A method and system for evaluating tobacco field fertilization effect based on big data

Through big data and artificial intelligence technology, combined with knowledge graphs and deep learning, and comprehensive analysis of multi-source data, the efficiency and accuracy of traditional fertilization evaluation methods are solved, and a more accurate evaluation of the fertilization effect of tobacco fields is achieved.

CN118798491BActive Publication Date: 2024-11-15LIANGSHAN YI AUTONOMOUS PREFECTURE ACAD OF AGRI SCI +1

Patent Information

Application Number
CN202411273341.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-12
Publication Date
2024-11-15
Estimated Expiration
2044-09-12

AI Technical Summary

Technical Problem

Traditional tobacco field fertilization evaluation methods rely on manual observation and empirical judgment, and there are problems such as long test cycles, high cost, insufficient comprehensive data, inefficient efficiency and insufficient evaluation results, which affect crop yield and quality.

Method used

Using a big data-based method, comprehensive analysis is carried out to evaluate the fertilization effect by collecting and processing soil, plant physical samples, soil real-time monitoring data, hyperspectral remote sensing images and drone image data. .

Benefits of technology

It improves the comprehensiveness and accuracy of fertilization effect evaluation, improves data processing efficiency and feature extraction accuracy, and achieves a more accurate fertilization effect evaluation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118798491B_ABST
    Figure CN118798491B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of smart agricultural technology, and specifically discloses a tobacco field fertilization effect evaluation method and system based on big data. The method and system utilize multi-source data of soil physical samples, soil real-time monitoring data time series, plant physical samples, hyperspectral remote sensing image data series, and drone image data series to realize soil property evaluation, plant physical property evaluation, plant overall growth evaluation, plant individual growth evaluation, and plant disease and insect pest evaluation; utilize knowledge graph technology to evaluate and analyze soil properties, thereby improving the processing efficiency of a large number of data indicators; utilize the respective characteristics of hyperspectral remote sensing images and drone images, extract feature vectors in a targeted manner, utilize artificial intelligence technology to complete feature vector recognition and classification, select multiple feature values ​​to form feature vectors, thereby improving the accuracy and comprehensiveness of feature extraction, and ultimately improving the accuracy of evaluation results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of smart agriculture, and specifically relates to a tobacco field fertilization effect evaluation method and system based on big data. Background Art

[0002] Traditional tobacco field fertilization evaluation methods mainly rely on manual observation and empirical judgment, and the test samples are mainly derived from soil and plant tissue tests. These methods have problems such as long testing cycles, high costs, incomplete data, low efficiency, strong subjectivity, and inaccurate evaluation results. On the other hand, the traditional fertilization effect evaluation indicators that rely on manual experience are relatively single, and the evaluation method is not comprehensive enough, which will also affect the accuracy of the evaluation, thereby affecting the subsequent fertilization plan and ultimately affecting the yield and quality of crops. With the development of big data and artificial intelligence technology, the collection and processing of multi-source data has become more convenient and faster. By processing and analyzing multi-source data, the implicit correlation between data can be better discovered, thereby improving agricultural production efficiency and reducing resource consumption. How to use big data and artificial intelligence technology to collect multi-source data, process and analyze multi-source data, and thus conduct a more effective and comprehensive evaluation of the tobacco field fertilization effect is the technical problem to be solved by the present invention. The present invention aims to provide a tobacco field fertilization effect evaluation method and system based on big data. Summary of the Invention

[0003] The purpose of the present invention is to overcome the above-mentioned problems existing in the prior art and provide a tobacco field fertilization effect evaluation method and system based on big data. The method and system collect multi-source data affected by fertilization, and process and comprehensively analyze various data respectively according to the characteristics of multi-source data, and effectively and comprehensively evaluate the tobacco field fertilization effect through the comprehensive analysis results.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0005] A tobacco field fertilization effect evaluation method based on big data includes the following steps:

[0006] Collecting physical soil samples, the physical soil samples being obtained using a multi-point sampling method, and processing the physical soil samples to obtain soil data; the soil data including soil physical property data and soil chemical property data; wherein the soil physical property data includes a soil compaction index, a soil water-stable aggregate index, and a soil enzyme activity index; the soil chemical property data includes a soil pH value and a soil nutrient index; the soil nutrient index is calculated based on soil organic matter, total nitrogen, total phosphorus, total potassium, alkaline-hydrolyzable nitrogen, available phosphorus, available potassium, and soil microbial biomass carbon and nitrogen;

[0007] A time series of real-time soil monitoring data is collected, wherein the time series includes a real-time soil moisture sequence, a real-time soil temperature sequence, and a real-time soil pH value sequence; the real-time soil pH value sequence and the pH value in the soil chemical property data are fused in the following manner: the pH value in the soil chemical property data is added to the end of the real-time soil pH value sequence to obtain a fused soil pH value sequence; and the real-time soil moisture sequence, the real-time soil temperature sequence, and the fused soil pH value sequence are averaged to obtain a soil moisture index, a soil temperature index, and a soil pH value index.

[0008] Soil property evaluation: A soil property knowledge graph is established, and a triple (Ei, Ej, R) is established, where Ei represents the soil physical property entity, which is a first-level node and has second-level nodes under it, including soil compaction index, soil water-stable aggregate index, and soil enzyme activity index; Ej represents the soil chemical property entity, which is a first-level node and has second-level nodes under it, including soil nutrient index, soil moisture index, soil temperature index, and soil pH index; R represents the soil evaluation entity, with values ​​including excellent, good, and poor; entity matching is performed in the soil property knowledge graph to obtain a triple whose similarity with the current soil index is greater than a preset threshold, and the value of the soil evaluation entity corresponding to the triple is used as the current soil property evaluation result;

[0009] Collecting plant samples, processing the plant samples, and obtaining plant data; the plant data including plant chemical trait data and plant agronomic trait data; the plant chemical trait data including plant nutrient indexes, which are calculated based on the total nitrogen, total phosphorus, and total potassium of the plant; the plant agronomic trait data including plant height index, stem girth index, effective leaf number index, and maximum leaf area index, where the maximum leaf area index is equal to the product of the maximum leaf length and the maximum leaf width multiplied by the leaf area coefficient;

[0010] Plant physical trait evaluation: A plant trait evaluation mapping table is pre-established, the plant trait evaluation mapping table including grades corresponding to different plant index value ranges, including excellent, good, and poor grades; the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index are obtained from the plant physical sample, and the indexes are matched in the pre-established plant trait evaluation mapping table to obtain the grade corresponding to each index; the number of indicators corresponding to each grade is counted, and the grade with the largest number of indicators is used as the current plant physical trait evaluation result;

[0011] Collect tobacco field image data sequence, the tobacco field image data sequence includes hyperspectral remote sensing image data sequence IS=((ISN1,ISR1,ISG1),(ISN2,ISR2,ISG2),...,(ISNn,ISRn ,ISG n ), the IS is composed of n image triplets, each image size is 1024×1024, (ISN1, ISR1, ISG1)~(ISNn, ISR n , ISGn) represent the two-dimensional spatial information of the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images respectively. The gray value of each pixel in ISN1~ISNn represents the reflectivity of the pixel in the near-infrared band, the gray value of each pixel in ISR1~ISRn represents the reflectivity of the pixel in the red band; the gray value of each pixel in ISG1~ISGn represents the reflectivity of the pixel in the green band;

[0012] It also includes a UAV image data sequence IU=(IU1, IU2, ..., IUm), where IU1~IUm represent the 1st to mth UAV image data;

[0013] The hyperspectral remote sensing image data sequence IS is used to analyze the overall growth and characteristics of the plants, and the drone image data sequence IU is used to analyze the growth and characteristics of individual plants;

