Agricultural monitoring decision method and system based on multi-source perception and image analysis

Abstract

The present application relates to the technical field of image processing, and more particularly to a method and system for agricultural monitoring and decision-making based on multi-source perception and image analysis. The method first collects multi-source images of farmland, including foggy day crop tilt images, and obtains high-quality images through registration fusion and smoothing denoising processing. Potential boundary points are marked using pixel gradient calculation, and a refined boundary image is generated by combining edge detection algorithms to determine high-precision boundary positions through gradient feature analysis. Based on the boundary positions, the pixel gradient direction distribution is integrated to form a set of crop linear structures, and the stem tilt trend is tracked through linear fitting and direction statistics to generate a stem tilt direction trend graph. The water-sensitive spectral features in the hyperspectral image are extracted, and the morphological information of the stem tilt trend is fused to input a pre-trained soil moisture evaluation model to generate a soil moisture deviation evaluation graph representing the field water distribution condition. This technical solution provides data-driven decision support for agricultural precision irrigation and resource optimization.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an agricultural monitoring decision-making method and system based on multi-source sensing and image analysis. Background Technology

[0002] Precision agriculture is the core of modern agricultural development, and improving crop management and resource utilization efficiency is crucial for ensuring crop safety and sustainable development. In the cultivation of crops such as sugarcane, stalk tilt and soil moisture are key factors affecting yield, and accurate analysis of these two factors provides a scientific basis for field management. However, existing technologies suffer from poor image quality in complex environments, especially in low-visibility scenarios such as foggy days, leading to insufficient accuracy in subsequent analysis and posing significant challenges. Traditional methods for sugarcane stalk tilt analysis and soil moisture assessment rely on simple image processing techniques, such as threshold segmentation or manual feature extraction. However, these methods cannot effectively cope with noise interference in foggy environments, causing stalks to blend into the background and blurring boundary recognition. Furthermore, soil moisture assessment is usually based on single sensor data, lacking comprehensive analysis with stalk condition, and cannot reflect the actual field conditions, especially in complex environments where it is difficult to provide reliable decision support. Image noise has a particularly significant impact on the accuracy of stalk boundary detection. Light scattering in foggy days and cluttered backgrounds blur stalk edges, making it difficult for traditional edge detection methods to accurately extract stalk contours, resulting in boundary position offsets and affecting the determination of tilt direction. Noise interference further complicates the analysis of stem tilt trends, affecting the accurate assessment of growth anomalies caused by soil moisture deviations and leading to a lack of basis for field management decisions. Therefore, accurately extracting stem boundaries and correlating them with soil moisture deviations in complex environments is a key issue that precision agriculture urgently needs to address. Summary of the Invention

[0003] To achieve the above objectives, the present invention adopts the following technical solution:

[0004] An agricultural monitoring decision-making method based on multi-source sensing and image analysis includes the following steps:

[0005] Step S1: Simultaneously acquire multi-source image data of farmland using a visible light camera and a hyperspectral camera. The multi-source image data includes tilted images of the target crops in foggy weather. Perform registration and fusion processing on the multi-source image data, and use smoothing processing to remove noise interference to obtain a smooth image.

[0006] Step S2: Calculate the gradient values ​​of pixels based on the smoothed image and mark potential boundary points to form a preliminary boundary point set. Use an edge detection algorithm to process the preliminary boundary point set to obtain a refined boundary image of the target crop. Extract gradient features from the refined boundary image and determine the true boundary to obtain a high-precision boundary position.

[0007] Step S3: Analyze the pixel gradient direction distribution based on the high-precision boundary position and integrate it into a linear structure segment to obtain the target crop linear structure set, determine the dominant growth direction vector of the target crop and perform matching tracking to obtain the stem tilt direction trend map;

[0008] Step S4: Extract the water-related spectral features from the hyperspectral image and combine them with the morphological trend information reflected by the tilt direction trend map. Input the information into the pre-trained soil moisture assessment model for analysis to obtain a soil moisture deviation assessment map, which is used to represent the field water distribution.

[0009] As a preferred embodiment of the present invention, step S1, obtaining the smoothed image, includes:

[0010] Visible light cameras and hyperspectral cameras installed in the farmland simultaneously acquire multi-source image data sequences, including the target crops, from different angles;

[0011] The multi-source image data sequence is spatiotemporally registered, and an image fusion algorithm is used to fuse the texture details of the visible light image with the spectral information of the hyperspectral image to generate a fused image.

