Potato disease detection method and system based on multi-source data fusion

Abstract

The invention discloses a potato disease detection method and system based on multi-source data fusion, and the system comprises a data collection module which is used for collecting multi-source data and comprises an image collection unit, a meteorological data collection unit, a biological characteristic collection unit, a sensor data collection unit and a pesticide data collection unit; the data processing module is used for processing the multi-source data, and the data processing center comprises a data preprocessing unit, a data fusion unit and a disease identification and analysis unit; the user interaction module is used for interacting with a user, and the user interaction module comprises a monitoring and early warning unit, a diagnosis and prevention suggestion unit, a data visualization and report unit and a remote monitoring and regulation unit. Through multi-source data fusion, the disease detection model can learn a more comprehensive and complex disease characteristic mode, so that the accuracy of potato disease type judgment is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of potato disease detection, and in particular to a potato disease detection method and system based on multi-source data fusion. Background Art

[0002] Traditional potato disease detection and prevention methods mainly rely on manual field inspections. Farmers or agricultural technicians rely on visual observation and experience to identify whether potato plants are infected. This method has many limitations. First, the efficiency of manual inspections is extremely low, especially in the case of large-scale planting. It requires a lot of manpower and time costs, and it is difficult to achieve timely and comprehensive monitoring of the entire planting area. For example, in a large potato planting farm of thousands of acres, it may take several weeks to complete a round of inspections by manually checking each potato plant one by one. During this period, the disease may have spread rapidly. Secondly, the accuracy of manual judgment is seriously affected by subjective factors. Different people have different cognition and judgment criteria for disease symptoms, which is prone to misjudgment or omission. Some early disease symptoms may be relatively hidden and difficult to detect by manual observation alone, thus missing the best prevention and control time. Furthermore, manual inspections can only be carried out on-site in the field, and remote and real-time monitoring cannot be achieved, which is particularly passive when responding to sudden disease epidemics.

[0003] With the continuous advancement of science and technology, the application of information technology in the agricultural field has gradually emerged, bringing new opportunities for potato disease detection. In recent years, some disease detection technologies based on a single data source have begun to emerge. For example, image-based disease detection technology uses high-definition cameras to obtain images of potato plants, and then uses computer vision algorithms to identify and analyze features such as lesions in the image to determine whether there is a disease. This method has improved the detection efficiency and accuracy to a certain extent, but there are still obvious shortcomings. Because it relies only on image data, it cannot fully consider the impact of environmental factors on the occurrence of diseases. In fact, meteorological conditions such as temperature, humidity, and light are closely related to the occurrence and spread of potato diseases. For example, a high temperature and high humidity environment is very likely to induce fungal diseases such as late blight, while a drought environment may cause potato root diseases. In addition, factors such as soil fertility, physical properties, and the use of pesticides also play an important role in the healthy growth and disease susceptibility of potatoes. Disease detection based only on image data is prone to misdiagnosis or incomplete analysis of the causes of disease occurrence.

[0004] On the other hand, although information such as meteorological data, soil sensor data, and pesticide use data are also collected and applied to a certain extent in agricultural production, they are often analyzed and used in isolation, and fail to be effectively integrated with potato plant growth characteristic data (such as plant height, leaf area, chlorophyll content, etc.) and image data. This fragmentation of data makes it difficult for agricultural producers to grasp the growth status and disease risks of potatoes as a whole, and it is impossible to formulate scientific and accurate prevention and control strategies. For example, even if it is known that meteorological conditions are conducive to the occurrence of a certain disease, if it is not combined with the current growth status of the potato plant and the signs of the disease shown in the image, it is impossible to determine whether the disease has occurred and the severity of the disease, making it difficult to determine appropriate prevention and control measures, such as whether pesticides are needed, the type and dosage of pesticides, etc.

