CN102759510A - Spectral detection method of rape canopy information - Google Patents

Abstract

The invention discloses a rapeseed canopy information spectral detection method, which utilizes an information detection mode integrating multi-channel spectra, uses a pre-trained BP neural network to establish a model for the spectral information collected from the rapeseed canopy, and obtains a SPAD value And total nitrogen content, after testing, the prediction performance and effect of the model are better. The present invention utilizes the information detection mode that integrates multi-channel spectrum, not only will not cause mutual interference between spectral channels, but also some spectral channel information detection that has no obvious effect can also be used as an auxiliary channel for light correction and environmental impact correction. The BP neural network model improves the accuracy of spectrum detection of rapeseed canopy information, the detection method of the invention has high precision, convenient collection of sample information, and is suitable for agricultural popularization and application.

Description

A method for spectral detection of rapeseed canopy information

technical field

The invention relates to a plant information spectrum detection method, in particular to a rapeseed canopy information spectrum detection method. the

Background technique

Plant chlorophyll and total nitrogen content are the main components of plant nutrition and the most important factors related to plant photosynthesis. Chlorophyll is all organisms that can be transformed into plant nutrition photosynthesis. Total nitrogen is one of the essential nutrients for plants. It is the main component of substances such as amino acids, proteins, alkaloids, nucleic acids and chlorophyll. Their content can directly or indirectly reflect the vital signs of plants, so it is of great significance to study and detect the content of these parameters in plants. the

The color of almost all objects in nature is mainly determined by the spectral reflection characteristics in the 400-700nm region. For normal growing plants, the color of the leaves is determined by the spectral characteristics of chlorophyll, which has a strong reflection of green light, so the leaves are green. Chlorophyll is an important component of plant nitrogen content, and the depth of crop canopy color can reflect the level of total nitrogen metabolism in the plant. The spectral reflectance of different plants in the visible and near-infrared bands showed different degrees of increase under nitrogen deficiency. Scholars at home and abroad have studied various crops through different sensors and spectral data and have shown that spectral indices can be used to estimate the chlorophyll and total nitrogen content of major crops and vegetables such as wheat, corn, rice, and cotton. In summary, the reflectance spectrum and color characteristics in the visible region can be used to estimate the chlorophyll and total nitrogen content of crop leaves. the

The invention patent application with the publication number CN1746660A discloses a new method and a designed measuring instrument using the canopy reflection spectrum crop canopy pigment ratio, which can quickly and conveniently measure the SIPI value of the crop, and accurately measure the crop canopy characteristics. The evaluation of pigment ratio plays an important role in judging crop growth and guiding the use of nitrogen fertilizer. Using sunlight as the light source, through six identical photoelectric sensors with special spectral response characteristics, the incident light of sunlight and the reflected light of vegetation are respectively detected at three characteristic wavelengths of near infrared, red light and blue light, and the measured signal After A/D conversion, the microcontroller calculates the SIPI value according to the calculation formula of the SIPI value, and then calculates the result representing the growth state of the crop according to the SIPI value, and the obtained result is displayed on the liquid crystal display. The measurement results can be saved in the instrument, and can be transmitted to the PC through the RS232 serial port for further analysis. The detection method and the instrument have lower requirements on daylight lighting conditions, simple structure, light weight, low cost and convenient use, and are suitable for mass production and application. the

The invention patent application with the publication number CN101403689A discloses a non-destructive detection method for physiological indicators of plant leaves based on visible-near infrared spectroscopy, which can quickly and simultaneously detect the contents of chlorophyll, nitrogen, lutein, water and other components . The invention collects the spectrum of the calibration set samples, and establishes a calibration model between the spectral value and the plant component content standard after preprocessing the spectral data and optimizing the band; collects the spectrum of the unknown sample, and after processing the spectral data, the The selected band data is substituted into the calibration model to predict the content of the component to be measured. The technical scheme of the invention uses full-spectrum information, the measured parameters are highly scalable, and the prediction accuracy and model adaptability of the correction model are improved; the transmission-reflection detection method adopted in the invention increases spectral sensitivity, and is more adaptable to leaf types. Strong; the invention uses an improved wavelet analysis method to simultaneously perform noise removal and baseline correction preprocessing on the leaf spectral data, which can effectively improve the prediction accuracy. the

