Intelligent diagnosis method and system for fruit tree diseases and insect pests based on image recognition

Abstract

The invention relates to the technical field of disease and insect pest diagnosis, in particular to an intelligent fruit tree disease and insect pest diagnosis method and system based on image recognition. The method comprises the steps of obtaining an image of a target object and environment parameters corresponding to the image for at least one time; the image of one time and the environment parameters corresponding to the image are analyzed; and outputting the prediction result in at least one form. According to the method, defective user pictures are not directly processed, and the internal knowledge base is dynamically modulated by using the environmental parameters, so that the knowledge base simulates the proper performance in the current environment, and matching and decision making are carried out on the same condition reference as user input. This mechanism ensures that cognition of input data is established on the basis of explicit understanding and compensation of environmental interference. And the operation threshold of the diagnosis method is greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of disease and pest diagnosis technology, and more specifically, to an intelligent diagnostic method and system for fruit tree diseases and pests based on image recognition. Background Technology

[0002] In recent years, with the continuous deepening of multimodal research in deep learning, the automatic diagnosis of agricultural pests and diseases and the delivery of agricultural technology can be accomplished with the help of computer vision and natural language processing technologies.

[0003] In the development of smart agriculture, the identification and control of fruit tree diseases are crucial for inexperienced digital users. Existing technologies, by constructing a knowledge graph of fruit tree diseases and pests, effectively link fragmented knowledge and can perform high-precision matching based on user-retrieved and uploaded images, providing recommendations for prevention and control resources, significantly improving the accuracy of information services. This method relies on the accurate extraction of disease and pest features from images to support the retrieval and reasoning of the knowledge graph.

[0004] However, existing technologies suffer from a critical flaw in practical applications: their high-precision diagnosis relies on idealized image quality. Farmers, limited by their expertise and equipment operation skills, often have photos affected by environmental factors such as insufficient light, angular deviations, or uneven brightness. When processing these photos, the system cannot distinguish between visual artifacts caused by environmental parameters and genuine disease characteristics. For example, a leaf appearing dull due to dim lighting might be misjudged as pathological darkening, triggering an incorrect diagnosis. This cognitive bias prevents even highly accurate models from operating reliably in real-world scenarios because the interpretation of the input data is fundamentally flawed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent diagnostic method and system for fruit tree diseases and pests based on image recognition, so as to solve the problem of high operational threshold of existing diagnostic methods.

[0006] Firstly, an intelligent diagnostic method and system for fruit tree diseases and pests based on image recognition is provided, including: Acquire at least one image of the target object and the corresponding environmental parameters; The image and its corresponding environmental parameters are analyzed to obtain at least one form of prediction result; the environmental parameters are used to dynamically modulate the internal knowledge base to simulate its performance under the current environment; the knowledge base is a structured information storage module that provides a decision basis for the output of the prediction result. The prediction results are output in at least one form.

[0007] As a further improvement to this technical solution, acquiring at least one image and the corresponding environmental parameters includes: Only one image and its corresponding environmental parameters are captured; or, The system acquires multiple images and their corresponding environmental parameters. Based on each acquired image and its corresponding environmental parameters, it provides correction parameters for the next data acquisition until it acquires an image and its corresponding environmental parameters that meet the expectations. The correction parameters are used to guide users to obtain images that meet expectations and the corresponding environmental parameters.

[0008] As a further improvement to this technical solution, the environmental parameters corresponding to the image refer to: It can reflect the parameters of the environment in which the target object is located when the image is generated.

[0009] As a further improvement to this technical solution, the analysis includes one of the images and the corresponding environmental parameters: The analysis is performed using the last acquired image and its corresponding environmental parameters. The steps are as follows: Extract at least one type of feature from the image; By filtering the features from all types, a feature set corresponding to each type is obtained; The feature set and environmental parameters are used as input features for analysis.

[0010] As a further improvement to this technical solution, the filtering of features across all types includes: Features are categorized according to type; Remove incomplete features; Integrate the remaining features belonging to the same type into a feature set; Alternatively, the remaining features can be filtered at least once more, and after the final filtering, the remaining features belonging to the same type can be integrated into a feature set.

[0011] As a further improvement to this technical solution, the remaining features are screened at least once more, including: Screening criteria are determined based on type and environmental parameters; Features that do not meet the screening criteria will be removed.

[0012] As a further improvement to this technical solution, the steps for analyzing the image and the corresponding environmental parameters include: Obtain input features; Using environmental parameters as prior information, a correction weight vector with the same dimension as the environmental parameters is output. The correction weight is then used to modulate the template in the knowledge base element by element to obtain the aligned template. Based on the aligned template, aggregation matching is used to perform calculations to obtain the prediction results; Output the prediction results.