[0012] Extract the grayscale image corresponding to the visible light band from the fused image, and use it as the tilted image;

[0013] The tilted image is smoothed using a Gaussian filtering algorithm to remove noise interference, resulting in a smoothed image with noise suppression.

[0014] As a preferred embodiment of the present invention, step S2, obtaining the refined boundary image, includes:

[0015] The preliminary boundary point set is processed using an edge detection algorithm. First, non-maximum suppression is performed to obtain a preliminary suppressed image by suppressing the gradient values ​​of non-edge pixels.

[0016] The preliminary suppressed image is subjected to dual threshold connection processing to effectively connect edge pixels and generate a refined boundary image;

[0017] Based on the refined boundary image, the gradient sharpness features of local regions are extracted, the rate of change of gradient sharpness is calculated, and the results are compared with a preset sharpness threshold.

[0018] If the rate of change of the gradient sharpness is greater than a preset sharpness threshold, then the local region is determined to be the true boundary, thereby determining the high-precision boundary position.

[0019] As a preferred embodiment of the present invention, step S2, obtaining the high-precision boundary position, includes:

[0020] Based on the smoothed image, the gradient value of each pixel is calculated, and potential boundary points are marked according to the gradient values, thereby forming a preliminary set of boundary points;

[0021] An edge detection algorithm is used to process the preliminary boundary point set to obtain a refined boundary image of the target crop;

[0022] Extract gradient sharpness features of local regions from the refined boundary image and calculate the rate of change of the gradient sharpness features;

[0023] If the rate of change is greater than a preset sharpness threshold, then the local region is marked as the true boundary;

[0024] Generate high-precision boundary positions, and organize the coordinates of the high-precision boundary positions to generate a set containing the coordinates of the boundary points;

[0025] The continuity of the high-precision boundary positions is verified using a pixel neighborhood analysis method, resulting in a set of verified boundary positions.

[0026] As a preferred embodiment of the present invention, step S3, integrating the linear structure segment, includes:

[0027] Based on the high-precision boundary position, the gradient angles of adjacent pixels are extracted, and the corresponding gradient angle distribution is generated;

[0028] Calculate the histogram of the gradient angle distribution and determine the directional consistency of the histogram;

[0029] If the directional consistency of the histogram is higher than a preset consistency threshold, then adjacent pixels are integrated into a linear structure segment through pixel neighborhood analysis.

[0030] As a preferred embodiment of the present invention, step S3, obtaining the crop linear structure set, includes:

[0031] Based on the high-precision boundary position, the gradient direction of adjacent pixels is extracted, and the distribution histogram of the gradient direction is calculated.

[0032] If the directional consistency of the distribution histogram is higher than a preset consistency threshold, then adjacent pixels are integrated through pixel neighborhood analysis to form a linear structure segment.

[0033] A noise filtering mechanism is applied to the linear structure segment to remove outliers, ensuring a more accurate set of crop linear structures.

[0034] As a preferred embodiment of the present invention, step S3, determining the dominant growth direction vector, includes:

[0035] The Hough transform is used to fit a straight line to the set of linear structures of the crops, generating a set of fitted straight lines.

[0036] For the set of fitted straight lines, directional statistics are calculated to generate a directional distribution histogram, which is used to describe the dominant growth direction of crop stems;

[0037] Based on the directional distribution histogram, determine the dominant growth direction vector, i.e., the dominant growth trend direction;

[0038] The dominant growth direction vector is normalized to generate a standardized dominant growth direction vector, ensuring that the direction vector remains consistent under different conditions.

[0039] As a preferred embodiment of the present invention, step S3, obtaining the stem tilt direction trend diagram, includes:

[0040] The Hough transform is used to fit a straight line to the set of linear structures of the crops, generating a set of fitted straight lines.

[0041] For the set of fitted lines, the direction statistics of the fitted lines are calculated, and a direction distribution histogram is generated to determine the dominant growth direction of the crop stem;

[0042] Based on the directional distribution histogram, determine the dominant growth direction vector, which is the overall growth trend direction of the crop stem;

[0043] The overall growth trend of sugarcane stalks is tracked by matching the dominant growth direction vector with the linear structure segments in the crop linear structure set.

[0044] If a directional deviation is detected during the tracking process, the matching results are adjusted through deviation correction to improve matching accuracy.

