Potato leaf water content staging monitoring method based on optimized spectral band

Abstract

The invention relates to the field of crop irrigation technology and remote sensing, in particular to a potato leaf water content staging monitoring method based on an optimized spectral band. Comprising the following steps: acquiring hyperspectral measured data of leaves; preprocessing the hyperspectral measured data of the leaves respectively; carrying out fusion screening by adopting a continuous projection algorithm SPA and an interval variable iteration space contraction method IVISSA; the optimized spectral band combination is input into a seedling stage leaf water content SVM prediction model, and a leaf water content prediction value in the seedling stage is obtained; inputting into a tuber formation period leaf water content SVM prediction model to obtain a leaf water content prediction value in a tuber formation period; and inputting into a tuber expansion period leaf water content SVM prediction model to obtain a leaf water content prediction value in a tuber expansion period.

Description

Technical Field

[0001] The present invention relates to the field of crop irrigation technology and remote sensing, and in particular to a method for monitoring the moisture content of potato leaves by stages based on optimized spectral bands. Background Art

[0002] Potatoes are annual herbaceous plants belonging to the Solanaceae family. Their tubers are edible and they are the fourth most important food crop in the world, second only to wheat, rice and corn. Inner Mongolia is one of the main potato producing areas in my country. The region belongs to the northern part of my country with a dry climate and little rainfall, resulting in a serious shortage of water resources in the region. However, potatoes require a large amount of water, and there are significant differences in the demand for water in different growth periods. In particular, water shortages during the seedling stage, tuber formation stage and tuber swelling stage will directly lead to small tubers and small numbers, which will seriously affect their yield.

[0003] Against the backdrop of increasingly scarce water resources, how to achieve efficient utilization of limited water resources and achieve the goal of saving agricultural water has become a key technical issue that needs to be urgently addressed in the agricultural field. The traditional method of regular irrigation has technical defects such as serious waste of water resources and high costs, especially in the northern region. Due to the dry climate, sufficient sunshine hours, and fast evaporation rate of soil moisture, most of the irrigation water is lost by sunlight evaporation in a short period of time after adopting traditional regular irrigation, resulting in its technical effect on increasing potato yields being extremely limited. If the irrigation frequency is increased based on the water demand characteristics of the key growth stage of potatoes, and timely and appropriate irrigation is carried out in combination with real-time monitoring, it will not only effectively reduce the cost input of potatoes as an important food crop in the cultivation process, but also significantly improve water use efficiency, increase potato yield and quality, help reduce the cost of living of the people, and have significant economic and social benefits.

[0004] If there is a method that can monitor the moisture status of potato plants in real time during key growth stages and supply reasonable and appropriate amounts of water, it will be possible to ensure the coordinated realization of high yield and high quality of potatoes and efficient use of water, which is of great significance to solving the above problems. Summary of the Invention

[0005] In order to address the deficiencies in the prior art, the present invention provides a method for monitoring the moisture content of potato leaves in different stages based on optimized spectral bands.

[0006] The present invention provides

[0007] A method for monitoring potato leaf moisture content by stages based on optimized spectral bands comprises the following steps:

[0008] Step 1: Obtain leaf hyperspectral measured data;

[0009] Step 2: After the original band spectra of the hyperspectral measured data of the seedling stage, the tuber formation stage, and the tuber expansion stage canopy leaf hyperspectral measured data were preprocessed in sequence by SG smoothing, the first-order derivative of SG, and the second-order derivative of SG smoothing, the continuous projection algorithm SPA and the interval variable iterative space shrinkage method IVISSA were used for fusion screening to extract 66 optimized spectral bands for the seedling stage, 66 optimized spectral bands for the tuber formation stage, and 64 optimized spectral bands for the tuber expansion stage canopy leaf;

[0010] Step 3: 14 optimized characteristic band combinations for predicting leaf water content at the seedling stage were extracted from the 66 seedling-stage optimized spectral bands;

[0011] Step 4: Input the 14 seedling stage optimized characteristic band combinations into the seedling stage leaf water content SVM prediction model to obtain the leaf water content prediction value at the seedling stage;

[0012] Step 5: Optimize the spectral bands of 66 tuber formation stages and extract 19 optimized characteristic band combinations for predicting leaf water content during the tuber formation stage;

[0013] Step 6: Input the 19 optimized spectral band combinations during the tuber formation period into the SVM prediction model for leaf water content during the tuber formation period to obtain the predicted value of leaf water content during the tuber formation period;

[0014] Step 7, optimizing the spectral bands of 64 tuber bulking stages, and extracting 22 optimized characteristic band combinations for predicting the leaf water content during the tuber bulking stage;

[0015] Step 8: The optimized spectral wave combination of 22 tuber canopy leaves during the tuber expansion period is input into the SVM prediction model of leaf water content during the tuber expansion period to obtain the predicted value of leaf water content during the tuber expansion period.

