by applying spectral analysis data, shape characteristics, and color property data for grain samples to an artificial intelligence algorithm-based germination rate prediction program. have.

Description

Prediction System and Prediction Method for Germination Rate of Grain

The present invention relates to a grain germination rate prediction system and a method for predicting grain germination rate using the system.

Germination is the start of life for germination or breeding of the species under a suitable environment. As the genetic information of inactive DNA and various enzymes and nutrients in the seed's eyes and endosperm are activated when the external environment conditions improve, it becomes active as a plant. It is the process of starting life.

Rice releases the plant's growth material, gibberellin, under moderate temperature and moisture (30-35%), resulting in biosynthesis in the genes of the blastocyst tissue, resulting in massive replication of hydrolytic enzymes and carbohydrates from embryos to embryos. Breathing and metabolism become active while moving in large quantities, increasing the weight of the building and germinating.

The percentage of seeds that can germinate in a certain amount of rice is called the germination rate, and the method of measuring the germination rate of rice measures the number of granules germinated within 7 days under the temperature condition of 20. In addition, in order for germination to occur, the enzyme must be activated in the embryo, and this enzyme activity phenomenon can be measured by absorbance at a specific wavelength by applying a spectroscopic method. That is, although there is a difference in degree in the general raw material state, the quality degradation by the enzyme continues to occur in the same manner as the germination process, and the content of atomic groups and trace components can be measured by irradiating light of various wavelength bands. Therefore, a high germination rate in rice means that the proportion of grains in which the embryo is alive is high, and that the grains are alive has enzymatic activity. It is possible.

The present inventors made extensive efforts to develop a method for predicting germination rates with improved accuracy. As a result, the present invention was completed by proving that the germination rate of the grain sample can be more accurately predicted by applying the spectral analysis data of the grain sample and the shape and color characteristic data of the embryo part to the germination rate prediction program based on the artificial intelligence algorithm. .

Accordingly, an object of the present invention is to provide a system for predicting germination rate of grain.

Another object of the present invention is to provide a method for predicting the germination rate of grain.

One aspect of the present invention relates to a system for predicting germination rates of grains comprising:

A sample sorting unit in which grain samples are aligned;

A light source unit including a light source;

A reflected light detector configured to detect reflected light reflected from the grain sample by a light source irradiated to the grain sample from the light source unit;

An image photographing unit for photographing an image of a grain sample;

An image analysis unit analyzing shape characteristics and color characteristics from the image obtained by the image photographing unit; And

Germination rate prediction unit to predict the germination rate of grain by applying the data obtained from the reflected light detection unit and the image analysis unit.

The present inventors have made earnest research efforts to develop a method for predicting germination rates with improved accuracy, and as a result, spectral analysis data for grain samples and shape and color characteristic data of embryo parts are applied to a germination rate prediction program based on an artificial intelligence algorithm. It was found that the germination rate of the sample can be predicted more accurately.

In the germination rate prediction system of the grain of the present invention, the light source may be one or more light sources selected from the group consisting of a light source having an ultraviolet wavelength range, a light source having a visible light wavelength range, and a light source having a near infrared wavelength range.

The light source may be located above the sample alignment unit.

The wavelength range of the light source may be 200 to 2,200 nm.

The reflected light detector may include a sensor for spectrum measurement in a near infrared (NIR) band.

The sensor for spectrum measurement in the near infrared (NIR) band may be located above the sample alignment unit.

The sensor for spectrum measurement in the near infrared (NIR) band detects the near infrared light spectrum in the 900-1,700 nm band from reflected light reflected from the sample, and photoelectrically converts it to output.

The shape characteristics obtained from the image obtained by the image photographing unit are the area, perimeter, minor axis, major axis, long and short axis ratio, and contour ratio of the grain sample embryo. ) And one or more shape characteristics selected from the group consisting of roundness.

The color characteristic obtained from the image obtained by the image photographing unit may be a Lab color space value of a grain sample embryo.

In the Lab color space value, L is a luminance axis, a is a red/green complementary color axis, and b is a yellow/blue complementary axis.

The germination rate of grain can be predicted by applying the spectral analysis data obtained from the reflected light detection unit and the shape characteristic and color characteristic data obtained from the image analysis unit to the germination rate prediction program based on the artificial intelligence algorithm.

The grain is selected from the group consisting of brown rice, mung bean, sorghum, yulmu, soybean, barley, green tea, corn, wheat and oats.

Another aspect of the invention relates to a method for predicting the germination rate of grains comprising the following steps:

A light source irradiation step of irradiating a light source to the grain sample;

A reflected light detection step of detecting reflected light reflected from the grain sample;

An image analysis step of analyzing shape characteristics and color characteristics from images of grain samples; And

Germination rate prediction step of predicting the germination rate of grain by applying the data obtained in the reflected light detection step and the image analysis step to the germination rate prediction program based on an artificial intelligence algorithm.