[0045] Based on the adjusted matching results, a trend map of the stem tilt direction is generated, and this map is used to show the growth trend of the stem.

[0046] As a preferred embodiment of the present invention, step S4, which obtains the soil moisture deviation assessment map, includes:

[0047] From the hyperspectral image, spectral indices constructed based on water-sensitive bands are extracted as water-related spectral features;

[0048] From the stem tilt direction trend map, the consistency index between the average tilt angle and the tilt direction is quantitatively extracted as morphological trend information;

[0049] The moisture-related spectral features and the morphological trend information are fused at the feature level to construct a comprehensive feature vector.

[0050] The comprehensive feature vector is input into the pre-trained soil moisture assessment model to calculate the relative deviation value of soil moisture.

[0051] Spatial interpolation and visualization rendering are performed on the relative deviation values ​​to generate the soil moisture deviation assessment map.

[0052] The present invention also provides an agricultural monitoring and decision-making system, comprising:

[0053] The acquisition module is used to simultaneously acquire multi-source image data of farmland using a visible light camera and a hyperspectral camera. The multi-source image data includes tilted images of target crops in foggy weather. The multi-source image data is registered and fused, and smoothing is used to remove noise interference to obtain a smooth image.

[0054] The extraction module is used to calculate the gradient values ​​of pixels and mark potential boundary points based on the smoothed image to form a preliminary boundary point set. The preliminary boundary point set is processed by an edge detection algorithm to obtain a refined boundary image of the target crop. Gradient features are extracted from the refined boundary image, and the true boundary is determined to obtain a high-precision boundary position.

[0055] The analysis module is used to analyze the pixel gradient direction distribution based on the high-precision boundary position and integrate it into a linear structure segment to obtain the target crop linear structure set, determine the dominant growth direction vector of the target crop and perform matching tracking to obtain the stem tilt direction trend map.

[0056] The evaluation module is used to extract water-related spectral features from the hyperspectral image and combine them with the morphological trend information reflected by the tilt direction trend map. The results are then input into a pre-trained soil moisture assessment model for analysis to obtain a soil moisture deviation assessment map, which is used to represent the field moisture distribution.

[0057] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0058] This invention discloses an agricultural monitoring decision-making method and system based on multi-source sensing and image analysis. Addressing the challenges of noise interference, low boundary detection accuracy, and difficulty in quantifying soil moisture deviation in sugarcane stalk tilt images during foggy weather, this invention achieves high-precision stalk tilt trend analysis and soil moisture assessment through multi-stage image processing and analysis techniques. First, the invention smooths the tilt images acquired during foggy weather using Gaussian filtering, effectively removing noise interference and generating clear, smooth images. Next, based on pixel gradient calculation and edge detection algorithms, the stalk boundaries are accurately extracted, generating high-precision boundary positions. Through gradient direction distribution analysis and Hough transform linear fitting, linear structure segments are integrated and the dominant growth direction vector is determined, generating a stalk tilt trend map. Finally, the trend information and gradient features are fused to generate a soil moisture deviation assessment map. This invention, through the organic combination of smoothing, edge detection, linear structure integration, and trend analysis, significantly improves the detection accuracy of sugarcane stalk tilt direction and the reliability of soil moisture assessment under foggy conditions, providing efficient technical support for precision agriculture. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of an embodiment of the agricultural monitoring decision-making method based on multi-source sensing and image analysis in this application.

[0060] Figure 2 This is a schematic diagram of one embodiment of an agricultural monitoring and decision-making system based on multi-source sensing and image analysis in this application. Detailed Implementation

[0061] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0062] like Figure 1 The agricultural monitoring decision-making method based on multi-source sensing and image analysis in this embodiment may specifically include:

[0063] Step S1: Simultaneously acquire multi-source image data of farmland using a visible light camera and a hyperspectral camera. The multi-source image data includes tilted images of the target crops in foggy weather. Perform registration and fusion processing on the multi-source image data, and use smoothing processing to remove noise interference to obtain a smooth image.

[0064] The process of obtaining the smoothed image in step S1 includes: simultaneously acquiring multi-source image data sequences containing the target crop from different angles using a visible light camera and a hyperspectral camera set up at the farmland site; performing spatiotemporal registration on the multi-source image data sequences; and using an image fusion algorithm to fuse the texture details of the visible light image with the spectral information of the hyperspectral image to generate a fused image; extracting the grayscale image corresponding to the visible light band from the fused image as a tilted image; and using a Gaussian filtering algorithm to smooth the tilted image, removing noise interference, and obtaining a noise-suppressed smoothed image.