[0016] Furthermore, the 14 seedling stage optimized characteristic band groups include: SGSD696, SGFD1503, SGSD1420, SGSD1203, SGFD1749, SGFD1082, SGSD699, SGFD1738, SGSD532, SGSD1429, SGSD1811, SGSD1636, SGFD1519, and SGFD1426.

[0017] Furthermore, the 19 optimized spectral band combinations for the tuber formation period include: SGFD555, SGFD556, SGSD509, SGSD538, SGFD1540, SGSD1422, SGFD1511, SGSD1603, SGSD1970, SGFD1730, SGFD1541, SGSD1796, SGFD1465, SGFD1731, SGSD1420, SGSD699, SGSD836, SGSD696, and SGSD341.

[0018] Furthermore, the optimized spectral wave combinations of 22 canopy leaves during the tuber expansion period include: SGSD1201, SGFD538, SGFD1712, SGSD1583, SGSD696, SGFD1068, SGFD1759, SGFD1503, SGSD1424, SGFD1573, SGSD1463, SGFD561, SGSD1812, SGFD531, SGSD551, SGFD1538, SGFD2049, SGSD514, SGSD2420, SGFD1424, SGFD1505, SGSD1720;

[0019] Furthermore, the SVM prediction model for leaf water content in the seedling stage, the SVM prediction model for leaf water content in the tuber formation stage, and the SVM prediction model for leaf water content in the tuber expansion stage were constructed specifically by the following method:

[0020] The hyperspectral measured data of canopy leaves at the seedling stage, tuber formation stage and tuber expansion stage, namely the hyperspectral reflectance and the corresponding water content data, were obtained respectively.

[0021] The original band spectra of the seedling stage, tuber formation stage, and tuber expansion stage were preprocessed with SG smoothing, SG first-order derivative, and SG smoothed second-order derivative to obtain the optimized spectral bands of each growth period.

[0022] The three pre-processed optimized spectral bands were screened for characteristic bands using the continuous projection algorithm (SPA) and the interval variable iterative space shrinkage method (IVISSA) for fusion screening. Separate combinations were made for each growth period, namely 66 optimized spectral bands for the seedling stage, 66 optimized spectral bands for the tuber formation stage, and 64 optimized spectral bands for the canopy and leaves during the tuber expansion stage.

[0023] Lasso regression analysis was performed on the optimized characteristic band combinations of each growth period, and the optimized characteristic band combinations were arranged in descending order according to the absolute value of the Lasso regression coefficient. 14 optimized characteristic band combinations for the seedling stage, 19 optimized spectral band combinations for the tuber formation stage, and 22 optimized spectral band combinations for canopy leaves during the tuber expansion stage were obtained.

[0024] The 14 seedling-stage optimized characteristic band combinations were used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral characteristics and moisture content, and obtained the seedling leaf moisture content SVM prediction model;

[0025] Nineteen optimized spectral band combinations during the tuber formation period were used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral features and moisture content, and obtained the SVM prediction model for leaf moisture content during the tuber formation period.

[0026] The optimized spectral wave combination of 22 canopy leaves during the tuber swelling period was used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral characteristics and moisture content, and obtained the SVM prediction model of leaf moisture content during the tuber swelling period.

[0027] Furthermore, the 66 seedling-optimized spectral bands, 66 tuber-forming-stage optimized spectral bands, and 64 tuber-swelling-stage canopy leaf-optimized spectral bands are combined in the following manner: 23 SG smoothed spectral bands, 20 SG smoothed first-order derivative bands, and 23 SG smoothed second-order derivative bands that are screened for high correlation with leaf water content and low collinearity in the seedling stage; 23 SG smoothed spectral bands R, 22 SG smoothed first-order derivative bands, and 21 SG smoothed second-order derivative bands that are screened in the tuber-forming stage; 22 SG smoothed spectral bands R, 20 SG smoothed first-order derivative bands, and 22 SG smoothed second-order derivative bands that are screened in the tuber-swelling stage; and each growth period is combined separately.

