Maize staging fertilization control method based on Internet of Things

Abstract

The invention relates to the technical field of intelligent mechanical fertilization control of the Internet of Things, in particular to a corn staging fertilization control method based on the Internet of Things. The method comprises the following steps: firstly, acquiring various nutrition data, hyperspectral images and growth stages of corn plants; further fusing the overall nutrition data and the overall NDVI value in each corn area to obtain growth suitability; further obtaining a real-time growth influence coefficient of each plant pixel point according to the time sequence fluctuation of the growth suitability of the same corn area in combination with the difference between the growth suitability of the corn area and the growth suitability of the adjacent corn area and the area of the corn area; further analyzing the fluctuation of the growth influence coefficient and the spatial offset of the corn area, and obtaining a real-time growth interference accumulation coefficient in combination with the growth delay condition of the corn plant; and finally, according to all the real-time growth interference accumulation coefficients corresponding to each corn region, and in combination with the low degree of each kind of nutrition data, realizing regional and staged accurate control fertilization.

Description

Technical Field

[0001] This invention relates to the field of intelligent Internet of Things (IoT) mechanical fertilization control technology, specifically to an IoT-based method for phased fertilization control of corn. Background Technology

[0002] Modern agriculture uses data-driven precision management of corn fertilization. For example, cornfields are monitored in zones, and soil fertility is analyzed by collecting real-time data on soil moisture, nitrogen, phosphorus, and potassium content. Combined with the growth of corn plants, their growth stages are identified, and their nutrient requirements are assessed. Based on these nutrient requirements, drip irrigation systems or drones are used to apply fertilizer to the cornfield in zones.

[0003] However, existing fertilization control methods have poor control precision in abiotic stress soils. For example, when using existing methods to regulate fertilization in arid regions (such as Gansu), the nutrients in the soil are greatly affected by the water content of different areas. For instance, when the water content is low, the dissolution and migration of nutrient molecules may be interfered with, which is not conducive to the absorption of nutrients by the maize plant roots. Furthermore, the influence of topography and rainfall on the transport of water in the soil may cause some migration in arid areas, further affecting the content of absorbable nutrients in the soil. As a result, it is difficult to accurately supplement fertilization according to the actual soil conditions and crop growth status in different regions. Summary of the Invention

[0004] To address the problem of inaccurate fertilization control in existing technologies for arid agricultural soils, the present invention aims to provide an Internet of Things-based method for phased fertilization control of maize, the specific technical solution of which is as follows:

[0005] The system acquires various nutrient data, hyperspectral images, and NDVI values ​​of cornfields at a preset acquisition frequency, and identifies the growth stages of different corn plants.

[0006] In each hyperspectral image of the corn plant, the corn region is obtained based on the similarity features between the pixels and the adjacent pixels; the overall nutrient data and the overall NDVI value of each corn region are fused to obtain the growth suitability of each corn region.

[0007] The corn regions between adjacent frames of the hyperspectral images are matched; based on the temporal fluctuation of the growth suitability of the same corn region, combined with the difference in growth suitability between the corn region and adjacent corn regions, and the area of ​​the corn region, the real-time growth influence coefficient of each plant pixel is obtained.

[0008] Based on the fluctuation of the growth influence coefficient of a single plant pixel and the change of the center of gravity of the corn region to which it belongs, as well as the degree of growth delay and the growth stage of the corn plant, the real-time growth interference accumulation coefficient of the corresponding plant pixel is obtained; based on all the growth interference accumulation coefficients corresponding to each corn region in real time, combined with the degree of lowness of each nutrient data, fertilization is controlled.

[0009] Furthermore, the method for obtaining the corn region includes:

[0010] In any of the hyperspectral images, the hyperspectral curve of each pixel is converted into a hyperspectral vector; based on region growth, among all the pixels of the corn plants, the cosine similarity of the hyperspectral vectors of the pixel and its neighboring pixels in the preset neighborhood is less than or equal to a preset similarity threshold as the growth termination condition, and the cornfield is divided into regions.

[0011] Furthermore, the method for obtaining the growth suitability includes:

[0012] At any acquisition time corresponding to any of the hyperspectral images, the average value of each nutrient data of all monitoring points and the average value of NDVI of all pixels in each corn region are used as PCA input. The load vector of PC1 is extracted as the weight of each index and the weighted sum is performed to obtain the growth suitability of each corn region at the corresponding acquisition time.

[0013] Furthermore, the method for obtaining the growth influence coefficient includes:

[0014] Select any of the corn regions mentioned above as the target region, and divide the fluctuation period based on the maximum values ​​of the growth suitability and the period between the maximum values ​​and the first collection time;

[0015] Based on the duration of the target region in each fluctuation period, the growth suitability at the start time, and the maximum growth suitability, the effective growth coefficient for each fluctuation period is obtained; at the current time, based on the number of negative difference values ​​and the absolute value of the mean in the first difference sequence of the effective growth coefficient, combined with the first effective growth coefficient in the time series, the first influencing factor is obtained.

