Pineapple hydroheart disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data

Abstract

The invention relates to the field of agricultural image processing and computer vision, in particular to a pineapple hydroheart disease morbidity prediction method based on unmanned aerial vehicle multi-source remote sensing data, and the technical scheme comprises the steps: obtaining a canopy RGB image, a multispectral image and LiDAR point cloud data of a pineapple in a growth period based on an unmanned aerial vehicle platform, so as to extract more comprehensive plant feature information; the pineapple nutrition parameters are inverted by using the multi-source remote sensing data, and the pineapple hydroheart disease incidence prediction model is constructed by mining the quantitative relationship between the nutrition parameters and the pineapple hydroheart disease incidence, so that the precision and robustness of the pineapple nutrition parameter inversion model are remarkably improved, and the large-area nondestructive detection of the pineapple hydroheart disease incidence is realized. The method has the advantages of high efficiency, flexibility, low cost, high data acquisition speed and the like, the plant nutrition state is reflected in time through images, and an effective means is provided for plant disease detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural image processing and computer vision, and particularly relates to a pineapple water core disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data. BACKGROUND

[0002] Water core disease is a physiological disease that occurs universally in the current pineapple industry, and has become a major limiting factor for determining the quality of pineapple fruits, and an important factor affecting the healthy development of China's pineapple industry. The occurrence of pineapple water core disease is affected by natural and human factors, such as temperature changes before harvest, fruit sunburn, excessive rainfall, and excessive use of nitrogen fertilizer, which can all lead to the occurrence of pineapple water core disease. The nutritional status of pineapple plants is closely related to the incidence of water core disease, especially in the case of excessive nitrogen nutrition, the risk of disease occurrence increases significantly. Traditional monitoring of the incidence of pineapple water core disease mainly relies on field sampling and manual observation of fruit samples by cutting, which is not only time-consuming and labor-intensive, but also costly, making it difficult to meet the monitoring needs of large-scale planting areas and limiting its application in agriculture.

[0003] In view of this, we propose a pineapple water core disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data to solve the existing problems. SUMMARY

[0004] The purpose of the present application is to provide a pineapple water core disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solution: a pineapple water core disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data, the operation steps comprising: S1: obtaining multi-source remote sensing data of pineapple growth period based on unmanned aerial vehicle platform, the multi-source remote sensing data including crown layer RGB image, multi-spectral image and LiDAR point cloud data of pineapple growth period; S2: obtaining regional pineapple remote sensing data by image stitching, calibration and geographic registration of multiple single images collected by the unmanned aerial vehicle, so as to pre-process the unmanned aerial vehicle remote sensing data; S3: extracting features from the pre-processed multi-source remote sensing data, including texture features, vegetation index and pineapple crown height features from RGB image, multi-spectral image and LiDAR point cloud data; S4: inverting pineapple nutrition parameters based on the extracted features, including spectrum, texture and plant height, and the pineapple nutrition parameters including leaf yellowing ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass; S5: fitting the water core disease incidence according to a single pineapple nutrition parameter or a combination of multiple pineapple nutrition parameters.

[0006] Further, in S3, the gray level co-occurrence matrix method is used to extract the texture features of the RGB image.

[0007] Further, 8 texture feature parameters are obtained by calculating the gray level co-occurrence matrix, including contrast, correlation, energy, homogeneity, entropy, variance, difference and mean, and 8 texture features are extracted for each waveband of RGB, and a total of 24 texture features are obtained.

[0008] Further, in S3, the red, green, blue, red edge and near-infrared waveband data of the pineapple canopy are obtained by using the multispectral sensor, the red, green, blue, red edge and near-infrared waveband reflectance of the pineapple planting area are extracted, and the vegetation index is calculated.

[0009] Further, in S3, the digital surface model and the digital elevation model are generated by using the point cloud data obtained by the LiDAR sensor, the maximum value of the digital surface model is subtracted from the mean value of the digital elevation model, and the pineapple canopy height information is obtained.