[0028] The beneficial effects achieved by the present invention are as follows: by adopting the technical solution of the present invention, the water status of potato plants in the three key growth periods of seedling, tuber formation and tuber swelling can be monitored in a timely and accurate manner according to the actual situation of water shortage in the region and the water requirement characteristics of potatoes in different growth periods, and precise irrigation control can be implemented according to the water shortage information. This is not only of great practical significance for the efficient water management of potatoes in the region, but also of great strategic significance.

[0029] The present invention preprocesses raw band data using SG smoothing, its first derivative, and its second derivative, facilitating accurate selection of optimized bands. Next, after initially integrating the SPA and IVISSA methods to select characteristic bands, duplicate bands are removed and the remaining characteristic bands are combined, further improving the performance of the optimized characteristic band combination. Finally, the optimized characteristic bands are sorted using Lasso regression features, gradually increasing the number of bands to determine the optimal number of characteristic bands that can accurately monitor potato leaf moisture content during three key growth stages. This method is characterized by the organic combination of data preprocessing, characteristic band screening, and Lasso regression feature sorting, forming a complete and systematic optimized characteristic band combination and monitoring method system that can meet the needs of potato leaf moisture content monitoring at different growth stages.

[0030] The present invention proposes to use support vector machine method to construct leaf water content monitoring models for potato in three growth stages respectively; the method is characterized by high accuracy, fast calculation and good prediction performance for new data. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is the prediction model for potato leaf water content at the seedling stage of the present invention and its verification results.

[0032] Figure 2 The present invention is a calculation code for the predicted value of water content of potato leaves at the seedling stage.

[0033] Figure 3 This is a prediction model for potato leaf water content during the tuber formation period of the present invention and its verification results.

[0034] Figure 4 The present invention provides a code for calculating the predicted value of water content of potato leaves during the tuber formation period.

[0035] Figure 5 This is a prediction model for potato leaf water content during the tuber swelling period and its verification results.

[0036] Figure 6 The present invention provides a code for calculating the predicted value of water content of potato leaves during the tuber expansion period.

[0037] Figure 7 This is a flowchart of the potato leaf moisture content stage monitoring method based on optimized spectral bands of the present invention. DETAILED DESCRIPTION

[0038] In the field of potato planting technology, there is still a lack of a set of precise irrigation methods based on scientific theories, with quantitative standards and systematic operating specifications. Specifically, due to the failure to accurately control the irrigation timing and accurately regulate the water volume according to the water demand patterns of potatoes in different growth stages, it is difficult to break through the bottleneck of potato yield. At the same time, the efficiency of water resource utilization is at a low level, making it difficult to achieve a coordinated improvement in agricultural production benefits and ecological benefits. Based on this, an embodiment of the present invention provides a potato leaf water content stage monitoring method based on optimized spectral bands. This method combines three original spectral band preprocessing methods to obtain optimized spectral bands, and for the first time integrates the continuous projection algorithm and interval variable The iterative space shrinkage method is used to screen the optimized characteristic bands, and Lasso regression is used to calculate the Lasso regression coefficient of each band. The optimized characteristic band combinations are arranged in descending order according to the absolute value of the coefficient. On the basis of gradually increasing the number of bands, the support vector machine method can be used to predict the leaf water content of potatoes in the seedling stage, tuber formation stage and tuber swelling stage in real time, quickly and losslessly. It can be used for accurate monitoring of the water content of potato plants in different growth stages, thereby realizing scientific irrigation and ensuring maximum per-acre yield. To facilitate understanding of this embodiment, the potato leaf water content stage monitoring method based on optimized spectral bands disclosed in the embodiment of the present invention is first introduced in detail.

[0039] like Figure 7 As shown, a potato leaf moisture content stage monitoring method based on optimized spectral bands can monitor the leaf moisture content of potatoes in key growth periods in real time, quickly and accurately, including the following steps:

[0040] Step 1: Obtain leaf hyperspectral measured data;

[0041] Hyperspectral measured data, also known as hyperspectral reflectance, can be acquired using a ground object spectrometer (SVC-1024i). This data includes seedling stage hyperspectral data, tuber formation stage hyperspectral data, and canopy and leaf hyperspectral data during tuber expansion. (How many supplementary basic data are there?)