[0016] The second influencing factor is obtained based on the difference in growth suitability between the target area and its adjacent maize areas at all collection times and the area of ​​the target area;

[0017] By combining the current first influence factor and the second influence factor, the real-time growth influence coefficient of the target region is obtained, which is used as the real-time growth influence coefficient of each pixel in the target region.

[0018] Furthermore, the method for obtaining the second impact factor includes:

[0019] At each collection time, a growth difference coefficient is obtained based on the difference in growth suitability between the target area and its adjacent maize areas; the average growth difference coefficient of the target area at all current collection times is combined with the average area of ​​the area to obtain a second influencing factor.

[0020] Furthermore, the method for obtaining the growth interference accumulation coefficient includes:

[0021] Select any plant pixel as the target pixel; obtain the mean of the first difference values ​​of the growth influence coefficient of the target pixel as the growth influencing factor;

[0022] All corn regions to which the target pixel once belonged are obtained as regions to be analyzed. The displacement factor is obtained based on the distance between the centroid of all regions to be analyzed and the corn region to which they belonged at the first acquisition time.

[0023] Based on the duration of all complete growth stages of the corn plant corresponding to the target pixel, and the degree of deviation from the corresponding preset standard growth time, the growth interference accumulation coefficient of the target pixel is obtained in real time by combining the growth influence factor and the displacement factor.

[0024] Furthermore, the method for controlling fertilization includes:

[0025] In the current hyperspectral image, the growth interference accumulation coefficients of all pixels in each corn region are fused to obtain the real-time accumulation coefficient of each corn region.

[0026] Nutrients are selected as target nutrients one by one; the fertilization requirement is obtained by combining the real-time accumulation coefficient with the difference between the preset minimum concentration and the measured concentration required by the plants for the target nutrient in any of the corn regions.

[0027] When the fertilization demand exceeds a preset demand threshold, the preset minimum concentration of fertilizer for the corresponding corn region at the current growth stage is adjusted based on the fertilization demand to obtain the corrected fertilization concentration of the target nutrient and control fertilization.

[0028] Furthermore, during the controlled fertilization process, drones are used to apply fertilizer to the corn areas that require supplemental fertilization.

[0029] Furthermore, the method for matching the corn regions between adjacent frames of the hyperspectral images includes:

[0030] The two corn regions with the most overlapping pixels in coordinates between adjacent frames of the hyperspectral image are matched.

[0031] Furthermore, the method for identifying the growth stages of different maize plants includes:

[0032] The corn plant regions are obtained based on NDVI values. The hyperspectral detection results of the corn plant regions are input into a CNN network to identify the growth stages of different corn plant regions.

[0033] The present invention has the following beneficial effects:

[0034] This invention first acquires various nutrient data, hyperspectral images, and growth stages of maize plants in a maize field, and divides the field into maize regions to provide a foundation for subsequent analysis. It then integrates the overall nutrient data and overall NDVI value within each maize region to obtain the growth suitability of each region, achieving multi-dimensional data fusion and accurately assessing the suitability of each region for maize growth. Furthermore, it reveals the impact of drought stress on growth suitability based on the temporal fluctuations of growth suitability within the same maize region. Combining the differences in growth suitability between a maize region and adjacent regions, as well as the area of ​​the maize region, it reflects the range of water content influence, obtaining the real-time growth influence coefficient for each plant pixel, providing a basis for subsequent precise fertilization control. Next, it analyzes the spatial diffusion of drought stress based on the fluctuations of the growth influence coefficient of individual plant pixels and the changes in the center of gravity of the maize region to which it belongs. Combined with the growth delay of maize plants, it accurately represents the cumulative impact of drought stress on maize plants, providing a basis for final precise fertilization control. Finally, it controls fertilization based on the real-time cumulative coefficients of all growth disturbances corresponding to each maize region, combined with the degree of low levels of each nutrient data. This solution integrates hyperspectral images, NDVI values, and nutrient data to dynamically assess the impact and accumulation of drought on plant pixels, enabling precise fertilization by region and stage. This effectively addresses the technical challenges of inaccurate fertilization control caused by the spatial spread of drought stress and growth retardation. Attached Figure Description

[0035] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 A flowchart illustrating an IoT-based method for controlling phased fertilization of corn, as provided in one embodiment of the present invention;

[0037] Figure 2This is a flowchart illustrating a method for obtaining a growth influence coefficient, as provided in one embodiment of the present invention. Detailed Implementation

[0038] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an Internet of Things-based corn staggered fertilization control method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0040] The following description, in conjunction with the accompanying drawings, details a specific scheme for a corn phased fertilization control method based on the Internet of Things provided by this invention.

[0041] Please see Figure 1 The diagram illustrates a flowchart of a corn staggered fertilization control method based on the Internet of Things (IoT) according to an embodiment of the present invention, specifically including:

[0042] Step S1: Acquire various nutrient data, hyperspectral images and NDVI values ​​of the cornfield at a preset acquisition frequency, and identify the growth stages of different corn plants.