[0005] In summary, the existing potato disease detection technology, whether it is traditional manual inspection or single data source detection method, is difficult to meet the needs of large-scale, precise and intelligent production in modern agriculture. Therefore, it is necessary to develop a potato disease detection system based on multi-source data fusion, organically combining image data, meteorological data, biometric data, sensor data and pesticide data, and realizing comprehensive, accurate and real-time detection and analysis of potato diseases through advanced data fusion algorithms and intelligent analysis models, which is of great significance for improving potato yield and quality and ensuring the economic benefits and sustainable development of agricultural production. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a potato disease detection system based on multi-source data fusion, comprising:

[0007] A data acquisition module, used for collecting multi-source data, the data acquisition module includes an image acquisition unit, a meteorological data acquisition unit, a biometrics acquisition unit, a sensor data acquisition unit and a pesticide data acquisition unit;

[0008] The multi-source data includes image data, meteorological data, biometric data, sensor data and pesticide data;

[0009] The image acquisition unit is used to obtain image data of potato plants, wherein the image data includes leaf image data, appearance detail image data of stems and appearance detail image data of tubers of potato plants;

[0010] The meteorological data acquisition unit is used to obtain meteorological data of the planting area, and the meteorological data includes temperature data, humidity data, light intensity data, rainfall data and wind speed data;

[0011] The biometric characteristic collection unit is used to obtain biometric characteristic data of potato plants, wherein the biometric characteristic data includes plant height data, leaf area data, leaf thickness data, chlorophyll content data and root development index data;

[0012] The sensor data acquisition unit is used to obtain sensor data of the planting area, wherein the sensor data includes soil moisture data, soil fertility data, atmospheric pressure data and light intensity data;

[0013] The pesticide data collection unit is used to obtain pesticide usage data, usage dosage data, usage time data, usage frequency data and pesticide residue detection data;

[0014] A data processing module, used for processing the multi-source data, the data processing module includes a data preprocessing unit, a data fusion unit, and a disease identification and analysis unit;

[0015] The user interaction module is used to interact with the user, and the user interaction module includes a monitoring and early warning unit, a diagnosis and prevention and control suggestion unit, a data visualization and reporting unit, and a remote monitoring and control unit.

[0016] Furthermore, the data preprocessing unit is used to preprocess the multi-source data, including:

[0017] The image data is cropped, scaled, and normalized, and an adaptive filtering algorithm is used for denoising, color enhancement, and feature highlighting.

[0018] The missing values ​​in meteorological data and sensor data are filled using methods based on data trends and physical models, and the outliers are processed using an anomaly detection algorithm based on machine learning and standardized;

[0019] Biometric data is processed using time-frequency analysis methods such as wavelet transform;

[0020] The pesticide data is cleaned, converted and standardized, and a pesticide composition and effect database is established.

[0021] Furthermore, the data fusion unit is used to fuse the preprocessed multi-source data, including:

[0022] Feature-level fusion: First, principal component analysis is performed on each data to reduce its dimension, and then features are extracted from image data, meteorological data, biometric data, sensor data, and pesticide data, respectively, and then combined into a comprehensive feature vector using a feature combination strategy based on the kernel method;

[0023] Decision-level fusion: image data, meteorological data, biometric data, sensor data, and pesticide data are processed and decided separately, and the decision results are integrated by using dynamic weighted average method combined with fuzzy reasoning method;

[0024] Model fusion: build models for image data, meteorological data, biometric data, sensor data and pesticide data for training and prediction respectively, and then fuse the prediction results.

[0025] Furthermore, the disease identification and analysis unit is used to identify and analyze potato diseases based on the fused data, including:

[0026] Disease identification: Build a hybrid intelligent disease identification model that combines deep learning algorithms, machine learning algorithms, and rule-based expert systems, and use a multi-model fusion confidence assessment method to give a confidence score;

[0027] Disease analysis: Use the time series model in deep learning to analyze the changing trends of disease spot area and biological characteristic data, combine meteorological data to determine the stage of disease development, use causal analysis algorithm to analyze the causal relationship between multi-source data to determine the cause of disease occurrence, and use the disease propagation model based on data assimilation to predict the development trend of disease.

[0028] Furthermore, the monitoring and early warning unit is used to display the growth status and disease detection results of potatoes in real time, and send early warning information to the user when the risk of disease occurrence exceeds a preset threshold;

[0029] The diagnosis and prevention and control recommendation unit is used to display the accurate diagnosis results of the disease and recommend the appropriate type of pesticide, dosage, time of use, method of use and non-pesticide prevention and control measures according to the type and severity of the disease;

[0030] The data visualization and reporting unit is used to display multi-source data and disease detection results in intuitive charts and graphs, and generate data analysis reports;

[0031] The remote monitoring and control unit is used to enable users to remotely access the system, view real-time data and disease information, and remotely control field equipment.