The detection effect in the prior art is susceptible to systematic errors caused by detection instruments and the like. the

Contents of the invention

The invention discloses a rape canopy information spectrum detection method, which solves the problem in the prior art that the detection effect is easily affected by system errors caused by detection instruments and the like. the

A rapeseed canopy information spectral detection method, comprising the following steps:

A, collect the spectral information of some rape leaf samples, obtain its SPAD value;

B, for the rape leaf sample in step A, detect its total nitrogen content;

C, with the spectral information in step A as input variable, with the SPAD value corresponding to described spectral information and total nitrogen content as output variable, set up neural network model;

D. Collect the spectral information of the rapeseed to be detected and use the model to obtain the SPAD value and total nitrogen content of the rapeseed to be detected, so as to determine the growth state of the rapeseed to be detected. the

The SPAD value is a parameter to measure the relative content of chlorophyll in a plant, and is a sign of chlorophyll content. the

Total nitrogen content refers to the total nitrogen content in plants. Nitrogen is an essential macroelement for plants and a component of important living substances such as proteins, chlorophyll, nucleic acids, enzymes, and biological hormones. Different crops require different amounts of nitrogen in the same period, and the same crop The amount of nitrogen required in different periods is different. At the same time, different nitrogen application treatments have different effects on crop nitrogen absorption and yield. The detection of total nitrogen content has a great impact on crop growth and yield. the

The spectral information in the step A and step D is the spectral information of some leaves of the canopy of rapeseed. The detection of canopy information is convenient for instrument operation, and it is beneficial for the instrument to dynamically and quickly detect the spectral information of plants. the

Unless otherwise specified, the following spectral information generally refers to the spectral information of some leaves in the rape canopy. the

In step A and step D, the probe light of 450-1200nm is emitted to the rape leaf, and the probe light reflected by the rape leaf is collected to obtain the corresponding spectral information, and the SPAD value and the total nitrogen content are detected in the range of 450-1200nm. Sensitive characteristic effect, the information in this area is very helpful for the model detection of rape leaf SPAD value and total nitrogen content. the

16 channels of the probe light in the range of 450-1200nm, the wavelengths of the 16 probe lights are 450nm, 480nm, 550nm, 640nm, 680nm, 720nm, 780nm, 820nm, 860nm, 880nm, 940nm, 960nm, 1040nm, 1100nm, 1150nm , 1200nm. the

The neural network model is a BP neural network model. the

In the present invention, the rape leaf sample information is collected first, and the SPAD value and the total nitrogen content value of the rape leaf sample are measured and calculated by an instrument. Rapeseed leaf samples are scanned for spectral data, and the reflectance of each band in the characteristic band is recorded, and the spectral reflectance (that is, the spectral information) is used as the input variable of the neural network, and the corresponding SPAD value obtained by instrument measurement and The total nitrogen content was used as the output variable to establish the BP neural network model. the

After the BP neural network model is established, in order to improve its reliability, it generally needs to be trained. Through training, the weights and thresholds in the corrected BP neural network are obtained, and the error between the measured value of the instrument and the actual output of the BP neural network model is calculated. If the error is not greater than the minimum expected error, or the maximum number of cycles has been reached, the training ends, otherwise continue. the

Compared with prior art, the beneficial effect of the present invention is:

(1) The BP neural network obtains its contribution rate to the parameter by modifying the weight value of each input variable on the output, and the input variable that has nothing to do with the output will be trained by the BP network and its corresponding weight will be reduced to It is close to 0, which improves the accuracy of spectral detection of rapeseed canopy information, and is beneficial for managers to fertilize and cultivate correctly. the