[0010] Further, in S4, the pineapple leaf yellowing ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass are inversed by using the machine learning method.

[0011] Further, in S5, for any one of the pineapple nutrition parameters, the water heart disease incidence can be fitted.

[0012] Further, in S5, for any combination of two pineapple nutrition parameters, the water heart disease incidence can be fitted.

[0013] Further, in S5, for any combination of three pineapple nutrition parameters, the water heart disease incidence can be fitted.

[0014] Further, in S5, for the combination of four pineapple nutrition parameters, the water heart disease incidence can be fitted.

[0015] Compared with the prior art, the beneficial effects of the present application are: The present application is based on the unmanned aerial vehicle platform to obtain the canopy RGB image, multispectral image and LiDAR point cloud data in the pineapple growth period, to extract more comprehensive plant feature information, to use multi-source remote sensing data to inverse the pineapple nutrition parameters, and to construct the pineapple water heart disease incidence prediction model by mining the quantitative relationship between the nutrition parameters and the pineapple water heart disease incidence, which significantly improves the precision and robustness of the pineapple nutrition parameter inversion model, realizes the non-destructive detection of the pineapple water heart disease incidence on a large scale, has the advantages of high efficiency, flexibility, low cost, fast data acquisition speed, etc., and timely reflects the plant nutrition state through the image, which provides an effective means for plant disease detection. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1Workflow diagram of the pineapple water core disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data of the present application; Figure 2 Device diagram of the unmanned aerial vehicle platform and the sensors carried thereon of the present application; Figure 3 Flowchart of the texture feature, plant height index and spectral index calculation of the present application; Figure 4 Vegetation index table used in the present application; Figure 5 Table diagram of the pineapple leaf yellowing ratio inversion by different machine learning methods of the present application; Figure 6 Effect comparison diagram of the pineapple plant nitrogen content inversion by different machine learning methods of the present application; Figure 7 Table diagram of the pineapple plant nitrogen accumulation amount inversion by different machine learning methods of the present application; Figure 8 Effect comparison diagram of the pineapple plant biomass inversion by different machine learning methods of the present application; Figure 9 Curve diagram of the relationship between the pineapple leaf yellowing ratio and the water core disease incidence of the present application; Figure 10 Curve diagram of the relationship between the pineapple plant nitrogen accumulation amount and the water core disease incidence of the present application; Figure 11 Curve diagram of the relationship between the pineapple aboveground biomass and the water core disease incidence of the present application; Figure 12 Fitting relationship table diagram of the multivariate nutrition parameters and the water core disease incidence of the present application.

[0017] Figure 4 R, R, R, R and R are the reflectance of multispectral red, green, blue, red edge and near-infrared bands, respectively. R , R G , R B , R RED and R NIR are the reflectance of multispectral red, green, blue, red edge and near-infrared bands, respectively. DETAILED DESCRIPTION

[0018] The technical solutions of the present application are further described below in combination with the drawings and specific embodiments. Embodiment one

[0019] Current statistics of pineapple water core disease incidence still rely on artificial regional investigation, which has problems such as long cycle, high cost, large subjective error and destructive sampling. It is not only time-consuming and labor-intensive, but also difficult to meet the dynamic monitoring needs of large-scale planting areas, which limits the development of pineapple industry. Developing a pineapple water core disease prediction model based on unmanned aerial vehicle platform can solve the drawbacks of traditional investigation methods, quickly cover a wide range of planting areas, significantly reduce labor costs, and provide technical support for disease prediction and prevention in pineapple planting areas and sustainable development of pineapple industry.