[0042] Step 2, the original band spectra of the hyperspectral measured data of the seedling stage, the hyperspectral measured data of the tuber formation stage, and the hyperspectral measured data of the canopy leaves during the tuber expansion stage are respectively preprocessed by SG smoothing, the first derivative of SG, and the second derivative of SG smoothing;

[0043] Specifically, a Savitzky-Golay (SG) filter with a window size of 11 and a polynomial order of 2 was used to smooth and de-noise the raw spectral reflectance, effectively eliminating high-frequency noise interference. Subsequently, first- and second-order derivative transformations were applied to the raw spectral reflectance, highlighting the slope and curvature characteristics of the spectral curve through differential operations. SG smoothing was combined with first- and second-order derivative (FD) and SD transformations to generate three pre-processed optimized spectral bands: SG-smoothed spectral reflectance (R), SG-smoothed first-order derivative (SG+FD), and SG-smoothed second-order derivative (SG+SD). Furthermore, the SPA and IVISSA screening methods were integrated to remove duplicate bands and combine the remaining bands to extract 66 optimized spectral bands for the seedling stage, 66 optimized spectral bands for the tuber formation stage, and 64 optimized spectral bands for canopy leaves during the tuber expansion stage.

[0044] This application adopts three preprocessing methods: SG smoothing, the first-order derivative of SG smoothing, and the second-order derivative of SG smoothing, which can effectively eliminate high-frequency noise interference and highlight the slope change characteristics and curvature characteristics of the spectral curve, which is more conducive to the subsequent screening of characteristic bands; the fusion screening of SPA's "de-redundancy" capability and IVISSA's "precision" can cover sensitive bands in different growth periods, ensuring that the screened characteristic bands are representative in each growth period, while greatly reducing the data dimension, laying the foundation for simplifying the model.

[0045] Step 3: The 66 seedling-stage optimized spectral bands were sorted using Lasso regression features. The Lasso regression coefficients of each band were calculated and arranged in descending order. At the same time, the number of bands was gradually increased and the input was added to the 14 optimized feature band combinations. The accuracy of the support vector machine (SVM) prediction model for seedling leaf water content tended to be stable, that is, the 14 optimized feature band combinations for predicting seedling leaf water content were extracted.

[0046] Step 4: In the METLAB environment, the 14 seedling stage optimized characteristic band combinations were input into the seedling stage leaf water content SVM prediction model to obtain the leaf water content prediction value at the seedling stage;

[0047] The SVM prediction model for leaf water content at the seedling stage was constructed based on a combination of 14 optimized characteristic bands at the seedling stage;

[0048] The 14 seedling stage optimized characteristic band groups include: SGSD696, SGFD1503, SGSD1420, SGSD1203, SGFD1749, SGFD1082, SGSD699, SGFD1738, SGSD532, SGSD1429, SGSD1811, SGSD1636, SGFD1519, SGFD1426;

[0049] Among them, R represents the hyperspectral reflectance of the band after SG smoothing preprocessing; SGFD represents the first-order derivative of the reflectance of the band after SG smoothing first-order derivative preprocessing; SGSD represents the second-order derivative of the reflectance of the band after SG smoothing second-order derivative preprocessing; for example, SGSD696 represents the band data after SG smoothing and then second-order derivative preprocessing, and its central wavelength is 696nm; SGFD1503 represents the band data after SG smoothing and then first-order derivative preprocessing, and its central wavelength is 1503nm.

[0050] Step 5: 66 optimized spectral bands during the tuber formation period were sorted using Lasso regression features. The Lasso regression coefficients of each band were calculated and arranged in descending order. At the same time, the number of bands was gradually increased and the input was added to the 19 optimized feature band combinations. The accuracy of the support vector machine (SVM) prediction model for leaf moisture content during the tuber formation period tended to be stable, and the 19 optimized feature band combinations for predicting leaf moisture content during the tuber formation period were extracted.

[0051] Step 6: In the METLAB environment, the 19 optimized spectral band combinations for the tuber formation period were input into the SVM prediction model for leaf moisture content during the tuber formation period to obtain the predicted leaf moisture content during the tuber formation period, which was used for real-time, non-destructive, and accurate monitoring of the moisture status of potato plants.

[0052] The SVM prediction model for leaf water content during tuber formation was constructed based on a combination of 19 optimized spectral bands during tuber formation;

[0053] The 19 optimized spectral band combinations for tuber formation period include: SGFD555, SGFD556, SGSD509, SGSD538, SGFD1540, SGSD1422, SGFD1511, SGSD1603, SGSD1970, SGFD1730, SGFD1541, SGSD1796, SGFD1465, SGFD1731, SGSD1420, SGSD699, SGSD836, SGSD696, SGSD341;

[0054] Among them, R represents the hyperspectral reflectance of the band after SG smoothing preprocessing; SGFD represents the first-order derivative of the reflectance of the band after SG smoothing first-order derivative preprocessing; SGSD represents the second-order derivative of the reflectance of the band after SG smoothing second-order derivative preprocessing; for example, SGFD555 represents the band data after SG smoothing and then first-order derivative preprocessing, and its central wavelength is 555nm; SGSD509 represents the band data after SG smoothing and then second-order derivative preprocessing, and its central wavelength is 509nm.

