Method for recommending fruit disease prescription

Abstract

The invention relates to a method for recommending a fruit disease prescription. Comprising the steps of obtaining real-time state image data in the early stage of fruit growth; performing disease occurrence monitoring on the real-time state image data based on a neural network model; after the occurrence of the disease is judged, the neural network model generates disease information corresponding to the real-time state image data; matching a plurality of prescriptions in a fruit disease prescription library based on a semantic understanding deep learning model according to the disease information; and recommending the plurality of matched prescriptions to the user, and recording according to the adoption condition of the user. According to the method, a fruit disease prescription library is constructed, and fruit disease prescriptions are recommended through a disease information descriptor embedding method, so that scientific prevention and control measure recommendation on early-warned and monitored diseases is completed.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of fruit tree planting, and in particular to a method for recommending fruit disease prescriptions. Background Art

[0002] Traditional orchard cultivation models have certain negative environmental impacts. Excessive use of chemical pesticides and fertilizers can lead to soil and water pollution, damaging the ecological environment. Furthermore, traditional water resource management methods can be wasteful and inappropriate, necessitating more environmentally friendly and sustainable cultivation methods.

[0003] In related technologies, only crop diseases and insect pests are monitored and processed, or only pests are identified through neural network models, and it is impossible to recommend disease prescriptions for pests.

[0004] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0005] It should be noted that this section is intended to provide background or context for the technical solutions of the present invention as stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the Invention

[0006] The object of the present invention is to provide a method for recommending fruit disease prescriptions, thereby at least to a certain extent solving one or more problems caused by the limitations and defects of the related art.

[0007] The present invention provides a method for recommending fruit disease prescriptions, comprising:

[0008] Acquire real-time image data of the early stage of fruit growth;

[0009] Performing disease occurrence monitoring on the real-time status image data based on a neural network model;

[0010] When it is determined that a disease has occurred, the neural network model generates disease information corresponding to the real-time state image data;

[0011] Matching multiple prescriptions in a fruit disease prescription library based on a deep learning model based on semantic understanding of the disease information;

[0012] Recommend multiple matched prescriptions to users and record them based on the users' adoption.

[0013] Optionally, the step of generating, by the neural network model, disease information corresponding to the real-time state image data after determining that a disease has occurred includes:

[0014] The disease information at least includes the name of the disease and the severity of the disease.

[0015] Optionally, the step of matching multiple prescriptions in a fruit disease prescription library based on a deep learning model based on semantic understanding of the disease information includes:

[0016] The fruit disease prescription library classifies and stores prescriptions in sequence based on the disease name and the disease severity.

[0017] Optionally, the step of matching multiple prescriptions in a fruit disease prescription library based on a deep learning model based on semantic understanding of the disease information includes:

[0018] The disease name and the disease severity are semantically understood based on a deep learning model, and the prescriptions in the fruit disease prescription library are sorted by matching degree, and the prescriptions with the highest matching degree are recommended in turn.

[0019] Optionally, the step of recommending the matched multiple prescriptions to the user and recording the results based on the user's adoption status includes:

[0020] The collected prescriptions are managed in a favorites organization structure based on the user's adoption.

[0021] Optionally, the step of obtaining real-time image data of the fruit's early growth state includes:

[0022] The real-time status image data is collected by a drone equipped with a multispectral imager.

[0023] Optionally, the step of monitoring the occurrence of diseases on the real-time status image data based on a neural network model includes:

[0024] The neural network model uses historical state image data of the early stage of fruit growth as input and generates diseases in the later stage as output for optimization training.

[0025] Optionally, the step of generating, by the neural network model, disease information corresponding to the real-time state image data after determining that a disease has occurred includes:

[0026] The neural network model extracts features from real-time image data of the early stage of fruit growth and detects and locates diseases through multi-scale lesion comparison.