[0014] The hyperspectral remote sensing image data sequence IS is processed as follows: the normalized vegetation index NDVI value of each pixel in the 1st to nth hyperspectral remote sensing images is calculated, the NDVI value is mapped to the image, and n NDVI distribution matrices are generated; for each NDVI distribution matrix, the following operations are performed: the distribution matrix is ​​divided into 64 non-overlapping 16×16 pixel blocks Pb, each pixel block Pb is used as a segmentation unit, and the histogram statistics of the NDVI value are performed in each Pb. The abscissa of the histogram represents the NDVI value, and the ordinate represents the number of pixels corresponding to the NDVI value. The NDVI value distribution in each pixel block Pb is counted; the formula is used Calculate the inter-class variance, where p0 and p1 are the pixel ratios of the tobacco field area and the non-tobacco field area, respectively, and u0 and u1 are the average grayscale of the tobacco field area and the non-tobacco field area, respectively. Traverse all possible thresholds and find the threshold T that maximizes the inter-class variance. Use the threshold T corresponding to the pixel block in each pixel block Pb to achieve regional segmentation. Splice the segmentation results of all pixel blocks and finally obtain the tobacco field area coordinates corresponding to n hyperspectral remote sensing images. Use the tobacco field area coordinates to process the original hyperspectral remote sensing images to obtain the two-dimensional spatial information of the tobacco field area in the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images. Use the data of the three different bands to synthesize a color image to obtain a data sequence I of n tobacco field area remote sensing images. reg =(I reg1 ,I reg2 ,...,I regn), feature extraction is performed on each image data in the tobacco field area remote sensing image data sequence to obtain the color feature, texture feature and shape feature corresponding to each image; the specific operations are as follows: convert the image to RGB color space, calculate the 5 eigenvalues ​​in each color channel (R, G, B) in the image, cluster the pixel values ​​in the image, select the number of categories as 2, calculate the proportion a1 and a2 of the number of pixels in each category, calculate the pixel mean μ and variance σ, calculate the difference d between the maximum pixel value and the minimum pixel value, normalize and map the μ, σ and d to the range of (0,1), and concatenate the a1, a2, normalized μ, σ and d to form a color feature vector; use the gray level co-occurrence matrix method to calculate the texture feature vector; calculate the remote sensing image data of each tobacco field area The tobacco field outline in the image is constructed, a grid mask with an area size of 1024×1024 is constructed, wherein the grid size is 4×4, and the total area value area of ​​the tobacco field area is calculated using the grid mask; the grid mask is evenly divided into 64×64 areas, a grid is randomly selected in each area, the plant area in the grid is identified, the minimum circumscribed rectangle of the plant is calculated to obtain the area occupied by each plant, the number of plants contained in each grid is calculated, the mean ka of the number of plants contained in all the selected grids is calculated, the total number of grids occupied by the tobacco field area in the current image is calculated, and finally the total number of plants kt contained in the tobacco field area in the current image is obtained, and the total area value area, the mean ka of the number of plants contained in the grid, and the total number of plants kt are spliced ​​to form a shape feature vector;

[0015] Evaluation of overall plant growth: The color feature vector, texture feature vector, and shape feature vector are concatenated and fused to obtain a hyperspectral remote sensing image feature vector, and the hyperspectral remote sensing image feature vector is input into a pre-trained artificial convolutional neural network model A to classify the overall growth of tobacco plants, wherein the categories include excellent, good, and poor. The pre-trained artificial convolutional neural network A includes 1 input layer, 1 output layer, 3 convolutional layers, 1 pooling layer, and 2 hidden layers, and uses relu as an activation function and cross entropy as a loss function. The feature vectors of n hyperspectral remote sensing images are respectively input into the above-mentioned model A for identification, and the tobacco plant growth category corresponding to each image is counted, and the category with the largest number is used as the final evaluation result of the overall plant growth.

[0016] The UAV image data sequence IU is processed as follows:

[0017] Identify the plant regions in the 1st to mth drone images, and further identify the plant leaf region and plant stem region of each plant. Process the plant leaf region to obtain the transverse and longitudinal diameters of each leaf, thereby obtaining the area of ​​each leaf. Process the plant stem region to obtain the plant height and stem girth of each plant. Calculate the mean leaf area, maximum leaf area, mean plant height, and mean stem girth of each drone image, and splice them into a drone image plant feature vector.

[0018] Individual plant growth assessment: The plant feature vectors of the drone images are input into a pre-trained artificial convolutional neural network model B to classify the individual growth conditions of tobacco plants in the field into three categories: excellent, good, and poor. The pre-trained artificial convolutional neural network B includes one input layer, one output layer, four convolutional layers, one pooling layer, and one hidden layer, uses relu as the activation function, and uses cross entropy as the loss function. The feature vectors of m drone images are respectively input into the above-mentioned model B for identification, and the individual growth condition categories of tobacco plants corresponding to each image are counted. The category with the largest number of categories is used as the final individual plant growth assessment result.

[0019] Identify the plant regions in the 1st to mth drone images, and further identify the plant leaf regions and plant stem regions of each plant. Process the plant leaf regions to obtain the integrity of each leaf. Determine whether the leaf is damaged by detecting the edge features of the leaf region. If the leaf is damaged, the integrity value is 0, and if it is not damaged, the integrity value is 1. Count the integrity of all leaves in each drone image, and add the integrity of all leaves to obtain the leaf integrity value c corresponding to the drone image. Perform color feature extraction on the leaf and stem regions, and count the percentage pe of black and yellow pixel regions in all pixels in the leaf and stem regions. Concatenate c and pe to form a pest and disease feature vector.

[0020] Plant pest and disease assessment: The pest and disease feature vector is input into a pre-trained artificial convolutional neural network model C to classify the tobacco field plant pest and disease situation, and the categories include severe, general, and mild. The pre-trained artificial convolutional neural network C includes 1 input layer, 1 output layer, 2 convolutional layers, 1 pooling layer, and 1 hidden layer, using ReLU as the activation function and cross entropy as the loss function. The pest and disease feature vectors of m drone images are respectively input into the above model C for identification, and the tobacco field plant pest and disease category corresponding to each image is counted. The category with the largest number of categories is used as the final plant pest and disease assessment result.

[0021] Output soil property evaluation results, plant physical property evaluation results, plant overall growth evaluation results, plant individual growth evaluation results and plant disease and pest evaluation results.

[0022] A tobacco field fertilization effect evaluation system based on big data includes the following functional modules:

[0023] Soil data acquisition and processing module: used to collect physical soil samples, the physical soil samples are obtained using a multi-point sampling method, and the physical soil samples are processed to obtain soil data; the soil data includes soil physical property data and soil chemical property data; collect soil real-time monitoring data time series, the real-time soil monitoring data time series includes soil real-time moisture series, soil real-time temperature series, and soil real-time pH series; fuse the real-time soil pH series with the pH value in the soil chemical property data to obtain soil moisture index, soil temperature index, and soil pH index;

[0024] Soil property evaluation module: used to establish a soil property knowledge graph and a triple (Ei, Ej, R), where Ei represents the soil physical property entity, which is a first-level node and has second-level nodes under it, including soil compaction index, soil water-stable aggregate index, and soil enzyme activity index; Ej represents the soil chemical property entity, which is a first-level node and has second-level nodes under it, including soil nutrient index, soil moisture index, soil temperature index, and soil pH index; R represents the soil evaluation entity, and its values ​​include excellent, good, and poor; entity matching is performed in the soil property knowledge graph to obtain a triple whose similarity with the current soil index is greater than a preset threshold, and the value of the soil evaluation entity corresponding to the triple is used as the current soil property evaluation result;

[0025] Plant data collection and processing module: used to collect plant physical samples, process the plant physical samples, and obtain plant data; the plant data includes plant chemical trait data and plant agronomic trait data;

