Corn disease monitoring and early warning system based on agricultural Internet of Things

Abstract

The invention relates to the technical field of intelligent early warning, in particular to a corn disease monitoring and early warning system based on agricultural Internet of Things, which comprises a data fusion calibration module, a multi-dimensional feature extraction module, a self-adaptive fusion reference construction module, a feature evaluation and fusion module, a classification and recognition module and an early warning decision generation module, carrying out fusion calibration on the visible light image data and the corresponding environment data to obtain standard monitoring data; performing multi-dimensional feature extraction on the standard monitoring data to obtain image feature information and environment feature information; constructing a multi-source feature adaptive fusion reference; evaluating the disease influence degree of the image feature information and the environment feature information to obtain a comprehensive feature index; identifying a disease type and a disease grade as a disease identification result of the corn; according to a disease early warning mapping relation in a disease early warning rule base and a continuous monitoring period condition required for triggering early warning, obtaining early warning information; according to the invention, the efficiency of corn disease monitoring and early warning can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent early warning technology, and in particular to a corn disease monitoring and early warning system based on the Internet of Things in agriculture. Background Technology

[0002] In the field of maize disease monitoring and early warning, existing technologies have failed to achieve efficient fusion and calibration of visible light image data and environmental data. The lack of precise control over the consistency of data acquisition timing, the failure to effectively correct image data for changes in illumination, and the inability to reliably remove abnormal parameters from environmental data lead to biases and redundancy in the monitoring data. This prevents the formation of a standardized monitoring foundation, making it difficult to obtain accurate and effective raw data support for subsequent feature extraction work, directly impacting the basic reliability of disease monitoring.

[0003] Existing technologies lack scientific adaptive mechanisms in feature fusion and disease identification, failing to fully quantify the dynamic correlation between image and environmental features. The fixed weighting of these two types of features prevents dynamic adjustment based on synergistic or antagonistic relationships, resulting in comprehensive feature indicators that fail to accurately reflect the degree of disease impact. Furthermore, the matching of features with disease types and levels lacks rigor during disease identification, and early warning triggering is not combined with reasonable continuous monitoring cycle analysis, leading to insufficient accuracy in disease type and level identification. Early warning information generation is also delayed or prone to misjudgment, failing to meet the timeliness and accuracy requirements for early control of maize diseases. Therefore, improving the efficiency of maize disease monitoring and early warning based on agricultural IoT has become an urgent problem to be solved. Summary of the Invention

[0004] To achieve the above objectives, this invention provides a corn disease monitoring and early warning system based on agricultural Internet of Things (IoT), characterized in that the system includes a data fusion calibration module, a multi-dimensional feature extraction module, an adaptive fusion benchmark construction module, a feature evaluation and fusion module, a classification and recognition module, and an early warning decision generation module, wherein: The data fusion calibration module is used to fuse and calibrate the visible light image data of the target area with the corresponding environmental data to obtain the standard monitoring data of the target area; The multi-dimensional feature extraction module is used to extract multi-dimensional features from the standard monitoring data to obtain image feature information and environmental feature information of the target area; The adaptive fusion benchmark construction module is used to adaptively adjust the weights of the influence of image data and environmental data on the disease in the target area according to the dynamic correlation between the image feature information and the environmental feature information, and construct a multi-source feature adaptive fusion benchmark for the target area. The feature evaluation and fusion module is used to evaluate the degree of disease impact of the image feature information and the environmental feature information under the multi-source feature adaptive fusion benchmark, so as to obtain the comprehensive feature index of the target area; The classification and identification module is used to map the comprehensive feature index to the corn disease classification knowledge base, and identify the disease type and disease level of the target area as the disease identification result of the corn. The early warning decision generation module is used to obtain early warning information for corn based on the corn disease early warning rule base, according to the disease early warning mapping relationship in the corn disease early warning rule base and the continuous monitoring period conditions required to trigger the early warning.

[0005] In a preferred embodiment, when the data fusion calibration module performs fusion calibration of visible light image data of the target area with corresponding environmental data to obtain standard monitoring data of the target area, it is specifically used for: Based on the acquisition timestamp of the visible light image data in the target area, a time-series environmental data segment matching the acquisition timestamp is selected from the corresponding environmental data of the target area to obtain the spatiotemporal aligned data of the target area; Illumination compensation is performed on the visible light image data to obtain standardized visible light image data of the target area; By removing dimensionless abnormal environmental parameters from the spatiotemporal aligned data, uniform environmental data for the target region is obtained. The standardized visible light image data and the consistent environmental data are stitched together to obtain the standard monitoring data for the target area.

[0006] In a preferred embodiment, when the multidimensional feature extraction module performs multidimensional feature extraction on the standard monitoring data to obtain image feature information and environmental feature information of the target area, it is specifically used for: The color distribution statistics and texture roughness information of the visible light image data in the standard monitoring data are used as the initial image features of the target region. A sliding time window analysis is performed on the environmental data in the standard monitoring data, and the mean, fluctuation range and trend information in the analysis results are integrated into the initial environmental characteristics of the target area. The initial image features are locally saliency-enhanced, and the features describing the color anomalies and texture structure changes of the corn canopy in the enhanced image are collected as the image feature information of the target region; The initial environmental characteristics are screened for disease relevance, and the disease-prone environmental groups selected are used as the environmental characteristic information of the target area.

[0007] In a preferred embodiment, when the adaptive fusion benchmark construction module adaptively adjusts the weights of the influence of image data and environmental data on the disease in the target region based on the dynamic correlation between the image feature information and the environmental feature information, and constructs a multi-source feature adaptive fusion benchmark for the target region, it is specifically used for: The image feature information is correlated and compared with the environmental feature information to obtain the image-environment feature correlation of the target area; Based on the consistency and magnitude matching of feature change directions in the image-environment feature correlation, the synergistic and antagonistic relationships between the image feature information and the environmental feature information are quantitatively evaluated to construct the feature correlation strength matrix of the target region. Based on the feature correlation strength matrix, the standard disease severity weight coefficient of maize in the target region is adjusted to obtain the dynamic feature weight allocation conditions of the target region; According to the dynamic feature weight allocation conditions, the sub-features in the image feature information and the sub-features in the environment feature information are weighted and fused, and the weighted feature values ​​are projected onto a unified feature vector space to construct a multi-source feature adaptive fusion benchmark for the target region.

[0008] In a preferred embodiment, when the adaptive fusion benchmark construction module performs a quantitative evaluation of the synergistic and antagonistic relationships between the image feature information and the environmental feature information based on the consistency and magnitude matching degree of feature change directions in the image-environment feature correlation to construct the feature correlation strength matrix of the target region, it is specifically used for: Monitor the changing trends of the image feature information and the environmental feature information over a continuous time series to obtain the image feature change vector and environmental feature change vector of the standard monitoring data; Based on the directions of the image feature change vector and the environmental feature change vector, the consistency of their change directions is evaluated, and the cooperative and antagonistic change relationships between the image feature information and the environmental feature information are identified. Amplitude change matching analysis is performed on the image feature change vector and the environmental feature change vector to determine the strength level of the cooperative change relationship and the antagonistic change relationship; By combining the cooperative change relationship, the antagonistic change relationship, and the strength level, a feature association strength matrix is ​​constructed to represent the dynamic correlation between the image feature information and the environmental feature information.

[0009] In a preferred embodiment, when the adaptive fusion benchmark construction module adjusts the standard disease severity weight coefficients of maize in the target region according to the feature correlation strength matrix to obtain the dynamic feature weight allocation conditions for the target region, it is specifically used for: Based on the feature association strength matrix, the quantized association values ​​of key feature categories in the image feature information and the environmental feature information are analyzed to form a key feature association set of the target region; In the key feature association set, the quantitative association values ​​with the same direction of change of the key feature category are grouped into the cooperative group, and the quantitative association values ​​with the opposite direction of change of the key feature category are grouped into the antagonistic group. Based on the comparison relationship between the synergistic group and the antagonistic group, the baseline weight coefficient of the standard disease severity in maize is dynamically adjusted to obtain the adjusted weight coefficient set of the target region; The adjusted set of weight coefficients is subjected to proportional balancing to ensure that the sum of the weight coefficients remains constant, thereby obtaining the dynamic feature weight allocation conditions for the target region.

[0010] In a preferred embodiment, when the feature evaluation and fusion module evaluates the degree of disease impact of the image feature information and the environmental feature information under the multi-source feature adaptive fusion benchmark to obtain the comprehensive feature index of the target region, it is specifically used for: Based on the feature weight relationship in the multi-source feature adaptive fusion benchmark, the sub-features in the image feature information and the sub-features in the environment feature information are weighted by importance to obtain the weighted feature vector of the target region; For the feature dimension of the weighted feature vector, the ability of the feature dimension to distinguish different disease levels in the historical disease sample set of the target area is analyzed to obtain the discriminative score of the feature dimension; The discriminative score is compared with the standard threshold of the corn to filter out the feature dimensions whose discriminative scores are higher than the standard threshold, so as to obtain a high-discriminative feature subset of the target region; The feature dimensions in the highly discriminative feature subset are aggregated to obtain the comprehensive feature index of the target region.

[0011] In a preferred embodiment, when the classification and identification module maps the comprehensive feature indicators to the corn disease classification knowledge base and identifies the disease type and disease level of the target area as the corn disease identification result, it is specifically used for: Extract disease feature templates corresponding to different disease types and disease levels from the corn disease classification knowledge base; The comprehensive feature index is matched with the standard feature template in the disease feature template to obtain the similarity comparison result of the disease feature template; Based on the similarity comparison results, the target feature templates for the corn are selected; Output the disease type identifier and disease level identifier corresponding to the target feature template as the disease identification result of the corn.

[0012] In a preferred embodiment, when the early warning decision generation module executes the early warning rule base based on the corn disease early warning rule base, and obtains the early warning information for the corn based on the disease identification results, the module specifically performs the following: Using the disease type and disease level in the identification results as the query basis, access the disease early warning mapping relationship in the disease early warning rule base of corn, and retrieve the target early warning level and continuous monitoring cycle conditions corresponding to the query basis; By retrospectively analyzing the comprehensive characteristic index sequence of the target area under the continuous monitoring cycle conditions, it is determined whether the disease status in the disease identification result meets the state threshold range corresponding to the target early warning level; When the judgment result meets the state threshold range, the disease type, disease level, target area identifier and corresponding prevention and control measures suggestions in the disease early warning rule base are combined and encoded according to the target early warning level to obtain the early warning instruction for the target area; When the judgment result does not meet the state threshold range, the disease identification result and the comprehensive characteristic index obtained in this monitoring cycle are stored as historical monitoring data in the associated database of the disease early warning rule base, which is used to update the trend analysis baseline in the disease early warning rule base.