[0032] Furthermore, the image acquisition unit includes a high-definition camera carried by the drone and a ground fixed-point shooting device;

[0033] The soil fertility sensor in the sensor data acquisition unit is used to detect the key nutrient content of nitrogen, phosphorus and potassium in the soil, and also detects the content of other trace elements in the soil and soil pH information. The sensor data acquisition unit also includes an air humidity sensor and a carbon dioxide concentration sensor.

[0034] Furthermore, the feature level fusion in the data fusion unit specifically includes:

[0035] Extract lesion texture, color distribution and shape features from image data;

[0036] Extract temperature fluctuation range and humidity change rate characteristics from meteorological data;

[0037] Extract the plant height growth rate and chlorophyll content change trend characteristics from the biological characteristic data;

[0038] Extract soil moisture change curve and light intensity mean features from sensor data;

[0039] Extract pesticide use interval characteristics, pesticide type and disease association characteristics, and pesticide residue change characteristics from pesticide data;

[0040] The above extracted features are combined into a comprehensive feature vector.

[0041] Furthermore, the decision-level fusion in the data fusion unit specifically includes:

[0042] The image data processing system determines the disease type based on the image features;

[0043] The meteorological data analysis module assesses disease risks based on meteorological conditions;

[0044] The biometric data evaluation subsystem analyzes the association between abnormal plant growth and disease;

[0045] The sensor data processing unit infers the degree to which the environment is conducive to the growth of diseases;

[0046] The pesticide data processing module analyzes the rationality of pesticide use and the control effect;

[0047] The decision results of the above modules are integrated through voting, weighted average or Bayesian reasoning.

[0048] Furthermore, the model fusion in the data fusion unit specifically includes:

[0049] The convolutional neural network model is used for training and prediction of image data. The improved convolutional neural network model adds an attention mechanism.

[0050] Gradient boosting decision tree model is used for meteorological data and biometric data processing;

[0051] The kernel support vector machine model is used for sensor data processing;

[0052] Multilayer perceptron model is used for pesticide data processing;

[0053] The prediction results of the above models are fused through the integrated learning method Stacking.

[0054] Furthermore, in the hybrid intelligent disease recognition model in the disease recognition and analysis unit, the expert system stores a large amount of potato disease knowledge and empirical rules, and provides auxiliary judgment when the deep learning and machine learning models are uncertain.

[0055] Furthermore, the disease analysis in the disease identification and analysis unit uses a long short-term memory network to analyze the changing trend of the lesion area and biological characteristic data, adopts a Bayesian network to analyze the causal relationship between multi-source data, and uses a disease propagation model based on data assimilation to predict the development trend of the disease.

[0056] Furthermore, the automated pesticide use recording system of the pesticide data collection unit also records parameters such as pesticide spraying uniformity.

[0057] The present invention provides a potato disease detection method and system based on multi-source data fusion, which has the following beneficial effects:

[0058] By integrating image data, meteorological data, biometric data, sensor data, and pesticide data, the system can evaluate the health status of potato plants from multiple dimensions. For example, image data can intuitively present the characteristics of lesions, meteorological data can reflect the inducing factors of the environment on the occurrence of diseases, biometric data reflects the relationship between the growth status of the plant itself and the disease, sensor data monitors changes in the soil and atmospheric environment, and pesticide data involves the impact of prevention and control measures. These data complement each other and avoid misjudgments that may be caused by detection from a single data source. For example, leaf abnormalities caused by pesticide damage may be mistaken for diseases based on image judgment alone, but they can be accurately distinguished by combining pesticide use data. Multi-source data fusion enables the disease detection model to learn more comprehensive and complex disease feature patterns, thereby significantly improving the accuracy of judging potato disease types, whether it is common diseases such as late blight, early blight, and mole disease, or some relatively rare diseases or diseases with unclear initial symptoms, they can be more accurately identified.