[0026] Plant physical trait evaluation module: used to pre-establish a plant trait evaluation mapping table, the plant trait evaluation mapping table includes grades corresponding to different plant index value ranges, and the grades include excellent, good and poor; the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index are obtained through the plant physical sample, and matched in the pre-established plant trait evaluation mapping table to obtain the grade corresponding to each index, and the number of indicators corresponding to each grade is counted, and the grade with the largest number of indicators is used as the current plant physical trait evaluation result;

[0027] Image data acquisition and processing module: used to acquire tobacco field image data sequence, the tobacco field image data sequence includes hyperspectral remote sensing image data sequence IS=((ISN1,ISR1,ISG1),(ISN2,ISR2,ISG2),...,(ISNn,ISR n ,ISG n ), the IS is composed of n image triplets, each image size is 1024×1024, (ISN1, ISR1, ISG1)~(ISNn, ISR n , ISGn) respectively represent the two-dimensional spatial information of the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images, the grayscale value of each pixel in ISN1~ISNn represents the reflectivity of the pixel in the near-infrared band, the grayscale value of each pixel in ISR1~ISRn represents the reflectivity of the pixel in the red band; the grayscale value of each pixel in ISG1~ISGn represents the reflectivity of the pixel in the green band; it also includes a drone image data sequence IU=(IU1,IU2,...,IUm), where IU1~IUm represent the 1st to mth drone image data; the hyperspectral remote sensing image data sequence IS is used to analyze the overall growth and characteristics of the plant, and the drone image data sequence IU is used to analyze the growth and characteristics of the individual plant; the hyperspectral remote sensing image data sequence is processed to obtain a color feature vector, a texture feature vector and a shape feature vector; the drone image data sequence IU is processed to obtain a drone image plant feature vector and a disease and pest feature vector;

[0028] The overall plant growth assessment module is used to splice and fuse the color feature vector, texture feature vector and shape feature vector to obtain a hyperspectral remote sensing image feature vector, and input the hyperspectral remote sensing image feature vector into a pre-trained artificial convolutional neural network model A to identify the overall growth category of tobacco field plants, which includes three categories: excellent, good and poor. The pre-trained artificial convolutional neural network A includes 1 input layer, 1 output layer, 3 convolutional layers, 1 pooling layer and 2 hidden layers, uses relu as the activation function and cross entropy as the loss function. The feature vectors of n hyperspectral remote sensing images are respectively input into the above-mentioned model A for identification, and the tobacco field plant growth category corresponding to each image is counted, and the category with the largest number is used as the final plant overall growth assessment result.

[0029] The plant individual growth assessment module is used to input the plant feature vectors of the drone image into a pre-trained artificial convolutional neural network model B to identify the individual growth category of tobacco field plants, which includes excellent, good and poor categories. The pre-trained artificial convolutional neural network B includes 1 input layer, 1 output layer, 4 convolutional layers, 1 pooling layer and 1 hidden layer, uses ReLU as the activation function and cross entropy as the loss function. The feature vectors of m drone images are respectively input into the above-mentioned model B for identification, and the individual growth category of tobacco field plants corresponding to each image is counted. The category with the largest number of categories is used as the final individual plant growth assessment result.

[0030] Plant pest and disease assessment module: used to input the pest and disease feature vector into a pre-trained artificial convolutional neural network model C to identify the category of tobacco field plant pest and disease conditions, which includes three categories: severe, general, and mild; the pre-trained artificial convolutional neural network C includes 1 input layer, 1 output layer, 2 convolution layers, 1 pooling layer, and 1 hidden layer, uses ReLU as the activation function, and uses cross entropy as the loss function; the pest and disease feature vectors of m drone images are respectively input into the above-mentioned model C for identification, and the tobacco field plant pest and disease category corresponding to each image is counted, and the category with the largest number is used as the final plant pest and disease assessment result;

[0031] Output module: used to output soil property evaluation results, plant physical property evaluation results, plant overall growth evaluation results, plant individual growth evaluation results and plant disease and pest evaluation results.

[0032] By adopting the above technology, compared with the prior art, the beneficial effects of the present invention are:

[0033] This application collects, processes and analyzes multi-source data such as physical soil samples, physical plant samples, real-time soil monitoring data, hyperspectral remote sensing image data and drone image data to comprehensively evaluate the effects of fertilization, using richer data to improve the comprehensiveness of the evaluation of fertilization effects.

[0034] This application uses knowledge graph technology to evaluate and analyze soil properties, improving the processing efficiency of a large number of data indicators. It also uses knowledge graphs to explore the correlation between different indicators, thereby improving the accuracy of fertilization effect evaluation.

[0035] This application utilizes the respective characteristics of hyperspectral remote sensing images and drone images to extract feature vectors in a targeted manner, and uses artificial intelligence technology to realize the recognition and classification of feature vectors.

[0036] When processing hyperspectral remote sensing images and drone images, this application improves existing technologies, effectively improves the efficiency of image processing, selects multiple eigenvalues ​​to form feature vectors, improves the accuracy and comprehensiveness of feature extraction, and ultimately improves the accuracy of evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without making any creative efforts.

[0038] Figure 1 This application presents a flow chart of a tobacco field fertilization effect evaluation method based on big data;

[0039] Figure 2 Flowchart of soil property assessment for this application;

[0040] Figure 3 This is a flow chart for evaluating the physical traits of the plants for this application;

[0041] Figure 4 This is a flow chart for evaluating the overall growth of the plants for this application;

[0042] Figure 5 This is a flow chart for evaluating the growth of individual plants in this application;

[0043] Figure 6 Flowchart for assessing plant pests and diseases for this application;

[0044] Figure 7 This application provides a module diagram of a tobacco field fertilization effect evaluation system based on big data. DETAILED DESCRIPTION

[0045] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is in no way intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] Example 1

[0047] like Figure 1 As shown, this embodiment provides a tobacco field fertilization effect evaluation method based on big data, comprising the following steps:

[0048] Collecting physical soil samples, wherein the physical soil samples are obtained using a multi-point sampling method, and processing the physical soil samples to obtain soil data; the soil data includes soil physical property data and soil chemical property data;

[0049] Collecting a time series of real-time soil monitoring data, the time series comprising a real-time soil moisture sequence, a real-time soil temperature sequence, and a real-time soil pH value sequence; fusing the real-time soil pH value sequence with the pH value in the soil chemical property data to obtain a soil moisture index, a soil temperature index, and a soil pH value index;

[0050] Soil property evaluation: A soil property knowledge graph is established, and a triple (Ei, Ej, R) is established, where Ei represents the soil physical property entity, which is a first-level node and has second-level nodes under it, including soil compaction index, soil water-stable aggregate index, and soil enzyme activity index; Ej represents the soil chemical property entity, which is a first-level node and has second-level nodes under it, including soil nutrient index, soil moisture index, soil temperature index, and soil pH index; R represents the soil evaluation entity, with values ​​including excellent, good, and poor; entity matching is performed in the soil property knowledge graph to obtain a triple whose similarity with the current soil index is greater than a preset threshold, and the value of the soil evaluation entity corresponding to the triple is used as the current soil property evaluation result;

[0051] Collecting plant physical samples and processing the plant physical samples to obtain plant data; the plant data includes plant chemical trait data and plant agronomic trait data;

[0052] Plant physical trait evaluation: A plant trait evaluation mapping table is pre-established, the plant trait evaluation mapping table including grades corresponding to different plant index value ranges, including excellent, good, and poor grades; the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index are obtained from the plant physical sample, and the indexes are matched in the pre-established plant trait evaluation mapping table to obtain the grade corresponding to each index; the number of indicators corresponding to each grade is counted, and the grade with the largest number of indicators is used as the current plant physical trait evaluation result;