(2) Using the information detection mode integrating multi-channel spectrum and adopting BP neural network modeling method, the impact of environmental factors on the accuracy and reliability of detection can be minimized, and not only will there be no interaction between spectral channels interference, and some spectral channel information detection that has no obvious effect can also be used as an auxiliary channel for light correction and environmental impact correction. It is helpful to improve the adaptability of the instrument. the

Description of drawings

Fig. 1 is the relation diagram of the SPAD value of the method of the present invention and the SPAD value of the instrument measurement. the

Fig. 2 is a graph showing the relationship between the total nitrogen content obtained by the method of the present invention and the total nitrogen content obtained by instrumental measurement. the

Detailed ways

The invention discloses a rapeseed canopy information spectrum detection method, which is used for measuring the SPAD and total nitrogen content of the rapeseed, thereby determining the growth state of the plant. the

model building

(1) Select 60 plots as the training sample set at the green leaf stage of rapeseed, carry out the canopy spectrum experiment on the plot, and select 15 plots as the prediction sample set in addition, use the spectral detection probe of the rapeseed canopy information spectral detector to detect each Canopy spectrum scanning is carried out on the rape canopy of each plot. The spectral information reflected by sunlight on the rape canopy first passes through 16 spectral detection information channels, and then the light beam is focused on the center of the filter through the optical channel of the focusing mirror. The light sheet transmits the reflectance spectrum information of a specific band to the photoelectric sensor. After the weak signal is amplified and processed, the photoelectric sensor obtains the 16-band spectral reflectance value of the rapeseed canopy in each plot, and the DSP processor will obtain the 16-band spectrum The reflection value is displayed on the display unit of the MCU through the interface circuit. the

In order to implement the method of the present invention, can adopt rapeseed canopy information spectral detection instrument, comprise spectral detection head and be used to receive and process the MCU of output signal of spectral detection head, MCU is also respectively connected with instruction input unit, display unit, storage unit and wireless transceiver unit. the

The spectral detection head includes a spectral detection information channel, a filter, a photoelectric sensor arranged in sequence on the optical path, and a DSP processor connected to the photoelectric sensor circuit, and the DSP processor is connected to the MCU through an interface circuit. A preprocessing unit for conditioning and amplifying the output signal of the photoelectric sensor is arranged between the photoelectric sensor circuit and the DSP processor. the

(2) measure each plot canopy SPAD value 30 times by SPAD-502 detection instrument for each plot, obtain the SPAD value of this plot by calculating the mean value, by picking the rape leaves in each plot area, by Kjeldahl The nitrogen method measures its total nitrogen content. the

Among them, SPAD-502 is a chlorophyll meter produced by Japan MINOLTA Company. This detection instrument measures the chlorophyll content of plants by measuring the spectral absorption characteristics of plant leaves. The Kjeldahl method is a method for determining the total nitrogen in a compound or mixture, that is, under the condition of a catalyst, the sample is digested with concentrated sulfuric acid to convert the organic nitrogen into an inorganic ammonium salt, and then the ammonium is converted to an inorganic ammonium salt under alkaline conditions. The salt is converted into ammonia, distilled out with water vapor and absorbed by excess acid solution, and then titrated with standard alkali to calculate the amount of nitrogen in the sample. the

(3) Use the 16-band spectral reflectance values in the training sample set in step (1) as input variables, and use the SPAD value and total nitrogen content measured by the detection instrument in step (2) as output variables to establish a BP neural network model. the

In the BP neural network, there are 16 input neurons, which are the characteristic bands 450nm, 480nm, 550nm, 640nm, 680nm, 720nm, 780nm, 820nm, 860nm, 880nm, 940nm, 960nm, 1040nm, Spectral reflectance values at 1100nm, 1150nm, and 1200nm, take the reflectance values of 16 bands as x ₁ , x ₂ ... x ₁₆ 16 input variables, and the SPAD value and total nitrogen content measured by the detection instrument as y ₁ , y ₂ 2 Output neuron, after the data is standardized, the Sigmoid parameter in the BP network is selected as 0.9, 6 hidden neurons are selected, and the training is performed 1000 times. The 15 samples are listed in Table 1. Its fitting residual value is 0.00019.