In the germination rate prediction program based on the artificial intelligence algorithm, the input values are data quantifying shape characteristics and color characteristics of the sample embryo part of the grain embryo, and spectroscopic analysis data for reflected light reflected from the grain sample, and the learning target is germination or not. It can be 0 (not germinated) and 1 (germinated).

The grain is selected from the group consisting of brown rice, mung bean, sorghum, yulmu, soybean, barley, green tea, corn, wheat and oats.

The present invention relates to a germination rate prediction system and a germination rate prediction method for grains, and it is possible to more accurately predict the germination rate of grain samples by applying spectral analysis data, shape characteristics, and color property data for grain samples to an artificial intelligence algorithm-based germination rate prediction program. have.

1 is a result of comparing germination rate measurement and prediction in the 200-400 nm UV region.
2 is a result of comparing germination rate measurement and prediction in the 900-1,700 nm NIR region.
3 is a result of comparing germination rate measured values and predicted values in the UV-NIR region.
4 is a result of checking the number of samples (N = 190 points) germinated by germination force (date when germination starts).
Figure 5a shows the image acquisition step in the embryo shape recognition algorithm flowchart.
Figure 5b shows the factor analysis and extraction steps in the embryo shape recognition algorithm flowchart.
Figure 5c shows the shape acquisition step in the embryo shape recognition algorithm flowchart.
Figure 5d shows the shape analysis step in the embryo shape recognition algorithm flowchart.
6A is a raw and CBB image of germinated brown rice on day 1.
6B is a raw and CBB image of germinated brown rice on day 2.
6C is a raw and CBB image of germinated brown rice on day 3.
6D is a raw and CBB image of brown rice that cannot germinate.
7 is a learning flow chart of artificial intelligence.
8A shows an ANN (Artificial Neural Network) program.
8B shows the learning process of the ANN program.
9A is a result of measuring germination rate through PLSR analysis by NIR.
9B is a result of measuring germination rate through image data and ANN analysis.
Figure 9c is the result of measuring the germination rate through NIR data, image data and ANN analysis.

Hereinafter, the present invention will be described in more detail by the following examples. However, these examples are only for illustrating the present invention, and the scope of the present invention is not limited by these examples.

Experimental Example 1: Measurement of germination rate of rice by spectroscopic analysis

In order to measure the germination rate of rice, ultraviolet (UV), visible and near-infrared (NIR) light sources are irradiated to the sample and the ultraviolet light spectrum in the 200-400 nm band and 900-1,700 nm band in the reflected light reflected from the sample The near infrared light spectrum was detected.

The spectrum was obtained in a total of 460 wavelength bands in the wavelength range of 2-30 nm difference in the 200-400 nm ultraviolet region and the 900-1,700 nm near infrared region.

As can be seen in Figure 1, the principal component (principal component) analysis of the spectrum of a total of 204 wavelength bands obtained in the ultraviolet (UV) region of 200-400 nm showed the lowest variance when there are 7 main components, As a result of analyzing the correlation of a total of 204 wavelength bands, it was found that there was no correlation in a specific area, but a correlation in a wide range. The coefficient of determination was 0.85.

In addition, as can be seen in FIG. 2, in the principal component analysis for a total of 256 wavelength band spectrums obtained in the near infrared (NIR) region of 900-1,700 nm, the dispersion was the lowest when there were 5 principal components, and the correlation of the total 204 wavelength bands As a result of analyzing the relationship, it was found that there was no correlation in a specific region, similar to that in the UV region, but a correlation in a wide range. As a result of analyzing the measured germination rate and the predicted germination rate by spectrum using a calibration graph, the determination coefficient Was 0.88.

As can be seen in Figure 3, the principal component (principal component) analysis for the spectrum of a total of 460 wavelength bands obtained in the ultraviolet (UV) region of 200-400 nm and near-infrared (NIR) region of 900-1,700 nm is 6 days When the variance was the lowest, and a correlation analysis of a total of 460 wavelength bands was found, there was a correlation (5% level) in more than 250 wavelength bands. As a result of analyzing the calibration graph, the determination coefficient was 0.87.

As can be seen in Figure 4, the results of measuring the time of germination for 190 grains of brown rice showed that there is a difference in germination ability, that is, the germination capacity, which is the sprouting force for each granule, and thus it is necessary to compensate for binarization. I judged it.

Experimental Example 2: Improve germination rate measurement accuracy

In order to supplement the binarization analysis method for ambiguous values between 1 (germination) and 0 (unemergence) in germination rate analysis, embryo shape and color factor information were collected, and based on the collected factors, spectral analysis data and artificial intelligence program Through the effective application of the learning technique weights, the improvement of germination rate measurement technique was promoted.