[0020] As shown in Figure 1 , the pineapple water core disease incidence prediction method based on unmanned aerial vehicle multi-source remote sensing data includes the following operation steps: S1: Obtain multi-source remote sensing data of pineapple growth period based on unmanned aerial vehicle platform, including canopy RGB image, multispectral image and LiDAR point cloud data of pineapple growth period; S2: Obtain regional pineapple remote sensing data by image stitching, calibration and geographic registration of multiple single images collected by unmanned aerial vehicle, so as to preprocess the unmanned aerial vehicle remote sensing data; S3: Feature extraction is performed on the preprocessed multi-source remote sensing data, and texture features, vegetation index and pineapple canopy height features are extracted from RGB image, multispectral image and LiDAR point cloud data respectively; S4: Inverse pineapple nutrition parameters based on the extracted features, including spectrum, texture and plant height, pineapple nutrition parameters including leaf yellowing ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass; S5: Fit the water core disease incidence according to a single pineapple nutrition parameter or a combination of multiple pineapple nutrition parameters.

[0021] In S1, the unmanned aerial vehicle platform and the sensors carried thereon are as shown in Figure 2 , DJI M300 RTK is used to carry DJI Zenmuse P1, RedEdge-P and DJI Zenmuse L1 sensors to obtain canopy RGB image, multispectral image and LiDAR point cloud data of pineapple key growth period. Each flight task is carried out at 11:00-14:00, under the weather condition of direct sunlight on the ground and clear sky, so as to reduce the influence of cloud coverage and change of solar elevation angle on the quality of remote sensing data. The flight height of the unmanned aerial vehicle is 30 m, the speed is 2.1 m / s, the lateral overlap rate and the heading overlap rate are 78% and 85% respectively.

[0022] Compared with the previous invention using only a single sensor to obtain remote sensing images, different sensors are carried based on the unmanned aerial vehicle platform, including DJI Zenmuse P1, RedEdge-P and DJI Zenmuse L1, and multi-source remote sensing data of pineapples are obtained, more comprehensive plant feature information is extracted, including spectral information of multispectral images, texture information of RGB images and crown height structure information of LiDAR point clouds, more effective features are provided for the inversion of pineapple nutrition parameters, and the precision and robustness of the pineapple nutrition parameter inversion model are significantly improved.

[0023] In S2, Agisoft Metashape Professional software is used to stitch the original aerial RGB and multispectral images, the multispectral images are calibrated in the stitching process, and the orthographic mosaic image of the pineapple planting area is generated. The stitched RGB and multispectral images are georeferenced using ArcMap 10.8 software, the orthographic image is cropped in combination with the pineapple test plot boundary vector data, the image of a single plot is extracted, the acquired LiDAR data is reconstructed using DJI Terra software, the point cloud density is set according to the percentage, and after noise reduction, the point cloud data is output in the.LAS format.

[0024] The process of obtaining texture features, plant height indexes and spectral indexes from RGB images, LiDAR point clouds and multispectral data images is shown in Figure 3

[0025] In S3, the gray level co-occurrence matrix (GLCM) method is used to extract the texture features of the RGB image. First, the three bands (red, green and blue) of the RGB image are converted into gray images, then 8 texture feature parameters are obtained by calculating the gray level co-occurrence matrix, including contrast, correlation, energy, homogeneity, entropy, variance, difference and mean, 8 texture features are extracted for each band of the RGB image, and a total of 24 texture features are obtained.

[0026] The red, green, blue, red edge and near-infrared band data of the pineapple canopy are obtained using the multispectral sensor, the red, green, blue, red edge and near-infrared band reflectance of the pineapple planting area is extracted using the "Zonal Statistics As Table" tool in ArcMap, and 24 vegetation indices are calculated, as shown in Figure 4

[0027] ​​LiDAR (Light Detection and Ranging) was used to obtain the canopy height information of pineapples. Firstly, the ground points and non-ground points were separated by filtering algorithm, and then the digital surface model (DSM) and digital elevation model (DEM) were generated respectively. DSM represents the elevation information of the ground surface and the objects above it, while DEM only represents the elevation information of the ground surface. Using the "Zonal Statistics As Table" tool in ArcMap, the values of DSM and DEM of each test plot were extracted, and the canopy height characteristics of pineapples were obtained by subtracting the maximum value of DSM from the mean value of DEM.