[0053] Collect tobacco field image data sequence, the tobacco field image data sequence includes hyperspectral remote sensing image data sequence IS=((ISN1,ISR1,ISG1),(ISN2,ISR2,ISG2),...,(ISNn,ISR n ,ISG n), the IS is composed of n image triplets, each image size is 1024×1024, (ISN1, ISR1, ISG1)~(ISNn, ISR n , ISGn) represent the two-dimensional spatial information of the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images respectively. The gray value of each pixel in ISN1~ISNn represents the reflectivity of the pixel in the near-infrared band, the gray value of each pixel in ISR1~ISRn represents the reflectivity of the pixel in the red band; the gray value of each pixel in ISG1~ISGn represents the reflectivity of the pixel in the green band;

[0054] It also includes a UAV image data sequence IU=(IU1, IU2, ..., IUm), where IU1~IUm represent the 1st to mth UAV image data;

[0055] The hyperspectral remote sensing image data sequence IS is used to analyze the overall growth and characteristics of the plants, and the drone image data sequence IU is used to analyze the growth and characteristics of individual plants;

[0056] Processing the hyperspectral remote sensing image data sequence to obtain color feature vectors, texture feature vectors, and shape feature vectors;

[0057] Evaluation of overall plant growth: The color feature vector, texture feature vector, and shape feature vector are concatenated and fused to obtain a hyperspectral remote sensing image feature vector, and the hyperspectral remote sensing image feature vector is input into a pre-trained artificial convolutional neural network model A to classify the overall growth of tobacco plants, wherein the categories include excellent, good, and poor. The pre-trained artificial convolutional neural network A includes 1 input layer, 1 output layer, 3 convolutional layers, 1 pooling layer, and 2 hidden layers, and uses relu as an activation function and cross entropy as a loss function. The feature vectors of n hyperspectral remote sensing images are respectively input into the above-mentioned model A for identification, and the tobacco plant growth category corresponding to each image is counted, and the category with the largest number is used as the final evaluation result of the overall plant growth.

[0058] Processing the UAV image data sequence IU to obtain UAV image plant feature vectors and pest and disease feature vectors;

[0059] Individual plant growth assessment: The plant feature vectors of the drone images are input into a pre-trained artificial convolutional neural network model B to classify the individual growth conditions of tobacco plants in the field into three categories: excellent, good, and poor. The pre-trained artificial convolutional neural network B includes one input layer, one output layer, four convolutional layers, one pooling layer, and one hidden layer, uses relu as the activation function, and uses cross entropy as the loss function. The feature vectors of m drone images are respectively input into the above-mentioned model B for identification, and the individual growth condition categories of tobacco plants corresponding to each image are counted. The category with the largest number of categories is used as the final individual plant growth assessment result.

[0060] Plant pest and disease assessment: The pest and disease feature vector is input into a pre-trained artificial convolutional neural network model C to classify the tobacco field plant pest and disease situation, and the categories include severe, general, and mild. The pre-trained artificial convolutional neural network C includes 1 input layer, 1 output layer, 2 convolutional layers, 1 pooling layer, and 1 hidden layer, using ReLU as the activation function and cross entropy as the loss function. The pest and disease feature vectors of m drone images are respectively input into the above model C for identification, and the tobacco field plant pest and disease category corresponding to each image is counted. The category with the largest number of categories is used as the final plant pest and disease assessment result.

[0061] Output soil property evaluation results, plant physical property evaluation results, plant overall growth evaluation results, plant individual growth evaluation results and plant disease and pest evaluation results.

[0062] Example 2

[0063] like Figure 2 As shown, this embodiment provides a tobacco field fertilization effect evaluation method based on big data, wherein when performing soil property evaluation, the method specifically includes the following steps:

[0064] The soil physical property data include soil compaction index, soil water-stable aggregate index and soil enzyme activity index; the soil chemical property data include soil pH value and soil nutrient index; the soil nutrient index is calculated through soil organic matter, total nitrogen, total phosphorus, total potassium, alkaline-hydrolyzable nitrogen, available phosphorus, available potassium, and soil microbial biomass carbon and nitrogen.

[0065] Soil compaction was measured using a soil compaction meter. The soil depth was recorded during the measurement and divided into soil layer compaction (0-15 cm) and soil subsurface compaction (15-30 cm).

[0066] Water-stable soil aggregates were determined using a soil aggregate analyzer. Air-dried soil was placed in a sieve set and sieved into five grades: >2 mm, 2-1 mm, 1-0.5 mm, 0.5-0.25 mm, and <0.25 mm. A 50-g sample was then prepared proportionally. Sieves of 2 mm, 1 mm, 0.5 mm, 0.25 mm, and 0.053 mm were placed in a bucket, and the prepared dry soil sample was placed on the top layer of the sieve set. Water was added to the edge of the top sieve of the sieve set. After shaking for 30 minutes, the sieve set was gently removed. The soil from each sieve layer was rinsed with distilled water and placed in a drying container. The soil was then air-dried and weighed. The results were calculated as follows: % aggregate size = dry weight of each aggregate layer / total sample weight * 100; mean weight diameter (MWD). , where X i represents the average diameter of each particle size component, and K represents the proportion of the particle size component.

[0067] The soil nutrient index is calculated as follows: After chemical analysis of the physical soil samples, the organic matter, total nitrogen, total phosphorus, total potassium, alkaline-hydrolyzable nitrogen, available phosphorus, available potassium, soil microbial biomass carbon and nitrogen contents of the physical soil samples are obtained. The above contents are normalized so that each content is mapped to a value in the range of (0, 1). In the order of (organic matter, total nitrogen, total phosphorus, total potassium, alkaline-hydrolyzable nitrogen, available phosphorus, available potassium, soil microbial biomass carbon and nitrogen content), the values ​​of each content are binary-coded as 0 and 1 and spliced ​​to obtain the soil nutrient index; for example, if the organic matter content value is 0.1, its corresponding binary code is 0001, if the total nitrogen content value is 0.2, its corresponding binary code is 0010, and so on. The final soil nutrient index is a string of binary code vectors.

[0068] The fusion processing method is as follows: the pH value in the soil chemical property data is added to the end of the soil real-time pH value sequence to obtain a fused soil pH value sequence; the soil real-time moisture sequence, the soil real-time temperature sequence and the fused soil pH value sequence are averaged to obtain a soil moisture index, a soil temperature index and a soil pH value index.

[0069] Example 3

[0070] like Figure 3 As shown, this embodiment provides a method for evaluating the effect of fertilization in tobacco fields based on big data, wherein when evaluating the physical traits of plants, the method specifically includes the following steps:

[0071] Physical plant samples are collected and processed to obtain plant data; the plant data include plant chemical trait data and plant agronomic trait data; the plant chemical trait data include plant nutrient indices, which are calculated based on the total nitrogen, total phosphorus, and total potassium of the plant; the plant agronomic trait data include a plant height index, a stem girth index, an effective leaf number index, and a maximum leaf area index, where the maximum leaf area index is equal to the product of the maximum leaf length and the maximum leaf width multiplied by the leaf area coefficient.

[0072] At maturity, three representative plants were selected and separated according to the flue-cured tobacco roots, stems, and leaves. The plants were placed in an oven and sterilized at 105°C for 30 min. The oven temperature was then adjusted to 65°C, and the plant samples were dried to a constant weight. The samples were then weighed with an electronic scale and the dry matter weight was recorded. The plants were then classified according to the flue-cured tobacco roots, stems, and leaves, crushed, sieved, and placed in self-sealing bags for plant nutrient testing to obtain the total nitrogen, total phosphorus, and total potassium content of the plants.