[0043] In one embodiment of the present invention, the base fertilizer is applied evenly in the cornfield and the sowing time of each corn plant is consistent. Monitoring begins from the time the base fertilizer is applied, and the preset collection frequency is once a day. The nutrient data include at least the concentrations of nitrogen, phosphorus and potassium.

[0044] Soil moisture content at different locations in the cornfield is monitored using soil moisture sensors (e.g., frequency domain reflectometers), with measurements taken every half hour.

[0045] Monitoring points were evenly selected in the cornfield. A multi-channel ion meter (connected to multiple ion-selective electrodes) was used to monitor the dissolved ions at each monitoring point (in-situ calibration was performed at the monitoring point) to reflect the nutrient content in the soil at the corresponding monitoring point (measured once every half hour).

[0046] For example: Electrodes measure nitrate nitrogen in the soil, reflecting the soil's nitrogen content. Phosphate electrodes reflect the soil's phosphorus content (e.g., ...). ). k + Electrodes that reflect the potassium content in the soil.

[0047] Ion concentrations are obtained from measured potential values ​​using the Nernst equation (existing method). Nutrient data collected multiple times throughout the day are averaged to obtain the overall nutrient data for the day.

[0048] A drone equipped with a hyperspectral camera was used to inspect the cornfield. To avoid light interference, the drone was used to acquire hyperspectral images of the field at fixed times each day, usually during the period of direct sunlight, such as 12:00-14:00. The hyperspectral images acquired at adjacent times were ensured to have a 30% overlap rate. Based on a feature matching algorithm, the hyperspectral images from different locations were stitched together to obtain a complete hyperspectral image of the cornfield. The image was then calibrated to correspond with the actual scene progress to obtain the position coordinates of each monitoring point in the hyperspectral image.

[0049] It should be noted that the size of the cornfield, the corn variety, and the setting of monitoring points (number, distribution, collection frequency, etc.) can all be set by the implementer and are not restricted here.

[0050] Considering that corn plants have different nutritional requirements at different growth stages, and that the NDVI value changes with corn growth stages in corn cultivation, it is necessary to extract the NDVI value of each point in the figure and identify the different growth stages of corn plants.

[0051] Preferably, in one embodiment of the present invention, the corn plant region is obtained based on the NDVI value, and the hyperspectral detection results of the corn plant region are input into the CNN network to identify the growth stage of different corn plant regions.

[0052] As an example, the NDVI threshold is set to 0.4. The binarized image distinguishes between vegetation and non-vegetation areas. Opening (denoising) and closing (hole filling) operations are used to optimize the binary mask. Each independent connected component is marked as a corn plant region. Overly small regions (noise) are removed, such as connected components with fewer than 10 pixels. The threshold for overly small regions can be adjusted as needed.

[0053] Furthermore, the average spectral curve of each plant region is extracted and normalized preprocessed. Then, the spectral data is input into a pre-trained 1D CNN model, which contains convolutional layers, pooling layers, and fully connected layers. By analyzing the nonlinear mapping relationship between spectral features and growth stages, the growth stage classification results of maize plants within the maize plant region are output.

[0054] It should be noted that normalization can be linear normalization, which is performed under the corresponding data dimension and is applicable to the normalization used in this embodiment. The growth stages of corn are divided into the existing V and R stages. The working principles of various sensors, UAVs, hyperspectral cameras, methods for extracting NDVI values, methods for obtaining corn plant regions, and methods for distinguishing growth stages using CNN networks are all commonly used techniques by those skilled in the art and will not be elaborated further.

[0055] Fertilization and leaf nutrition are both expressed through element content, while the NDVI value is used to reflect the nutritional status of the leaves.

[0056] Step S2: In each hyperspectral image, the corn region is obtained from the pixels of the corn plant based on the similarity features between the pixels and the adjacent pixels; the overall nutrient data and the overall NDVI value of each corn region are fused to obtain the growth suitability of each corn region.

[0057] Within the same cornfield, due to differences in soil conditions (such as different water content and different terrain) and the varying consumption of nutrients by different corn plants during their growth process, the degree to which the soil in different locations is suitable for corn plant growth varies. Therefore, it is necessary to obtain the corn region from the pixels of the corn plants in each hyperspectral image based on the similarity features between the pixels and their adjacent pixels.

[0058] In order to eliminate the interference of irrelevant pixels (such as empty spaces), the analysis is performed only on the pixels of the corn plants, and the pixels of the corn plants are called plant pixels.

[0059] Considering the different nutrient supply capacities in different maize regions and that the NDVI value reflects the health of maize growth, the overall nutrient data and overall NDVI value of each maize region were integrated to obtain the growth suitability of each maize region. This characterizes the suitability of the corresponding area in the maize field for maize growth, preparing for subsequent analysis of the impact of water content on maize growth and optimization of fertilization control.