[0027] Optionally, the step of generating, by the neural network model, disease information corresponding to the real-time state image data after determining that a disease has occurred includes:

[0028] The neural network model combines the extracted features with the disease patterns in the fruit disease knowledge base to analyze environmental factors and predict the probability of disease occurrence.

[0029] The technical solution provided by the present invention can have the following beneficial effects:

[0030] In the present invention, a fruit disease prescription database is constructed and a disease information description word embedding method is used to recommend fruit disease prescriptions to complete the recommendation of scientific prevention and control measures for the diseases that are warned and monitored. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings are incorporated into and constitute a part of this specification, illustrate embodiments consistent with the present invention, and together with the description, serve to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and it is clear that those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0032] Figure 1 A schematic flow chart illustrating a method for recommending fruit disease prescriptions in an exemplary embodiment of the present invention;

[0033] Figure 2 A schematic diagram illustrating the construction of a neural network model in an exemplary embodiment of the present invention is shown. DETAILED DESCRIPTION

[0034] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0035] In addition, the accompanying drawings are merely schematic illustrations of embodiments of the present invention and are not necessarily drawn to scale. Identical reference numerals in the drawings represent identical or similar parts, and thus repeated descriptions thereof will be omitted. Some of the blocks shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically separate entities.

[0036] The present invention provides a method for recommending fruit disease prescriptions, referring to Figure 1 As shown in , including:

[0037] Step S100: Acquire real-time state image data of the early stage of fruit growth.

[0038] Step S200: monitoring the occurrence of diseases on the real-time status image data based on a neural network model.

[0039] Step S300: When it is determined that a disease has occurred, the neural network model generates disease information corresponding to the real-time status image data.

[0040] Step S400: Match multiple prescriptions in the fruit disease prescription library according to the deep learning model based on semantic understanding of the disease information.

[0041] Step S500: recommend multiple matched prescriptions to the user and record them according to the user's adoption status.

[0042] It should be understood that the fruit disease prevention and control measures system consists of three functional parts: fruit disease prescription recommendation, historical prescription, and collection prescription, to complete the scientific prevention and control measures recommendation for the diseases that are warned and monitored.

[0043] It should also be understood that the fruit may be any fruit of any fruit tree, specifically apple. Diseases and pests may also be referred to as pests and diseases or insect pests.

[0044] It is also necessary to understand that, based on the user's adoption of the prescription in the apple pest and disease prescription recommendation function, the system will record the historical prescriptions recommended to the user, which can more conveniently realize the viewing of historical prescription recommendations from the historical records.

[0045] The above-mentioned method for recommending fruit disease prescriptions is adopted to complete the recommendation of scientific prevention and control measures for the diseases warned and monitored by constructing a fruit disease prescription database and recommending fruit disease prescriptions using the disease information description word embedding method.

[0046] Below, each step of the method for recommending fruit disease prescriptions in this exemplary embodiment will be described in more detail.

[0047] In some embodiments, step S300 includes:

[0048] The disease information at least includes the name of the disease and the severity of the disease.

[0049] It should be understood that disease information can also include information such as the time of disease and pest infestation, the name of the disease and pest, the category of disease severity, and location information. Real-time image data of the early stage of fruit growth is obtained from cameras, sensors, or databases. These image data can be used for subsequent disease detection. The acquired real-time status image data is input into a pre-trained neural network model. The model will analyze the image features to determine whether the fruit is diseased. If the model determines that the fruit is diseased, it will further analyze the image and generate disease information. This information at least includes the name of the disease (such as powdery mildew, anthracnose, etc.) and the severity of the disease (such as mild, moderate, severe). The generated disease information is input into a deep learning model based on semantic understanding. The model will search for multiple prescriptions that match the disease name and disease severity in the fruit disease prescription library. The multiple matched prescriptions are displayed to the user, allowing the user to choose whether to adopt them. Based on the user's choice, the prescription adopted by the user is recorded.

[0050] In some embodiments, step S400 includes:

[0051] The fruit disease prescription library classifies and stores prescriptions in sequence based on the disease name and the disease severity.