[0013] In a preferred embodiment, when the early warning decision generation module performs a retrospective analysis of the comprehensive characteristic index sequence of the target area under the continuous monitoring period conditions, and determines whether the disease status in the disease identification result meets the state threshold range corresponding to the target early warning level, it is specifically used for: Based on the continuous monitoring cycle conditions, obtain the comprehensive feature index sequence in the disease early warning rule base; In the disease early warning rule base, the lower limit value and upper limit value of the feature indicators associated with the target early warning level are used as the preliminary state threshold of the disease early warning rule base; The comprehensive feature index values ​​in the comprehensive historical feature index sequence are compared with the lower limit and upper limit of the feature index, and the statistical results of the preliminary state threshold are obtained by statistically analyzing whether all the historical comprehensive feature index values ​​are within the range of the defined values ​​of the preliminary state threshold. Based on the statistical results, a list of records showing whether the disease status in the comprehensive feature index sequence meets the preliminary status threshold is generated. Based on the list of states that meet the requirements, determine the state threshold range for the target warning level.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention achieves accurate fusion and calibration of visible light image data and environmental data through a data fusion calibration module. After spatiotemporal alignment, illumination compensation, and abnormal parameter removal, standardized and high-quality monitoring data is obtained. Relying on a multi-dimensional feature extraction module, through local saliency enhancement and disease correlation screening, image feature information and environmental feature information reflecting maize disease characteristics are accurately extracted, providing reliable and effective data support for subsequent disease analysis and significantly improving the quality of basic data for disease monitoring.

[0015] 2. This invention utilizes an adaptive fusion benchmark construction module to dynamically adjust feature weights, and combines a feature evaluation and fusion module to screen highly discriminative features and aggregate them into comprehensive feature indicators. This enables the classification and identification module to accurately match disease types and levels. The early warning decision generation module generates early warning information containing control measures suggestions based on a disease early warning rule base and continuous monitoring cycle analysis. At the same time, it continuously updates the trend analysis baseline, which not only improves the efficiency and accuracy of disease monitoring and early warning, but also provides targeted guidance for corn disease control, helping to reduce the impact of diseases. Attached Figure Description

[0016] Figure 1 This is a system architecture diagram of a corn disease monitoring and early warning system based on agricultural Internet of Things provided in an embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0020] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0021] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0022] In practice, the server-side equipment deployed in a corn disease monitoring and early warning system based on agricultural IoT may consist of one or more devices. This corn disease monitoring and early warning system based on agricultural IoT can be implemented as: a business instance, a virtual machine, and hardware devices. For example, this corn disease monitoring and early warning system based on agricultural IoT can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, this corn disease monitoring and early warning system based on agricultural IoT can be understood as software deployed on a cloud node, used to provide corn disease monitoring and early warning systems based on agricultural IoT to various user terminals. Alternatively, this corn disease monitoring and early warning system based on agricultural IoT can also be implemented as a virtual machine deployed on one or more devices in a cloud node. This virtual machine contains application software for managing various user terminals. Alternatively, this corn disease monitoring and early warning system based on agricultural IoT can also be implemented as a server composed of numerous identical or different types of hardware devices, with one or more hardware devices configured to provide corn disease monitoring and early warning systems based on agricultural IoT to various user terminals.

[0023] In terms of implementation, the corn disease monitoring and early warning system based on agricultural IoT and the user terminal are mutually compatible. That is, if the corn disease monitoring and early warning system based on agricultural IoT is implemented as an application installed on a cloud service platform, then the user terminal is a client that establishes a communication connection with the application; or if the corn disease monitoring and early warning system based on agricultural IoT is implemented as a website, then the user terminal is implemented as a webpage; or if the corn disease monitoring and early warning system based on agricultural IoT is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0024] like Figure 1 The figure shown is a system architecture diagram of a corn disease monitoring and early warning system based on agricultural Internet of Things provided in an embodiment of the present invention.

[0025] The corn disease monitoring and early warning system 100 based on agricultural Internet of Things (IoT) of this invention can be set up in a cloud server. In terms of implementation, it can be implemented as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed as a website. Depending on the functions implemented, the corn disease monitoring and early warning system 100 based on agricultural IoT may include a data fusion calibration module 101, a multi-dimensional feature extraction module 102, an adaptive fusion benchmark construction module 103, a feature evaluation and fusion module 104, a classification and recognition module 105, and an early warning decision generation module 106. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by an electronic device's processor and perform a fixed function, stored in the electronic device's memory.

[0026] In this embodiment of the invention, in a corn disease monitoring and early warning system based on agricultural Internet of Things (IoT), each of the above-mentioned modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. The corn disease monitoring and early warning system based on agricultural IoT provided by this embodiment of the invention allows for adjustment of the applicability of the system architecture without modifying the program code. This is achieved by adding modules and directly calling them, enabling cluster-based horizontal expansion and flexibly expanding the corn disease monitoring and early warning system based on agricultural IoT. In practical applications, the above modules can be set in the same device or different devices, or they can be set in virtual devices, such as service instances in a cloud server.

[0027] The following describes, with reference to specific embodiments, the various components and specific workflow of a corn disease monitoring and early warning system based on agricultural Internet of Things: The data fusion calibration module 101 is used to fuse and calibrate the visible light image data of the target area with the corresponding environmental data to obtain the standard monitoring data of the target area; In this embodiment of the invention, when the data fusion calibration module performs fusion calibration of the visible light image data of the target area with the corresponding environmental data to obtain standard monitoring data of the target area, it is specifically used for: Based on the acquisition timestamp of the visible light image data in the target area, a time-series environmental data segment matching the acquisition timestamp is selected from the corresponding environmental data of the target area to obtain the spatiotemporal aligned data of the target area; Illumination compensation is performed on the visible light image data to obtain standardized visible light image data of the target area; By removing dimensionless abnormal environmental parameters from the spatiotemporal aligned data, uniform environmental data for the target region is obtained. The standardized visible light image data and the consistent environmental data are stitched together to obtain the standard monitoring data for the target area.

[0028] The acquisition timestamp of the visible light image data is clearly defined. This timestamp is a complete time information synchronously recorded by the image acquisition device at the instant the visible light image of the target area is acquired, covering the specific moment at the year, month, day, hour, minute, second, and millisecond level, and corresponding one-to-one with the image data to ensure that each visible light image can be traced back to the precise acquisition time. Subsequently, environmental data corresponding to the target area is extracted. This type of environmental data is continuously collected and generated by environmental monitoring equipment deployed in and around the target area, including core indicators related to the image acquisition scene and corn disease monitoring, such as temperature, humidity, light intensity, and soil moisture content. All environmental data are stored sequentially according to the order of acquisition, and each data point is accompanied by a time record with the same precision as the image acquisition timestamp, ensuring comparability in the time dimension. Next, using the acquisition timestamp of the visible light image as a benchmark, it is compared item by item with the time record of the environmental data. Priority is given to selecting environmental data segments whose time records perfectly match the acquisition timestamp. If no perfect match is found, a symmetrical time range is defined centered on the image acquisition timestamp based on the preset acquisition frequency of the environmental monitoring equipment. This range can completely cover the environmental state before and after the image acquisition time, avoiding the loss of environmental data due to differences in acquisition frequency. If there is no valid data within the defined range, the range can be expanded by 1.5 times the acquisition frequency, and all relevant environmental data can be selected from the finally defined time range. The selected environmental data segments are then associated with the corresponding visible light image data to ensure that each visible light image data can correspond to an environmental data segment reflecting the environmental state of the same target area at the same time, achieving dual alignment in time and space, and finally forming spatiotemporally aligned data of the target area.

[0029] A comprehensive illumination state analysis is performed on the visible light image data of the target area. The brightness value of the image is read pixel by pixel, the mean and standard deviation of the brightness of all pixels are calculated, and the region type is determined based on the statistics: dark regions are defined as pixel brightness values ​​< μ-1.2σ, ​​overexposed regions are defined as pixel brightness values ​​> μ+1.2σ, ​​and normally lit regions are defined as pixel brightness values ​​in the range of [μ-1.2σ, ​​μ+1.2σ]. At the same time, the coordinate range and number of pixels of each region are recorded. For darker areas, the brightness of each pixel is compared with the average brightness of the overall image. The brightness of each pixel is gradually increased according to the degree of difference. After each adjustment, the difference between the pixel and the average brightness is compared again until the difference falls within the reasonable range of [μ-0.3σ, μ+0.3σ], ensuring that the image details in the darker areas are clearly presented and not blurred. For overexposed areas, the difference between each pixel and the average brightness is calculated, and the pixel brightness is gradually reduced proportionally until the difference is within the reasonable range mentioned above, avoiding the loss of image details due to excessive brightness, while maintaining the color gradation in the area. For areas with normal lighting, their original brightness is kept unchanged without additional adjustments. During the brightness adjustment process, the RGB color components of each pixel are simultaneously verified to ensure that the adjusted color ratio is consistent with the real scene and that there are no issues such as color cast or fading. For the edge transitions of different illuminated areas in the image, a progressive brightness adjustment method is adopted to make the brightness of adjacent areas smoothly connected without obvious light and dark segmentation marks. After precise adjustment, color verification, and edge transition processing of all pixels, standardized visible light image data of the target area with uniform brightness, true color, and complete details is finally obtained.