[0060] Preferably, in one embodiment of the present invention, the region division method is consistent for each hyperspectral image; one example is selected here for description.

[0061] In any hyperspectral image, convert the hyperspectral curve of each pixel into a hyperspectral vector;

[0062] As an example, for each pixel in the corn plant area in the image, the reflectance of the individual pixel in all bands is arranged in ascending order of bands to obtain the hyperspectral vector of that pixel.

[0063] Considering that the higher the cosine similarity between vectors, the higher the degree of similarity, the cornfield is divided into regions based on region growth. Among the pixels of all corn plants, the growth termination condition is that the cosine similarity between the hyperspectral vector of a pixel and its neighboring pixels in the preset neighborhood is less than or equal to a preset similarity threshold.

[0064] As an example, the preset neighborhood is eight neighborhoods, the preset similarity threshold is 0.7, and a fixed grid method is used to set seed points. For example, the horizontal axis resolution is 1920, the vertical axis resolution is 1080, the grid parameter is 1 / 40, the horizontal axis is divided into 48 equal parts, the vertical axis is divided into 27 equal parts, the horizontal and vertical dividing lines are distributed in a grid, and the plant pixels at the intersection of the dividing lines are used as seed points.

[0065] Create an independent growth region for each seed point. Starting from each seed point, check the plant pixels in its eight neighboring regions. If the cosine similarity is greater than 0.7, merge the pixel into the current growth region and continue to expand outward. If the similarity is less than or equal to 0.7, stop growth in that direction. When growth in all directions can no longer continue, the current corn region division is complete.

[0066] Among them, the cosine similarity metric is used to quantify the similarity between hyperspectral vectors, representing the similarity features between pixels and their neighboring pixels. Cosine similarity is an existing technology and will not be elaborated further.

[0067] Preferably, in one embodiment of the present invention, considering that there may be multiple monitoring points in a cornfield area and multiple data for the same nutrient, the average value is taken to represent the nutrient data of the entire cornfield area; similarly, the average NDVI value of all pixels is taken.

[0068] Considering that PCA is a common dimensionality reduction and feature extraction method, various nutrient data and the mean NDVI may be correlated with each other, and all reflect the growth of maize crops to a certain extent. PC1 represents the direction that explains the greatest variation in data and usually contains the most important information. The loading vector can be understood as the degree of contribution of each index to the principal component, and the weight allocation is more objective in this way.

[0069] Based on this, at any acquisition time corresponding to a hyperspectral image, the mean values ​​of various nutrient data of all monitoring points and the mean NDVI value of all pixels in each maize region are used as PCA inputs. The load vector of PC1 is extracted as the weight of each index and the weighted sum is performed to obtain the growth suitability of each maize region at the corresponding acquisition time. This achieves efficient fusion of multidimensional data and accurately assesses the suitability of each maize region for maize growth at the acquisition time corresponding to the hyperspectral image.

[0070] It should be noted that all nutritional data and NDVI values ​​underwent normalization preprocessing in their respective data dimensions before being input into PCA; the PCA algorithm is well understood by those skilled in the art and will not be elaborated upon further.

[0071] Step S3: Match the corn regions between hyperspectral images of adjacent frames; based on the temporal fluctuations of the growth suitability of the same corn region, combined with the difference in growth suitability between the corn region and adjacent corn regions, and the area of ​​the corn region, obtain the real-time growth influence coefficient of each plant pixel.

[0072] When soil moisture content fluctuates (especially in arid soils), the different moisture levels in different areas of the soil can reduce the solubility and migration of nutrients in the base fertilizer, which is not conducive to the absorption of nutrients by the roots of corn crops. Therefore, after applying base fertilizer in areas with a high degree of water deficiency, the increase in nutrients is smaller than in normal areas, and it may put continuous stress on the nutrient absorption of corn plants, resulting in persistently low effective nutrients in the area and slower plant growth in the corresponding area.

[0073] Unlike individual plant growth stages caused by genetic defects or pests, low soil moisture content (drought) has a wider range of inhibitory effects on plant growth than random, small-scale differences. Moreover, its influence is relatively stable in space. Since maize plant growth is affected by a variety of factors, the division of maize regions in hyperspectral images will also change over time. Therefore, it is necessary to match maize regions between adjacent frames of hyperspectral images to facilitate the analysis of the long-term impact on the plants.

[0074] Preferably, in one embodiment of the present invention, the two corn regions with the most overlapping pixels in the hyperspectral images of adjacent frames are matched.

[0075] It should be noted that the monitoring devices or some markers (such as boundary markers) in the field will remain unchanged to facilitate coordinate alignment of different hyperspectral images.

[0076] Considering that drought stress affects the suitability of land for growth, which is reflected in the temporal fluctuations of suitability, and that the difference in suitability between the maize region and adjacent maize regions spatially reflects the similar distribution of suitability, it indirectly reflects the extent of the drought's impact; while the area of ​​the maize region reflects the extent of the drought's impact.