[0059] With real-time monitoring and analysis of multi-source data, the system can keenly capture early signs of disease. Abnormal fluctuations in temperature and humidity in meteorological data, slowed plant height growth rate and slight decrease in chlorophyll content in biometric data, and subtle changes in soil moisture or fertility reflected by sensor data that deviate from the appropriate range may be detected by the system and comprehensively judged as an increased risk of disease before obvious visible lesions are formed. Based on this early warning information, growers can take preventive measures in advance, such as adjusting irrigation volume, improving ventilation conditions, and applying preventive pesticides, etc., to effectively curb the further development of diseases, greatly reduce the possibility of large-scale outbreaks of diseases, and reduce yield losses caused by diseases. DETAILED DESCRIPTION

[0060] The following specific embodiments further describe the present invention in detail. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0061] In the embodiment, the data collection phase includes:

[0062] Image data acquisition:

[0063] Collaborative operation of multiple drones: Assume that there are 3 drones responsible for monitoring a potato planting area of ​​1,000 mu. The drone is set to take a comprehensive photo every 2 days, with a flight altitude of 50 meters and a speed of 10 meters per second. The shooting angles include 90 degrees vertical downward and 45 degrees tilt. Its high-definition multispectral camera has a resolution of 1024×768 pixels in the visible light band (400-700nm) and a resolution of 800×600 pixels in the near-infrared band (700-1100nm). In one shot, a drone captured a leaf image of a suspected lesion in a specific area. The lesion area accounted for about 10% of the image, the color was dark green, the shape was approximately elliptical, the major axis was about 5 cm, and the minor axis was about 3 cm. The ground image acquisition robot supplemented the key areas (about 20% of the planting area) every day, and the detailed images of the lesions showed irregular textures on the edges of the lesions, and the color of some areas began to turn brown. All image data are transmitted to the data processing center through an encrypted 5G network within 5 seconds.

[0064] Meteorological data collection:

[0065] Combination of weather station and micro sensor: The weather station near the planting area provides data every 30 minutes, such as the temperature is 25.5℃, humidity is 70%, light intensity is 10000 lux, rainfall is 0 mm, and wind speed is 3 m / s during a certain period. The 20 micro weather sensors distributed in the planting area update data every 10 minutes, and the temperature of a certain sensor location is 26.2℃, humidity is 72%, light intensity is 9800 lux, and wind speed is 2.8 m / s at a specific moment. Within a week, the highest temperature reaches 30℃, the lowest temperature is 20℃, the temperature fluctuation range is 10℃, and the humidity rises from 60% to 80% in one day, with a change rate of 20% / day. These data are gathered to the data processing center within 2 minutes through the wireless sensor network.

[0066] Biometric Data Collection:

[0067] Non-contact measurement of plant height and leaf area: Non-contact LiDAR and spectrometer measure plant height and leaf area every 3 days. For example, the initial plant height of a potato plant is 30 cm, and it is measured to be 31.5 cm after 3 days, with a plant height growth rate of 0.5 cm / 3 days. The initial area of ​​a leaf was measured by a spectrometer to be 100 square centimeters, and 105 square centimeters after 3 days. The chlorophyll fluorescence meter measures the chlorophyll content every 5 days. The initial chlorophyll content is 40SPAD value, and it drops to 38SPAD value after 5 days, showing a downward trend. The soil penetrating radar evaluates the root system every 3 months and finds that the root depth of a plant is 30 cm, the average root diameter is 2 mm, the number of roots is about 50, and the root vitality is 0.6 (vitality index value).

[0068] Sensor data collection:

[0069] Comprehensive monitoring of soil moisture and fertility: The soil moisture sensor has a humidity value of 30% at a depth of 10 cm, 35% at a depth of 20 cm, and 40% at a depth of 30 cm. The data is updated every hour, and the soil moisture change curve shows two obvious fluctuations within a week. The soil fertility sensor is tested every 5 days. The soil nitrogen content in a certain area is 0.2%, the phosphorus content is 0.1%, the potassium content is 0.3%, the pH is 6.5, the trace element calcium content is 0.15%, and the magnesium content is 0.05%. The air humidity sensor and the carbon dioxide concentration sensor collect data every 20 minutes. At a certain moment, the air humidity is 65% and the carbon dioxide concentration is 400ppm.

[0070] Pesticide data collection:

[0071] Automated recording system: Metalaxyl mancozeb was used in a pesticide spraying, with a dosage of 100 grams of active ingredient per mu, the spraying time was 10:00 on May 10, 2024, and the frequency of use was once every 7 days. Pesticide residue detection data showed that 3 days after the application, the metalaxyl mancozeb residue in the soil was 0.5 mg / kg, and the residue on the leaves was 1 mg / kg. The automated system records the coefficient of variation of the uniformity of pesticide spraying as 10% (indicating the uniformity), and the data is stored in real time and transmitted to the data processing center within 1 minute.