[0030] Historical environmental data of the target area was collected, and the dimensional information of various core environmental parameters such as temperature (°C), relative humidity (%RH), light intensity (lux), soil moisture content (%), and wind speed (m / s) were recorded simultaneously. Statistical analysis was carried out for parameters of different dimensions—the mean, standard deviation, 25th percentile, 75th percentile, and 99th percentile of each parameter were calculated. At the same time, Z-score standardization was used to eliminate dimensional differences. The standardization formula is: Z=(x-μ) / σ, where x is the original value of the parameter, resulting in a dimensionless unified analysis dataset. Based on the climate characteristics, soil types, and local agricultural production practices of the target area, and using a standardized dataset as a foundation, a dimensionless value range |Z-score|≤3 is determined to cover 99% of normal environmental conditions. This range can accurately exclude extreme values ​​caused by natural disasters or sensor malfunctions, and objectively reflect the natural fluctuation patterns of parameters under normal agricultural production scenarios. The specific value of each parameter is checked one by one and compared with the determined normal value range. If the value of an environmental parameter exceeds the normal range, it is directly determined as an abnormal environmental parameter. Through data filtering operations, the abnormal parameter is completely removed from the spatiotemporally aligned data to avoid abnormal data interfering with the subsequent fusion results. After removing abnormal parameters, the integrity of the remaining environmental parameters is checked. If the remaining parameters can fully reflect the environmental state of the target area and no key information is missing, all remaining parameters are retained. If key environmental parameters are missing, environmental parameter values ​​of the same type as the missing parameters and within the normal range are selected from adjacent time points to supplement them, based on the normal distribution pattern of historical environmental data for this scene. This ensures that the supplemented parameters can truly reflect the environmental state at the time of image acquisition and are logically consistent with other normal parameters. Through the judgment and removal of abnormal parameters and the supplementation and improvement of key parameters, the final result is consistent environmental data of the target area where all parameters are within the normal range, logically coherent, and information complete.

[0031] Deep analysis is performed on standardized visible light image data of the target area to extract core information, including scene details carried by the pixel matrix, scene feature descriptions, and the spatial coordinate range covered by the image. During extraction, the original accuracy and detail integrity of the image are maintained without any compression or simplification. Simultaneously, the standardized environmental data is structured and categorized according to the type of environmental parameters, clearly defining the specific acquisition location, data unit, and accurate value for each parameter. This ensures the clarity of the environmental data and facilitates subsequent matching with the image data. Using the spatial coordinate range of the standardized visible light image data as the core correlation basis, the spatial coordinate information of the image is precisely matched with the acquisition location information of the standardized environmental data. This ensures that each spatial coordinate point in the image corresponds to various environmental parameters at that location. For example, the pixel features of a coordinate point in the image are bound to environmental parameters such as temperature and humidity at that location. Furthermore, the overall scene features of the image are correlated with the overall environmental parameter statistics of the target area, ensuring a complete spatial correspondence between the image information and the environmental data, without any misalignment or discrepancies. During the binding process, a data tag association method is adopted to add a unified identification tag to each set of successfully matched image data and environmental data, which facilitates subsequent data retrieval and analysis. At the same time, the original attributes and accuracy of the data are strictly maintained throughout the integration process, without changing any core information of the data. Through the structured integration, precise spatial matching and tag association of the two types of data, a complete data set that includes both intuitive scene information and quantitative environmental parameters is finally formed, namely the standard monitoring data of the target area.

[0032] The beneficial effects include: by precisely matching the acquisition time of visible light image data and environmental data, a high degree of spatiotemporal alignment is achieved, ensuring the correlation and consistency of data at both temporal and spatial levels, laying a solid foundation for subsequent fusion calibration. Targeted illumination compensation processing effectively corrects abnormal illumination issues such as underexposure and overexposure in images, resulting in standardized visible light image data with uniform brightness, true color, and complete detail, thus improving the quality and usability of the image data. By filtering out abnormal environmental parameters and supplementing key information, the accuracy, rationality, and completeness of the consistent environmental data are guaranteed, preventing abnormal data from interfering with the fusion results. Through structured integration and precise spatial matching of standardized visible light image data and consistent environmental data, deep fusion of the two types of data is achieved. The resulting standard monitoring data includes both image information that intuitively reflects the target area scene and parameter information that quantitatively describes the environmental state. The data is comprehensive, logically coherent, accurate, and reliable, providing high-quality data support for relevant monitoring analysis and decision-making in the target area, enhancing the scientific rigor and practicality of the monitoring results.

[0033] The multi-dimensional feature extraction module 102 is used to extract multi-dimensional features from the standard monitoring data to obtain image feature information and environmental feature information of the target area. In this embodiment of the invention, when the multi-dimensional feature extraction module performs multi-dimensional feature extraction on the standard monitoring data to obtain image feature information and environmental feature information of the target area, it is specifically used for: The color distribution statistics and texture roughness information of the visible light image data in the standard monitoring data are used as the initial image features of the target region. A sliding time window analysis is performed on the environmental data in the standard monitoring data, and the mean, fluctuation range and trend information in the analysis results are integrated into the initial environmental characteristics of the target area. The initial image features are locally saliency-enhanced, and the features describing the color anomalies and texture structure changes of the corn canopy in the enhanced image are collected as the image feature information of the target region; The initial environmental characteristics are screened for disease relevance, and the disease-prone environmental groups selected are used as the environmental characteristic information of the target area.

[0034] Visible light image data is separated from standard monitoring data. Pixel-by-pixel color information is extracted from this image data, and the specific performance of the red, green, and blue color components of each pixel is read. The distribution range, frequency, and proportion of each color component in the entire image are statistically analyzed. Simultaneously, the overall performance level of each color component is calculated. These statistical results are quantified into texture roughness information through a process of "difference calculation - statistical analysis - regional comparison - numerical assignment". First, the image pixels are traversed row by row and column by column, calculating the color difference and grayscale difference between each pixel and its neighboring pixels. Next, the mean and variance of the differences between all adjacent pixels are calculated. The larger the mean / variance, the more drastic the pixel change, and the coarser the texture of the corresponding area. Then, the difference statistical indicators of different regions are compared to clarify the differences in roughness in each region. Finally, a quantified value is assigned according to the magnitude of the difference statistical indicators. For example, the difference mean is divided into three levels: high, medium, and low, corresponding to high, medium, and low roughness quantification values, respectively, or the difference variance is directly used as the roughness quantification result, ultimately forming texture roughness information that accurately reflects the surface roughness characteristics of the image.

[0035] To analyze the texture information of the image, pixels are traversed row by row and column by column to analyze the changes in color and grayscale between adjacent pixels. The density, regularity, and undulation of pixel arrangement are observed to determine the roughness features of the image surface. By comparing the differences in pixel changes in different regions, the roughness of the texture is quantified, forming texture roughness information. The obtained color distribution statistics and texture roughness information are then correlated and integrated to ensure that both types of information completely correspond to the visible light image data of the same target area, ultimately yielding the initial image features of the target area.

[0036] All environmental data are extracted from standard monitoring data and arranged sequentially according to their collection time to ensure temporal continuity. A fixed-length time window is set to fully cover environmental data changes within a certain period. Starting from the initial moment of the environmental data, the time window is moved sequentially along the time axis, covering a new time period with each move, until all environmental data have been traversed. For the environmental data within each time window, the overall average level of all data values ​​is calculated to obtain the mean. The maximum and minimum values ​​within the window are identified, and the difference between them determines the fluctuation range. The direction of data change over time is observed to determine whether it is increasing, decreasing, or stable, thus obtaining trend information. The mean, fluctuation range, and trend information corresponding to each time window are summarized to ensure that all information is derived from time-series analysis of environmental data and fully reflects the changing characteristics of environmental data. Finally, these are integrated to form the initial environmental characteristics of the target area.

[0037] The visible light image information in the initial image features is used to locate the corn canopy region. By identifying areas in the image that conform to the morphological and color characteristics of the corn canopy, the specific distribution range of the corn canopy in the image is clarified, eliminating interference from non-canopy areas. For the located corn canopy region, color and texture structure standards for corn canopies under normal growth conditions are established. These standards are determined based on the general characteristics of healthy corn canopies, including normal color ranges and texture arrangement patterns. The color distribution statistics and texture roughness information in the initial image features are compared region by region with the above standards to identify areas where the color deviates from the normal range and abnormal areas of texture structure. Local signal enhancement processing is performed on these abnormal areas. By adjusting the color contrast and texture clarity of the abnormal areas, the features of color anomalies and texture structure changes are made more prominent, while the features of non-abnormal areas remain unchanged to avoid overall image feature distortion. All enhanced features describing corn canopy color anomalies and texture structure changes are collected and aggregated to ensure that the feature information comprehensively covers all abnormal areas and accurately reflects the specific manifestations of the anomalies, ultimately obtaining the image feature information of the target area.

[0038] The occurrence patterns and related research findings of common maize diseases were collected and compiled into a set of environmental conditions conducive to maize disease occurrence. This set includes the specific ranges and combinations of environmental parameters such as temperature, humidity, and light directly related to disease occurrence, serving as the basis for disease correlation screening. Each environmental parameter and its variation information in the initial environmental characteristics were broken down item by item, clarifying the specific values, fluctuations, and trends of each parameter. The decomposed initial environmental characteristics were compared item by item with the set of environmental conditions conducive to maize disease occurrence, analyzing the degree of fit between each environmental parameter and parameter combination and the disease-prone conditions. Environmental parameters with significant correlation to disease occurrence and their corresponding fluctuation amplitudes and trends were retained, while environmental information irrelevant to disease occurrence or with very weak correlation was eliminated. The retained environmental parameters were combined according to their role in the disease occurrence process to form environmental parameter combinations that reflect the risk of disease occurrence. These combinations accurately correspond to the environmental states conducive to maize disease occurrence. Finally, these combinations were aggregated into a disease-prone environmental group for the target area, thus obtaining the environmental characteristic information of the target area.

[0039] The beneficial effects include the accurate extraction of color distribution statistics and texture roughness information from standard monitoring data into visible light images as initial image features, providing a solid foundation for subsequent image feature analysis. A systematic analysis of environmental data using a sliding time window integrates mean, fluctuation amplitude, and trend into initial environmental features, comprehensively reflecting the temporal variation patterns of environmental data. Local saliency enhancement of the initial image features accurately aggregates characteristics of maize canopy color anomalies and texture structure changes, making the image feature information more targeted and effectively locating crop growth anomalies. The initial environmental features are screened based on maize disease-prone environmental conditions, resulting in disease-prone environmental groups focusing on disease-related factors, enhancing the correlation between environmental features and disease monitoring. Finally, the acquired image and environmental feature information accurately captures key features from the dimensions of crop morphology and growth environment, providing comprehensive and reliable feature data support for monitoring maize growth status and disease early warning in target areas, assisting in subsequent accurate analysis and decision-making.