[0077] Therefore, based on the temporal fluctuations of the growth suitability of the same maize region, combined with the differences in growth suitability between the maize region and adjacent maize regions, and the area of ​​the maize region, the real-time growth influence coefficient of each plant pixel is obtained, which characterizes the degree of influence of the current drought stress on each plant pixel, providing a basis for subsequent precise control of fertilization.

[0078] The plant pixels are the pixels of the corn plant.

[0079] Preferably, in one embodiment of the present invention, please refer to Figure 2 The flowchart illustrates a method for obtaining a growth influence coefficient according to an embodiment of the present invention, specifically including:

[0080] Step S301: Select any corn region as the target region, and divide the fluctuation period based on the maximum value of growth suitability and the period between the maximum value and the first collection time.

[0081] Considering that under drought stress, the growth suitability of maize plants will increase slightly and for a short period of time after fertilization (dissolution of basal fertilizer after rainfall or irrigation), and then decrease relatively steadily, the time domain is first divided based on the maximum value point and the first collection time point to obtain the fluctuation period.

[0082] Step S302: Based on the duration of each fluctuation period of the target region, the growth suitability at the start time, and the maximum growth suitability, obtain the effective growth coefficient for each fluctuation period; at the current time, based on the number of negative difference values ​​and the absolute value of the mean in the first difference sequence of the effective growth coefficient, combined with the first effective growth coefficient of the time series, obtain the first influencing factor.

[0083] For a single fluctuation period, the longer its duration, the higher the maximum growth suitability, indicating better effective growth of the maize plant. At the same time, the smaller the start time of the fluctuation period, the earlier the fluctuation period is in the time domain, the shorter the duration of drought stress on the maize plant, which is more conducive to maize growth. Therefore, the effective growth coefficient of each fluctuation period is obtained by combining these parameters.

[0084] As an example, for any fluctuation period, the starting time of the period is used as the denominator, and the product of the duration of the period and the maximum growth suitability within the period is used as the numerator. The ratio of the fractions is used as the corresponding effective growth coefficient, which shows the temporal fluctuation characteristics of growth suitability. The first collection time point starts from 1 to avoid the denominator being zero.

[0085] Considering that the larger the number of negative difference values ​​and the absolute value of the mean in the first difference sequence of the effective growth coefficient, the stronger the decreasing trend of the effective growth coefficient over time, reflecting the greater impact of drought stress on the target area; the smaller the first effective growth coefficient, the lower the initial suitability for maize growth, indicating the greater impact of (low) soil moisture content, so this is used to obtain the first influencing factor.

[0086] The product of the number of negative difference values ​​and the absolute value of the mean in the first-order difference sequence of the effective growth coefficient up to the present is used as the numerator, the first effective growth coefficient in the time series is used as the denominator, and the ratio of the fractions is used as the first real-time influencing factor of the target region.

[0087] Step S303: Based on the differences in growth suitability between the target area and its adjacent maize areas at all collection times, and the area of ​​the target area, obtain the second influencing factor.

[0088] Considering that, up to now, the larger the target area is, the larger the area affected by low soil moisture content is; at the same time, the smaller the difference in growth suitability between the target area and the adjacent maize area at each collection, the more similar the impact of low soil moisture content is, which indirectly reflects the larger range of drought impact.

[0089] Based on this, at each collection time, the growth difference coefficient is obtained according to the difference in growth suitability between the target area and its adjacent maize areas; the mean of the growth difference coefficients of the target area at all existing collection times and the mean of the area are combined to obtain the second influencing factor.

[0090] As an example, when the target area has adjacent edge pixels with other corn areas, the two corn areas are determined to be adjacent; at each acquisition time, the mean of the absolute values ​​of the difference between the growth suitability of the target area and its adjacent corn areas is used as the growth difference coefficient at each acquisition time. The difference between the growth suitability of the corn area and its adjacent corn areas is expressed in the form of the average of the absolute values ​​of the difference.

[0091] For the target region, the mean of the growth difference coefficients up to the current time is used as the denominator, the mean of the region area is used as the numerator, and the fractional ratio is used as the real-time second influencing factor of the target region.

[0092] The area of ​​the region is the number of pixels within that region.

[0093] Step S304: Merge the current first and second influence factors to obtain the real-time growth influence coefficient of the target area, which is used as the real-time growth influence coefficient of each pixel in the target area.

[0094] Considering that the first influencing factor reflects the degree of drought impact from the perspective of suitable growth conditions, and the second influencing factor reflects the degree of drought impact from the perspective of the affected area, the two are finally combined to obtain the real-time impact coefficient of the target area, which is then directly used as the real-time growth impact coefficient of each pixel in the target area.

[0095] As an example, the product of the first and second impact factors is used as the growth impact coefficient.

[0096] In other embodiments of the present invention, the implementer may also fuse the first influence factor and the second influence factor by positive correlation methods such as addition or weighted summation, which will not be elaborated further.