[0052] It should be understood that classifying and storing prescriptions in sequence based on the disease name and the disease severity can facilitate quick and easy query and use of prescriptions, providing strong support for the prevention and control of fruit diseases. Once the model determines that the fruit is diseased, it will further conduct an in-depth analysis of the image to generate detailed disease information. The disease information contains at least the name of the disease and the severity of the disease. Disease name: The model will compare the image features with various known disease features in the training data to determine the specific disease name, such as powdery mildew, anthracnose, downy mildew, etc. Different diseases will show different symptoms on the fruit. For example, powdery mildew will form a white powdery substance on the surface of the fruit, and anthracnose will cause black sunken spots on the fruit. Disease severity: The model will judge the disease severity based on factors such as the distribution range and severity of the disease symptoms on the fruit. It is generally divided into mild, moderate and severe. For example, a mild disease may only cause minor symptoms in a localized area of ​​the fruit; a moderate disease may cause more pronounced symptoms and have a wider distribution; and a severe disease may affect most of the fruit, potentially severely impacting its growth and quality. The fruit disease prescription database is a database that categorizes and stores prescriptions by disease name and severity. Its structure is as follows: disease name is used as the primary classification, with different diseases stored separately. For example, powdery mildew, anthracnose, and downy mildew each have a category. Within each disease name category, severity is further classified as mild, moderate, and severe. For each combination of disease name and severity, a corresponding treatment prescription is stored. These prescriptions may include detailed information such as the drug name, dosage, method of use, and treatment duration. For example, for a mild powdery mildew infection, the prescription might be "Apply 50% carbendazim wettable powder at a dilution of 800 times, spraying every seven days for 2-3 consecutive sprays." A deep learning model based on semantic understanding parses the disease information, extracts the disease name and severity, and then matches it with the fruit disease prescription database. It finds a list of prescriptions that fully match the input disease information and outputs these prescriptions as recommended results.

[0053] In some embodiments, step S400 includes:

[0054] The disease name and the disease severity are semantically understood based on a deep learning model, and the prescriptions in the fruit disease prescription library are sorted by matching degree, and the prescriptions with the highest matching degree are recommended in turn.

[0055] It's important to understand that while apple pest and disease monitoring can reveal the names of apple pests and diseases, more accurate prescription recommendations require users to further describe the severity and scale of the disease. Based on this description, a deep learning model is applied to achieve semantic understanding and recommend the three most compatible prescriptions.

[0056] Deep learning models require strong semantic understanding capabilities to accurately parse disease information and effectively match it with prescription databases. Typically, natural language processing (NLP) techniques are combined with specific deep learning architectures, such as recurrent neural networks (RNNs) and their variants (such as long short-term memory (LSTM) networks and gated recurrent units (GRUs), or the attention-based Transformer architecture. These models are pre-trained on large amounts of agricultural text data to learn semantic representations of disease names, disease severity, and related agricultural terminology.

[0057] After obtaining the disease information generated by the neural network model, including the disease name and severity, it must first be preprocessed. This includes removing unnecessary symbols and converting the text to lowercase to ensure data consistency. For example, "Powdery Mild" will be processed as "Powdery Mild."

[0058] A pre-trained deep learning model is used to convert disease information and each prescription in the fruit disease prescription library into a vector representation. During this process, the model captures the semantic information in the text, placing semantically similar text closer together in the vector space. For example, "mild powdery mildew" and "treatment plan for mild powdery mildew infection" will have a high degree of similarity in the vector space.

[0059] For disease information, pass it as input to the semantic understanding model to obtain the corresponding vector representation V disease For each prescription p in the fruit disease prescription library i , and its vector representation V is also obtained through the same model prescriptioni .