[0055] Step 7: 64 optimized spectral bands during the tuber bulking period were sorted using Lasso regression features. The Lasso regression coefficients of each band were calculated and arranged in descending order. At the same time, the number of bands was gradually increased and the input was added to the 22 optimized feature band combinations. The accuracy of the support vector machine (SVM) prediction model for leaf water content during the tuber bulking period tended to be stable, that is, 22 optimized feature band combinations for predicting leaf water content during the tuber bulking period were extracted.

[0056] Step 8: In the METLAB environment, the optimized spectral wave combination of 22 tuber expansion period canopy leaves is input into the tuber expansion period leaf water content SVM prediction model to obtain the leaf water content prediction value of the tuber expansion period;

[0057] The SVM prediction model for leaf water content during tuber bulking was constructed based on a combination of 22 optimized spectral bands during tuber bulking;

[0058] The optimized spectral wave combinations of 22 canopy leaves during tuber expansion period include: SGSD1201, SGFD538, SGFD1712, SGSD1583, SGSD696, SGFD1068, SGFD1759, SGFD1503, SGSD1424, SGFD1573, SGSD1463, SGFD561, SGSD1812, SGFD531, SGSD551, SGFD1538, SGFD2049, SGSD514, SGSD2420, SGFD1424, SGFD1505, SGSD1720;

[0059] R represents the hyperspectral reflectance of the band after SG smoothing preprocessing; FD represents the first-order derivative of the reflectance of the band after SG smoothing first-order derivative preprocessing; SD represents the second-order derivative of the reflectance of the band after SG smoothing second-order derivative preprocessing; for example, SGSD1201 represents the band data after SG smoothing and then second-order derivative preprocessing, with a central wavelength of 1201nm; SGFD538 represents the band data after SG smoothing and then first-order derivative preprocessing, with a central wavelength of 538nm.

[0060] The SVM prediction model for leaf water content in the seedling stage, the SVM prediction model for leaf water content in the tuber formation stage, and the SVM prediction model for leaf water content in the tuber expansion stage were constructed by the following method:

[0061] The hyperspectral measured data of canopy leaves at the seedling stage, tuber formation stage, and tuber expansion stage, i.e., hyperspectral reflectance, and the corresponding water content data were obtained respectively. For the acquisition of hyperspectral measured data, a ground object spectrometer (SVC-1024i) was used. The hyperspectral reflectance of potato canopy leaves was measured between 10:00 and 14:00 Beijing time when the weather was clear and without cloud cover, and there was no wind or very low wind speed. The working range of the spectrometer was 337-2521nm. During the measurement process, a standard whiteboard was used to calibrate each group of targets before and after observation.

[0062] Then, the original band spectra of the seedling stage, tuber formation stage and tuber swelling stage were preprocessed with SG smoothing, the first-order derivative of SG smoothing and the second-order derivative of SG smoothing to obtain the optimized spectral bands of each growth period; based on this, the continuous projection algorithm (SPA) and the interval variable iterative space shrinkage method (IVISSA) were integrated to screen the characteristic bands of the three preprocessed optimized spectral bands, remove the duplicate bands, and combine the rest; 23 SG smoothed spectral bands (R) with high correlation and low collinearity with leaf water content in the seedling stage, 20 SG smoothed first-order derivative (SG+FD) bands, and 23 SG smoothed second-order derivative (SG+SD) bands with high correlation and low collinearity with leaf water content in the tuber formation stage were obtained. There are 23 low-collinear SG smoothed spectral bands (R), 22 SG smoothed first-order derivative (SG+FD) bands, and 21 SG smoothed second-order derivative (SG+SD) bands, which are highly correlated with leaf water content during the tuber swelling period. There are 22 low-collinear SG smoothed spectral bands (R), 20 SG smoothed first-order derivative (SG+FD) bands, and 22 SG smoothed second-order derivative (SG+SD) bands, that is, 66 optimized characteristic bands were determined in the seedling stage and tuber formation stage, respectively, and 64 optimized characteristic bands were determined in the tuber swelling period. Then, Lasso regression analysis was performed on the optimized characteristic band combinations of each growth period, and the optimized characteristic band combinations were arranged in descending order according to the absolute value of the Lasso regression coefficient.