Among them, SY1 and SY2 are the actual measured values, which are used to compare with the results calculated by the instrument model. the

NO. _{x1 x2} ..... X ₁₆ Y ₁ sY ₁ Y ₂ Sy ₂ 1 0.52 0.46  … 0.47 21.3 20.1 2.56 2.33 2 0.64 0.57  … 0.64 25.6 24.4 3.78 3.47 3 0.54 0.42  … 0.54 32.4 31.8 4.28 5.16 4 0.71 0.56  … 0.41 33.2 33.9 6.63 4.93 5 0.69 0.49  … 0.45 35.9 34.3 7.48 7.58 6 0.56 0.34  … 0.67 35.1 35.8 8.14 7.82 7 0.48 0.39  … 0.56 38.4 39.2 7.45 7.35 8 0.34 0.30  … 0.32 41.5 42.1 12.58 13.18 9 0.38 0.27  … 0.54 42.8 43.4 13.36 15.02 10 0.24 0.20  … 0.28 52.3 51.8 16.25 15.12 11 0.36 0.27  … 0.35 52.0 52.6 14.36 14.73 12 0.17 0.15  … 0.62 58.9 57.4 15.25 14.79 13 0.23 0.11  … 0.37 55.4 57.2 13.32 12.80 14 0.15 0.10  … 0.25 63.7 61.6 14.61 13.58 15 0.26 0.12  … 0.37 50.1 47.5 13.32 14.51

Table 1

(4) After the neural network model training process, establish the detection model of the instrument, use the 16 band spectral reflectance values of the predicted sample set in step (1) as the input variables of the neural network, and use the BP neural network obtained from the training samples to obtain the prediction The SPAD value and total nitrogen content of the sample set are compared with the SPAD value and total nitrogen content measured by the instrument of the predicted sample set obtained in step (2). The relationship obtained is shown in Figure 1 and Figure 2. the

The correlation coefficient R ² between the predicted value and the measured value in the calculation of the model SPAD is 0.9361, and the correlation coefficient between the predicted value and the measured value of the total nitrogen content is R ² = 0.8237. nutrient testing requirements.

The correlation coefficient is obtained by using the least squares-support vector machine. The least squares-support vector machine is based on structural risk minimization, which can better improve the generalization ability of the learning machine, and can still obtain small Error mathematical statistics method. The closer the correlation coefficient of the model is, the smaller the root mean square error R is, and the better the predictive ability of the model is. the

Embodiment 1～2

After completing the establishment of the model, start the growth state of measured rape, select 15 sub-districts, carry out canopy spectrum scanning to the rape canopy of each sub-district with the spectral detection probe of rape canopy information spectrum detector, apply the method of the present invention to record The SPAD value and total nitrogen content of each plot were compared with the SPAD value and total nitrogen content obtained by SPAD502 and Kjeldahl method, and some data obtained are shown in Table 2 and Table 3. the

Among them, Y1 is the SPAD value calculated by using the BP network model, SY1 is the SPAD value obtained by using the SPAD502, Y2 is the total nitrogen content calculated by the BP network model, and SY1 is the total nitrogen content obtained by the Kjeldahl method. the

Table 2 (Example 1)