[0097] Step S4: Based on the fluctuation of the growth influence coefficient of a single plant pixel and the change of the center of gravity of the corn region to which it belongs, as well as the degree of growth delay and growth stage of the corn plant, obtain the real-time growth interference accumulation coefficient of the corresponding plant pixel; based on all the growth interference accumulation coefficients corresponding to each corn region in real time, combined with the degree of lowness of each nutrient data, control fertilization.

[0098] Within the same cornfield, the degree of nutrient absorption disturbance may be due to the following factors, which can cause a certain degree of migration: the difference in terrain between different areas of the cornfield causes water to migrate from high to low terrain, resulting in increased accumulation of nutrient disturbance in high-altitude areas; climatic factors such as rainfall erosion of the soil also have a certain impact on the degree of nutrient absorption disturbance in different areas.

[0099] Furthermore, since corn has standard growth times at different growth stages, such as VE corresponding to 0-5 days after emergence and V1 corresponding to 5-7 days after emergence, the degree of growth delay of corn plants can be analyzed, reflecting the degree of accumulated stress caused by low soil moisture content. Therefore, based on the fluctuation of the growth influence coefficient of a single plant pixel, the change in the center of gravity of the corn region, and the degree of growth delay and growth stage of the corn plant, the real-time growth interference accumulation coefficient of the corresponding plant pixel can be obtained, accurately representing the cumulative impact of drought stress on corn plants, and providing a basis for precise control of fertilization.

[0100] Preferably, in one embodiment of the present invention, any plant pixel is selected as the target pixel to analyze each plant pixel individually.

[0101] Considering that the first-order difference reflects the rate of change, the larger the mean of the first-order difference value, the greater the continued aggravation of the effect of low water content. Therefore, up to the present, the mean of the first-order difference values ​​of the growth influence coefficient of the target pixel is used as the influence growth factor to represent the fluctuation characteristics of the growth influence coefficient.

[0102] Considering that the target pixel may have been assigned to different corn regions, we take all the corn regions to which the target pixel has belonged up to the present as the region to be analyzed. Considering that changes in water content will cause changes in growth conditions, and that spatial migration of soil moisture will also cause spatial migration of corn regions, we obtain a displacement factor based on the distance between the centroid of all regions to be analyzed and the corn region to which they belong at the first acquisition time, to represent the change in the centroid of the corn region.

[0103] As an example, the average distance between the centroid of the region to be analyzed in the current hyperspectral image and the centroid of the corn region to which the target pixel belongs in the first acquired hyperspectral image is used as the real-time displacement factor of the target pixel.

[0104] When the target pixel does not belong to any corn region in the first acquired hyperspectral image, the coordinates of the target pixel itself are used as the centroid.

[0105] Considering that there is a standard growth duration for a complete growth stage of a maize plant, the longer the actual growth duration, the greater the degree of deviation, the higher the degree of growth delay, and the greater the cumulative impact of growth disturbance. The current growth stage can be used to obtain the complete growth stage. Furthermore, considering that the greater the growth factor and displacement factor, the greater the cumulative and lasting impact of growth disturbance.

[0106] Based on this, the growth interference accumulation coefficient of the target pixel is obtained in real time by taking into account the duration of all complete growth stages of the corn plant corresponding to the target pixel, the degree of lengthening relative to the corresponding preset standard growth time, and combining the growth factor and displacement factor.

[0107] As an example, the average of the ratio of the duration of the corn plant's complete growth stage corresponding to the target pixel to the corresponding preset standard growth time is used as the delay coefficient. The product of the growth factor and the displacement factor is used as the numerator, the sum of the delay coefficient and the preset positive parameter 1 (divided by zero) is used as the denominator, the ratio of the fractions is used as the water influence factor, and the water influence factor is used as the independent variable. After normalization by the sigmoid function, the real-time growth interference accumulation coefficient of the target pixel is obtained.

[0108] Among them, the delay coefficient, based on the preset standard growth time, represents the degree of growth delay of corn plants.

[0109] It should be noted that when the corn plant has not gone through a complete growth stage, the delay coefficient is taken as 0. The standard growth duration of each growth stage of the corn plant is affected by the corn variety selected by the implementer. For example, Zhengdan 958 has a shorter V stage; Xianyu 335 has a longer R stage. The specific details can be obtained by referring to the experimental production data or production manual of the selected corn variety, and will not be explained here.

[0110] In another embodiment of the present invention, it is also considered that the more growth stages the plant has experienced, the longer the duration of drought, and the greater the cumulative impact of growth disturbance, the more the number of growth stages experienced up to the present time is multiplied by the water influence factor and used as the independent variable of the sigmoid function to obtain the real-time growth disturbance accumulation coefficient of the target pixel.

[0111] Considering that some cornfields currently have high levels of drought stress and low levels of nutrient elements as indicated by measured data, it is necessary to promptly supplement fertilization in the corresponding areas. Therefore, based on the real-time growth disturbance accumulation coefficients for each cornfield area and the degree of low levels of each nutrient data, fertilization should be controlled to ensure the growth of corn plants.