[0028] By comprehensively using spectral analysis, statistical analysis and machine learning technology, the nutritional parameters of pineapples (leaf chlorosis ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass) were successfully retrieved using multi-source remote sensing data. By mining the quantitative relationship between nutritional parameters and pineapple heart disease incidence, a pineapple heart disease incidence prediction model was constructed, realizing the non-destructive detection of pineapple heart disease incidence and providing a new idea and method for pineapple heart disease incidence prediction. It can also help farmers to adjust the agronomic measures in time and improve the yield and quality of pineapples.

[0029] In S4, machine learning methods were used to retrieve pineapple nutritional parameters, including retrieving pineapple leaf chlorosis ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass. In the process of machine learning modeling, hyperparameter search and cross-validation methods were used to improve the performance of the retrieval model. Hyperparameter search lists all possible combinations of hyperparameters, evaluates the performance of the model for each combination, and selects the best hyperparameters. Cross-validation is to divide the data set into multiple subsets and train the model multiple times under different training-test splits to reduce the instability of the model caused by different data segmentation methods. Field measurement data was used to evaluate the retrieval model, and the evaluation indicators included the coefficient of determination (R 2 ), root mean square error (RMSE) and relative root mean square error (rRMSE).

[0030] For pineapple leaf chlorosis ratio (CLR) retrieval, based on spectral, texture and canopy height features, pineapple leaf chlorosis ratio was retrieved using multiple linear regression (MLR), partial least squares regression (PLSR), gradient boosting decision tree (GBDT), extreme gradient boosting tree (XGBoost), random forest (RFR) and artificial neural network (ANN). The accuracy of different methods of leaf chlorosis ratio retrieval model was compared as Figure 5 shown, and the XGBoost model was preferred for leaf chlorosis ratio retrieval, with R 2 up to 0.74.

[0031] For the inversion of pineapple plant nitrogen content (PNC), the same as the leaf yellowing ratio, the inversion of pineapple plant nitrogen content based on spectral, texture and canopy height characteristics, respectively, adopts six methods of multiple linear regression (MLR), partial least squares regression (PLSR), gradient boosting decision tree (GBDT), extreme gradient boosting tree (XGBoost), random forest (RFR) and artificial neural network (ANN) to invert the pineapple leaf yellowing ratio, and the model effect comparison is as shown in Figure 6 As shown in the figure, the XGBoost model is preferably used to invert the pineapple plant nitrogen content, R 2 reached 0.73.

[0032] For the inversion of pineapple plant nitrogen accumulation (PNA), the inversion of pineapple plant accumulation based on spectral, texture and canopy height characteristics, respectively, adopts six methods of multiple linear regression (MLR), partial least squares regression (PLSR), gradient boosting decision tree (GBDT), extreme gradient boosting tree (XGBoost), random forest (RFR) and artificial neural network (ANN) to invert the pineapple plant nitrogen accumulation, and the model precision comparison of different methods of plant nitrogen accumulation is as shown in Figure 7 As shown in the figure, the RF model is used to invert the pineapple plant nitrogen accumulation, R 2 reached 0.91.

[0033] For the inversion of pineapple plant biomass (AGB), the inversion of pineapple biomass based on spectral, texture and canopy height characteristics, respectively, adopts six methods of multiple linear regression (MLR), partial least squares regression (PLSR), gradient boosting decision tree (GBDT), extreme gradient boosting tree (XGBoost), random forest (RFR) and artificial neural network (ANN). The model effect comparison of pineapple plant biomass inversion is as shown in Figure 8 As shown in the figure, the XGBoost model is finally used to invert the biomass (AGB), R 2 reached 0.93, and the RMSE was 33.16 g / m 2 .

[0034] In S5, for the construction of water heart disease incidence model, according to the optimal estimation model of the inverted nutrition parameter, the quantitative relationship between it and the water heart disease incidence is analyzed, and the water heart disease incidence is predicted based on different pineapple nutrition parameters.