[0063] This application utilizes SG smoothing, the first-order derivative of SG smoothing, and the second-order derivative of SG smoothing as preprocessing methods. It also employs the SPA and IVISSA fusion methods to screen feature bands, remove duplicate bands, and combine the remaining bands. The Lasso regression feature ranking method is further used to determine the optimal number of band combinations for constructing a potato leaf moisture content monitoring model for different growth stages. This results in a complete set of innovative technical solutions for monitoring potato leaf moisture content during different growth stages. This technical combination strategy, through a progressive optimization process of "preprocessing and purifying data → fusion and screening to extract sensitive features → Lasso refinement of core features," ultimately improves data quality, reduces model complexity, and enhances adaptability to different growth stages, thereby constructing a highly accurate and stable potato leaf moisture content monitoring model for different growth stages.

[0064] The first method is to build an SVM prediction model for leaf water content in the seedling stage:

[0065] The support vector machine (SVM) model is not in the form of a conventional equation, but is saved in the form of a file; the modeling method is as follows:

[0066] When the number of optimized characteristic bands was gradually increased and the Lasso regression coefficient absolute values ​​were input into the 1-14 seedling stage optimized characteristic band combinations (as shown in Table 1) in descending order, the accuracy of the SVM model for predicting potato seedling leaf water content tended to be stable, and the R values ​​of the training set and the test set were 0. 2 were 0.912 and 0.758, and the RMSE were 0.822 and 1.228 ( Figure 1 a, b), that is, the model constructed by combining 14 optimized characteristic bands can accurately predict the water content of potato plant leaves at the seedling stage;

[0067]

[0068] Among them, R represents the hyperspectral reflectance of the band after SG smoothing preprocessing; SGFD represents the first-order derivative of the reflectance of the band after SG smoothing first-order derivative preprocessing; SGSD represents the second-order derivative of the reflectance of the band after SG smoothing second-order derivative preprocessing.

[0069] The 14 seedling-stage optimized characteristic band combinations were used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral characteristics and moisture content, and obtained the SVM prediction model for seedling leaf moisture content.

[0070] The above model was verified using independent samples, and the coefficient of determination (R-squared) between the predicted value and the measured value was calculated to be 0.835, and the root mean square error (RMSE) was 0.741 ( Figure 1 c).

[0071] According to the above model, write the code as follows; when using it, bring the measured original band spectral reflectance into the "seedling stage hyperspectral data.xlsx" in the figure and run ( Figure 2 ), the corresponding leaf moisture content prediction value can be obtained for accurate real-time monitoring of the moisture status of potato plants in the seedling stage. The advantage of using this method is that it automatically performs three preprocessing processes on the original band: SG smoothing, the first-order derivative of SG smoothing, and the second-order derivative of SG smoothing. It automatically locks the 14 optimized characteristic band combinations of the seedling stage, and has high prediction accuracy and fast calculation for the potato seedling leaf moisture content.

[0072] The second method is to construct an SVM prediction model for leaf water content during the tuber formation period;

[0073] This method also uses the support vector machine method to build a model; the modeling method is as follows:

[0074] When the number of optimized characteristic bands was gradually increased and the Lasso regression coefficient absolute values ​​were input into the optimized characteristic band combinations of 1-19 tuber formation periods (as shown in Table 2), the accuracy of the SVM model for predicting leaf water content in potato tuber formation period tended to be stable, and the R values ​​of the training set and the test set were 0. 2 were 0.922 and 0.711, respectively, and the RMSE were 0.740 and 1.307 ( Figure 3 a, b), that is, the model constructed by combining 19 optimized characteristic bands can accurately predict the water content of potato plant leaves during the tuber formation period.

[0075]

[0076] Among them, R represents the hyperspectral reflectance of the band after SG smoothing preprocessing; SGFD represents the first-order derivative of the reflectance of the band after SG smoothing first-order derivative preprocessing; SGSD represents the second-order derivative of the reflectance of the band after SG smoothing second-order derivative preprocessing.

[0077] The optimized spectral band combinations of 19 tuber formation periods were used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral features and moisture content, and obtained the SVM prediction model of leaf moisture content in the tuber formation period.

[0078] The above model was verified using independent samples, and the coefficient of determination (R-squared) between the predicted value and the measured value was calculated to be 0.746, and the root mean square error (RMSE) was 0.771 ( Figure 3 c).