[0112] Preferably, in one embodiment of the present invention: considering that the larger the growth interference accumulation coefficient of all pixels in the current corn region, the greater the real-time accumulation of the corn region, the growth interference accumulation coefficient of all pixels in each corn region is fused in the current hyperspectral image to obtain the real-time accumulation coefficient of each corn region.

[0113] As an example, in the current hyperspectral image, the average of the growth disturbance accumulation coefficients of all pixels in each maize region is used as the corresponding real-time accumulation coefficient.

[0114] Selecting nutrients by category as target nutrients facilitates the analysis of each nutrient. In this example, the nutrients only include nitrogen, phosphorus, and potassium.

[0115] Considering that the greater the difference between the preset minimum concentration required by the corn area and the measured concentration, the greater the degree of the measured concentration being too low; and that the greater the real-time accumulation coefficient, the greater the degree of drought impact accumulation; therefore, the fertilization requirement is obtained by combining the difference between the preset minimum concentration required by the plants for the target nutrients and the measured concentration in any corn area with the real-time accumulation coefficient.

[0116] As an example, the product of the difference between the preset minimum concentration and the measured concentration and the real-time accumulation coefficient is used as the independent variable and mapped through the premnmx(x) function, where x is the independent variable and the mapping result is used as the fertilizer demand degree.

[0117] When the fertilization demand exceeds the preset demand threshold, the preset minimum concentration of fertilizer for the corresponding corn area at the current growth stage is adjusted based on the fertilization demand to obtain the corrected fertilization concentration of the target nutrients and control fertilization.

[0118] As an example, the preset demand threshold is 0.5. The sum of the fertilization demand and the constant 1 is used as the correction coefficient. The product of the preset minimum concentration of fertilizer in the current corn area and the correction coefficient is used as the corrected fertilization concentration. Drones are used to fertilize the corn areas that need supplemental fertilization.

[0119] It should be noted that when there are corn plants at different growth stages in a corn area, the required minimum concentration for each stage is weighted and summed according to the proportion of corn plants at each growth stage to obtain the required minimum concentration of the target nutrient for the corresponding corn area; similarly, the required minimum concentration of the target nutrient is obtained by weighting and summing according to the proportion of corn plants at each growth stage to obtain the required minimum concentration of the target nutrient for the corresponding corn area.

[0120] The required concentrations of various nutrients and fertilizer concentration ranges for corn plants at different growth stages are related to the corn variety and planting area. These can be obtained by consulting experimental production data or production manuals for corn. This will allow you to obtain the preset minimum and maximum concentrations for fertilizer application. The fertilizer concentration should be adjusted so that it does not exceed the preset maximum concentration. If it does, only the preset maximum concentration should be used to avoid burning the seedlings.

[0121] Implementers can also change the calculation method, for example, by multiplying the range of fertilizer concentrations of the target nutrient at the current growth stage by the fertilizer requirement, and using the sum of the product and the preset minimum concentration as the corrected fertilizer concentration.

[0122] The analysis process is the same for each corn region; only one example will be described here.

[0123] In another embodiment of the present invention, it is also considered that in the pre-fertilization plan, there may be no fertilization plan for some growth stages, that is, the preset minimum concentration of fertilizer may be 0. At this time, the higher the fertilization demand, the stronger the demand for fertilizer in the corn area. In this case, a preset high demand threshold of 0.75 is set, and the difference between the real-time fertilization demand and 0.75 is used as a supplementary correction coefficient. The product of the supplementary correction coefficient and the fertilizer concentration corresponding to the last fertilization is used as the supplementary fertilization concentration, so as to carry out supplementary fertilization and ease the transition to the next fertilization. A minimum time interval, such as 15 days, can also be set between two adjacent supplementary fertilizations to avoid burning the seedlings.

[0124] In summary, addressing the problem of inaccurate fertilization control in existing technologies for stress-affected agricultural soils, this invention proposes an IoT-based method for phased fertilization control of maize. This invention first acquires various nutrient data, hyperspectral images, and the growth stages and regions of maize plants. It then integrates overall nutrient data and overall NDVI values ​​within each maize region to obtain growth suitability. Furthermore, based on the temporal fluctuations of growth suitability within the same maize region, combined with the differences in growth suitability between adjacent maize regions and the area of ​​the maize region, it obtains the real-time growth influence coefficient for each plant pixel. It further analyzes the fluctuations of the growth influence coefficient and the spatial offset of the maize region, combined with the growth delay of the maize plant, to obtain the real-time growth disturbance accumulation coefficient. Finally, based on the real-time growth disturbance accumulation coefficients corresponding to all maize regions, and the degree of low levels of each nutrient data, it achieves precise regional and phased fertilization control.