[0073] Similar to the calculation method of soil nutrient indicators, the total nitrogen, total phosphorus, and total potassium contents of the plants are normalized so that each content is mapped to a value in the range of (0, 1). In the order of (total nitrogen, total phosphorus, total potassium), the values ​​of each content are binary-encoded as 0 and 1 and concatenated to obtain the soil nutrient index; for example, if the total nitrogen content value is 0.1, its corresponding binary code is 0001; if the total phosphorus content value is 0.2, its corresponding binary code is 0010, and so on. The final plant nutrient index is a string of binary code vectors.

[0074] During the cluster and maturity stages of flue-cured tobacco, five plants representing representative plot growth were selected and their agronomic traits were measured according to the flue-cured tobacco industry standard YC / T142-1998. The measurements included plant height, stem girth, number of effective leaves, maximum leaf length, and maximum leaf width. Maximum leaf area was calculated based on the measurement results. Maximum leaf area is calculated by multiplying the product of maximum leaf length and maximum leaf width by the leaf area index (LAI). The LIA was determined using conventional methods, which are not limited to this method in the present invention.

[0075] Evaluation of plant physical traits: A plant trait evaluation mapping table is established in advance, and the plant trait evaluation mapping table includes grades corresponding to different plant indicator value ranges, and the grades include excellent, good and poor; the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index are obtained through the plant physical samples, and matched in the pre-established plant trait evaluation mapping table to obtain the grade corresponding to each indicator, and the number of indicators corresponding to each grade is counted, and the grade with the largest number of indicators is used as the current plant physical trait evaluation result.

[0076] Example 4

[0077] like Figure 4 As shown, this embodiment provides a method for evaluating the fertilization effect of tobacco fields based on big data, wherein when evaluating the overall growth of plants, the method specifically includes the following steps:

[0078] The hyperspectral remote sensing image data sequence IS is processed as follows: the normalized vegetation index NDVI value of each pixel in the 1st to nth hyperspectral remote sensing images is calculated, the NDVI value is mapped to the image, and n NDVI distribution matrices are generated; for each NDVI distribution matrix, the following operations are performed: the distribution matrix is ​​divided into 64 non-overlapping 16×16 pixel blocks Pb, each pixel block Pb is used as a segmentation unit, and the histogram statistics of the NDVI value are performed in each Pb. The abscissa of the histogram represents the NDVI value, and the ordinate represents the number of pixels corresponding to the NDVI value. The NDVI value distribution in each pixel block Pb is counted; the formula is used Calculate the inter-class variance, where p0 and p1 are the pixel ratios of the tobacco field area and the non-tobacco field area, respectively, and u0 and u1 are the average grayscale of the tobacco field area and the non-tobacco field area, respectively. Traverse all possible thresholds and find the threshold T that maximizes the inter-class variance. Use the threshold T corresponding to the pixel block in each pixel block Pb to achieve regional segmentation. Splice the segmentation results of all pixel blocks and finally obtain the tobacco field area coordinates corresponding to n hyperspectral remote sensing images. Use the tobacco field area coordinates to process the original hyperspectral remote sensing images to obtain the two-dimensional spatial information of the tobacco field area in the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images. Use the data of the three different bands to synthesize a color image to obtain a data sequence I of n tobacco field area remote sensing images. reg =(I reg1 ,I reg2 ,...,I regn), feature extraction is performed on each image data in the tobacco field area remote sensing image data sequence to obtain the color feature, texture feature and shape feature corresponding to each image; the specific operations are as follows: convert the image to RGB color space, calculate the 5 eigenvalues ​​in each color channel (R, G, B) in the image, cluster the pixel values ​​in the image, select the number of categories as 2, calculate the proportion a1 and a2 of the number of pixels in each category, calculate the pixel mean μ and variance σ, calculate the difference d between the maximum pixel value and the minimum pixel value, normalize and map the μ, σ and d to the range of (0,1), and concatenate the a1, a2, normalized μ, σ and d to form a color feature vector; use the gray level co-occurrence matrix method to calculate the texture feature vector; calculate the remote sensing image data of each tobacco field area The tobacco field outline in the image is constructed, a grid mask with an area size of 1024×1024 is constructed, where the grid size is 4×4, and the total area value area of ​​the tobacco field area is calculated using the grid mask; the grid mask is evenly divided into 64×64 areas, a grid is randomly selected in each area, the plant area in the grid is identified, the minimum circumscribed rectangle of the plant is calculated to obtain the area occupied by each plant, the number of plants contained in each grid is calculated, the mean ka of the number of plants contained in all the selected grids is calculated, and the total number of grids occupied by the tobacco field area in the current image is calculated. Finally, the total number of plants kt contained in the tobacco field area in the current image is obtained, and the total area value area, the mean ka of the number of plants contained in the grid, and the total number of plants kt are spliced ​​to form a shape feature vector.

[0079] Example 5

[0080] like Figure 5 As shown, this embodiment provides a method for evaluating the effect of fertilization in tobacco fields based on big data, wherein when evaluating the growth of individual plants, the method specifically includes the following steps:

[0081] Identify the plant areas in the 1st to mth drone images, and further identify the plant leaf area and plant stem area of ​​each plant. Process the plant leaf area to obtain the transverse and longitudinal diameters of each leaf, thereby obtaining the area of ​​each leaf. Process the plant stem area to obtain the plant height and stem girth of each plant. Calculate the mean plant leaf area, maximum leaf area, mean plant height, and mean stem girth in each drone image, and splice them into a drone image plant feature vector.

[0082] Each identified leaf is processed using the minimum circumscribed rectangle method, with the width of the minimum circumscribed rectangle being the transverse diameter of the leaf, and the length of the minimum circumscribed rectangle being the longitudinal diameter of the leaf. The area of ​​each leaf is calculated using the product of the transverse and longitudinal diameters and a predetermined scaling ratio between the image and the object. The contour of the plant stem area is identified to obtain the top endpoint coordinates and the bottom endpoint coordinates corresponding to the stem area. The height of each plant is calculated using the difference between the two and a predetermined scaling ratio between the image and the object. The stem area is evenly divided into three sections: upper, middle, and lower. A measurement point is selected from each section to obtain the diameter of the stem section. The average of the diameter values ​​corresponding to the three sections is taken as the stem diameter of the plant. The cross section of the stem is regarded as a standard circle. The stem circumference of the plant is calculated according to the circumference calculation formula of the circle and a predetermined scaling ratio between the image and the object.

[0083] Example 6

[0084] like Figure 6 As shown, this embodiment provides a method for evaluating the effect of fertilization in tobacco fields based on big data, wherein when evaluating the plant disease and insect pest situation, the method specifically includes the following steps:

[0085] Identify the plant areas in the 1st to mth drone images, and further identify the plant leaf areas and plant stem areas of each plant. Process the plant leaf areas to obtain the integrity of each leaf. Determine whether the leaf is incomplete by detecting the edge features of the leaf area. If there is incompleteness, the integrity value is 0, and if there is no incompleteness, the integrity value is 1. Count the integrity of all leaves in each drone image, and add up the integrity of all leaves to obtain the leaf integrity value c corresponding to the drone image. Perform color feature extraction on the leaf area and stem area, and count the percentage pe of black and yellow pixel areas in all pixels in the leaf area and stem area. Concatenate the c and pe to form a pest and disease feature vector.

[0086] By detecting the edge features of the leaf area, it is determined whether the leaf is defective. First, the edge contour line features of the leaf are identified. The edge contour line features include the edge lines around the leaf and the edge lines on the leaf surface. The edge lines around the leaf reflect whether there are defects around the leaf, and the edge lines on the leaf surface reflect whether there are defects on the leaf surface. In this embodiment, an artificial convolutional neural network is selected to detect whether the leaf is defective. The input is the extracted edge contour line features, which are input into a pre-trained artificial convolutional neural network model. The output is a probability value vector of the presence of a defect on the leaf. The artificial convolutional neural network model adopts a conventional model in the prior art, and the present invention is not limited to this.