[0013] As a further improvement to this technical solution, at least one form of prediction result includes having only one prediction result; Alternatively, multiple prediction results sorted in a preset order.

[0014] Secondly, an intelligent diagnostic method for fruit tree diseases and pests based on image recognition is provided, including: The image acquisition module is used to acquire at least one image of the target object and the corresponding environmental parameters of the image; The prediction result analysis module is used to analyze one of the images and the corresponding environmental parameters to obtain at least one form of prediction result; as well as, The prediction result output module outputs the prediction results to the outside world in at least one form.

[0015] Thirdly, a computer storage medium is provided, on which a computer program is stored, which, when executed by a processing device, implements the above-mentioned intelligent diagnosis method for fruit tree diseases and pests based on image recognition.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This image recognition-based intelligent diagnosis method and system for fruit tree diseases and pests does not directly process flawed user images. Instead, it dynamically modulates an internal knowledge base (i.e., high-quality baseline features) using environmental parameters, simulating its expected behavior under the current environment. This allows for matching and decision-making with user input on the same baseline conditions. This mechanism ensures that the understanding of input data is based on an explicit understanding and compensation for environmental disturbances, significantly lowering the operational threshold of the diagnostic method. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the intelligent diagnosis method for fruit tree diseases and pests based on image recognition according to the present invention. Figure 2 A flowchart illustrating the steps involved in analyzing images and corresponding environmental parameters in this invention. Figure 3 This is a flowchart of the steps for filtering remaining features according to the present invention; Figure 4 This is a schematic diagram of the image recognition-based intelligent diagnostic system for fruit tree diseases and pests according to the present invention. Detailed Implementation

[0018] 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 are only some embodiments of the present invention, and not all embodiments. 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] In intelligent diagnosis of fruit tree diseases for non-professional farmers, the core challenge lies in the contradiction between high-precision models relying on high-quality input and the low-quality, variable shooting conditions of actual user images. Existing solutions enhance user images through post-processing or simply improve model robustness, failing to fundamentally address the problem of models misinterpreting the nature of data due to environmental interference.

[0020] The core innovation of this application's technical solution lies in its approach: instead of directly processing flawed user images, it dynamically modulates an internal knowledge base (i.e., high-quality baseline features) using environmental parameters, simulating its expected performance under the current environment. This allows for matching and decision-making with user input based on the same baseline conditions. This mechanism ensures that the understanding of input data is built upon an explicit understanding and compensation for environmental interference, significantly lowering the operational threshold of the diagnostic method.

[0021] Specifically, this application provides an image recognition-based intelligent diagnosis method for fruit tree diseases and pests, including: Acquire at least one image of the target object and the corresponding environmental parameters; By analyzing one of the images and the corresponding environmental parameters, at least one form of prediction result can be obtained; The prediction results are output in at least one form.

[0022] The first embodiment provides an image recognition-based intelligent diagnosis method for fruit tree diseases and pests, specifically for scenarios involving intelligent diagnosis of fruit tree diseases among non-professional farmers. (See [link to relevant documentation]). Figure 1 ,include: Step S1.1: Acquire at least one image of the target object and the corresponding environmental parameters; This includes acquiring an image and the corresponding environmental parameters only once. This is the most common implementation method, where farmers only need to take one picture. For example, farmers only need to use a smartphone or other mobile device to take one picture of the part of the fruit tree suspected of being infected (i.e., the target object, such as leaves, branches, or fruit).

[0023] For smartphones or other mobile devices equipped with corresponding sensors, after user authorization, the device API can be invoked simultaneously at the moment of shooting to read a series of physical environment data in real time, which is used to obtain environmental parameters corresponding to the image. This typically includes: illuminance values ​​(unit: lux) provided by the ambient light sensor, directly reflecting the brightness of the shooting scene; device attitude data (such as three-axis accelerometers and gyroscopes) provided by the inertial measurement unit (IMU), which can be used to calculate the pitch angle, yaw angle, etc. of the device relative to the horizontal plane at the time of shooting, thereby determining the shooting angle; in addition, some devices can also provide parameters such as ambient temperature and humidity. These sensor data directly constitute an important part of the environmental parameters in the form of high-precision numerical values, and their advantage lies in the objectivity, directness, and independence from the image content.

[0024] It is possible that in some embodiments, step S1.1 requires acquiring multiple images and their corresponding environmental parameters. Furthermore, based on each acquired image and its corresponding environmental parameters, correction parameters are provided for the next data acquisition, until an image and its corresponding environmental parameters that meet the expectations are obtained. Among them, the correction parameters are used to guide users to obtain images that meet expectations and the corresponding environmental parameters.