[0060] Calculate the similarity between the disease information vector and each prescription vector as a measure of matching. Commonly used similarity calculation methods include cosine similarity and Euclidean distance. Taking cosine similarity as an example, its calculation formula is:

[0061]

[0062] Among them, the value range of cosine similarity is between [-1, 1]. The closer the value is to 1, the more similar the directions of the two vectors are, that is, the higher the semantic matching degree between the disease information and the prescription.

[0063] Based on the calculated matching degree, the prescriptions in the fruit disease prescription database are sorted in descending order. The prescription with the highest matching degree is placed at the top, and so on. The sorted prescription list is presented to the user, with the prescription with the highest matching degree being recommended first.

[0064] For example, for the disease information "mild powdery mildew," the calculated prescription matching degrees are as follows: Prescription A: Matching degree 0.9; Prescription B: Matching degree 0.7; Prescription C: Matching degree 0.6. The recommended order is Prescription A, Prescription B, Prescription C. The user will first see Prescription A, which best matches the current disease situation.

[0065] As time passes and data accumulates, the semantic understanding model can be continuously optimized. For example, user feedback on recommended prescriptions can be collected. If a user finds a recommended prescription inappropriate, it will be marked as a low-quality recommendation. This feedback data can be used to fine-tune the model, enabling it to better understand the semantic relationship between disease information and prescriptions, improving the accuracy of matching ranking. At the same time, the fruit disease prescription library is regularly updated to incorporate new treatment options and drug information to ensure the timeliness and effectiveness of recommended prescriptions.

[0066] In some embodiments, step S500 includes:

[0067] The collected prescriptions are managed in a favorites organization structure based on the user's adoption.

[0068] It is important to understand that based on the user's acceptance of prescription recommendations in the Apple Pest and Disease Prescription Recommendation and Apple Pest and Disease Prescription Recommendation History functions, the user can decide whether to save a recommended prescription. Saved prescriptions are organized into a favorites folder with the disease name as the favorite, making it easier for users to view them later.

[0069] When a user adopts a prescription, the system offers a "Favorite" option. Clicking this option adds the prescription's related information (such as the prescription name, corresponding disease name, disease severity, and prescription details) to the user's favorites. This process is achieved through interaction on the front-end interface, and the corresponding favorites are recorded in the back-end database.

[0070] Favorites can be organized in a hierarchical structure to help users better manage and find prescriptions. Common organizational structures can be designed based on the following dimensions:

[0071] Classification by disease name

[0072] Favorites can be categorized by disease name. For example, all prescriptions related to "powdery mildew" could be grouped into one category, and those related to "anthracnose" into another. This way, when users search for prescriptions for a specific disease, they can directly navigate to the corresponding category, improving search efficiency.

[0073] Classification by disease severity

[0074] Each disease category can be further categorized by severity. For example, the "Powdery Mildew" category can be further divided into "Mild Powdery Mildew Prescription," "Moderate Powdery Mildew Prescription," and "Severe Powdery Mildew Prescription." This categorization allows users to more accurately find the prescription appropriate for the current disease severity.

[0075] Custom tag classification

[0076] In addition to categorizing prescriptions based on disease name and severity, users can also add custom tags to prescriptions. For example, users can add tags like "Low-Cost Prescription" or "High-Efficiency Prescription" based on factors such as the cost of the prescription and its therapeutic effect. Custom tags allow users to categorize and filter prescriptions more flexibly.

[0077] In addition, in the favorites, the display of prescriptions should be clear and easy for users to view. Each prescription can display key information, such as the prescription name, the corresponding disease name and the disease severity. At the same time, a sorting function can be provided to allow users to sort prescriptions according to their needs. Common sorting methods include:

[0078] Sort by collection time

[0079] The prescriptions can be sorted in the order of the time when the user collected them, with the most recently collected prescriptions at the front. In this way, the user can see the most recently collected prescriptions in a timely manner.

[0080] Sort by frequency of use

[0081] The frequency of use of each prescription is counted and the most frequently used prescriptions are ranked first. This sorting method allows users to quickly find frequently used prescriptions.