[0087] Common tobacco plant diseases and pests manifest as yellowing and even black spots on leaves and stems. To identify these diseases and pests, we extract the color features of the leaves and stems and use the three color channels of the RGB color image to calculate the proportion of black and yellow pixels in the total number of pixels. This pixel ratio reflects the health of the plant and is therefore extracted as a feature value for subsequent identification of the severity of the disease and pest.

[0088] Example 7

[0089] like Figure 7 As shown, this embodiment provides a tobacco field fertilization effect evaluation system based on big data, including the following functional modules:

[0090] Soil data acquisition and processing module: used to collect physical soil samples, the physical soil samples are obtained using a multi-point sampling method, and the physical soil samples are processed to obtain soil data; the soil data includes soil physical property data and soil chemical property data; collect soil real-time monitoring data time series, the real-time soil monitoring data time series includes soil real-time moisture series, soil real-time temperature series, and soil real-time pH series; fuse the real-time soil pH series with the pH value in the soil chemical property data to obtain soil moisture index, soil temperature index, and soil pH index;

[0091] Soil property evaluation module: used to establish a soil property knowledge graph and a triple (Ei, Ej, R), where Ei represents the soil physical property entity, which is a first-level node and has second-level nodes under it, including soil compaction index, soil water-stable aggregate index, and soil enzyme activity index; Ej represents the soil chemical property entity, which is a first-level node and has second-level nodes under it, including soil nutrient index, soil moisture index, soil temperature index, and soil pH index; R represents the soil evaluation entity, and its values ​​include excellent, good, and poor; entity matching is performed in the soil property knowledge graph to obtain a triple whose similarity with the current soil index is greater than a preset threshold, and the value of the soil evaluation entity corresponding to the triple is used as the current soil property evaluation result;

[0092] Plant data collection and processing module: used to collect plant physical samples, process the plant physical samples, and obtain plant data; the plant data includes plant chemical trait data and plant agronomic trait data;

[0093] Plant physical trait evaluation module: used to pre-establish a plant trait evaluation mapping table, the plant trait evaluation mapping table includes grades corresponding to different plant index value ranges, and the grades include excellent, good and poor; the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index are obtained through the plant physical sample, and matched in the pre-established plant trait evaluation mapping table to obtain the grade corresponding to each index, and the number of indicators corresponding to each grade is counted, and the grade with the largest number of indicators is used as the current plant physical trait evaluation result;

[0094] Image data acquisition and processing module: used to acquire tobacco field image data sequence, the tobacco field image data sequence includes hyperspectral remote sensing image data sequence IS=((ISN1,ISR1,ISG1),(ISN2,ISR2,ISG2),...,(ISNn,ISR n ,ISG n ), the IS is composed of n image triplets, each image size is 1024×1024, (ISN1, ISR1, ISG1)~(ISNn, ISR n , ISGn) respectively represent the two-dimensional spatial information of the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images, the grayscale value of each pixel in ISN1~ISNn represents the reflectivity of the pixel in the near-infrared band, the grayscale value of each pixel in ISR1~ISRn represents the reflectivity of the pixel in the red band; the grayscale value of each pixel in ISG1~ISGn represents the reflectivity of the pixel in the green band; it also includes a drone image data sequence IU=(IU1,IU2,...,IUm), where IU1~IUm represent the 1st to mth drone image data; the hyperspectral remote sensing image data sequence IS is used to analyze the overall growth and characteristics of the plant, and the drone image data sequence IU is used to analyze the growth and characteristics of the individual plant; the hyperspectral remote sensing image data sequence is processed to obtain a color feature vector, a texture feature vector and a shape feature vector; the drone image data sequence IU is processed to obtain a drone image plant feature vector and a disease and pest feature vector;

[0095] The overall plant growth assessment module is used to splice and fuse the color feature vector, texture feature vector and shape feature vector to obtain a hyperspectral remote sensing image feature vector, and input the hyperspectral remote sensing image feature vector into a pre-trained artificial convolutional neural network model A to identify the overall growth category of tobacco field plants, which includes three categories: excellent, good and poor. The pre-trained artificial convolutional neural network A includes 1 input layer, 1 output layer, 3 convolutional layers, 1 pooling layer and 2 hidden layers, uses relu as the activation function and cross entropy as the loss function. The feature vectors of n hyperspectral remote sensing images are respectively input into the above-mentioned model A for identification, and the tobacco field plant growth category corresponding to each image is counted, and the category with the largest number is used as the final plant overall growth assessment result.

[0096] The plant individual growth assessment module is used to input the plant feature vectors of the drone image into a pre-trained artificial convolutional neural network model B to identify the individual growth category of tobacco field plants, which includes excellent, good and poor categories. The pre-trained artificial convolutional neural network B includes 1 input layer, 1 output layer, 4 convolutional layers, 1 pooling layer and 1 hidden layer, uses ReLU as the activation function and cross entropy as the loss function. The feature vectors of m drone images are respectively input into the above-mentioned model B for identification, and the individual growth category of tobacco field plants corresponding to each image is counted. The category with the largest number of categories is used as the final individual plant growth assessment result.

[0097] Plant pest and disease assessment module: used to input the pest and disease feature vector into a pre-trained artificial convolutional neural network model C to identify the category of tobacco field plant pest and disease conditions, which includes three categories: severe, general, and mild; the pre-trained artificial convolutional neural network C includes 1 input layer, 1 output layer, 2 convolution layers, 1 pooling layer, and 1 hidden layer, uses ReLU as the activation function, and uses cross entropy as the loss function; the pest and disease feature vectors of m drone images are respectively input into the above-mentioned model C for identification, and the tobacco field plant pest and disease category corresponding to each image is counted, and the category with the largest number is used as the final plant pest and disease assessment result;

[0098] Output module: used to output soil property evaluation results, plant physical property evaluation results, plant overall growth evaluation results, plant individual growth evaluation results and plant disease and pest evaluation results.