[0040] The adaptive fusion benchmark construction module 103 is used to adaptively adjust the weights of the influence of image data and environmental data on the disease in the target area according to the dynamic correlation between the image feature information and the environmental feature information, and construct a multi-source feature adaptive fusion benchmark for the target area. In this embodiment of the invention, when the adaptive fusion benchmark construction module adaptively adjusts the weights of the influence of image data and environmental data on the disease in the target area based on the dynamic correlation between the image feature information and the environmental feature information, and constructs a multi-source feature adaptive fusion benchmark for the target area, it is specifically used for: The image feature information is correlated and compared with the environmental feature information to obtain the image-environment feature correlation of the target area; Based on the consistency and magnitude matching of feature change directions in the image-environment feature correlation, the synergistic and antagonistic relationships between the image feature information and the environmental feature information are quantitatively evaluated to construct the feature correlation strength matrix of the target region. Based on the feature correlation strength matrix, the standard disease severity weight coefficient of maize in the target region is adjusted to obtain the dynamic feature weight allocation conditions of the target region; According to the dynamic feature weight allocation conditions, the sub-features in the image feature information and the sub-features in the environment feature information are weighted and fused, and the weighted feature values ​​are projected onto a unified feature vector space to construct a multi-source feature adaptive fusion benchmark for the target region.

[0041] The adaptive fusion benchmark construction module, when performing a quantitative evaluation of the synergistic and antagonistic relationship between the image feature information and the environmental feature information based on the consistency and magnitude matching degree of feature change direction in the image-environment feature correlation to construct the feature correlation strength matrix of the target region, is specifically used for: Monitor the changing trends of the image feature information and the environmental feature information over a continuous time series to obtain the image feature change vector and environmental feature change vector of the standard monitoring data; Based on the directions of the image feature change vector and the environmental feature change vector, the consistency of their change directions is evaluated, and the cooperative and antagonistic change relationships between the image feature information and the environmental feature information are identified. Amplitude change matching analysis is performed on the image feature change vector and the environmental feature change vector to determine the strength level of the cooperative change relationship and the antagonistic change relationship; By combining the cooperative change relationship, the antagonistic change relationship, and the strength level, a feature association strength matrix is ​​constructed to represent the dynamic correlation between the image feature information and the environmental feature information.

[0042] When the adaptive fusion benchmark construction module adjusts the standard disease severity weight coefficients of maize in the target region according to the feature correlation strength matrix to obtain the dynamic feature weight allocation conditions for the target region, it is specifically used for: Based on the feature association strength matrix, the quantized association values ​​of key feature categories in the image feature information and the environmental feature information are analyzed to form a key feature association set of the target region; In the key feature association set, the quantitative association values ​​with the same direction of change of the key feature category are grouped into the cooperative group, and the quantitative association values ​​with the opposite direction of change of the key feature category are grouped into the antagonistic group. Based on the comparison relationship between the synergistic group and the antagonistic group, the baseline weight coefficient of the standard disease severity in maize is dynamically adjusted to obtain the adjusted weight coefficient set of the target region; The adjusted set of weight coefficients is subjected to proportional balancing to ensure that the sum of the weight coefficients remains constant, thereby obtaining the dynamic feature weight allocation conditions for the target region.

[0043] All sub-features were extracted from the image feature information of the target area, including specific manifestations of abnormal corn canopy color and detailed features of texture structure changes. Each sub-feature clearly corresponds to a specific observable phenomenon in the image. Simultaneously, all sub-features were extracted from the environmental feature information, covering the values, fluctuation ranges, and trends of various environmental parameters in the disease-prone environment group, ensuring that both types of sub-features are complete and without omission. A correspondence was established between image sub-features and environmental sub-features according to the disease's impact logic. For example, the image sub-feature of yellowing canopy was paired with the environmental sub-feature of persistently high humidity, and the image sub-feature of texture breakage was paired with the environmental sub-feature of abnormal temperature fluctuations, ensuring that each paired sub-feature pair could potentially contribute to the occurrence or development of the disease. Correlation analysis was performed on each paired sub-feature pair to observe whether the two types of sub-features appeared within the same monitoring period, determining whether their changes were synchronous. If the corresponding environmental sub-feature changed when a certain image sub-feature appeared, a correlation was determined; if the corresponding environmental sub-feature did not change when the image sub-feature appeared, the correlation was determined to be weak or non-corresponding. The association analysis results of all paired sub-features are summarized to clarify the association between various sub-features, including the number of associated sub-feature pairs and the degree of association, and finally the image environment feature association of the target region is formed.

[0044] The criteria for judging the consistency of feature change directions are clearly defined. If an image sub-feature shows an increasing trend of disease signs, and the corresponding environmental sub-feature also shows a trend conducive to disease occurrence, then the two are judged to have consistent change directions. If an image sub-feature shows an increasing trend of disease signs, while the corresponding environmental sub-feature shows a trend inhibiting disease occurrence, then the change directions are judged to be inconsistent. Simultaneously, the criteria for judging the matching degree of feature change amplitude are determined. The obviousness of image sub-feature changes is compared with the drasticness of environmental sub-feature changes. If the image sub-feature shows significant abnormalities, and the corresponding environmental sub-feature changes are also relatively drastic, then the amplitude matching degree is judged to be high. If the image sub-feature abnormalities are slight while the environmental sub-feature changes drastically, or vice versa, then the amplitude matching degree is judged to be low. Based on the above criteria, each pair of paired sub-features in the image-environment feature correlation is evaluated one by one. If the change directions are consistent and the amplitude matching degree is high, it is judged to be a strong synergistic relationship; if the change directions are consistent but the amplitude matching degree is low, it is judged to be a weak synergistic relationship; if the change directions are inconsistent but the amplitude matching degree is low, it is judged to be a weak antagonistic relationship; if the change directions are inconsistent but the amplitude matching degree is high, it is judged to be a strong antagonistic relationship. Each relationship is assigned a specific quantitative representation: strong collaboration corresponds to the highest representation value, weak collaboration to a medium-high representation value, no association to a medium representation value, weak antagonism to a medium-low representation value, and strong antagonism to the lowest representation value. The quantitative representation results of all paired sub-features are arranged in matrix form, with rows corresponding to image sub-features and columns corresponding to environment sub-features. Each element in the matrix is ​​the quantitative value of the association strength of the corresponding paired sub-feature, ultimately constructing a feature association strength matrix for the target region.

[0045] The initial values ​​for the standard disease severity weighting coefficients of maize are determined. These initial values ​​are set based on conventional experience in maize disease monitoring. The overall initial weights of image feature information and environmental feature information, as well as the initial weight proportion of each type of sub-feature within its respective feature information, are clearly defined to ensure that the initial weight allocation conforms to the logic of general disease monitoring. All quantified values ​​in the feature association strength matrix are extracted, and the association strength between each image sub-feature and all environmental sub-features is analyzed one by one. If a certain image sub-feature shows a strong synergistic relationship with most environmental sub-features, it indicates that the image sub-feature has high reference value for disease judgment, and its weight coefficient should be increased. If a certain image sub-feature shows a weak or antagonistic relationship with most environmental sub-features, its weight coefficient is decreased. Similarly, the association strength between each environmental sub-feature and all image sub-features is analyzed, and its weight coefficient is adjusted according to the proportion of synergistic or antagonistic relationships. A high proportion of synergistic relationships increases the weight, while a high proportion of antagonistic relationships decreases the weight. During the adjustment process, the sum of all sub-feature weight coefficients is kept constant to avoid overall weight imbalance caused by adjusting a single feature weight. The basis for each sub-feature weight adjustment is recorded to ensure a direct correspondence between the adjustment result and the quantization result of the feature association strength matrix. The adjusted weight coefficients of all image and environment sub-features are categorized and organized, clarifying the final weight value for each sub-feature and the corresponding logic between weight adjustment and feature association strength. This ultimately yields the dynamic feature weight allocation conditions for the target region.

[0046] The system comprehensively breaks down image and environmental features, ensuring that all sub-features are fully extracted. Image sub-features include specific manifestations of various color anomalies in the corn canopy and detailed features of different types of texture structure changes. Environmental sub-features include the value range, fluctuation patterns, and trends of various disease-prone environmental parameters. Each sub-feature has a clear descriptive dimension and judgment criteria. According to dynamic feature weight allocation conditions, each sub-feature is assigned a corresponding weight value. The weight value directly reflects the degree of influence of the sub-feature on disease judgment; the higher the weight, the stronger its decisive influence on disease monitoring results. The feature performance of each sub-feature is quantitatively described. For example, the "canopy browning area" in the image sub-feature is given a specific description based on the size of the coverage area, and the "humidity fluctuation" in the environmental sub-feature is given a specific description based on the frequency of fluctuations. This ensures that the performance of each sub-feature can be transformed into concrete information that can be weighted and calculated. The quantitative description results of each sub-feature are integrated with their corresponding weight values. Weighting is achieved through direct association; that is, the weighted feature value of each sub-feature is a direct correspondence between its quantitative performance and weight, without the need for additional complex calculations, ensuring that the weighting process is transparent and traceable. A unified feature vector space is established, with the number of dimensions matching the total number of extracted sub-features. Each dimension corresponds to a unique sub-feature, and the value range of each dimension is uniformly set based on the quantization range of all sub-features, ensuring that all weighted feature values ​​can be included in this space. The weighted feature values ​​of each sub-feature are sequentially filled into the corresponding dimensions of the feature vector space in a preset order, ensuring that each weighted feature value has a unique and fixed position in the space, forming a complete multi-source feature vector. In this way, the weighted information of the originally scattered image sub-features and environmental sub-features is integrated into the same analytical framework, enabling different types of features to be correlated and synergistically interact, ultimately constructing a multi-source feature adaptive fusion benchmark for the target area that comprehensively reflects the combined impact of image features and environmental features on the disease.

[0047] Continuous time-series information is extracted from standard monitoring data. This time series is arranged naturally according to the chronological order of data collection, fully covering the entire monitoring period of the target area to ensure the capture of dynamic changes in features over time. Image feature information is decomposed into all sub-features, including different types of color anomalies in the corn canopy and various texture structure changes, each corresponding to a specific observable phenomenon in the image. Simultaneously, environmental feature information is also decomposed into all sub-features, covering various environmental parameters, fluctuation states, and trends in disease-prone environments, ensuring no omissions and clear definitions for both types of sub-features. For each image sub-feature, its specific manifestation at each monitoring moment is extracted along the continuous time series. For example, the coverage area and color intensity of a certain canopy color anomaly sub-feature at different times are tracked, and its changes from the initial moment to subsequent moments are arranged chronologically to form the change sequence of that image sub-feature. The same operation is performed on all image sub-features, integrating all image sub-feature change sequences to form an image feature change vector for the target area. Each element in the vector corresponds to the complete change trajectory of an image sub-feature over the continuous time series. Each environmental sub-feature is processed in the same way, and its value, fluctuation range and trend at each monitoring time are extracted along the time series. Its evolution over time is tracked and a change sequence is formed. The change sequences of all environmental sub-features are integrated to obtain the environmental feature change vector of the target area. This vector is completely consistent with the time dimension of the image feature change vector, ensuring that the two can be analyzed in a time-level correspondence.