The image of 300 grains of germinated or ungerminated brown rice is acquired with a COMS camera (Flea2, Flir, CAN) of 1280×960 resolution, and then the contour of the embryo area is extracted by considering color, position and shape characteristics. Was recognized, and the area, circumference, minor axis, long axis, long axis ratio, outline ratio, circularity, and color factor of the embryo were analyzed.

As a result of analyzing the shape and color characteristics, as can be seen in Table 1, significant differences were analyzed in the a and b values among the color characteristics such as area, circumference, and long and short axis.

Germination
(n=293) Area ^** (Area) Perimeter ^** (Peri
Meter) Shortened ^** (Minor axis)(A) Major axis ^* (B) Long and short axis ratio ^* (Aspect ratio) (A/B) Contour ratio Circularity
(Round
ness) L a ^** b ^*** Ungerminated brown rice
(n=105) 70338.2 ^a 462.4 ^a 112.7 ^a 691.2 ^a 0.17 ^a 3.2 4.2 94.1 0.5 ^a 1.9 ^a Germinated brown rice
(n=188) 76309.8 ^b 467.3 ^b 124.5 ^b 706.7 ^b 0.16 ^b 3.1 4.4 94.0 0.2 ^b 1.2 ^b

The artificial intelligence program developed a program for improving the germination rate measurement technology using a back propagation algorithm (C++ language, window-based interface). As a learning method, supervised learning rule was used, and the learning rate in progress was graphically expressed, and deviations and times according to learning were implemented, and a predicted value (Recall value) was implemented after the end of learning. It was developed as much as possible.

This program quantifies 10 factors, such as embryo shape and color characteristics, which are predicted to correlate with the germination rate, which is a quality factor, and uses them as network input data, and the learning target is germination 0 (non-emergence). And 1 (germination) was used, and learning (12,000 times of learning) was stopped when network errors converged in the learning process.

As a result of analyzing the effect of artificial intelligence application on germination rate accuracy for 300 samples, as shown in FIGS. 9A to 9C, the accuracy of the PSR (Partial Least Square Regression) analysis by NIR is 90.3%, and randomization It was confirmed that it is the cause of the deterioration of accuracy, and the accuracy of the image data and AI analysis was 74.3%, and there was a limit to measure the germination rate using image data alone.

As a result of analyzing by applying NIR data and image data as input data to the AI learning program, the germination rate accuracy was 95.9%. When NIR data and image data were applied to an artificial intelligence learning program, the accuracy was improved as the sample error distributed between germination (1) and non-emergence (0) decreased.

NIR data Video data + ANN analysis NIR data + video data + ANN analysis sample water 300 points 300 points 300 points Error sample 29 points 76 points 12 points Analysis accuracy 90.3% 74.3% 95.9%

Claims (8)

A sample sorting unit in which grain samples are aligned;
A light source unit including a light source;
A reflected light detector configured to detect reflected light reflected from the grain sample by a light source irradiated to the grain sample from the light source unit;
An image photographing unit for photographing an image of a grain sample;
The area, perimeter, minor axis, major axis, long and short axis ratio, contour ratio and circularity of the grain sample embryo section from the image obtained by the image capturing unit ( an image analysis unit analyzing one or more shape characteristics and color characteristics selected from the group consisting of roundness); And
Germination rate prediction system of a grain including a germination rate prediction unit for predicting the germination rate of the grain by applying the data obtained from the reflected light detection unit and the image analysis unit.
The system of claim 1, wherein the grain is selected from the group consisting of brown rice, mung bean, sorghum, yulmu, soybean, barley, green tea, corn, wheat, and oats.
The system of claim 1, wherein the light source is at least one light source selected from the group consisting of a light source having an ultraviolet wavelength range, a light source having a visible light wavelength range, and a light source having a near infrared wavelength range.
delete The system for predicting germination rate of grains according to claim 1, wherein the color property is a Lab color space value of a grain sample embryo.
The system of claim 1, wherein the data obtained from the reflected light detection unit and the image analysis unit is applied to an artificial intelligence learning program to predict the germination rate of grain.
A light source irradiation step of irradiating a light source to the grain sample;
A reflected light detection step of detecting reflected light reflected from the grain sample;
An image analysis step of analyzing shape characteristics and color characteristics from images of grain samples; And
A germination rate prediction method comprising a germination rate prediction step of predicting the germination rate of grain by applying the data obtained in the reflected light detection step and the image analysis step to an artificial intelligence learning program.
The method of claim 7, wherein the grain is selected from the group consisting of brown rice, mung bean, sorghum, yulmu, soybean, barley, green tea, corn, wheat, and oats.

Images