[0099] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, computer devices, or computer-readable storage media. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0100] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Based on the technical essence of the present invention and within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement of the above embodiment shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for evaluating the effect of fertilization in tobacco fields based on big data, characterized in that: The steps include: Collecting physical soil samples, wherein the physical soil samples are obtained by using a multi-point sampling method, and processing the physical soil samples to obtain soil data; the soil data includes soil physical property data and soil chemical property data; Collecting a time series of soil real-time monitoring data, wherein the time series of soil real-time monitoring data includes a soil real-time moisture series, a soil real-time temperature series, and a soil real-time pH value series; fusing the soil real-time pH value series with the pH value in the soil chemical property data to obtain a soil moisture index, a soil temperature index, and a soil pH value index; Soil property evaluation: Establish a soil property knowledge graph and establish a triple (Ei, Ej, R), where Ei represents the soil physical property entity, which is a first-level node, and has second-level nodes, including soil compaction index, soil water-stable aggregate index and soil enzyme activity index; Ej represents the soil chemical property entity, which is a first-level node, and has second-level nodes, including soil nutrient index, soil moisture index, soil temperature index and soil pH index; R represents the soil evaluation entity, and its values ​​include excellent, good and poor; perform entity matching in the soil property knowledge graph to obtain a triple whose similarity with the current soil index is greater than a preset threshold, and use the value of the soil evaluation entity corresponding to the triple as the current soil property evaluation result; Collecting plant physical samples, processing the plant physical samples, and obtaining plant data; the plant data includes plant chemical trait data and plant agronomic trait data; Plant physical trait evaluation: pre-establish a plant trait evaluation mapping table, the plant trait evaluation mapping table includes grades corresponding to different plant index value ranges, and the grades include excellent, good and poor; obtain the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index through the plant physical sample, match them in the pre-established plant trait evaluation mapping table, obtain the grade corresponding to each indicator, count the number of indicators corresponding to each grade, and take the grade with the largest number of indicators as the current plant physical trait evaluation result; Collect tobacco field image data sequence, the tobacco field image data sequence includes a hyperspectral remote sensing image data sequence IS=((ISN1, ISR1, ISG1), (ISN2, ISR2, ISG2), ..., (ISNn, ISR n ,ISG n )), the IS is composed of n image triplets, each image size is 1024×1024, (ISN1, ISR1, ISG1)~(ISNn, ISR n , ISGn) represent the two-dimensional spatial information of the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images respectively. The gray value of each pixel in ISN1~ISNn represents the reflectivity of the pixel in the near-infrared band, the gray value of each pixel in ISR1~ISRn represents the reflectivity of the pixel in the red band; the gray value of each pixel in ISG1~ISGn represents the reflectivity of the pixel in the green band; It also includes a UAV image data sequence IU=(IU1, IU2, ..., IUm), where IU1~IUm represent the 1st to mth UAV image data; The hyperspectral remote sensing image data sequence IS is used to analyze the overall growth and characteristics of the plants, and the drone image data sequence IU is used to analyze the individual growth and characteristics of the plants; Processing the hyperspectral remote sensing image data sequence to obtain a color feature vector, a texture feature vector and a shape feature vector; The hyperspectral remote sensing image data sequence IS is processed as follows: the normalized difference vegetation index NDVI value of each pixel in the 1st to nth hyperspectral remote sensing images is calculated, the NDVI value is mapped to the image, and n NDVI distribution matrices are generated; for each NDVI distribution matrix, the following operations are performed: the distribution matrix is ​​divided into 64 non-overlapping 16×16 pixel blocks Pb, each pixel block Pb is used as a segmentation unit, and the histogram statistics of the NDVI value are performed in each Pb, the abscissa of the histogram represents the NDVI value, and the ordinate represents the number of pixels corresponding to the NDVI value, and the NDVI value distribution in each pixel block Pb is counted; Using the formula Calculate the inter-class variance, where p0 and p1 are the pixel ratios of the tobacco field area and the non-tobacco field area, respectively, and u0 and u1 are the average grayscales of the tobacco field area and the non-tobacco field area, respectively. Traverse all possible thresholds to find the threshold T that maximizes the inter-class variance. Use the threshold T corresponding to the pixel block in each pixel block Pb to achieve regional segmentation, splice the segmentation results of all pixel blocks, and finally obtain the tobacco field area coordinates corresponding to n hyperspectral remote sensing images. Use the tobacco field area coordinates to process the original hyperspectral remote sensing images to obtain the two-dimensional spatial information of the tobacco field area in the near-infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images. Use the data of three different bands to synthesize a color image to obtain a data sequence I of n tobacco field area remote sensing images. reg =(I reg1 ,I reg2 ,...,I regn ); Evaluation of the overall growth of plants: The color feature vector, texture feature vector and shape feature vector are spliced ​​and fused to obtain a hyperspectral remote sensing image feature vector, and the hyperspectral remote sensing image feature vector is input into a pre-trained artificial convolutional neural network model A to identify the category of the overall growth of tobacco field plants, and the category includes three categories: excellent, good and poor; the pre-trained artificial convolutional neural network A includes 1 input layer, 1 output layer, 3 convolutional layers, 1 pooling layer and 2 hidden layers, and relu is used as the activation function and cross entropy is used as the loss function; the feature vectors of n hyperspectral remote sensing images are respectively input into the above-mentioned model A for identification, and the category of tobacco field plant growth corresponding to each image is counted, and the category with the largest number is used as the final evaluation result of the overall growth of plants; Processing the drone image data sequence IU to obtain drone image plant feature vectors and pest and disease feature vectors; Evaluation of individual plant growth: The plant feature vectors of the drone image are input into a pre-trained artificial convolutional neural network model B to identify the categories of individual growth conditions of tobacco field plants, which include excellent, good and poor categories; the pre-trained artificial convolutional neural network B includes 1 input layer, 1 output layer, 4 convolutional layers, 1 pooling layer and 1 hidden layer, and uses relu as the activation function and cross entropy as the loss function; the feature vectors of m drone images are respectively input into the above model B for identification, and the category of individual growth conditions of tobacco field plants corresponding to each image is counted, and the category with the largest number is taken as the final evaluation result of individual plant growth conditions; Plant pest and disease situation assessment: the pest and disease feature vector is input into a pre-trained artificial convolutional neural network model C to identify the category of tobacco field plant pest and disease situation, which includes three categories: severe, general and mild; the pre-trained artificial convolutional neural network C includes 1 input layer, 1 output layer, 2 convolutional layers, 1 pooling layer and 1 hidden layer, and relu is used as the activation function and cross entropy is used as the loss function; the pest and disease feature vectors of m drone images are respectively input into the above model C for identification, and the category of tobacco field plant pest and disease situation corresponding to each image is counted, and the category with the largest number is taken as the final plant pest and disease situation assessment result; Output soil property assessment results, plant physical property assessment results, plant overall growth assessment results, plant individual growth assessment results and plant disease and pest assessment results.

2. A tobacco field fertilization effect evaluation method based on big data as claimed in claim 1, characterized in that: The soil physical property data include soil compaction index, soil water-stable aggregate index and soil enzyme activity index; the soil chemical property data include soil pH value and soil nutrient index; the soil nutrient index is calculated through soil organic matter, total nitrogen, total phosphorus, total potassium, alkaline nitrogen, available phosphorus, available potassium, soil microbial biomass carbon and nitrogen; the fusion processing method is as follows: the pH value in the soil chemical property data is added to the end of the soil real-time pH value sequence to obtain a fused soil pH value sequence; the soil real-time moisture sequence, the soil real-time temperature sequence and the fused soil pH value sequence are averaged to obtain a soil moisture index, a soil temperature index and a soil pH value index.

3. A tobacco field fertilization effect evaluation method based on big data as claimed in claim 1, characterized in that: The plant chemical trait data include plant nutrient indicators, which are calculated through the total nitrogen, total phosphorus and total potassium of the plant; the plant agronomic trait data include plant height index, stem girth index, effective leaf number index and maximum leaf area index, and the maximum leaf area index is equal to the product of the maximum leaf length and the maximum leaf width multiplied by the leaf area coefficient.

4. A tobacco field fertilization effect evaluation method based on big data as claimed in claim 3, characterized in that: The remote sensing image data sequence I of the tobacco field area reg Feature extraction is performed on each image data in the image to obtain the color feature vector, texture feature vector and shape feature vector corresponding to each image.

5. The method for evaluating the fertilization effect of tobacco fields based on big data as claimed in claim 4, characterized in that: The calculation method of the color feature vector and the texture feature vector specifically includes: converting the image into the RGB color space, calculating the five eigenvalues ​​in each color channel (R, G, B) in the image, including: clustering the pixel values ​​in the image, selecting the number of categories as 2, calculating the proportion a1 and a2 of the number of pixels in each category, calculating the pixel mean μ and variance σ, calculating the difference d between the maximum pixel value and the minimum pixel value, normalizing and mapping the μ, σ and d to the range of (0,1), and splicing the a1, a2, normalized μ, σ and d to form a color feature vector; and calculating the texture feature vector using the gray level co-occurrence matrix method.

6. A tobacco field fertilization effect evaluation method based on big data as claimed in claim 5, characterized in that: The calculation method of the shape feature vector specifically includes: calculating the tobacco field contour in each tobacco field area remote sensing image data, constructing a grid mask with an area size of 1024×1024, wherein the grid size is 4×4, and using the grid mask to calculate the total area value area of ​​the tobacco field area; evenly dividing the grid mask into 64×64 areas, randomly selecting a grid in each area, identifying the plant area in the grid, calculating the minimum circumscribed rectangle of the plant to obtain the area occupied by each plant, calculating the number of plants contained in each grid, calculating the mean ka of the number of plants contained in all selected grids, calculating the total number of grids occupied by the tobacco field area of ​​the current image, and finally obtaining the total number of plants kt contained in the tobacco field area of ​​the current image, and splicing the total area value area, the mean ka of the number of plants contained in the grid and the total number of plants kt to form a shape feature vector.