[0065] Specifically, by using visible light and hyperspectral cameras installed in the farmland, multi-source image data sequences were simultaneously acquired from different angles, including tilted images of the target crops in foggy conditions. The visible light camera provided detailed surface textures of the crops, while the hyperspectral camera captured reflectance information of the crops across multiple spectral bands. The combination of these data provided more comprehensive information for subsequent analysis, especially under hazy conditions, where simultaneous acquisition from multiple angles effectively reduced the impact of haze on image quality. First, the acquired multi-source image data sequences were spatiotemporally registered to ensure that images taken at different times and angles were aligned spatially and temporally. Spatiotemporal registration eliminates errors caused by differences in shooting angle or time by precisely aligning identical areas in the images, ensuring data consistency. After registration, an image fusion algorithm was used to fuse the texture details of the visible light image with the spectral information of the hyperspectral image, generating a fused image. This fused image combines the high-resolution texture information of the visible light image with the rich spectral information of the hyperspectral image, resulting in an image that not only has detailed texture representation but also presents the spectral characteristics of the crops, thereby improving the crop's condition. Next, the grayscale image corresponding to the visible light band is extracted from the fused image and used as a tilted image. The tilted image can effectively show the shape and tilt of the target crop under foggy conditions. In order to further improve the image quality, the tilted image is smoothed by a Gaussian filtering algorithm. Gaussian filtering reduces noise in the image through convolution operation, especially noise caused by external factors such as fog and light changes, making the image smoother and noise suppression effective. After this smoothing process, the smoothed image with noise suppression provides clear and reliable image data support for subsequent crop analysis, boundary detection and other farmland monitoring tasks.

[0066] Step S2: Calculate the gradient values ​​of pixels based on the smoothed image and mark potential boundary points to form a preliminary boundary point set. Use an edge detection algorithm to process the preliminary boundary point set to obtain a refined boundary image of the target crop. Extract gradient features from the refined boundary image and determine the true boundary to obtain a high-precision boundary position.

[0067] Step S2, which obtains the refined boundary image, includes: processing the initial boundary point set using an edge detection algorithm; firstly, performing non-maximum suppression by suppressing the gradient values ​​of non-edge pixels to obtain an initial suppressed image; and then performing double threshold connection processing on the initial suppressed image to effectively connect the edge pixels and generate the refined boundary image.

[0068] Specifically, the process of obtaining the refined boundary image begins with smoothing the image. This involves calculating the gradient value of each pixel and marking potential boundary points to form a preliminary boundary point set. Next, an edge detection algorithm is used to process this preliminary boundary point set. The first step of the Cannibal algorithm is non-maximum suppression. This process suppresses the gradient values ​​of non-edge pixels, ensuring that only pixels with strong edge features are retained, thus obtaining a preliminary suppressed image. In this process, the gradient values ​​of non-edge pixels are effectively reduced to highlight edge regions and reduce noise interference with edge detection. Then, a dual-threshold connection process is applied to the preliminary suppressed image. This process sets two thresholds to identify strong and weak edges respectively. Strong edge pixels are directly retained, while weak edge pixels are only retained when connected to strong edges; others are suppressed. This strategy ensures the coherence and accuracy of edges, avoiding isolated noise points from affecting image quality. Finally, through this dual-threshold connection process, the edges in the preliminary suppressed image are effectively connected, thus generating the refined boundary image.

[0069] Step S2, obtaining the high-precision boundary position, includes: calculating the gradient value of each pixel based on the smoothed image, and marking potential boundary points according to the gradient values ​​to form a preliminary boundary point set; processing the preliminary boundary point set using an edge detection algorithm to obtain a refined boundary image of the target crop; extracting the gradient sharpness features of local regions from the refined boundary image and calculating the rate of change of the gradient sharpness features; if the rate of change is greater than a preset sharpness threshold, then marking the local region as the true boundary; generating high-precision boundary positions, and organizing the coordinates of the high-precision boundary positions to generate a set containing the coordinates of the boundary points; verifying the continuity of the high-precision boundary positions using a pixel neighborhood analysis method to obtain a verified boundary position set.