[0048] The criteria for judging the direction of change of each sub-feature in the image feature change vector are clearly defined. If the change of an image sub-feature shows a trend that is conducive to disease judgment, such as the expansion of abnormal areas in the corn canopy or the deepening of abnormality, then the direction of change of that sub-feature is judged as positive; if it shows a trend that inhibits disease judgment, such as the shrinkage of abnormal areas or the reduction of abnormality, then it is judged as negative. Simultaneously, the criteria for judging the direction of change of each sub-feature in the environmental feature change vector are also established. Each sub-feature change direction in the image feature change vector is compared one by one with the corresponding paired sub-feature change direction in the environmental feature change vector. If both are positive or both are negative, it indicates that the two types of feature changes have the same trend of influence on disease judgment, and a synergistic change relationship is determined. If one is positive and the other is negative, it indicates that the two types of feature changes have opposite trends of influence on disease judgment, and an antagonistic change relationship is determined. After comparing all paired sub-feature change directions, the judgment results of all synergistic and antagonistic change relationships are summarized, and the number and specific combination of sub-feature pairs corresponding to each relationship are determined to ensure that the relationship identification results are comprehensive and accurate.

[0049] A metric for measuring the amplitude changes of sub-features in the image feature change vector was established. This was done by comparing the differences in the performance of sub-features at the initial and subsequent times in a continuous time series. The amplitude changes of image feature sub-features were determined using a dual dimension: the coverage area of ​​the abnormal region and the degree of abnormality. A large amplitude change was defined as the abnormal region's coverage area expanding by 50% or more compared to the initial time, with a significant degree of abnormality—meaning a clear color difference between the abnormal region and the healthy corn canopy, and prominent texture damage. A moderate amplitude change was defined as the abnormal region's coverage area expanding by 20% to 49%, with moderate color differences and texture damage, and no drastic changes. A small amplitude change was defined as the abnormal region's coverage area expanding by less than 20% or not expanding at all, with color and texture resembling the healthy corn canopy, and no significant deepening of abnormality. Simultaneously, a metric for measuring the amplitude changes of sub-features in the environmental feature change vector was also established. This involved comparing the differences in values ​​and fluctuation amplitudes of sub-features at the initial and subsequent times in the time series. The amplitude changes of environmental feature sub-features were judged based on the shift of parameters towards disease-prone areas and the severity of fluctuations. When a parameter value shifts 40% or more towards the disease-prone area, and the parameter's temporal fluctuations are significantly enhanced, indicating poor value stability, it is judged as a large amplitude change. When the parameter's shift towards the disease-prone area is between 20% and 39%, it is judged as a moderate amplitude change. When the parameter's shift towards the disease-prone area is less than 20%, it is judged as a small amplitude change. The same operation is performed on sub-feature pairs corresponding to antagonistic changes. Based on the combination of amplitude change levels of the two sub-features, they are sequentially judged into three levels: strong, medium, and weak. This ensures that each synergistic or antagonistic change relationship has a clear strong / weak level classification, and the classification criteria are directly related to the characteristic amplitude change.

[0050] The absolute value of the change in each feature parameter in the image feature change vector is taken to obtain the amplitude value of the change in each image feature parameter. Similarly, the absolute value of the change in each feature parameter in the environment feature change vector is taken to obtain the amplitude value of the change in each environment feature parameter. The amplitude values ​​of the corresponding feature parameter changes in the image feature change vector and the environment feature change vector are compared, and the ratio of the amplitude values ​​of the corresponding feature parameter changes is calculated. A fixed amplitude matching threshold range is set. If the ratio falls within the preset amplitude matching threshold range, the amplitude change of the corresponding feature parameter is determined to match. The number of feature parameters with matching amplitude changes is counted, and this number is divided by the total number of feature parameters. The quantity of data is used to obtain the amplitude matching ratio. Combined with the previously obtained consistency of change direction, three clear strength / weakness level judgment criteria are established. For cooperative change relationships, when both the consistency ratio and amplitude matching ratio are greater than or equal to 80%, it is judged as a strong cooperative level; when both are between 60% and 80%, it is judged as a medium cooperative level; when both are below 60%, it is judged as a weak cooperative level. For antagonistic change relationships, when both the inconsistency in change direction ratio and the amplitude mismatch ratio are greater than or equal to 80%, it is judged as a strong antagonistic level; when both are between 60% and 80%, it is judged as a medium antagonistic level; when both are below 60%, it is judged as a weak antagonistic level. A matrix framework is constructed, where each row corresponds to an image sub-feature in the image feature change vector, and each column corresponds to an environmental sub-feature in the environmental feature change vector. The number of rows in the matrix is ​​the same as the total number of image sub-features, and the number of columns is the same as the total number of environmental sub-features, ensuring that the matrix covers the association of all sub-feature pairs. Iterate through all paired sub-feature pairs, and based on the previously determined cooperative or antagonistic change relationships and their corresponding strength levels, find the corresponding quantization value and fill it into the intersection of the corresponding row and column in the matrix. After filling all sub-feature pairs with quantization values, perform a comprehensive validation of the matrix to ensure that the quantization value at each position accurately matches the relationship type and strength level of the sub-feature pair, with no errors or omissions. The matrix should fully represent the strength and type distribution of the dynamic correlation between image feature information and environmental feature information, ultimately forming a feature correlation strength matrix for the target region.

[0051] All quantified correlation values ​​were extracted from the feature correlation strength matrix. These values ​​clearly correspond to the correlation strength and relationship type of each sub-feature pair in the image and environmental feature information. Subsequently, based on the core needs of maize disease monitoring, key feature categories that directly influence disease occurrence and development were screened using clearly defined criteria. The criteria for direct influence are: features that are direct causes of disease with clear pathological mechanisms, providing necessary conditions for pathogen reproduction and spread, directly representing the direct morphological and structural changes produced after the disease acts on the maize plant, and serving as intuitive evidence of disease occurrence. The criteria for indirect influence are: features that act on the disease through one or more intermediate links, have no direct pathological support for disease association, or exhibit significant lag; such indirect influence features were directly excluded during the screening process. Specifically, key feature categories in image feature information include the core manifestations of maize canopy color abnormalities and the main types of texture structure changes, both direct representations of disease effects. Key feature categories in environmental feature information include environmental parameters and their changing trends that play a decisive role in disease-prone environments, both direct causes of disease occurrence, ensuring that key feature categories accurately reflect the core influencing factors related to the disease. For each key feature category, locate its intersection with all other key feature categories in the feature association strength matrix, extract the corresponding quantitative association value at that position, complete the extraction of quantitative association values ​​between all key feature categories one by one, bind the extracted quantitative association values ​​with the corresponding key feature category combination relationship, clarify the two key feature categories corresponding to each quantitative association value, ensure that the association relationship is clear and traceable, and finally summarize the quantitative association values ​​and corresponding combination relationships between all key feature categories to form the key feature association set of the target region.

[0052] The direction of change for the two key feature categories corresponding to each quantified correlation value in the key feature association set is clearly defined. This direction of change is determined based on the previous judgment of the direction of image feature change vectors and environmental feature change vectors, meaning that each key feature category has been clearly identified as either a positive or negative change. The direction of change for the two key feature categories corresponding to each quantified correlation value is checked one by one. If the two key feature categories have the same direction of change, whether both are positive or both are negative, it indicates that their influence on the disease is consistent, and the quantified correlation value is separated from the key feature association and assigned to the synergistic group. If the two key feature categories have opposite directions of change, one positive and the other negative, it indicates that their influence on the disease is contradictory, and the quantified correlation value is separated and assigned to the antagonistic group. During the classification process, each quantified correlation value is individually labeled, clearly defining its group and corresponding key feature category combination to ensure no quantified correlation values ​​are missed or misclassified. After classification, all quantified correlation values ​​of the synergistic and antagonistic groups are integrated to form clearly structured and clearly assigned synergistic and antagonistic groups.

[0053] A baseline weighting coefficient for the standard severity of maize diseases is determined. This coefficient is set based on routine experience and general patterns in maize disease monitoring, clarifying the initial weight percentage of each key characteristic category in disease assessment to ensure that the initial weight allocation meets the needs of general monitoring scenarios. A comprehensive analysis is conducted on the quantitative correlation values ​​in the synergistic and antagonistic groups. The overall strength level of the quantitative correlation values ​​in the synergistic group is statistically analyzed. If the overall quantitative correlation values ​​in the synergistic group are high, it indicates a strong synergistic effect between the corresponding key characteristic categories, which has greater reference value for disease assessment, and the baseline weighting coefficients for these key characteristic categories need to be increased. The overall strength level of the quantitative correlation values ​​in the antagonistic group is also statistically analyzed. In specific adjustments, the adjustment range is determined based on the differences in the strength of the quantitative correlation values; the higher the correlation strength, the larger the adjustment range, ensuring that the adjusted weighting coefficients accurately reflect the actual impact value of the key characteristic categories. After targeted adjustments to the baseline weighting coefficients for all key characteristic categories, all adjusted weighting coefficients are summarized to form an adjusted weighting coefficient set for the target area.