[0082] In some embodiments, step S100 includes:

[0083] The real-time status image data is collected by a drone equipped with a multispectral imager.

[0084] It is important to understand that a multispectral imager equipped with a drone is used to obtain remote sensing image data of the orchard, while a fixed-point camera is used to collect large-scale visible light images, and finally a large-scale multi-source image database of orchard diseases is jointly constructed.

[0085] In some embodiments, reference Figure 2 As shown, step S200 includes:

[0086] The neural network model uses historical state image data of the early stage of fruit growth as input and generates diseases in the later stage as output for optimization training.

[0087] It is important to understand that during the optimization and training process, the neural network model takes the historical state image data from multiple time points in the early stages of fruit growth, organizes them according to a specific time series and data format, and uses them as input to the neural network model. These image data cover various feature information such as the appearance, color, and texture of the fruit. The model uses convolution operations to slide convolution kernels of different sizes across the image to extract local features in the fruit image and generate feature maps, allowing the model to learn basic features such as the edges and shapes of the fruit. Maximum pooling or average pooling is used to reduce the dimensionality of the feature map, reducing the amount of computation while retaining key features. In the fully connected layer, the previously extracted features are integrated and mapped to the output space for disease prediction through matrix multiplication and bias operations.

[0088] During parameter adjustment, a backpropagation algorithm is used to calculate the error between the predicted results and the actual disease situation. Using the chain rule, the error is backpropagated from the output layer to calculate the gradient of each parameter to determine the direction and magnitude of the parameter update. Stochastic gradient descent is used as the optimizer to update the model parameters based on the calculated gradient. The learning rate of each parameter is adaptively adjusted, and parameters are dynamically updated during training, continuously optimizing the model towards reducing error. Ultimately, detailed information such as whether the fruit will develop disease later in the growth cycle, as well as the type and severity of the disease, is provided as the output, thereby continuously improving the accuracy and reliability of the model's predictions of fruit disease occurrence.

[0089] In some embodiments, reference Figure 2 As shown, step S300 includes:

[0090] The neural network model extracts features from real-time image data of the early stage of fruit growth and detects and locates diseases through multi-scale lesion comparison.

[0091] It is important to understand that the multi-scale lesion comparison phase is required to detect and locate the disease. The model will divide and analyze the extracted feature maps according to different scales. By constructing a feature pyramid at multiple scales, the features of possible lesions are compared at different levels. At a large scale, the approximate area of ​​the lesion can be quickly located, while at a small scale, the model focuses on identifying subtle features of the lesion, not missing any early or minor signs of disease. For example, for tiny lesions that appear in the early stages, in the small-scale feature map, the model can rely on the lesion feature patterns learned in the early stages to accurately identify their edges and internal texture features; and for larger lesions that have already spread, the large-scale feature map can quickly outline the overall range of the lesion.

[0092] The neural network model receives real-time image data of the early stages of fruit growth from various image acquisition devices, such as cameras in the orchard and drone-mounted cameras. These images are typically presented in the form of digital matrices, containing rich visual information such as color, texture, and shape.

[0093] The model first uses a convolutional layer for feature extraction. The convolutional layer contains multiple convolution kernels, each of which can be regarded as a small filter. These convolution kernels slide on the image and interact with local areas of the image through convolution operations. For example, a 3x3 convolution kernel moves one pixel at a time on the image (with a step size of 1), multiplies the weight of the convolution kernel with the pixel value of the corresponding image area, and sums the results to generate a new feature value. Different convolution kernels can capture different types of features in the image. For example, some convolution kernels are good at capturing edge information, while others are more sensitive to texture patterns. As the convolution layers are stacked, the model can gradually extract more advanced and abstract features from the original pixel information of the image.