[0070] Specifically, based on the smoothed image, the gradient value of each pixel is calculated, and potential boundary points are marked to form a preliminary boundary point set. This step identifies regions with significant changes in the image, i.e., possible boundaries, by calculating the gradient values ​​of pixels, thus obtaining a preliminary boundary point set. An edge detection algorithm is then used to process these preliminary boundary points, extracting the refined boundaries of the target crop in the image. This refinement process removes unimportant or blurry parts, highlighting true boundary features and improving the clarity and accuracy of the boundaries. From the refined boundary image, the gradient sharpness features of local regions are extracted, and their rate of change is calculated. By comparing the rate of change with a preset sharpness threshold, true boundaries are identified. If the calculated rate of change is greater than the sharpness threshold, the region is considered a true boundary. This process further improves the reliability and accuracy of the boundaries. The generated high-precision boundary positions are then reorganized to ensure the precise location of the boundary points and form a set containing the coordinates of all boundary points. This coordinate set provides accurate reference data for subsequent applications. The continuity of the high-precision boundary positions is verified using pixel neighborhood analysis, which ensures the consistency of the boundary's position and shape. Through verification, potential errors and isolated points were removed, making the final set of boundary locations more stable and reliable, thus providing higher quality data support for crop monitoring and precise positioning.

[0071] Step S3: Analyze the pixel gradient direction distribution based on the high-precision boundary position and integrate it into a linear structure segment to obtain the target crop linear structure set. Determine the dominant growth direction vector of the target crop and perform matching tracking to obtain the stem tilt direction trend map.

[0072] The step S3, which integrates the pixels into a linear structure segment, includes: extracting the gradient angles of adjacent pixels based on the high-precision boundary position and generating the corresponding gradient angle distribution; calculating the histogram of the gradient angle distribution and determining the directional consistency of the histogram; if the directional consistency of the histogram is higher than a preset consistency threshold, then integrating the adjacent pixels into a linear structure segment through pixel neighborhood analysis.

[0073] Specifically, based on the high-precision boundary position, the gradient angles of adjacent pixels are extracted, and a gradient angle distribution is generated to reflect the edge direction of each region in the image. By calculating the histogram of gradient angles, the consistency of direction is evaluated, and the stability of the edge region is judged. Regions with high consistency of direction show that the edges have strong clustering in the same direction, indicating that these regions belong to the same structural segment. If the consistency of direction of the histogram exceeds a preset threshold, adjacent pixels are integrated into linear structural segments through pixel neighborhood analysis. Pixels with similar directions are merged to form continuous linear structural segments. This process improves the accuracy of boundary recognition and provides structural information for analyzing the growth trend and spatial distribution of crops. The linear structural segments provide the basis for determining the dominant growth direction vector. Combined with matching and tracking, a stem tilt trend map is generated to realize dynamic monitoring and analysis of crop status.

[0074] The step S3 to obtain the crop linear structure set includes: extracting the gradient direction of adjacent pixels based on the high-precision boundary position and calculating the distribution histogram of the gradient direction; if the direction consistency of the distribution histogram is higher than the preset consistency threshold, then integrating adjacent pixels through pixel neighborhood analysis to form a linear structure segment; applying a noise filtering mechanism to the linear structure segment to remove outliers and ensure a more accurate crop linear structure set is obtained.

[0075] Specifically, firstly, the gradient directions of adjacent pixels are extracted based on high-precision boundary locations, and a gradient direction distribution histogram is generated to reveal the directional characteristics of edges in the image. This histogram is used to analyze the consistency of edge directions in the image. When the consistency exceeds a preset threshold, it indicates that the edge directions of the region are unified, which may represent the linear structure of crop growth. Based on this, a pixel neighborhood analysis method is used to integrate adjacent pixels to form linear structure segments, ensuring that these structural segments have continuity and coherence. In order to remove noise in the image, a noise filtering mechanism is applied to identify and remove outliers, ensuring that the extracted linear structure segments are smoother and more accurate. By removing noise points that are inconsistent with surrounding pixels, the quality of the linear structure is further improved. The entire process extracts the gradient directions of high-precision boundary locations, combines them with neighborhood analysis to extract linear structures, and ensures accuracy through noise filtering. Finally, the resulting set of crop linear structures not only reflects the growth direction of crops but also provides high-quality data support for subsequent growth trend analysis and spatial distribution analysis, improving the accuracy and reliability of agricultural image processing.