[0054] A constant standard for the sum of weight coefficients is defined. This standard is based on the conventional logic of feature weight allocation in disease diagnosis, ensuring that the sum of weight coefficients for all key feature categories remains fixed, thus avoiding fluctuations in the total weight sum from affecting the accuracy of subsequent feature fusion. The sum of all weight coefficients in the adjusted weight coefficient set is calculated, and the result is compared with the constant standard to confirm the constant standard for the sum of weight coefficients. This standard, based on the conventional logic of feature weight allocation in disease diagnosis, uniformly fixes the sum of weight coefficients for all key feature categories to a standard value of 1.0. This standard ensures that the weight allocation conforms to normalization logic, avoiding fluctuations in the total weight sum from affecting the accuracy of subsequent feature fusion. The sum of all weight coefficients in the adjusted weight coefficient set is calculated, and the result is precisely compared with the standard value of 1.0, while setting an allowable small deviation range of ±0.01 to determine whether the sum meets the constant requirement. If the adjusted sum falls within the deviation range of 1.0 ± 0.01, it is considered to meet the constant standard and no further adjustment is needed. If the adjusted sum is higher than 1.001 or lower than 0.999, it is considered to deviate from the constant standard and a proportional balancing process needs to be initiated. During the proportional balancing process, each weight coefficient is scaled proportionally according to the original relative proportions of all key feature categories to ensure that the relative importance of each feature category remains unchanged after adjustment. The adjustment ratio, the original value before adjustment, and the final value after adjustment for each weight coefficient are recorded in detail to ensure the traceability of the entire process. The final result is a dynamic feature weight allocation bar for the target area where the sum of the weight coefficients of all key feature categories strictly conforms to the 1.0 ± 0.01 standard range and the relative proportions are reasonable. During the proportional balancing process, the adjustment process of each weight coefficient is recorded, clearly specifying the adjustment ratio and the numerical changes before and after adjustment to ensure the traceability of the process. The final result is a dynamic feature weight allocation condition for the target area where the sum of the weight coefficients of all key feature categories is constant and the relative proportions are reasonable.

[0055] The feature evaluation and fusion module 104 is used to evaluate the degree of disease impact of the image feature information and the environmental feature information under the multi-source feature adaptive fusion benchmark, so as to obtain the comprehensive feature index of the target area. In this embodiment of the invention, when the feature evaluation and fusion module evaluates the degree of disease impact of the image feature information and the environmental feature information under the multi-source feature adaptive fusion benchmark to obtain the comprehensive feature index of the target region, it is specifically used for: Based on the feature weight relationship in the multi-source feature adaptive fusion benchmark, the sub-features in the image feature information and the sub-features in the environment feature information are weighted by importance to obtain the weighted feature vector of the target region; For the feature dimension of the weighted feature vector, the ability of the feature dimension to distinguish different disease levels in the historical disease sample set of the target area is analyzed to obtain the discriminative score of the feature dimension; The discriminative score is compared with the standard threshold of the corn to filter out the feature dimensions whose discriminative scores are higher than the standard threshold, so as to obtain a high-discriminative feature subset of the target region; The feature dimensions in the highly discriminative feature subset are aggregated to obtain the comprehensive feature index of the target region.

[0056] Based on the dynamic feature weight allocation conditions explicitly defined in the multi-source feature adaptive fusion benchmark, adaptive weight coefficients corresponding to sub-features in each image feature information and environmental feature information are first extracted. These coefficients are adjusted and determined according to the synergistic and antagonistic relationships quantized by the feature association strength matrix, directly reflecting the importance of each sub-feature to disease judgment. Subsequently, all sub-features are uniformly quantized, transforming the original representations of image sub-features and the original values ​​of environmental sub-features into standardized quantized values, ensuring that the value ranges of different types of sub-features are consistent and can directly participate in weighted calculations. Following the rule of "sub-feature weighted value = sub-feature standardized quantized value × corresponding adaptive weight coefficient," the weighted value of each sub-feature is calculated one by one. The higher the weight coefficient of a sub-feature, the greater its contribution to the final result. Finally, according to the preset feature dimension order, the weighted values ​​of all sub-features are sequentially integrated and arranged to form a target region weighted feature vector with dimensions consistent with the total number of sub-features, comprehensively reflecting the differences in the importance of each sub-feature.

[0057] The core of constructing the feature correlation strength matrix is ​​to calculate the correlation between each dimension of image feature vectors and environmental feature vectors using a mutual information algorithm. This algorithm can effectively capture both linear and nonlinear correlations between features, making it highly suitable for representing the correlation of multi-source data in complex field environments. In practice, the data of each dimension of the image and environmental feature vectors are first standardized to eliminate the interference of different dimensions on the calculation results. Then, based on the calculation logic of mutual information entropy, the probability distribution of each dimension's feature data is statistically analyzed to quantify the total amount of information contained between the two feature dimensions. The quantified result of the degree of information sharing is the element value of the feature correlation strength matrix. The value at each position in the matrix directly corresponds to the degree of correlation between a set of image feature dimensions and environmental feature dimensions.

[0058] Dynamic weight adjustment achieves precise control through a gradient descent optimization algorithm. The entire adjustment process uses the discriminative loss function of the comprehensive feature vector as a clear optimization objective. This function reflects the adaptability of the current feature fusion result to the disease identification task. During operation, the initial weight parameters of image features and environmental features are first initialized. Then, the discriminative loss value of the comprehensive feature vector under the current weight combination is calculated. Based on the core logic of gradient descent, the weight parameters are iteratively updated along the direction of decreasing loss function value, thereby dynamically adjusting the proportion of image features and environmental features in the fusion process. A dynamic judgment mechanism is embedded in the iteration process to continuously monitor the variance contribution of environmental data. When this contribution exceeds a preset judgment standard, the algorithm will automatically trigger the enhancement program of environmental feature weights. Simultaneously, the inter-class distance and intra-class distance relationship of image features are analyzed in real time. When the inter-class distance of image features is detected to be greater than the intra-class distance, it indicates that the image features have a better distinguishing effect on disease categories. At this time, the algorithm will focus on increasing the weight allocation ratio of image features.

[0059] Discriminant scoring is calculated using the F-score, which quantitatively evaluates the distinguishing ability of features by organically combining precision and recall. In practice, sample data is first categorized and organized according to disease type. For each feature dimension, the judgment result under a specific disease type is statistically analyzed: precision refers to the proportion of samples that actually belong to that disease type when the feature classifies them as such, out of all samples classified as belonging to that type; recall refers to the proportion of samples that actually belong to that disease type and are correctly classified as belonging to that type by the feature.

[0060] The similarity matching process utilizes the cosine similarity algorithm. The core principle of this algorithm is to measure the similarity between two vectors by calculating the cosine of the angle between them in space. In practice, the previously obtained comprehensive feature vector and the preset disease feature template are first treated as two vectors in vector space. Then, based on the calculation logic of vector dot product and magnitude, the cosine of the angle between the two vectors is obtained. The magnitude of this cosine value directly reflects the similarity between the comprehensive feature vector and the disease feature template. A fixed similarity threshold is set in the judgment stage. When the calculated cosine similarity value is higher than this threshold, the comprehensive feature vector is determined to match the corresponding disease feature template, thus completing the preliminary identification of the disease category.

[0061] The formula for calculating the comprehensive characteristic index is as follows: ; In the formula, As a comprehensive feature index, A subset of highly discriminative features. For the first The adaptive feature weights corresponding to each feature dimension For in set The feature dimension index is used for traversal. For the first Discriminative scores for each feature dimension The preset standard threshold, It is a function with maximum value. For the first The original feature values ​​of each feature dimension.

[0062] The high-discriminative feature subset is obtained through feature screening. During screening, all original feature dimensions of the samples are collected first. For each feature dimension, its discriminative ability among different categories of samples is calculated. Specifically, the distribution difference of feature values ​​of different categories of samples under that feature dimension is statistically analyzed. The greater the difference, the stronger the discriminative ability. A fixed discriminative ability threshold is set, and all feature dimensions with discriminative ability exceeding the threshold are included in the set. The resulting set is the high-discriminative feature subset.

[0063] No. The adaptive feature weights corresponding to each feature dimension are determined based on the contribution of that feature dimension to the target analysis task. This is done by collecting a large amount of sample data, fixing other feature dimensions, and only changing the weight of the first feature dimension. The numerical value of each feature dimension records the range of change in the target analysis results. A larger range of change indicates a higher contribution from that feature dimension. The contributions of all feature dimensions are normalized so that the sum of all feature dimension weights is a fixed baseline. Feature dimensions with higher contributions correspond to larger normalized values; this value is the value of the first feature dimension. Adaptive feature weights for each feature dimension.

[0064] No. The discriminative score of each feature dimension is the result of quantifying the distinguishing ability of that feature dimension. The calculation first obtains the first... For each feature dimension, calculate the feature value in all different categories of samples, calculate the mean and standard deviation of the feature value in each category, and then calculate the difference between the mean values ​​of different categories. Divide this difference by the mean of the standard deviations of all categories. The result is the discriminative score of the feature dimension. The higher the score, the better the discriminative effect of the feature dimension in different categories of samples.

[0065] The preset standard threshold is a fixed value set in advance. When setting it, the distribution of discriminative scores for all original feature dimensions is statistically analyzed, the minimum range in the score distribution is identified, and a fixed value within that range is selected as the standard threshold. Its purpose is to avoid the... If the discriminant score of each feature dimension is too small, it will cause anomalies in the calculation process. It is necessary to ensure that the denominator is always within a reasonable range.

[0066] No. The original feature values ​​of the first feature dimension are extracted directly from the sample, based on the first feature dimension. For each feature dimension attribute, select the corresponding detection or measurement method. For example, if it is a physical feature, use physical detection equipment; if it is a data feature, read it directly from the sample data and record the detected or read value directly. This value is the first value. The original feature values ​​of each feature dimension.

[0067] The comprehensive feature index is an index that comprehensively quantifies the information of all feature dimensions in a high-discriminative feature subset. Its value directly reflects the comprehensive performance level of the sample in the core feature dimensions.

[0068] During the calculation, for each feature dimension in the high-discriminative feature subset, the adaptive feature weight of that feature dimension is multiplied by the discriminative score to obtain the weighted discriminative value of that feature dimension, highlighting the role of important and highly discriminative feature dimensions.

[0069] The first The discriminant score of each feature dimension is compared with a preset standard threshold, and the larger of the two values ​​is selected as the denominator. The weighted discriminant value of that feature dimension is then divided by the denominator. This process uses a maximum value function to avoid abnormal calculation results caused by excessively small discriminant scores, while ensuring that the numerical variation is within a reasonable range.

[0070] Multiply the above calculation result by the original feature value of the feature dimension to obtain the final contribution value of the feature dimension to the comprehensive feature index. The larger the original feature value and the stronger the feature importance and discrimination ability, the greater the contribution value.

[0071] The final contribution values ​​of all feature dimensions in the high-discriminative feature subset are summed, and then the sum is divided by the total number of feature dimensions contained in the high-discriminative feature subset. The average value is the comprehensive feature index. This average value eliminates the influence of the number of feature dimensions on the result, making the comprehensive feature indexes of different samples comparable.