[0025] In practice, an initial image capture and environmental parameter acquisition are performed. Unlike the single-shot mode, a lightweight evaluation function is added here, which quickly and jointly evaluates the current input image and its corresponding environmental parameters. The evaluation mainly revolves around two dimensions: first, image quality, judged by calculating indicators such as image sharpness (e.g., Laplacian variance), the proportion and integrity of the target area in the image, and illumination uniformity (avoiding overexposure or underexposure); second, environmental adaptability, that is, judging whether the current environmental parameters (e.g., light intensity, shooting angle) fall within the ideal environment window corresponding to the optimal performance of the pre-trained model. For example, the model may perform best when the illuminance is 500-2000 lux and the shooting angle deviation is less than 30 degrees.

[0026] Based on the output of the evaluation function, a set of structured correction parameters are generated. These parameters are not internal model parameters, but rather translated into explicit user-oriented operational guidelines. For example, if the evaluation detects a blurry image (low sharpness score), the correction parameter might be: "Blurry image detected. Please hold the device steady, tap the screen to focus, and then take the picture." If the evaluation detects that the ambient light is too low (illuminance value below the ideal window lower limit), the correction parameter might be: "Insufficient light. Please turn on the flash or move to a brighter location to shoot." If the evaluation detects that the shooting angle is too tilted (angle deviation too large), the correction parameter might be: "The shooting angle affects recognition. Please try to make the camera directly facing the leaf plane."

[0027] These instructions are presented instantly via a user interface (UI) in the form of text, images, or voice. After adjusting according to the instructions, the user takes the next shot. New images and environmental parameters are received, and the evaluation function is run again. This evaluation takes into account the previous state, achieving progressive optimization. This process iterates until the evaluation score exceeds a preset quality threshold, indicating that the currently acquired data meets expectations. At this point, the acquisition guidance loop automatically exits.

[0028] It should be noted that the environmental parameters corresponding to the image refer to parameters that reflect the environment in which the target object was located when the image was generated (which can be understood as when the image was taken), including: Lighting parameters are the most critical environmental factors. The lighting environment of the target object (including light source intensity, spectral distribution, and direction) directly determines the intensity and color of the light reflected to the camera. For example, the same lesion may lose detail due to reflection under strong midday light, while it may become blurry due to increased noise under dim twilight light.

[0029] The observation geometry parameters mainly refer to the spatial pose relationship between the camera and the target surface. The actual position and orientation of the target, combined with the camera's shooting angle and distance, jointly determine the perspective distortion, scale, and focal area of ​​the target in the image. For example, when the blade plane is not parallel to the camera's imaging plane, it will cause trapezoidal distortion in the image and may cause some areas to be out of focus.

[0030] Step S1.2: Analyze one of the images and the corresponding environmental parameters to obtain at least one form of prediction result.

[0031] Possible forms of prediction results include single diagnostic conclusions and confidence levels, ranked lists and probability distributions, risk warnings and meta-judgments, or structured diagnostic reports.

[0032] Specifically, a single diagnostic conclusion with confidence level is the most straightforward form, which outputs the most probable category (such as "rust") and attaches a quantified confidence score (e.g., 92%).

[0033] The sorted list and probability distribution offer a more comprehensive and information-rich output format. It not only outputs the most likely category but also a list of the top K (e.g., K=3 or 5) candidate categories, ordered by probability in descending order, with each candidate accompanied by its corresponding probability or similarity score. For example: 1. Anthracnose (65%), 2. Brown spot (28%), 3. Physiological nutrient deficiency (7%). This format explicitly reveals the uncertainty of the model's decision-making. It is extremely valuable to farmers because the most similar disease, "anthracnose," and the second most similar disease, "brown spot," may be visually very similar, but their control measures differ. This list partially returns decision-making power to the user, who can combine their field experience ("recent heavy rainfall, high anthracnose incidence"), environmental parameters ("current humidity extremely high"), and other external knowledge to select more reasonable options or decisions for deeper review.

[0034] Structured diagnostic reports are the most detailed output format, transforming predictions into a comprehensive report containing multiple components. These reports typically include: primary diagnosis, differential diagnosis, characteristic evidence, environmental correlation analysis, and confidence level assessment.

[0035] Risk warning and meta-judgment: In some cases, the primary task of diagnosis is not to give a specific disease name, but to determine "whether there is an abnormality" or "risk level".

[0036] Considering that the final acquisition was a targeted optimization based on all previous evaluations, its output image data theoretically represents the optimal observation sample that can be obtained within the system's capabilities under the current field conditions. As a preferred option, see [reference needed]. Figure 2 The analysis is based on the last acquired image and the corresponding environmental parameters, specifically including: S2.1 Obtain at least one type of feature from the image, where the type refers to the feature classification of the fruit tree, such as: leaves, branches, or fruits; S2.2 Filter the features from all types to obtain the feature set corresponding to each type; specifically, filtering the features from all types includes: Features are categorized according to type; Remove incomplete features; The remaining features belonging to the same type are integrated into a feature set.