[0076] Step S3, determining the dominant growth direction vector, includes: using Hough transform to fit a straight line to the crop's linear structure set, generating a set of fitted straight lines; calculating direction statistics for the fitted straight line set, generating a direction distribution histogram, which describes the dominant growth direction of the crop stem; determining the dominant growth direction vector, i.e., the dominant growth trend direction, based on the direction distribution histogram; and performing vector normalization on the dominant growth direction vector to generate a standardized dominant growth direction vector, ensuring that the direction vector has consistency under different conditions.

[0077] Specifically, in determining the dominant growth direction vector, the Hough transform is used to fit the crop's linear structure set to generate a set of fitted lines. The Hough transform identifies and extracts the main linear structure of crop growth by transforming edge points in the image into points in the parameter space, thus reflecting the crop's growth direction. These fitted lines are then used to further analyze the crop's growth trend. Based on the set of fitted lines, direction statistics are calculated and a direction distribution histogram is generated. The histogram statistically analyzes the angular distribution of the fitted lines, reflecting the dominant growth direction of the crop stem. By analyzing the distribution characteristics of the histogram, the dominant growth trend direction can be determined, which is the main direction of crop growth. Based on the direction distribution histogram, the dominant growth direction vector is determined. To ensure the consistency of the direction vector under different conditions, it is normalized to eliminate magnitude differences and standardize it, ensuring the consistency and reliability of data analysis. Through this process, the crop's growth trend is transformed into a unified direction vector, providing a stable reference for subsequent crop growth monitoring and prediction.

[0078] Step S3, obtaining the stem tilt trend map, includes: using Hough transform to fit a straight line to the crop linear structure set, generating a set of fitted straight lines; calculating the direction statistics of the fitted straight lines for the set of fitted straight lines, generating a direction distribution histogram to determine the dominant growth direction of the crop stem; determining the dominant growth direction vector, i.e., the overall growth trend direction of the crop stem, based on the direction distribution histogram; matching the dominant growth direction vector with the direction vectors of each linear structure segment in the crop linear structure set to track the overall growth trend of the sugarcane stem; if a direction deviation is found during the tracking process, adjusting the matching result through deviation correction to improve the matching accuracy; generating a stem tilt trend map based on the adjusted matching result, and using this map to display the growth trend of the stem.

[0079] Specifically, in obtaining the stem tilt trend map, Hough transform is used to fit the crop linear structure set to generate a set of fitted lines. Hough transform maps edge points in the image to parameter space, identifies linear structures related to crop growth direction, and extracts information from these lines. By analyzing the fitted line set, the direction statistics of the fitted lines are calculated, generating a direction distribution histogram. This histogram reflects the direction distribution of the fitted lines, helping to determine the dominant growth direction of the crop stem. Analysis of this histogram clarifies the overall growth trend of the crop stem and determines the dominant growth direction vector. Based on the dominant growth direction vector, direction vector matching is performed with each linear structure segment in the crop linear structure set to track the overall growth trend of the sugarcane stem, ensuring accurate tracking of stem changes. If a direction deviation is found during tracking, deviation correction is used to adjust the matching results, improving matching accuracy. This correction process corrects the deviation, ensuring the accuracy of the stem direction trend and eliminating errors caused by external factors or image quality. Based on the adjusted matching results, a stem tilt trend map is generated, accurately displaying the growth trend of the sugarcane stem and providing a stable reference for subsequent crop growth monitoring and prediction. This technical process allows for precise tracking of crop growth trends, enhancing the application effects of agricultural image processing and precision agriculture.

[0080] Step S4: Extract the water-related spectral features from the hyperspectral image and combine them with the morphological trend information reflected by the tilt direction trend map. Input the information into the pre-trained soil moisture assessment model for analysis to obtain the soil moisture deviation assessment map, which is used to represent the field water distribution.

[0081] Step S4, which obtains the soil moisture deviation assessment map, includes: extracting spectral indices based on water-sensitive bands from hyperspectral images as water-related spectral features; quantifying and extracting the consistency index between the average tilt angle and tilt direction from the stem tilt direction trend map as morphological trend information; fusing the water-related spectral features and morphological trend information at the feature level to construct a comprehensive feature vector; inputting the comprehensive feature vector into a pre-trained soil moisture assessment model to calculate the relative deviation value of soil moisture; and performing spatial interpolation and visualization rendering on the relative deviation value to generate the soil moisture deviation assessment map.