[0079] According to the above model, write the code as follows; when using it, bring the measured original band spectral reflectance into the "tuber formation period hyperspectral data.xlsx" in the figure and run ( Figure 4 ) can obtain the corresponding leaf moisture content prediction value, which is used for accurate real-time monitoring of the water status of potato plants during the tuber formation period. The advantage of using this method is that it automatically performs three preprocessings on the original bands: SG smoothing, the first-order derivative of SG smoothing, and the second-order derivative of SG smoothing, and automatically locks the 19 optimized characteristic band combinations of the tuber formation period, which has high prediction accuracy and fast calculation for the leaf moisture content of potato during the tuber formation period.

[0080] The third method is to construct an SVM prediction model for leaf water content during the tuber bulking period;

[0081] This method also uses the support vector machine method to build a model; the modeling method is as follows:

[0082] When the number of optimized characteristic bands was gradually increased and the Lasso regression coefficient absolute values ​​were input into the optimized characteristic band combinations of 1-22 tuber bulking periods (as shown in Table 3), the accuracy of the SVM model for predicting leaf water content in the tuber bulking period of potato tended to be stable, and the R values ​​of the training set and the test set were 0. 2 0.912 and 0.720 respectively, and RMSE are 0.846 and 1.356 respectively ( Figure 5 a, b), that is, the model constructed by combining 22 optimized characteristic bands can accurately predict the water content of potato plant leaves during the tuber swelling period.

[0083]

[0084] Among them, R represents the hyperspectral reflectance of the band after SG smoothing preprocessing; SGFD represents the first-order derivative of the reflectance of the band after SG smoothing first-order derivative preprocessing; SGSD represents the second-order derivative of the reflectance of the band after SG smoothing second-order derivative preprocessing.

[0085] The optimized spectral wave combination of 22 canopy leaves during the tuber swelling period was used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral characteristics and moisture content, and obtained the SVM prediction model of leaf moisture content during the tuber swelling period.

[0086] The above model was verified using independent samples, and the coefficient of determination (R-squared) between the predicted value and the measured value was calculated to be 0.801, and the root mean square error (RMSE) was 0.915 ( Figure 5 c).

[0087] According to the above model, write the code as follows. When using it, bring the measured original band spectral reflectance into the "tuber bulking period hyperspectral data.xlsx" in the figure and run ( Figure 6 ) can obtain the corresponding leaf moisture content prediction value, which is used for accurate real-time monitoring of the water status of potato plants during the tuber swelling period. The advantage of using this method is that it automatically performs three preprocessings on the original bands: SG smoothing, the first-order derivative of SG smoothing, and the second-order derivative of SG smoothing, and automatically locks the 22 optimized characteristic band combinations of the tuber swelling period, which has high prediction accuracy and fast calculation for the leaf moisture content of potato during the tuber swelling period.

[0088] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for monitoring potato leaf moisture content by stages based on optimized spectral bands, characterized in that: The following steps are included: Step 1: Obtain leaf hyperspectral measured data; Step 2: After the original band spectra of the hyperspectral measured data of the seedling stage, the tuber formation stage, and the tuber expansion stage canopy leaf hyperspectral measured data were preprocessed by SG smoothing, the first-order derivative of SG smoothing, and the second-order derivative of SG smoothing, the continuous projection algorithm SPA and the interval variable iterative space shrinkage method IVISSA were used for fusion screening to extract 66 seedling stage optimized spectral bands, 66 tuber formation stage optimized spectral bands, and 64 tuber expansion stage canopy leaf optimized spectral bands, respectively. Step 3: 14 optimized characteristic band combinations for predicting leaf water content at the seedling stage were extracted from the 66 seedling-stage optimized spectral bands; Step 4: Input the 14 seedling stage optimized characteristic band combinations into the seedling stage leaf water content SVM prediction model to obtain the leaf water content prediction value at the seedling stage; Step 5: Optimize the spectral bands of 66 tuber formation stages and extract 19 optimized characteristic band combinations for predicting leaf water content during the tuber formation stage; Step 6: Input the 19 optimized spectral band combinations during the tuber formation period into the SVM prediction model for leaf water content during the tuber formation period to obtain the predicted value of leaf water content during the tuber formation period; Step 7, optimizing the spectral bands of 64 tuber bulking stages, and extracting 22 optimized characteristic band combinations for predicting the leaf water content during the tuber bulking stage; Step 8: The optimized spectral wave combination of 22 tuber canopy leaves during the tuber expansion period is input into the SVM prediction model of leaf water content during the tuber expansion period to obtain the predicted value of leaf water content during the tuber expansion period.

