[0026] reference figure 1 As shown in the flow chart of the steps, the embodiment of the present invention provides a method for multispectral estimation of tomato chlorophyll content, which takes greenhouse-grown tomatoes as the research object, through crop canopy multispectral image acquisition, maximum leaf area leaf extraction, and reflectivity Reconstruction, multi-spectral image feature definition and selection, partial least squares modeling, and finally a set of tomato chlorophyll content estimation method is proposed. The method includes the following specific steps:
[0027] (1) Place the standard emission board and the leaves in the same field of view, and use the camera to obtain 8 multispectral images of different wavebands on the crop canopy in the same scene according to the 8 wavebands sensitive to crop chlorophyll content.
[0028] Among them, the 8 bands sensitive to crop chlorophyll content are 520, 560, 640, 670, 700, 760, 800 and 870nm respectively. The reason is: according to the inventor’s previous spectrum test, the correlation analysis between the spectrum de-envelope line and the chlorophyll content was carried out, and the results are as follows figure 2 As shown, the study found that the blue edge (about 520 nm), yellow edge (about 640 nm), red edge (about 700 nm), green peak (about 560 nm), red valley (about 670 nm), near infrared 760, There are extreme points at 800 and 870nm. Therefore, the above-mentioned 8 bands are selected to construct a multi-spectral imaging system to better reflect the chlorophyll content, so as to overcome the disadvantages of traditional digital image data sets that are limited and may not be able to well characterize the nutritional status of crops.
[0029] Specifically, the Labsphere MS100 standard reflector and the blade are placed in the same field of view, and the FLI MLx205 camera is used with the CFW-1-8-28 automatic filter wheel to obtain 8 crop crowns in the same scene under different wavelengths. Layer image (ie, multispectral image).
[0030] Table 1 shows the average reflectivity of four color blocks (T1, T2, T3, T4) in the R, G, B, and NIR channels of the standard reflector from light to dark.
[0031]
[0032] (2) Select the leaf with the largest leaf area among the crop canopy leaves and determine its chlorophyll content.
[0033] Specifically, the leaf veins were removed, cut into pieces, placed in a 2:1 extract solution of acetone and absolute ethanol, and after soaking in a dark place for 24 hours, the absorbance value at 652nm was measured with a 752 UV-visible spectrophotometer. press Calculate the chlorophyll content. Among them, the chlorophyll content can understand the plant's nitro demand, and can also calculate the nitrogen fertilizer content. In addition, it can be seen from the correlation between the steps that the determination of the chlorophyll content actually only needs to be completed before step (5).
[0034] (3) Extract the leaf with the largest leaf area in each multispectral image, and for the three color blocks T2~T4 in the standard reflector, reconstruct the reflectance of the leaf with the largest leaf area corresponding to each multispectral image and get the average value. .
[0035] For the extraction of leaves with the largest leaf area, first perform morphological gradient calculation on each multispectral image, and then use the watershed algorithm based on wavelet transform to extract the leaves with the largest leaf area in each multispectral image; see below for details.
[0036] ① Morphological gradient calculation
[0037] Select the NIR component for morphological gradient calculation,
[0038]
[0039] among them, Is a morphological gradient image, Is the original image, Is a morphological structural element, Is the morphological expansion operation, Corrosion calculation for morphology. As a result, a large amount of internal leaf textures caused by leaf veins, light, etc. are removed.
[0040] ② Wavelet marking
[0041] First, use db4 wavelet function to perform 4-layer wavelet decomposition on the NIR image from the complex background, take the low-frequency coefficients to reconstruct the approximate image, and then go through the H-maxima transform (H maximum value transform), and finally select the threshold value of 18, which will be greater than 18 The maximum value is used as a mark.