[0082] Specifically, in obtaining the soil moisture deviation assessment map, spectral indices based on water-sensitive bands are first extracted from hyperspectral images as water-related spectral features. These features, by reflecting the reflectance changes in different bands, can effectively capture information related to water content, providing basic data for moisture analysis. Next, morphological trend information is extracted from the stem tilt direction trend map to quantify the average tilt angle and tilt direction consistency index. This information reflects the crop growth trend and helps identify areas that may be affected by water. The extracted water-related spectral features and morphological trend information are fused at the feature level to construct a comprehensive feature vector. This feature vector, which integrates water information and crop growth trends, serves as input for subsequent analysis. It is then input into a pre-trained soil moisture assessment model. The model uses these features to calculate the relative deviation value of soil moisture. By combining multiple data sources and the model's precise assessment, the accuracy of moisture prediction is improved. Spatial interpolation and visualization rendering are performed on the calculated relative deviation value to ensure the spatial consistency and accuracy of the assessment results. The generated soil moisture deviation assessment map displays the field water distribution, providing reliable support for water management and soil moisture monitoring in precision agriculture. This scheme achieves efficient and accurate soil moisture deviation assessment by comprehensively utilizing spectral features, morphological information, and machine learning models.

[0083] This invention also provides an agricultural monitoring and decision-making system based on multi-source sensing and image analysis, used to implement the above-mentioned methods, such as... Figure 2 As shown, the system includes:

[0084] The acquisition module is used to simultaneously acquire multi-source image data of farmland using a visible light camera and a hyperspectral camera. The multi-source image data includes tilted images of the target crops in foggy weather. The multi-source image data is registered and fused, and smoothing is used to remove noise interference to obtain a smooth image.

[0085] The extraction module is used to calculate the gradient values ​​of pixels and mark potential boundary points based on the smooth image to form a preliminary boundary point set. The preliminary boundary point set is processed by the edge detection algorithm to obtain a refined boundary image of the target crop. Gradient features are extracted from the refined boundary image, and the true boundary is determined to obtain the high-precision boundary position.

[0086] The analysis module is used to analyze the pixel gradient direction distribution based on the high-precision boundary position and integrate it into linear structure segments to obtain the linear structure set of the target crop, determine the dominant growth direction vector of the target crop and perform matching and tracking to obtain the stem tilt direction trend map.

[0087] The evaluation module is used to extract water-related spectral features from hyperspectral images and combine them with morphological trend information reflected by tilt direction trend map. The results are then input into a pre-trained soil moisture assessment model for analysis to obtain a soil moisture deviation assessment map, which is used to represent the field moisture distribution.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0089] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0090] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An agricultural monitoring decision-making method based on multi-source sensing and image analysis, characterized in that, Includes the following steps: Step S1: Simultaneously acquire multi-source image data of farmland using a visible light camera and a hyperspectral camera. The multi-source image data includes tilted images of the target crops in foggy weather. Perform registration and fusion processing on the multi-source image data, and use smoothing processing to remove noise interference to obtain a smooth image. Step S2: Calculate the gradient values ​​of pixels based on the smoothed image and mark potential boundary points to form a preliminary boundary point set. Use an edge detection algorithm to process the preliminary boundary point set to obtain a refined boundary image of the target crop. Extract gradient features from the refined boundary image and determine the true boundary to obtain a high-precision boundary position. Step S3: Analyze the pixel gradient direction distribution based on the high-precision boundary position and integrate it into a linear structure segment to obtain the target crop linear structure set. Determine the dominant growth direction vector of the target crop and perform matching tracking to obtain a stem tilt trend map. The linear structure segment includes: extracting the gradient angles of adjacent pixels based on the high-precision boundary position and generating corresponding gradient angle distributions; calculating the histogram of the gradient angle distribution and determining the direction consistency of the histogram; if the direction consistency of the histogram is higher than a preset consistency threshold, then adjacent pixels are integrated into a linear structure segment through pixel neighborhood analysis. A noise filtering mechanism is applied to the linear structure segment to remove outliers, ensuring a more accurate crop linear structure set. Step S3, determining the dominant growth direction vector, includes: using Hough transform to perform straight line fitting on the crop linear structure set to generate... A set of fitted straight lines is fitted; for the fitted set of straight lines, direction statistics are calculated to generate a direction distribution histogram, which is used to describe the dominant growth direction of the crop stem; based on the direction distribution histogram, the dominant growth direction vector is determined, that is, the dominant growth trend direction; the dominant growth direction vector is normalized to generate a standardized dominant growth direction vector to ensure that the direction vector has consistency under different conditions; step S3 to obtain the stem tilt direction trend map includes: matching the dominant growth direction vector with each linear structure segment in the crop linear structure set to track the overall growth trend of the sugarcane stem; if a direction deviation is found during the tracking process, the matching result is adjusted through deviation correction to improve the matching accuracy; based on the adjusted matching result, a stem tilt direction trend map is generated, and this map is used to display the growth trend of the stem; Step S4: Extract the water-related spectral features from the hyperspectral image and combine them with the morphological trend information reflected by the tilt direction trend map. Input the information into the pre-trained soil moisture assessment model for analysis to obtain a soil moisture deviation assessment map, which is used to represent the field water distribution.