[0035] For the construction of single variable fitting model, four pineapple nutrition parameters are respectively constructed to have a quantitative relationship with the water heart disease incidence.

[0036] For the relationship between the yellowing ratio of pineapple leaves and the incidence of water core disease, the yellowing ratio of leaves can directly reflect the nutritional status of pineapple and is the most easily observed phenotypic trait by pineapple producers. When excessive fertilizers are applied, the leaves of pineapple grow excessively, remain green for a long time, and cannot yellow normally, delaying the aging of the plant and increasing the incidence of water core disease in pineapple. By quantifying the relationship between the yellowing ratio of leaves and water core disease, the incidence of water core disease in pineapple can be quickly predicted. As shown in Figure 9 , the yellowing ratio of leaves and the incidence of water core disease are exponentially related, y = 1743.7e -0.194x , y is the incidence of water core disease, x is the yellowing ratio of leaves, the model accuracy is R 2 = 0.837, indicating that the yellowing ratio of leaves retrieved by the unmanned aerial vehicle can well explain the trend of the incidence. When the yellowing ratio is less than 20%, the incidence is relatively high (> 30%); as the yellowing ratio increases, the incidence shows a rapid downward trend, and when the yellowing ratio reaches more than 30%, the incidence drops to less than 5%.

[0037] For the relationship between the nitrogen content of pineapple plants and the incidence of water core disease, the nitrogen content of pineapple plants is an important parameter reflecting the nutritional status of pineapple, but it is not recommended to use the nitrogen content of plants alone to predict the incidence of water core disease.

[0038] For the relationship between the nitrogen accumulation of pineapple plants and the incidence of water core disease, the incidence of water core disease in pineapple increases with the increase of nitrogen accumulation. As shown in Figure 10 , the fitting result of the nitrogen accumulation of plants and the incidence is y = 1.014x - 11.486, the fitting accuracy R 2 = 0.761, indicating that the nitrogen accumulation of plants can well explain the change of the incidence. When the nitrogen accumulation of plants is less than 30 g / m 2 , the incidence of water core disease is relatively low, maintaining below 25%; as the nitrogen accumulation of plants increases, the incidence of water core disease increases significantly, and when the nitrogen accumulation of plants exceeds 30 g / m 2 , the incidence of water core disease can reach more than 35%.

[0039] For the aboveground biomass and the incidence of water core disease, as shown in Figure 11 , the aboveground biomass and the incidence of water core disease are exponentially related, the fitting result is y = 0.2554e 0.0154x , the fitting accuracy R 2 = 0.842, when the aboveground biomass is low (less than 250 g / m 2 ), the incidence of water core disease is low, usually below 10%; as the aboveground biomass increases, the incidence of water core disease increases rapidly, and when the aboveground biomass reaches more than 300 g / m 2 , the incidence of water core disease can exceed 30%.

[0040] As shown inFigure 12 For the fitting relationship between the multivariate nutrition parameters and the incidence of heart disease, linear fitting equations of 11 different combinations of nutrition parameters were constructed to explore the effect of multivariate combinations on the incidence of heart disease in pineapple. In the multivariate fitting equation, the combination of biomass and nitrogen-related parameters can improve the monitoring accuracy of the incidence of heart disease. However, the combination of PNC+PNA+AGB (R 2 =0.790) did not improve the fitting accuracy of the model, indicating that PNA and PNC may have high collinearity. When PNA+AGB was added to CLR (PNA+AGB+CLR, R 2 =0.878), the model accuracy reached the optimal. However, after adding all four variables (PNC+PNA+CLR+AGB), R 2 still maintained at 0.878, indicating that PNC has limited contribution in the multivariate model. In addition, the linear model fitting accuracy of the combination of PNA+CLR also reached R 2 =0.878. Therefore, the optimal model for predicting the incidence of heart disease is the combination of plant nitrogen accumulation (PNA) and leaf yellowing ratio (CLR), which has the best fitting effect and the relatively simple model. If only one nutrition parameter is used, the above ground biomass (AGB) of pineapple can be selected to predict the incidence of heart disease in pineapple.