[0072] The higher the value of the comprehensive feature index, the more outstanding the sample's performance in the dimension of high discriminative features, and the better it reflects the core features of the category to which the sample belongs; the lower the value, the weaker the sample's performance in the dimension of core features, providing accurate comprehensive quantitative basis for tasks such as sample classification, identification or evaluation.

[0073] The beneficial effects include weighting feature sub-items based on the weighting relationship of the multi-source feature adaptive fusion benchmark, ensuring that the influence of each sub-feature in the weighted feature vector accurately matches its correlation value with the actual disease, avoiding interference from secondary features, and providing core-focused basic data for disease assessment. By analyzing the discriminative scores of feature dimensions through historical disease sample sets, the ability of each dimension to distinguish different disease levels is clearly quantified, allowing for precise measurement of feature value and providing an objective basis for subsequent screening. Comparing the discriminative scores with standard thresholds filters out high-discriminative feature subsets, effectively eliminating feature dimensions with weak discriminative ability, simplifying assessment dimensions while ensuring feature quality, and improving the accuracy and efficiency of disease assessment. The dimensions in the high-discriminative feature subsets are integrated and aggregated according to weights, resulting in a comprehensive feature index that fully integrates the key information of each core feature. This highlights the dominant role of high-weight features while also considering the supporting value of complementary features, systematically and accurately reflecting the degree of maize disease impact in the target area, providing a comprehensive and reliable basis for subsequent disease judgment and decision-making, and improving the scientific rigor and practicality of maize disease monitoring.

[0074] The classification and identification module 105 is used to map the comprehensive feature index to the corn disease classification knowledge base, and identify the disease type and disease level of the target area as the disease identification result of the corn. In this embodiment of the invention, when the classification and recognition module maps the comprehensive feature index to the corn disease classification knowledge base and identifies the disease type and disease level of the target area as the corn disease identification result, it is specifically used for: Extract disease feature templates corresponding to different disease types and disease levels from the corn disease classification knowledge base; The comprehensive feature index is matched with the standard feature template in the disease feature template to obtain the similarity comparison result of the disease feature template; Based on the similarity comparison results, the target feature templates for the corn are selected; Output the disease type identifier and disease level identifier corresponding to the target feature template as the disease identification result of the corn.

[0075] The maize disease classification knowledge base includes common disease types encountered during cultivation, covering the characteristics of each disease at each stage. Each disease level is accompanied by standard feature descriptions verified through actual cases. The system is categorized by disease type, first identifying the specific disease type, then extracting its core features for different levels—including abnormal coloration of the maize canopy, details of changes in texture structure, corresponding environmental parameter ranges and fluctuation patterns, etc.—to comprehensively reflect the typical state of each disease level. The core features of each level are integrated into a unified structure, clarifying the logical relationships between features, forming standardized and clearly defined disease feature templates. This ensures that all disease types and corresponding level templates are included and correspond one-to-one with the information in the knowledge base.

[0076] The comprehensive characteristic indicators are broken down into core dimensions such as canopy color anomaly morphology, details of texture structure changes, and environmental parameter values ​​and trends, ensuring complete consistency with the corresponding dimensions of the standard characteristic template. Standard characteristic templates are selected one by one, and comparisons are conducted in the same order of dimensions: first, the matching of canopy color anomalies with the standard template is checked; then, using the preset indicators of texture structure-related sub-items in the standard characteristic template as a benchmark, the degree of conformity of texture structure changes is analyzed by comparing the proportion of texture damage area, the numerical range of fracture density, and the state of structural integrity in the comprehensive characteristic indicators—when all three completely match the preset indicators of the standard template, the texture structure conformity is high; when the core sub-item indicators are consistent with the standard template, and only the structural integrity has non-critical differences that do not affect the determination of disease type and grade, the texture structure conformity is medium; when the proportion of damage area, the numerical range of fracture density exceeds the preset range of the standard template, or when the core sub-item indicators have fundamental differences from the standard template, the texture structure conformity is low; finally, the consistency of environmental parameter conditions with the standard template is verified.

[0077] The selection rules are clearly defined: based on overall similarity, templates with high overall similarity are prioritized; if multiple templates exist, further comparisons are made on core dimensions such as specific canopy anomalies and key environmental parameters, selecting the one with the highest fit. If no template has high overall similarity, the best template is selected from those with medium similarity based on the fit of core dimensions; if all overall similarities are low, no directly matching template is determined. During the selection process, the similarity and core dimension fit of candidate templates are recorded in detail to ensure traceability and avoid subjective bias, ultimately determining the target feature template that best matches the comprehensive characteristic indicators and reflects the actual disease status of maize in the target area.

[0078] Retrieve the original association information of the target feature template in the knowledge base to clarify the corresponding specific disease type and disease level, ensuring information accuracy. Extract the unique identifiers of disease type and level, and integrate them to form a complete identification result, clearly presenting the specific type and severity of maize diseases in the target area. Output the results in a fixed format to ensure the completeness and accuracy of the information and facilitate subsequent applications.

[0079] The beneficial effects include: the disease feature templates systematically extracted from the maize disease classification knowledge base cover core image and environmental features verified by actual cases, providing comprehensive and accurate standards for similarity matching; by comparing comprehensive feature indicators with standard templates one by one according to corresponding dimensions, and combining dimensional fit to determine overall similarity, the matching process is transparent and traceable, avoiding subjective bias and ensuring objective results; target templates are selected based on similarity and core dimensional fit to ensure accurate correspondence with actual disease conditions and eliminate interference; the output identification results are linked to authoritative information in the knowledge base, clearly presenting the disease situation, providing accurate basis for subsequent disease prevention and control and intervention measures, and improving the scientificity, accuracy and practicality of maize disease identification.

[0080] The early warning decision generation module 106 is used to obtain early warning information for corn based on the corn disease early warning rule base, according to the disease early warning mapping relationship in the corn disease early warning rule base and the continuous monitoring cycle conditions required to trigger the early warning.

[0081] In this embodiment of the invention, when the early warning decision generation module executes the early warning rule base based on the corn disease early warning rule base, and obtains the early warning information of the corn according to the disease early warning mapping relationship in the corn disease early warning rule base and the continuous monitoring period conditions required to trigger the early warning based on the disease identification results, it is specifically used for: Using the disease type and disease level in the identification results as the query basis, access the disease early warning mapping relationship in the disease early warning rule base of corn, and retrieve the target early warning level and continuous monitoring cycle conditions corresponding to the query basis; By retrospectively analyzing the comprehensive characteristic index sequence of the target area under the continuous monitoring cycle conditions, it is determined whether the disease status in the disease identification result meets the state threshold range corresponding to the target early warning level; When the judgment result meets the state threshold range, the disease type, disease level, target area identifier and corresponding prevention and control measures suggestions in the disease early warning rule base are combined and encoded according to the target early warning level to obtain the early warning instruction for the target area; When the judgment result does not meet the state threshold range, the disease identification result and the comprehensive characteristic index obtained in this monitoring cycle are stored as historical monitoring data in the associated database of the disease early warning rule base, which is used to update the trend analysis baseline in the disease early warning rule base.

[0082] When the early warning decision generation module performs a retrospective analysis of the comprehensive characteristic index sequence of the target area under the continuous monitoring period conditions, and determines whether the disease status in the disease identification result meets the state threshold range corresponding to the target early warning level, it is specifically used for: Based on the continuous monitoring cycle conditions, obtain the comprehensive feature index sequence in the disease early warning rule base; In the disease early warning rule base, the lower limit value and upper limit value of the feature indicators associated with the target early warning level are used as the preliminary state threshold of the disease early warning rule base; The comprehensive feature index values ​​in the comprehensive historical feature index sequence are compared with the lower limit and upper limit of the feature index, and the statistical results of the preliminary state threshold are obtained by statistically analyzing whether all the historical comprehensive feature index values ​​are within the range of the defined values ​​of the preliminary state threshold. Based on the statistical results, a list of records showing whether the disease status in the comprehensive feature index sequence meets the preliminary status threshold is generated. Based on the list of states that meet the requirements, determine the state threshold range for the target warning level.

[0083] Corn disease early warning starts with disease identification results, extracting clear disease types and severity levels as the basis for precise queries, ensuring unambiguous query direction. Based on this, the construction of core parameters and a database provides scientific support for early warning: the determination of the "standard threshold" relies on historical labeled data of corn disease and healthy samples covering different regions and growth stages. Through statistical analysis of the distribution patterns of characteristic indicators, quantiles covering the vast majority of abnormal conditions are selected as the benchmark threshold; the "state threshold range" is calibrated using a combination of expert experience and machine learning—agricultural disease experts set initial intervals based on practical experience, and then use dynamic monitoring data from different growth stages and environmental conditions to train a random forest model, optimizing it to obtain clear upper and lower limits for mild, moderate, and severe early warnings. Meanwhile, the "Disease Classification Knowledge Base" and the "Disease Early Warning Rule Base" are based on labeled dataset training and expert rule optimization. The labeled dataset contains massive amounts of corn visible light images, time-series environmental data such as temperature and humidity, and disease labeling information. The core features of diseases are extracted through a lightweight convolutional neural network to generate standardized templates. Combined with experts' experience in disease transmission patterns and prevention and control windows, a mapping relationship between disease type, severity, and early warning level is established to form a structured early warning rule base.

[0084] The query criteria are compared with the mapping relationship in the rule base. First, the category entries are located according to the disease type. Then, the corresponding target warning level and continuous monitoring cycle conditions are extracted by matching the disease level. This cycle is set differently according to the severity of the disease. Mild diseases require multiple consecutive monitoring results to meet the threshold to trigger a warning, while moderate to severe diseases only require a single monitoring to meet the standard. Then, all monitoring data within the target area within the cycle are retrieved, comprehensive feature indicators are extracted and sorted by time to form a sequence, completely recording the trajectory of disease status changes. Next, the feature parameters associated with the target warning level are retrieved, and preset upper and lower limits are extracted to form preliminary state thresholds. The fit between the values ​​in the indicator sequence and the thresholds is compared one by one. After statistically analyzing the satisfaction status, a list of state satisfaction records is generated. If all records are "satisfied", the threshold is determined as the final state threshold range; otherwise, full satisfaction is strictly used as the judgment criterion.