2. The method according to claim 1, characterized in that, Step S1, which yields a smoothed image, includes: Visible light cameras and hyperspectral cameras set up in the farmland simultaneously acquire multi-source image data sequences containing the target crops from different angles; The multi-source image data sequence is spatiotemporally registered, and an image fusion algorithm is used to fuse the texture details of the visible light image with the spectral information of the hyperspectral image to generate a fused image. Extract the grayscale image corresponding to the visible light band from the fused image, and use it as the tilted image; The tilted image is smoothed using a Gaussian filtering algorithm to remove noise interference, resulting in a smoothed image with noise suppression.

3. The method according to claim 1, characterized in that, Step S2 yields the refined boundary image, which includes: The preliminary boundary point set is processed using an edge detection algorithm. First, non-maximum suppression is performed to obtain a preliminary suppressed image by suppressing the gradient values ​​of non-edge pixels. The preliminary suppressed image is subjected to dual threshold connection processing to effectively connect edge pixels and generate a refined boundary image.

4. The method according to claim 1, characterized in that, Step S2 obtains the high-precision boundary position, including: Extract gradient sharpness features of local regions from the refined boundary image and calculate the rate of change of the gradient sharpness features; If the rate of change is greater than a preset sharpness threshold, then the local region is marked as the true boundary; Generate high-precision boundary positions, and organize the coordinates of the high-precision boundary positions to generate a set containing the coordinates of the boundary points; The continuity of the high-precision boundary positions is verified using a pixel neighborhood analysis method, resulting in a set of verified boundary positions.

5. The method according to claim 1, characterized in that, Step S4 yields the soil moisture deviation assessment map, which includes: From the hyperspectral image, spectral indices constructed based on water-sensitive bands are extracted as water-related spectral features; From the stem tilt direction trend map, the consistency index between the average tilt angle and the tilt direction is quantitatively extracted as morphological trend information; The moisture-related spectral features and the morphological trend information are fused at the feature level to construct a comprehensive feature vector. The comprehensive feature vector is input into the pre-trained soil moisture assessment model to calculate the relative deviation value of soil moisture. Spatial interpolation and visualization rendering are performed on the relative deviation values ​​to generate the soil moisture deviation assessment map.

6. An agricultural monitoring and decision-making system for implementing the method of any one of claims 1-5, characterized in that, include: The acquisition module is used to simultaneously acquire multi-source image data of farmland using a visible light camera and a hyperspectral camera. The multi-source image data includes tilted images of target crops in foggy weather. The multi-source image data is registered and fused, and smoothing is used to remove noise interference to obtain a smooth image. The extraction module is used to calculate the gradient values ​​of pixels and mark potential boundary points based on the smoothed image to form a preliminary boundary point set. The preliminary boundary point set is processed by an edge detection algorithm to obtain a refined boundary image of the target crop. Gradient features are extracted from the refined boundary image, and the true boundary is determined to obtain a high-precision boundary position. The analysis module is used to analyze the pixel gradient direction distribution based on the high-precision boundary position and integrate it into a linear structure segment to obtain the target crop linear structure set, determine the dominant growth direction vector of the target crop and perform matching tracking to obtain the stem tilt direction trend map. The evaluation module is used to extract water-related spectral features from the hyperspectral image and combine them with the morphological trend information reflected by the tilt direction trend map. The results are then input into a pre-trained soil moisture assessment model for analysis to obtain a soil moisture deviation assessment map, which is used to represent the field moisture distribution.

Images