7. The method for evaluating the fertilization effect of tobacco fields based on big data according to claim 1, characterized in that: The method specifically includes: identifying plant areas in the 1st to mth drone images, and further identifying the plant leaf area and plant stem area of ​​each plant, processing the plant leaf area to obtain the horizontal diameter and vertical diameter of each leaf, thereby obtaining the area of ​​each leaf; processing the plant stem area to obtain the plant height and stem girth of each plant; calculating the mean plant leaf area, maximum leaf area value, plant height mean and stem girth mean in each drone image, and splicing them into a drone image plant feature vector.

8. The method for evaluating tobacco field fertilization effect based on big data according to claim 1, characterized in that: The method specifically includes: identifying plant areas in the 1st to mth drone images, and further identifying plant leaf areas and plant stem areas of each plant, processing the plant leaf areas to obtain the integrity of each leaf, judging whether the leaf is incomplete by detecting the edge features of the leaf area, if there is incompleteness, the integrity value is 0, if there is no incompleteness, the integrity value is 1, counting the integrity of all leaves in each drone image, adding the integrity of all leaves to obtain a leaf integrity value c corresponding to the drone image; extracting color features of the leaf area and stem area, counting the percentage pe of black and yellow pixel areas in all pixels in the leaf area and stem area; splicing the c and pe to form a pest and disease feature vector.

9. A tobacco field fertilization effect evaluation system based on big data, used to implement the method according to any one of claims 1 to 8, characterized in that: Includes the following functional modules: Soil data acquisition and processing module: used to collect physical soil samples, the physical soil samples are obtained by multi-point sampling method, and the physical soil samples are processed to obtain soil data; the soil data includes soil physical property data and soil chemical property data; collect soil real-time monitoring data time series, the soil real-time monitoring data time series includes soil real-time moisture series, soil real-time temperature series and soil real-time pH value series; the soil real-time pH value series and the pH value in the soil chemical property data are fused to obtain soil moisture index, soil temperature index and soil pH value index; Soil property evaluation module: used to establish a soil property knowledge graph and establish a triple (Ei, Ej, R), where Ei represents the soil physical property entity, which is a first-level node, and has second-level nodes, including soil compaction index, soil water-stable aggregate index and soil enzyme activity index; Ej represents the soil chemical property entity, which is a first-level node, and has second-level nodes, including soil nutrient index, soil moisture index, soil temperature index and soil pH index; R represents the soil evaluation entity, and the values ​​include excellent, good and poor; entity matching is performed in the soil property knowledge graph to obtain a triple whose similarity with the current soil index is greater than a preset threshold, and the value of the soil evaluation entity corresponding to the triple is used as the current soil property evaluation result; Plant data collection and processing module: used to collect plant physical samples, process the plant physical samples, and obtain plant data; the plant data includes plant chemical trait data and plant agronomic trait data; Plant physical trait evaluation module: used to pre-establish a plant trait evaluation mapping table, the plant trait evaluation mapping table includes grades corresponding to different plant index value ranges, and the grades include excellent, good and poor; obtain the plant nutrient index, plant height index, stem girth index, effective leaf number index, and maximum leaf area index through the plant physical sample, match them in the pre-established plant trait evaluation mapping table, obtain the grade corresponding to each indicator, count the number of indicators corresponding to each grade, and take the grade with the largest number of indicators as the current plant physical trait evaluation result; Image data acquisition and processing module: used to collect tobacco field image data sequence, the tobacco field image data sequence includes hyperspectral remote sensing image data sequence IS=((ISN1,ISR1,ISG1),(ISN2,ISR2,ISG2),...,(ISNn,ISR n ,ISG n )), the IS is composed of n image triplets, each image size is 1024×1024, (ISN1, ISR1, ISG1)~(ISNn, ISR n , ISGn) respectively represent the two-dimensional spatial information of the near infrared band, red band and green band of the 1st to nth hyperspectral remote sensing images, the gray value of each pixel in ISN1~ISNn represents the reflectivity of the pixel in the near infrared band, the gray value of each pixel in ISR1~ISRn represents the reflectivity of the pixel in the red band; the gray value of each pixel in ISG1~ISGn represents the reflectivity of the pixel in the green band; it also includes a drone image data sequence IU=(IU1,IU2,...,IUm), wherein IU1~IUm represent the 1st to mth drone image data; the hyperspectral remote sensing image data sequence IS is used to analyze the overall growth and characteristics of the plant, and the drone image data sequence IU is used to analyze the individual growth and characteristics of the plant; the hyperspectral remote sensing image data sequence is processed to obtain a color feature vector, a texture feature vector and a shape feature vector; the drone image data sequence IU is processed to obtain a drone image plant feature vector and a pest and disease feature vector; Plant overall growth assessment module: used for splicing and fusing the color feature vector, texture feature vector and shape feature vector to obtain a hyperspectral remote sensing image feature vector, inputting the hyperspectral remote sensing image feature vector into a pre-trained artificial convolutional neural network model A to identify the category of overall growth of tobacco field plants, the category includes three categories: excellent, good and poor; the pre-trained artificial convolutional neural network A includes 1 input layer, 1 output layer, 3 convolutional layers, 1 pooling layer and 2 hidden layers, using relu as the activation function and cross entropy as the loss function; inputting the feature vectors of n hyperspectral remote sensing images into the above model A for identification, counting the category of tobacco field plant growth corresponding to each image, and taking the category with the largest number as the final plant overall growth assessment result; Plant individual growth assessment module: used to input the plant feature vector of the drone image into the pre-trained artificial convolutional neural network model B to identify the category of individual growth conditions of tobacco field plants, the category includes excellent, good and poor three categories; the pre-trained artificial convolutional neural network B includes 1 input layer, 1 output layer, 4 convolutional layers, 1 pooling layer and 1 hidden layer, uses relu as the activation function, and uses cross entropy as the loss function; input the feature vectors of m drone images into the above model B for identification, count the category of individual growth conditions of tobacco field plants corresponding to each image, and take the category with the largest number as the final evaluation result of individual plant growth conditions; Plant pest and disease situation assessment module: used to input the pest and disease feature vector into the pre-trained artificial convolutional neural network model C to identify the category of tobacco field plant pest and disease situation, which includes three categories: severe, general and mild; the pre-trained artificial convolutional neural network C includes 1 input layer, 1 output layer, 2 convolution layers, 1 pooling layer and 1 hidden layer, using relu as the activation function and cross entropy as the loss function; the pest and disease feature vectors of m drone images are respectively input into the above model C for identification, and the category of tobacco field plant pest and disease situation corresponding to each image is counted, and the category with the largest number is taken as the final plant pest and disease situation assessment result; Output module: used to output soil property assessment results, plant physical property assessment results, plant overall growth assessment results, plant individual growth assessment results and plant disease and pest assessment results.

Citation Information

Patent Citations

  • Remote sensing and geographic information-based tobacco field tobacco quality monitoring and prediction method

    CN105631401A

  • Digital tobacco field intelligent agricultural system

    CN116523673A

  • Tobacco field soil collection and analysis system

    CN117330353A

  • Wheat nitrogen fertilizer recommendation method suitable for agricultural park scale

    CN117347283A

  • Crop yield prediction algorithm based on multi-source heterogeneous agricultural data

    CN117455062A

Cited By

  • Fertilizer demand analysis method based on AI planting plan

    CN121638549A