[0042] ③Wavelet watershed transform and maximum area leaf extraction
[0043] Because the chlorophyll determination is for the leaves with the largest leaf area, the tomato canopy images have many and overlapping leaves. Therefore, it is necessary to use an appropriate image processing method to find the chlorophyll measurement leaves in the image as accurately as possible. Based on wavelet markers, the marker watershed leaves are segmented and extracted, and the extracted leaves are counted through pixels to obtain the leaves with the largest leaf area. Afterwards, the reflectance of the found blades is reconstructed, characteristic parameters are extracted, and modeling is performed. In this way, the measurement scale matches the modeling scale, and the model is credible.
[0044] For reflectivity reconstruction, considering that in the actual shooting process, the T1 color block of the Labsphere MS100 standard reflector is prone to oversaturation, so the reflectivity reconstruction is performed for the three color blocks T2, T3, T4 (that is, the The pixel gray value is transformed into relative reflectivity, the purpose of this is to reduce the adverse effect of uneven illumination on prediction). Specifically, an empirical linear formula is established according to the reflectance of the three color blocks and the gray level of the image respectively, and the gray value of each channel of the image is converted into the relative reflectance and the average value is calculated.
[0045] (4) A normalized vegetation index is constructed by combining 8 average reflectances in pairs, and the 28 vegetation indices obtained together with the 8 average reflectances are defined as 36 multispectral image features.
[0046] Specifically, while taking the average reflectances of the 8 bands as image features, 28 normalized vegetation indices are also constructed based on the pairwise combination of these 8 average reflectances, and a total of 36 multispectral image features are defined in total.
[0047] (5) Perform correlation analysis on 36 multi-spectral image features and chlorophyll content, and select multi-spectral image features with a correlation of more than 0.75 as the multi-spectral feature of chlorophyll content.
[0048] Among them, the correlation of more than 0.75 means that the corresponding multi-spectral image features can well represent the change of leaf chlorophyll content.
[0049] (6) Use the multi-spectral image feature of the chlorophyll content to perform partial least square regression modeling to obtain the chlorophyll content estimation model, and use the chlorophyll content estimation model to estimate the chlorophyll content.
[0050] Based on the disclosed method for multispectral estimation of tomato chlorophyll content, experiments were conducted on tomatoes in the fruiting period of greenhouse cultivation. The test data are as follows.
[0051] (I) Conditions: During 14:00-16:00 in the afternoon, under natural light conditions, use Labsphere MS100 standard reflector, FLI MLx205 camera and CFW-1-8-28 automatic filter wheel.
[0052] (II) Data analysis and modeling
[0053] Follow the steps (1) to (6) above. In step (5) multi-spectral image feature selection, the 36 multi-spectral image features are correlated with the chlorophyll content measured in step (2). The result is Such as image 3 As shown, the 4 features with correlation coefficients exceeding 0.75 are selected as input parameters of the chlorophyll estimation model, and they are denoted as NDVI 640,760 , NDVI 700,760 , NDVI 700,800 , NDVI 700,870.
[0054] Refer to step (6) to use the 4 features selected above to perform partial least squares regression modeling. The test data are shown in Table 2 below. Among them, the first 40 sets of test data are used for data modeling, and the last 33 sets of data are used for model verification. ; Modeling results such as Figure 4 Shown. Specifically, starting from a calibration data set composed of 40 sets of experimental data, a tomato chlorophyll content estimation model Spad=-8.23-38.95×NDVI was constructed 640,670 +470.63×NDVI 700,760 -616.51×NDVI 700,800 +233.04×NDVI 700,870 , The model correction determination coefficient Rc 2 It is 0.87; 33 sets of test data constitute the verification data set, and the model verification is performed to verify the determination coefficient Rv 2 Is 0.84. The model has high accuracy and meets the requirements of lossless, fast and efficient estimation.
[0055] Table 2 Test data
[0056] Chlorophyll content
[0057] The technical solution provided by the present invention is described in detail above. Specific examples are used in this article to illustrate the principle and implementation of the present invention. The description of the above examples is only used to help understand the method and core idea of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.