[0072] The data preprocessing stage includes:

[0073] Image data preprocessing:

[0074] Cropping and scaling: The images collected by drones and robots were cropped to remove the irrelevant background at the edges, reducing the image size by about 30%. The images were then scaled to 800×600 pixels and a bilinear interpolation algorithm was used to unify the image feature scale after processing, facilitating subsequent analysis.

[0075] Normalization and adaptive filtering: After the image is normalized, the pixel value range is adjusted to 0-1, and the brightness and contrast parameters meet the unified standard. Adaptive filtering algorithm is used, and the filter window size in the lesion area is 3×3 pixels and the smooth area is 5×5 pixels, which effectively removes noise while retaining the edge details of the lesion, and improves the clarity of the lesion texture by about 20%.

[0076] Color enhancement and feature highlighting: After color enhancement, the color contrast of the lesion area is improved by about 30%, and the dark green lesions are easier to distinguish. Through image morphological operations, the accuracy of the lesion shape parameters is improved by about 15%, for example, the circularity calculation error is reduced.

[0077] Meteorological and sensor data preprocessing:

[0078] Missing value filling: At a certain moment, the temperature data of a micro-meteorological sensor is missing. The filling value is 25.8℃ calculated based on the daily temperature change trend of the six data points before and after and the heat conduction equation. In the abnormal value detection, the humidity value reported by a sensor is 95%, which exceeds the historical data range. After analysis, it is found that the equipment is faulty and is corrected to 80% by smoothing the adjacent data.

[0079] Standardization: Using the Z-score standardization method, the temperature data is standardized to a mean of 0 and a standard deviation of 1. For example, the original temperature data of 25°C becomes -0.5 after standardization (assuming the mean is 26°C and the standard deviation is 2°C).

[0080] Biometric data preprocessing:

[0081] Wavelet transform processing: After performing wavelet transform processing on the plant height data and removing high-frequency noise, the fluctuation range of the plant height data is reduced by about 25%, the data trend is more stable, and it more accurately reflects the growth status of the plant.

[0082] Pesticide data preprocessing:

[0083] Cleaning and unit conversion: During the cleaning of pesticide data, an incorrect pesticide dosage record was found and corrected. The pesticide dosage was uniformly converted to 1500 grams of active ingredient per hectare, and the pesticide residue was converted to the standard unit of milligrams per kilogram to facilitate data fusion analysis.

[0084] Establishment of pesticide database: In the pesticide database, the effect of metalaxyl mancozeb on the prevention and control of late blight is marked as high, and the recommended dosage range is 80-120 grams of active ingredient per mu. The dosage used this time is within a reasonable range, but its effect needs to be further analyzed in combination with the occurrence of the disease.

[0085] The data fusion phase includes:

[0086] Feature-level fusion:

[0087] Dimensionality reduction by principal component analysis: After PCA dimensionality reduction of the image data, three main characteristic components were retained, which were related to the lesion texture, color and shape, respectively. The data dimension was reduced from 10 dimensions to 3 dimensions, and the explained variance ratios were 40%, 35% and 20%, respectively. The temperature fluctuation range and humidity change rate features extracted from the meteorological data were combined with other data features. For example, when combined with the lesion texture features in the image data, they were mapped to a high-dimensional space through the radial basis function kernel. The dimension of the combined comprehensive feature vector was 5 dimensions. In this high-dimensional space, the correlation between the lesion texture and the temperature fluctuation was calculated to be 0.4 (correlation numerical example).

[0088] Decision-level fusion:

[0089] Decisions of each module: The image data processing system determines that the disease type in a certain area may be late blight based on image features, with a confidence level of 0.6. The meteorological data analysis module estimates the risk probability of late blight to be 0.4 based on meteorological conditions. The biometric data evaluation subsystem analyzes the correlation between abnormal plant growth and late blight to be 0.3. The sensor data processing unit infers that the environment is conducive to the breeding of late blight to a degree of 0.5. The pesticide data processing module analyzes the rationality of pesticide use and the control effect, and believes that the current use of pesticides is basically reasonable, but the effect needs further observation, and the rationality score is 0.6.

[0090] Dynamic weighted average and fuzzy reasoning fusion: Initially, the weight of each module is 0.2. According to the performance evaluation over a period of time, the image data processing system has a high accuracy rate, and its weight is increased to 0.3, and the weight of the meteorological data analysis module is reduced to 0.15. Using the fuzzy reasoning method, the decision results of each module are comprehensively considered, and the final fusion decision is that the probability of late blight is 0.45 and the confidence level is 0.5.