[0041] The above specific embodiments are only several preferred embodiments of the present application, and based on the technical solutions of the present application and the related inspiration of the above embodiments, those skilled in the art can make various alternative improvements and combinations on the above specific embodiments.

Claims

1. A method for predicting the incidence of pineapple water heart disease based on multi-source remote sensing data of a UAV, characterized in that, The operation steps comprise: S1: obtaining multi-source remote sensing data of the growth period of pineapples based on a UAV platform, wherein the multi-source remote sensing data comprises a canopy RGB image, multi-spectral image and LiDAR point cloud data of the growth period of the pineapples; S2: obtaining regional pineapple remote sensing data by image stitching, calibration and geographic registration of a plurality of single images collected by the UAV, so as to pre-process the UAV remote sensing data; S3: extracting features from the pre-processed multi-source remote sensing data, wherein texture features, vegetation index and pineapple canopy height features are extracted from the RGB image, multi-spectral image and LiDAR point cloud data respectively; S4: inverting pineapple nutrition parameters based on the extracted features, wherein the extracted features comprise spectrum, texture and plant height, and the pineapple nutrition parameters comprise leaf yellowing ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass, wherein a machine learning method is used to invert the pineapple leaf yellowing ratio, plant nitrogen content, plant nitrogen accumulation and aboveground biomass; S5: fitting the incidence of water core disease according to a single pineapple nutrition parameter or a combination of a plurality of pineapple nutrition parameters.

2. The method for predicting the incidence of pineapple water heart disease based on multi-source remote sensing data of unmanned aerial vehicles according to claim 1, characterized in that: In S3, the gray level co-occurrence matrix method is used to extract the texture features of the RGB image.

3. The method for predicting the incidence of pineapple water heart disease based on multi-source remote sensing data of unmanned aerial vehicles according to claim 2, characterized in that: Eight texture feature parameters are obtained by calculating the gray level co-occurrence matrix, including contrast, correlation, energy, homogeneity, entropy, variance, difference and mean, and 8 texture features are extracted for each waveband of the RGB, and a total of 24 texture features are obtained.

4. The method for predicting the incidence of pineapple water heart disease based on multi-source remote sensing data of unmanned aerial vehicles according to claim 1, characterized in that: In S3, the multi-spectral sensor is used to obtain red, green, blue, red edge and near-infrared waveband data of the pineapple canopy, and the red, green, blue, red edge and near-infrared waveband reflectance of the pineapple planting area is extracted, and the vegetation index is calculated.

5. The method for predicting the incidence of pineapple water heart disease based on multi-source remote sensing data of unmanned aerial vehicles according to claim 1, characterized in that: In S3, the point cloud data obtained by the LiDAR sensor is used to generate a digital surface model and a digital elevation model, and the maximum value of the digital surface model is subtracted from the mean value of the digital elevation model to obtain the pineapple canopy height information.

6. The method for predicting the incidence of pineapple water heart disease based on multi-source remote sensing data of unmanned aerial vehicles according to claim 1, characterized in that: In S5, the incidence of water core disease can be fitted for any one of the pineapple nutrition parameters.

7. The method for predicting the incidence of pineapple water heart disease based on unmanned aerial vehicle multi-source remote sensing data according to claim 1, characterized in that: In S5, the incidence of water core disease can be fitted for any combination of two pineapple nutrition parameters.

8. The method for predicting the incidence of pineapple water heart disease based on unmanned aerial vehicle multi-source remote sensing data according to claim 1, characterized in that: In S5, the incidence of water core disease can be fitted for any combination of three pineapple nutrition parameters.

9. The method for predicting the incidence of pineapple water heart disease based on unmanned aerial vehicle multi-source remote sensing data according to claim 1, characterized in that: In S5, the incidence of water core disease can be fitted for any combination of four pineapple nutrition parameters.

Images