[0094] After the convolution operation, the activation function ReLU is applied. The ReLU function is defined as f(x) = max(0, x) and introduces nonlinearity into the model. Without an activation function, no matter how many layers of linear convolution are applied, their combined effect is equivalent to a single linear transformation, making it ineffective for modeling complex image data. The ReLU function ensures that when the input value is less than 0, the output is 0; when the input value is greater than 0, the output is equal to the input value. This allows the model to learn more complex feature relationships.

[0095] To reduce the dimensionality of feature maps, minimize computational effort, and enhance model robustness, a pooling layer is typically used after the convolutional layer. Common pooling methods include max pooling and average pooling. Taking max pooling as an example, it divides the feature map into small, non-overlapping blocks (such as 2x2 blocks) and takes the maximum value in each block as the pooled output value for that block. This not only preserves the most important features in the image but also provides a certain degree of invariance to changes in the image, such as translation and rotation.

[0096] After processing through multiple convolutional and pooling layers, the neural network model generates feature maps of varying scales. Shallower-level feature maps retain more detailed image information because they are derived from the original image with fewer convolution and pooling operations. Their higher resolution allows them to capture subtle features of the lesion, such as the fine structure of its edges. Deeper-level feature maps, on the other hand, have lower resolution but contain more global and abstract information, enabling them to capture the differences between the lesion and surrounding healthy areas, as well as its position within the image.

[0097] For each scale feature map, the model analyzes the features of areas likely to contain lesions. It learns the differences in features at different scales between normal fruit and lesion areas. For example, the color of lesions may differ significantly in hue and saturation from that of normal fruit. Furthermore, lesions may have unique texture patterns, such as roughness and mottled textures. These characteristics manifest themselves differently in feature maps at different scales.

[0098] By comparing lesion features on feature maps at different scales, the model can more accurately detect and locate diseases. This is because lesions of different sizes appear more clearly on feature maps at different scales. Smaller lesions are easier to identify on high-resolution shallow-level feature maps, where their detailed features can be clearly captured. Larger lesions, on the other hand, are more accurately determined on low-resolution deep-level feature maps, as they can display their overall shape and location from a more macroscopic perspective. The model comprehensively considers lesion information on feature maps at all scales, comparing the consistency and differences in lesion features at different scales to determine the presence and specific location of the disease.

[0099] Ultimately, the model outputs disease detection and location information based on the results of multi-scale lesion comparison. This information can be presented in various forms, such as marking the lesion's location with a rectangular or polygonal box on the original image, along with the disease type (if the model has classification capabilities) and related lesion attributes (such as area and shape). This information provides a crucial basis for subsequent disease prevention and control measures, helping agricultural workers quickly and accurately understand the disease status of fruit and implement targeted treatments.

[0100] In some embodiments, reference Figure 2 As shown, step S300 includes:

[0101] The neural network model combines the extracted features with the disease patterns in the fruit disease knowledge base to analyze environmental factors and predict the probability of disease occurrence.

[0102] It should be understood that the alarm threshold for each disease and pest is determined based on the disease patterns in the apple disease knowledge base. When the predicted probability of a certain apple leaf disease exceeds the set threshold, the alarm is triggered and the relevant plant protection personnel are notified.

[0103] The Fruit Disease Knowledge Base is a comprehensive and extensive collection of information on various fruit diseases. The disease patterns are particularly critical, encompassing the characteristics of different diseases at different growth stages. For example, some diseases are more likely to occur during the young fruit stage, while others are more likely to develop during the mature stage. Disease patterns also encompass the close connection between diseases and environmental factors. For example, high temperatures and high humidity favor the growth of some fungal diseases, while drought conditions can make certain viral diseases more susceptible to transmission. This knowledge, gathered through long-term agricultural research, field observations, and extensive experimental data, provides a solid theoretical basis for subsequent disease prediction.