[0091] Model Fusion:

[0092] Model training: After 100 epochs of training for the improved CNN model for image data, the accuracy on the test set reached 90%. After training the gradient boosting decision tree model for meteorological data and biometric data, the accuracy of disease risk prediction was 85%. The accuracy of the kernel SVM model for sensor data was 88%. The accuracy of the multi-layer perceptron model for pesticide data was 80%.

[0093] Integrated learning fusion: The prediction results of the above models are fused through the Stacking method, and the prediction results of each model are used as new features to build a meta-model (using the SVM meta-model). In the cross-validation, the average accuracy of the 5-fold cross-validation reached 92%. The final fusion model's disease detection results showed that the disease type in a certain area was late blight, the severity was mild, and the probability of occurrence was 0.4.

[0094] The disease identification and analysis phase includes:

[0095] Disease identification:

[0096] The constructed hybrid intelligent disease recognition model combines deep learning algorithms, machine learning algorithms and expert systems. For example, when the CNN model is uncertain about the identification of a certain lesion, the expert system makes a comprehensive judgment based on the characteristics of the lesion, meteorological conditions, pesticide usage, etc., to help improve the recognition accuracy. The confidence score given by the model comprehensively considers the results of each model. For example, the confidence score for the identification of a certain disease is 0.7, which indicates a high reliability.

[0097] Disease analysis:

[0098] The long short-term memory network was used to analyze the trend of the lesion area, and it was predicted that the lesion area would expand by 20% in the next 5 days. Combined with meteorological data, it was determined that the current disease was in the early stages, because the lesion area was small and the meteorological conditions had not yet reached the conditions for a large-scale outbreak of the disease. The Bayesian network was used to analyze the causal relationship between multi-source data, and it was concluded that high soil moisture (probability 0.6) and recent improper use of pesticides (probability 0.4) were the main causes of the disease. The disease propagation model based on data assimilation was used to predict the development trend of the disease. It is expected that the disease will spread to a range of 10 meters in the next week, and the severity will increase to moderate. It is recommended to take measures such as improving soil drainage and adjusting the pesticide use plan for prevention and control.

[0099] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A potato disease detection system based on multi-source data fusion, characterized in that: include: A data acquisition module, used for collecting multi-source data, the data acquisition module includes an image acquisition unit, a meteorological data acquisition unit, a biometrics acquisition unit, a sensor data acquisition unit and a pesticide data acquisition unit; The multi-source data includes image data, meteorological data, biometric data, sensor data and pesticide data; The image acquisition unit is used to obtain image data of potato plants, wherein the image data includes leaf image data, appearance detail image data of stems and appearance detail image data of tubers of potato plants; The meteorological data acquisition unit is used to obtain meteorological data of the planting area, and the meteorological data includes temperature data, humidity data, light intensity data, rainfall data and wind speed data; The biometric characteristic collection unit is used to obtain biometric characteristic data of potato plants, wherein the biometric characteristic data includes plant height data, leaf area data, leaf thickness data, chlorophyll content data and root development index data; The sensor data acquisition unit is used to obtain sensor data of the planting area, wherein the sensor data includes soil moisture data, soil fertility data, atmospheric pressure data and light intensity data; The pesticide data collection unit is used to obtain pesticide usage data, usage dosage data, usage time data, usage frequency data and pesticide residue detection data; A data processing module, used for processing the multi-source data, the data processing module includes a data preprocessing unit, a data fusion unit, and a disease identification and analysis unit; The user interaction module is used to interact with the user, and the user interaction module includes a monitoring and early warning unit, a diagnosis and prevention and control suggestion unit, a data visualization and reporting unit, and a remote monitoring and control unit.

2. A potato disease detection system based on multi-source data fusion according to claim 1, characterized in that: The data preprocessing unit is used to preprocess the multi-source data, including: The image data is cropped, scaled, and normalized, and an adaptive filtering algorithm is used for denoising, color enhancement, and feature highlighting. The missing values ​​in meteorological data and sensor data are filled using methods based on data trends and physical models, and the outliers are processed using an anomaly detection algorithm based on machine learning and standardized; Biometric data is processed using time-frequency analysis methods such as wavelet transform; The pesticide data is cleaned, converted and standardized, and a pesticide composition and effect database is established.