[0085] If the disease status meets the threshold range, corresponding control measures are suggested, and these are combined with disease type, level, and target area identifiers to form a standardized code containing early warning information and control guidance. If the status does not meet the threshold, the identification results and indicator data are compiled, entered into the associated database of the disease early warning rule base, and auxiliary information is recorded. Then, the trend analysis baseline of the rule base is updated based on the new data, and the feature range and change interval are optimized to ensure the reliability of subsequent early warnings. The entire process, through precise parameter construction, rigorous rule matching, and dynamic data feedback, achieves the scientific rigor and timeliness of corn disease early warning.

[0086] The beneficial effects include: quickly identifying core information for early warning based on precise query criteria; combining complete indicator sequences with scientific threshold determination to avoid misjudgments and improve the accuracy of early warnings; providing actionable prevention and control guidance when conditions are met to assist in timely intervention; updating the rule base data when conditions are not met to continuously optimize the scientific nature and timeliness of early warnings; and a rigorous threshold determination process that closely matches actual monitoring conditions to provide a reliable basis for early warning decisions.

[0087] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0088] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A corn disease monitoring and early warning system based on agricultural Internet of Things, characterized in that, The system includes a data fusion calibration module, a multi-dimensional feature extraction module, an adaptive fusion benchmark construction module, a feature evaluation and fusion module, a classification and recognition module, and an early warning decision generation module, wherein: The data fusion calibration module is used to fuse and calibrate the visible light image data of the target area with the corresponding environmental data to obtain the standard monitoring data of the target area; The multi-dimensional feature extraction module is used to extract multi-dimensional features from the standard monitoring data to obtain image feature information and environmental feature information of the target area; The adaptive fusion benchmark construction module is used to adaptively adjust the weights of the influence of image data and environmental data on the disease in the target area according to the dynamic correlation between the image feature information and the environmental feature information, and construct a multi-source feature adaptive fusion benchmark for the target area. The feature evaluation and fusion module is used to evaluate the degree of disease impact of the image feature information and the environmental feature information under the multi-source feature adaptive fusion benchmark, so as to obtain the comprehensive feature index of the target area; The classification and identification module is used to map the comprehensive feature index to the corn disease classification knowledge base, and identify the disease type and disease level of the target area as the disease identification result of the corn. The early warning decision generation module is used to obtain early warning information for corn based on the corn disease early warning rule base, according to the disease early warning mapping relationship in the corn disease early warning rule base and the continuous monitoring period conditions required to trigger the early warning.

2. The corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 1, characterized in that, When the data fusion calibration module performs fusion calibration of the visible light image data of the target area with the corresponding environmental data to obtain standard monitoring data for the target area, it is specifically used for: Based on the acquisition timestamp of the visible light image data in the target area, a time-series environmental data segment matching the acquisition timestamp is selected from the corresponding environmental data of the target area to obtain the spatiotemporal aligned data of the target area; Illumination compensation is performed on the visible light image data to obtain standardized visible light image data of the target area; By removing dimensionless abnormal environmental parameters from the spatiotemporal aligned data, uniform environmental data for the target region is obtained. The standardized visible light image data and the consistent environmental data are stitched together to obtain the standard monitoring data for the target area.

3. The corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 1, characterized in that, When the multi-dimensional feature extraction module performs multi-dimensional feature extraction on the standard monitoring data to obtain image feature information and environmental feature information of the target area, it is specifically used for: The color distribution statistics and texture roughness information of the visible light image data in the standard monitoring data are used as the initial image features of the target region. A sliding time window analysis is performed on the environmental data in the standard monitoring data, and the mean, fluctuation range and trend information in the analysis results are integrated into the initial environmental characteristics of the target area. The initial image features are locally saliency-enhanced, and the features describing the color anomalies and texture structure changes of the corn canopy in the enhanced image are collected as the image feature information of the target region. The initial environmental characteristics are screened for disease relevance, and the disease-prone environmental groups selected are used as the environmental characteristic information of the target area.

4. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 1, characterized in that, When the adaptive fusion benchmark construction module adaptively adjusts the weights of the influence of image data and environmental data on the disease in the target region based on the dynamic correlation between the image feature information and the environmental feature information, and constructs a multi-source feature adaptive fusion benchmark for the target region, it is specifically used for: The image feature information is correlated and compared with the environmental feature information to obtain the image-environment feature correlation of the target area; Based on the consistency and magnitude matching of feature change directions in the image-environment feature correlation, the synergistic and antagonistic relationships between the image feature information and the environmental feature information are quantitatively evaluated to construct the feature correlation strength matrix of the target region. Based on the feature correlation strength matrix, the standard disease severity weight coefficient of maize in the target region is adjusted to obtain the dynamic feature weight allocation conditions of the target region; According to the dynamic feature weight allocation conditions, the sub-features in the image feature information and the sub-features in the environment feature information are weighted and fused, and the weighted feature values ​​are projected onto a unified feature vector space to construct a multi-source feature adaptive fusion benchmark for the target region.

5. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 4, characterized in that, The adaptive fusion benchmark construction module, when performing a quantitative evaluation of the synergistic and antagonistic relationship between the image feature information and the environmental feature information based on the consistency and magnitude matching degree of feature change direction in the image-environment feature correlation to construct the feature correlation strength matrix of the target region, is specifically used for: Monitor the changing trends of the image feature information and the environmental feature information over a continuous time series to obtain the image feature change vector and environmental feature change vector of the standard monitoring data; Based on the directions of the image feature change vector and the environmental feature change vector, the consistency of their change directions is evaluated, and the cooperative and antagonistic change relationships between the image feature information and the environmental feature information are identified. Amplitude change matching analysis is performed on the image feature change vector and the environmental feature change vector to determine the strength level of the cooperative change relationship and the antagonistic change relationship; By combining the cooperative change relationship, the antagonistic change relationship, and the strength level, a feature association strength matrix is ​​constructed to represent the dynamic correlation between the image feature information and the environmental feature information.

6. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 4, characterized in that, When the adaptive fusion benchmark construction module adjusts the standard disease severity weight coefficients of maize in the target region according to the feature correlation strength matrix to obtain the dynamic feature weight allocation conditions for the target region, it is specifically used for: Based on the feature association strength matrix, the quantized association values ​​of key feature categories in the image feature information and the environmental feature information are analyzed to form a key feature association set of the target region; In the key feature association set, the quantitative association values ​​with the same direction of change of the key feature category are grouped into the cooperative group, and the quantitative association values ​​with the opposite direction of change of the key feature category are grouped into the antagonistic group. Based on the comparison relationship between the synergistic group and the antagonistic group, the baseline weight coefficient of the standard disease severity in maize is dynamically adjusted to obtain the adjusted weight coefficient set of the target region; The adjusted set of weight coefficients is subjected to proportional balancing to ensure that the sum of the weight coefficients remains constant, thereby obtaining the dynamic feature weight allocation conditions for the target region.

7. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 1, characterized in that, When the feature evaluation and fusion module evaluates the degree of disease impact of the image feature information and the environmental feature information under the multi-source feature adaptive fusion benchmark to obtain the comprehensive feature index of the target area, it is specifically used for: Based on the feature weight relationship in the multi-source feature adaptive fusion benchmark, the sub-features in the image feature information and the sub-features in the environment feature information are weighted by importance to obtain the weighted feature vector of the target region; For the feature dimension of the weighted feature vector, the ability of the feature dimension to distinguish different disease levels in the historical disease sample set of the target area is analyzed to obtain the discriminative score of the feature dimension; The discriminative score is compared with the standard threshold of the corn to filter out the feature dimensions whose discriminative scores are higher than the standard threshold, so as to obtain a high-discriminative feature subset of the target region; The feature dimensions in the highly discriminative feature subset are aggregated to obtain the comprehensive feature index of the target region.

8. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 1, characterized in that, When the classification and recognition module maps the comprehensive feature indicators to the corn disease classification knowledge base and identifies the disease type and disease level of the target area as the corn disease identification result, it is specifically used for: Extract disease feature templates corresponding to different disease types and disease levels from the corn disease classification knowledge base; The comprehensive feature index is matched with the standard feature template in the disease feature template to obtain the similarity comparison result of the disease feature template; Based on the similarity comparison results, the target feature templates for the corn are selected; Output the disease type identifier and disease level identifier corresponding to the target feature template as the disease identification result of the corn.

9. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 1, characterized in that, The early warning decision generation module, when executing the disease early warning rule base based on the corn, and according to the disease early warning mapping relationship in the corn disease early warning rule base and the continuous monitoring period conditions required to trigger the early warning, to obtain the early warning information for the corn, is specifically used for: Using the disease type and disease level in the identification results as the query basis, access the disease early warning mapping relationship in the disease early warning rule base of corn, and retrieve the target early warning level and continuous monitoring cycle conditions corresponding to the query basis; By retrospectively analyzing the comprehensive characteristic index sequence of the target area under the continuous monitoring cycle conditions, it is determined whether the disease status in the disease identification result meets the state threshold range corresponding to the target early warning level; When the judgment result meets the state threshold range, the disease type, disease level, target area identifier and corresponding prevention and control measures suggestions in the disease early warning rule base are combined and encoded according to the target early warning level to obtain the early warning instruction for the target area; When the judgment result does not meet the state threshold range, the disease identification result and the comprehensive characteristic index obtained in this monitoring cycle are stored as historical monitoring data in the associated database of the disease early warning rule base, which is used to update the trend analysis baseline in the disease early warning rule base.

10. A corn disease monitoring and early warning system based on agricultural Internet of Things as described in claim 9, characterized in that, When the early warning decision generation module performs a retrospective analysis of the comprehensive characteristic index sequence of the target area under the continuous monitoring period conditions, and determines whether the disease status in the disease identification result meets the state threshold range corresponding to the target early warning level, it is specifically used for: Based on the continuous monitoring cycle conditions, obtain the comprehensive feature index sequence in the disease early warning rule base; In the disease early warning rule base, the lower limit value and upper limit value of the feature indicators associated with the target early warning level are used as the preliminary state threshold of the disease early warning rule base; The comprehensive feature index values ​​in the comprehensive historical feature index sequence are compared with the lower limit and upper limit of the feature index, and the statistical results of the preliminary state threshold are obtained by statistically analyzing whether all the historical comprehensive feature index values ​​are within the defined range of the preliminary state threshold. Based on the statistical results, a list of records showing whether the disease status in the comprehensive feature index sequence meets the preliminary status threshold is generated. Based on the list of states that meet the requirements, determine the state threshold range for the target warning level.

Images