[0104] To accurately analyze the impact of environmental factors on disease occurrence, real-time collection of diverse environmental data is necessary. Common environmental factors include temperature, humidity, light intensity, rainfall, and soil pH. This data can be collected using various sensors distributed throughout the orchard. Temperature sensors monitor the real-time temperature within the orchard, humidity sensors measure air humidity, light sensors record light intensity, rain sensors measure rainfall, and soil sensors monitor soil pH and other physical and chemical properties. These sensors transmit this collected environmental data in real time to the neural network model, serving as key input for model analysis.

[0105] The neural network model combines extracted fruit image features with disease patterns in a knowledge base of fruit diseases to conduct an in-depth analysis of the collected environmental factor data. The model looks for correlations between the fruit condition reflected by the image features and the environmental conditions in the disease patterns. For example, if the image features reveal tiny spots on the fruit surface that resemble lesions, the model will analyze whether the current environmental factors, such as temperature and humidity, meet the conditions for a particular disease to develop, based on the relationships between these symptoms and environmental factors in the disease knowledge base. If the disease patterns indicate that a particular disease is more likely to develop in a specific temperature range and high humidity, and the environmental data collected by the sensor falls within this temperature range and high humidity, the model will determine that these environmental conditions increase the likelihood of the disease developing.

[0106] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine different embodiments or examples described in this specification.

[0107] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the appended claims.

Claims

1. A method for recommending fruit disease prescriptions, characterized in that: include: Acquire real-time image data of the early stage of fruit growth; Performing disease occurrence monitoring on the real-time status image data based on a neural network model; When it is determined that a disease has occurred, the neural network model generates disease information corresponding to the real-time state image data; Matching multiple prescriptions in a fruit disease prescription library based on a deep learning model based on semantic understanding of the disease information; Recommend multiple matched prescriptions to users and record them based on the users' adoption.

2. The method for recommending fruit disease prescriptions according to claim 1, characterized in that: The step of generating the disease information corresponding to the real-time state image data by the neural network model after determining that a disease has occurred includes: The disease information at least includes the name of the disease and the severity of the disease.

3. The method for recommending fruit disease prescriptions according to claim 2, characterized in that: The step of matching multiple prescriptions in the fruit disease prescription library according to the deep learning model based on semantic understanding of the disease information includes: The fruit disease prescription library classifies and stores prescriptions in sequence based on the disease name and the disease severity.

4. The method for recommending fruit disease prescriptions according to claim 3, characterized in that: The step of matching multiple prescriptions in the fruit disease prescription library according to the deep learning model based on semantic understanding of the disease information includes: The disease name and the disease severity are semantically understood based on a deep learning model, and the prescriptions in the fruit disease prescription library are sorted by matching degree, and the prescriptions with the highest matching degree are recommended in turn.

5. The method for recommending fruit disease prescriptions according to claim 1, characterized in that: The steps of recommending the matched multiple prescriptions to the user and recording the user's adoption status include: The collected prescriptions are managed in a favorites organizational structure based on the user's adoption.

6. The method for recommending fruit disease prescriptions according to claim 1, characterized in that: The step of obtaining real-time state image data of the early stage of fruit growth comprises: The real-time status image data is collected by a drone equipped with a multispectral imager.

7. The method for recommending fruit disease prescriptions according to any one of claims 1 to 6, characterized in that: The step of monitoring the occurrence of diseases on the real-time state image data based on the neural network model includes: The neural network model uses historical state image data of the early stage of fruit growth as input and generates diseases in the later stage as output for optimization training.

8. The method for recommending fruit disease prescriptions according to claim 7, characterized in that: The step of generating the disease information corresponding to the real-time state image data by the neural network model after determining that a disease has occurred includes: The neural network model extracts features from real-time image data of the early stage of fruit growth and detects and locates diseases through multi-scale lesion comparison.

9. The method for recommending fruit disease prescriptions according to claim 8, characterized in that: The step of generating the disease information corresponding to the real-time state image data by the neural network model after determining that a disease has occurred includes: The neural network model combines the extracted features with the disease patterns in the fruit disease knowledge base to analyze environmental factors and predict the probability of disease occurrence.

Images