3. A potato disease detection system based on multi-source data fusion according to claim 2, characterized in that: The data fusion unit is used to fuse the preprocessed multi-source data, including: Feature-level fusion: First, principal component analysis is performed on each data to reduce its dimension, and then features are extracted from image data, meteorological data, biometric data, sensor data, and pesticide data, respectively, and then combined into a comprehensive feature vector using a feature combination strategy based on the kernel method; Decision-level fusion: image data, meteorological data, biometric data, sensor data, and pesticide data are processed and decided separately, and the decision results are integrated by using dynamic weighted average method combined with fuzzy reasoning method; Model fusion: build models for image data, meteorological data, biometric data, sensor data and pesticide data for training and prediction, and then fuse the prediction results.

4. A potato disease detection system based on multi-source data fusion according to claim 3, characterized in that: The disease identification and analysis unit is used to identify and analyze potato diseases based on the fused data, and includes: Disease identification: Build a hybrid intelligent disease identification model that combines deep learning algorithms, machine learning algorithms, and rule-based expert systems, and use a multi-model fusion confidence assessment method to give a confidence score; Disease analysis: Use the time series model in deep learning to analyze the changing trends of disease spot area and biological characteristic data, combine meteorological data to determine the stage of disease development, use causal analysis algorithm to analyze the causal relationship between multi-source data to determine the cause of disease occurrence, and use the disease propagation model based on data assimilation to predict the development trend of disease.

5. A potato disease detection system based on multi-source data fusion according to claim 4, characterized in that: The monitoring and early warning unit is used to display the growth status and disease detection results of potatoes in real time, and send early warning information to the user when the risk of disease occurrence exceeds a preset threshold; The diagnosis and prevention and control recommendation unit is used to display the accurate diagnosis results of the disease and recommend the appropriate type of pesticide, dosage, time of use, method of use and non-pesticide prevention and control measures according to the type and severity of the disease; The data visualization and reporting unit is used to display multi-source data and disease detection results in intuitive charts and graphs, and generate data analysis reports; The remote monitoring and control unit is used to enable users to remotely access the system, view real-time data and disease information, and remotely control field equipment.

6. A potato disease detection system based on multi-source data fusion according to claim 5, characterized in that: The image acquisition unit includes a high-definition camera carried by the drone and a ground fixed-point shooting device; The soil fertility sensor in the sensor data acquisition unit is used to detect the key nutrient content of nitrogen, phosphorus and potassium in the soil, and also detects the content of other trace elements in the soil and soil pH information. The sensor data acquisition unit also includes an air humidity sensor and a carbon dioxide concentration sensor.

7. A potato disease detection system based on multi-source data fusion according to claim 6, characterized in that: The feature level fusion in the data fusion unit specifically includes: Extract lesion texture, color distribution and shape features from image data; Extract temperature fluctuation range and humidity change rate characteristics from meteorological data; Extract the plant height growth rate and chlorophyll content change trend characteristics from the biological characteristic data; Extract soil moisture change curve and light intensity mean features from sensor data; Extract pesticide use interval characteristics, pesticide type and disease association characteristics, and pesticide residue change characteristics from pesticide data; The above extracted features are combined into a comprehensive feature vector.

8. The potato disease detection system based on multi-source data fusion according to claim 7, characterized in that: The decision-level fusion in the data fusion unit specifically includes: The image data processing system determines the disease type based on the image features; The meteorological data analysis module assesses disease risks based on meteorological conditions; The biometric data evaluation subsystem analyzes the association between abnormal plant growth and disease; The sensor data processing unit infers the degree to which the environment is conducive to the growth of diseases; The pesticide data processing module analyzes the rationality of pesticide use and the control effect; The decision results of the above modules are integrated through voting, weighted average or Bayesian reasoning.

9. The potato disease detection system based on multi-source data fusion according to claim 8, characterized in that: The model fusion in the data fusion unit specifically includes: The convolutional neural network model is used for training and prediction of image data. The improved convolutional neural network model adds an attention mechanism. Gradient boosting decision tree model is used for meteorological data and biometric data processing; The kernel support vector machine model is used for sensor data processing; Multilayer perceptron model is used for pesticide data processing; The prediction results of the above models are fused through the integrated learning method Stacking.

10. A potato disease detection method based on multi-source data fusion, characterized in that: The detection method of a potato disease detection system based on multi-source data fusion according to claims 1-9 is derived.