Community serial number _{x1 x2}  … X ₁₆ Y ₁ sY ₁ Y ₂ Sy ₂ 1 0.32 0.26  … 0.42 48.3 48.1 12.56 12.33 2 0.23 0.32  … 0.54 55.8 54.7 14.24 14.29 3 0.24 0.37  … 0.66 41.4 31.8 9.28 8.76 4 0.75 0.58  … 0.46 33.5 34.1 6.67 5.03 5 0.68 0.51  … 0.44 36.2 34.8 7.51 7.62 6 0.59 0.36  … 0.65 35.6 36.1 8.21 7.89 7 0.43 0.36  … 0.59 38.6 39.1 8.39 8.36 8 0.34 0.30  … 0.32 41.5 42.1 12.58 13.18 9 0.28 0.34  … 0.45 35.7 34.9 7.27 7.68 10 0.37 0.42  … 0.58 22.4 21.8 4.25 4.02 11 0.36 0.27  … 0.35 52.0 52.6 14.36 14.73 12 0.47 0.23  … 0.42 28.2 27.9 5.25 5.42 13 0.59 0.61  … 0.67 25.4 27.2 5.32 5.80 14 0.11 0.14  … 0.25 65.9 66.7 16.21 16.58 15 0.14 0.20  … 0.22 55.1 56.5 15.18 15.05

Table 3 (embodiment 2)

Community serial number _{x1 x2} ..... X ₁₆ Y ₁ sY ₁ Y ₂ Sy ₂ 1 0.72 0.66  … 0.77 7.1 6.9 0.87 0.93 2 0.68 0.65  … 0.72 8.3 8.9 1.23 1.31 3 0.59 0.52  … 0.64 12.4 11.8 2.05 2.13 4 0.61 0.53  … 0.55 14.6 13.9 3.68 3.91 5 0.62 0.59  … 0.47 15.9 16.2 4.28 4.51 6 0.56 0.49  … 0.62 15.2 15.8 4.76 4.89 7 0.44 0.49  … 0.42 22.6 23.1 5.43 5.15 8 0.41 0.45  … 0.52 24.5 24.7 6.08 6.12 9 0.39 0.23  … 0.51 39.8 38.7 9.36 9.01 10 0.24 0.20  … 0.28 52.3 51.8 16.25 15.12 11 0.36 0.27  … 0.35 52.0 52.6 14.36 14.73 12 0.21 0.35  … 0.37 48.9 48.1 13.27 13.87 13 0.29 0.17  … 0.22 58.1 58.7 14.19 14.27 14 0.21 0.17  … 0.25 61.4 61.9 15.01 15.87 15 0.23 0.28  … 0.17 52.1 51.5 13.79 14.21

As can be seen from Table 2 and Table 3, the SPAD value obtained by the present invention and the total nitrogen content are all very small in relation to the measured values obtained by using the SPAD502 instrument and the Kjeldahl nitrogen measurement method, which meets the nutrient detection requirements of crops for agricultural fertilizer and water management. the

Claims (6)

1. a rapeseed canopy information spectral detection method, is characterized in that, comprises the following steps: A, collect the spectral information of some rape leaf samples, obtain its SPAD value; B, for the rape leaf sample in step A, detect its total nitrogen content; C, with the spectral information in step A as input variable, with the SPAD value corresponding to described spectral information and total nitrogen content as output variable, set up neural network model; D. Collect the spectral information of the rapeseed leaf to be detected and use the model to obtain the SPAD value and total nitrogen content of the rapeseed leaf to be detected, so as to determine the growth state of the rapeseed to be detected. 2. The rapeseed canopy information spectrum detection method as claimed in claim 1, characterized in that, the spectral information in said step A and step D is the spectral information of rapeseed canopy part leaves. 3. rape canopy information spectrum detection method as claimed in claim 2, is characterized in that, in step A and step D, to the probe light of rape blade launch 450-1200nm, gathers the probe light after the reflection of rape blade, obtains Corresponding spectral information. 4. rape canopy information spectrum detection method as claimed in claim 3, is characterized in that, described detection light is 16 paths in the range of 450-1200nm. 5. rape canopy information spectrum detection method as claimed in claim 4, is characterized in that, described neural network model is BP neural network model. 6. rape canopy information spectrum detection method as claimed in claim 5, is characterized in that, described 16 road probe light wavelengths are respectively 450nm, 480nm, 550nm, 640nm, 680nm, 720nm, 780nm, 820nm, 860nm, 880nm, 940nm, 960nm, 1040nm, 1100nm, 1150nm, 1200nm.

Images