[0125] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0126] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for controlling phased fertilization of corn based on the Internet of Things, characterized in that, The method includes: The system acquires various nutrient data, hyperspectral images, and NDVI values ​​of cornfields at a preset acquisition frequency, and identifies the growth stages of different corn plants. In each hyperspectral image of the corn plant, the corn region is obtained based on the similarity features between the pixels and the adjacent pixels; the overall nutrient data and the overall NDVI value of each corn region are fused to obtain the growth suitability of each corn region. The corn regions between adjacent frames of the hyperspectral images are matched; based on the temporal fluctuation of the growth suitability of the same corn region, combined with the difference in growth suitability between the corn region and adjacent corn regions, and the area of ​​the corn region, the real-time growth influence coefficient of each plant pixel is obtained. Based on the fluctuation of the growth influence coefficient of a single plant pixel and the change of the center of gravity of the corn region to which it belongs, as well as the degree of growth delay and the growth stage of the corn plant, the real-time growth interference accumulation coefficient of the corresponding plant pixel is obtained; based on all the growth interference accumulation coefficients corresponding to each corn region in real time, combined with the degree of lowness of each nutrient data, fertilization is controlled.

2. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, The method for obtaining the corn region includes: In any of the hyperspectral images, the hyperspectral curve of each pixel is converted into a hyperspectral vector; based on region growth, among all the pixels of the corn plants, the cosine similarity of the hyperspectral vectors of the pixel and its neighboring pixels in the preset neighborhood is less than or equal to a preset similarity threshold as the growth termination condition, and the cornfield is divided into regions.

3. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, The method for obtaining the growth suitability includes: At any acquisition time corresponding to any of the hyperspectral images, the average value of each nutrient data of all monitoring points and the average value of NDVI of all pixels in each corn region are used as PCA input. The load vector of PC1 is extracted as the weight of each index and the weighted sum is performed to obtain the growth suitability of each corn region at the corresponding acquisition time.

4. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, The method for obtaining the growth influence coefficient includes: Select any of the corn regions mentioned above as the target region, and divide the fluctuation period based on the maximum values ​​of the growth suitability and the period between the maximum values ​​and the first collection time; Based on the duration of the target region in each fluctuation period, the growth suitability at the start time, and the maximum growth suitability, the effective growth coefficient for each fluctuation period is obtained; at the current time, based on the number of negative difference values ​​and the absolute value of the mean in the first difference sequence of the effective growth coefficient, combined with the first effective growth coefficient in the time series, the first influencing factor is obtained. The second influencing factor is obtained based on the difference in growth suitability between the target area and its adjacent maize areas at all collection times and the area of ​​the target area; By combining the current first influence factor and the second influence factor, the real-time growth influence coefficient of the target region is obtained, which is used as the real-time growth influence coefficient of each pixel in the target region.

5. The method for phased fertilization control of corn based on the Internet of Things according to claim 4, characterized in that, The methods for obtaining the second impact factor include: At each collection time, a growth difference coefficient is obtained based on the difference in growth suitability between the target area and its adjacent maize areas; the average growth difference coefficient of the target area at all current collection times is combined with the average area of ​​the area to obtain a second influencing factor.

6. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, The method for obtaining the growth interference accumulation coefficient includes: Select any plant pixel as the target pixel; obtain the mean of the first difference values ​​of the growth influence coefficient of the target pixel as the growth influencing factor; All corn regions to which the target pixel once belonged are obtained as regions to be analyzed. The displacement factor is obtained based on the distance between the centroid of all regions to be analyzed and the corn region to which they belonged at the first acquisition time. Based on the duration of all complete growth stages of the corn plant corresponding to the target pixel, and the degree of deviation from the corresponding preset standard growth time, the growth interference accumulation coefficient of the target pixel is obtained in real time by combining the growth influence factor and the displacement factor.

7. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, The method for controlling fertilization includes: In the current hyperspectral image, the growth interference accumulation coefficients of all pixels in each corn region are fused to obtain the real-time accumulation coefficient of each corn region. Nutrients are selected as target nutrients one by one; the fertilization requirement is obtained by combining the real-time accumulation coefficient with the difference between the preset minimum concentration and the measured concentration required by the plants for the target nutrient in any of the corn regions. When the fertilization demand exceeds a preset demand threshold, the preset minimum concentration of fertilizer for the corresponding corn region at the current growth stage is adjusted based on the fertilization demand to obtain the corrected fertilization concentration of the target nutrient and control fertilization.

8. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, During controlled fertilization, drones are used to apply fertilizer to corn areas that require supplemental fertilization.

9. The method for phased fertilization control of corn based on the Internet of Things according to claim 1, characterized in that, The method for matching the corn regions between adjacent frames of the hyperspectral images includes: The two corn regions with the most overlapping pixels in coordinates between adjacent frames of the hyperspectral image are matched.

10. A method for controlling phased fertilization of corn based on the Internet of Things according to claim 1, characterized in that, The method for identifying different growth stages of maize plants includes: The corn plant regions are obtained based on NDVI values. The hyperspectral detection results of the corn plant regions are input into a CNN network to identify the growth stages of different corn plant regions.

Images