2. The method for monitoring potato leaf moisture content by stages based on optimized spectral bands according to claim 1, characterized in that: The 14 seedling stage optimized characteristic band groups include: SGSD696, SGFD1503, SGSD1420, SGSD1203, SGFD1749, SGFD1082, SGSD699, SGFD1738, SGSD532, SGSD1429, SGSD1811, SGSD1636, SGFD1519, and SGFD1426.

3. The method for monitoring potato leaf moisture content by stages based on optimized spectral bands according to claim 1, characterized in that: The 19 optimized spectral band combinations for the tuber formation period include: SGFD555, SGFD556, SGSD509, SGSD538, SGFD1540, SGSD1422, SGFD1511, SGSD1603, SGSD1970, SGFD1730, SGFD1541, SGSD1796, SGFD1465, SGFD1731, SGSD1420, SGSD699, SGSD836, SGSD696, and SGSD341.

4. The method for monitoring potato leaf moisture content by stages based on optimized spectral bands according to claim 1, characterized in that: The 22 optimized spectral wave combinations for canopy leaves during the tuber bulking period include: SGSD1201, SGFD538, SGFD1712, SGSD1583, SGSD696, SGFD1068, SGFD1759, SGFD1503, SGSD1424, SGFD1573, SGSD1463, SGFD561, SGSD1812, SGFD531, SGSD551, SGFD1538, SGFD2049, SGSD514, SGSD2420, SGFD1424, SGFD1505, and SGSD1720.

5. The method for monitoring potato leaf moisture content by stages based on optimized spectral bands according to claim 1, characterized in that: The SVM prediction model for leaf moisture content in the seedling stage, the SVM prediction model for leaf moisture content in the tuber formation stage, and the SVM prediction model for leaf moisture content in the tuber expansion stage are specifically constructed by the following method: Obtain hyperspectral measured data of canopy leaves at the seedling stage, tuber formation stage, and tuber expansion stage, namely hyperspectral reflectance and corresponding water content data; The original band spectra of the seedling stage, tuber formation stage and tuber expansion stage were preprocessed by SG smoothing, SG smoothing first-order derivative and SG smoothing second-order derivative respectively to obtain the optimized spectral bands of each growth period. The three pre-processed optimized spectral bands were screened for characteristic bands using the continuous projection algorithm (SPA) and the interval variable iterative space shrinkage method (IVISSA) for fusion screening. Separate combinations were made for each growth period, namely 66 optimized spectral bands for the seedling stage, 66 optimized spectral bands for the tuber formation stage, and 64 optimized spectral bands for the canopy and leaves during the tuber expansion stage. Lasso regression analysis was performed on the optimized characteristic band combinations of each growth period, and the optimized characteristic band combinations were arranged in descending order according to the absolute value of the Lasso regression coefficient. 14 optimized characteristic band combinations for the seedling stage, 19 optimized spectral band combinations for the tuber formation stage, and 22 optimized spectral band combinations for canopy leaves during the tuber expansion stage were obtained. The 14 seedling-stage optimized characteristic band combinations were used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral characteristics and moisture content, and obtained the seedling leaf moisture content SVM prediction model; Nineteen optimized spectral band combinations during the tuber formation period were used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral features and moisture content, and obtained the SVM prediction model for leaf moisture content during the tuber formation period. The optimized spectral wave combination of 22 canopy leaves during the tuber swelling period was used as input and combined with the actual measured leaf moisture content data. Through training, the model learned the relationship between spectral characteristics and moisture content, and obtained the SVM prediction model of leaf moisture content during the tuber swelling period.

6. The method for monitoring potato leaf moisture content by stages based on optimized spectral bands according to claim 5, characterized in that: The 66 seedling-optimized spectral bands, 66 tuber-forming-stage optimized spectral bands and 64 tuber-swelling-stage canopy leaf-optimized spectral bands are combined in the following manner: 23 SG smoothed spectral bands, 20 SG smoothed first-order derivative bands, and 23 SG smoothed second-order derivative bands screened in the seedling stage with high correlation and low collinearity with their leaf water content; 23 SG smoothed spectral bands R, 22 SG smoothed first-order derivative bands, and 21 SG smoothed second-order derivative bands screened in the tuber-forming stage; 22 SG smoothed spectral bands R, 20 SG smoothed first-order derivative bands, and 22 SG smoothed second-order derivative bands screened in the tuber-swelling stage; and each growth period is combined separately.

Images