[0037] S2.3. Analyze the feature set and environmental parameters as input features. The steps include: Obtain input features; Using environmental parameters as prior information, a correction weight vector with the same dimension as the environmental parameters is output. This correction weight is then used to modulate the templates in the knowledge base element-by-element, resulting in aligned templates. The knowledge base refers to a structured information storage module that integrates core elements such as domain expert knowledge, historical data, disease feature databases, and model parameters, providing a decision-making basis for the prediction results. Essentially, it digitizes and standardizes scattered agricultural expertise (such as disease symptom descriptions, environmental influencing factors, and prevention experience) and the regularities obtained from training machine learning models, storing them to form a queryable, updatable, and reasonable static or dynamic database. Based on the aligned template, aggregation matching is used to perform calculations to obtain the prediction results; S2.4 Output the prediction results.

[0038] Taking the joint monitoring of early black spot disease and fruit maturity in apple trees as an example, this paper elaborates on how the scheme can achieve accurate analysis from an optimized image containing leaves and fruit.

[0039] Digital images of the apple tree canopy are automatically acquired in the orchard. These images are denoted as a matrix. Its dimensions are height ,width It has three color channels: red, green, and blue. Simultaneously, the sensor records environmental parameters at the moment of capture, forming a vector. .in, Represents temperature (unit: degrees Celsius). Represents relative humidity (unit: percentage). Represents light intensity (unit: lux). Environmental parameters need to be standardized for ease of subsequent processing: ; here, and It is a vector of the mean and standard deviation of various environmental parameters, pre-calculated on a large amount of training data. Operators This represents element-wise division. The standardized vector. The influence of different physical dimensions is eliminated, making it easier for the model to learn the relationship between environmental and feature changes.

[0040] Image Input a pre-trained instance segmentation network (such as Mask R-CNN). Network output. The detected instance object. For the first... One instance ( The output consists of three parts: Binary mask This is a binary matrix of the same size as the input image. A pixel with a value of 1 belongs to the instance (such as a leaf or an apple), and a pixel with a value of 0 does not belong to it.

[0041] Category Tags ,in This indicates the category "Leaf". It represents the category "Fruit".

[0042] Basic visual feature vectors This vector is derived from the backbone feature map of the segmentation network in the mask. The corresponding region is obtained through pooling using the ROI Align operation, which encodes the global color, texture, and shape outline information of the object.

[0043] This completes the transformation from pixels to a list of structured objects. .

[0044] Next, low-quality objects are removed, and a statistical feature descriptor reflecting the overall state is constructed for each type (leaf, fruit). Specifically, for each For each instance, calculate its mask integrity score: ; In the formula, It is the total number of pixels with a value of 1 in the mask. It is a function that calculates the bounding rectangle of an object. This is for calculating the area of ​​a rectangle. If... If a leaf is severely obscured or only a portion is captured in the photograph, it is considered to be missing or discarded. This is because early-stage black spot disease lesions may be small and scattered; if the leaf is incomplete, the extracted features will not represent the overall leaf condition, leading to missed detections or misdiagnosis. For each For example, calculate the circularity of its mask: ; In the formula, Calculate the perimeter of the mask outline. Simultaneously, check if its bounding rectangle touches the image boundary. If... If the fruit touches the boundary, it is considered to have an abnormal shape (possibly obscured or cut) and is discarded. This ensures that the apples used for maturity analysis are intact and unobstructed, making the measurement of characteristics such as color and size accurate and reliable.

[0045] Assuming that after filtering, we get Examples of high-quality blades and An example of high-quality fruit.

[0046] In addition, for each high-quality blade instance, in its masked area Within this process, a set of detailed features specifically designed for disease identification is calculated to form a vector. .For example: In the CIELab color space On the (red-green) channel, the mean and standard deviation of all pixel values ​​are calculated to capture the color difference and distribution between lesions (often dark) and healthy tissue (green).

[0047] Calculate the histogram entropy of local binary pattern (LBP) texture features to quantify the degree of disorder in leaf texture (the texture of lesion areas is usually coarser).

[0048] Calculate the proportion of green pixels within the mask (reflecting chlorophyll content).

[0049] For each high-quality fruit instance, in its masked area Internally, calculate the detailed feature vector. .For example: Calculate the ratio of the mean values ​​of the red channel to the green channel in the RGB color space (reflecting the degree of coloration).

[0050] Calculate the statistical kurtosis of hue (H) in the HSI color space (reflecting color purity; ripe fruit is purer).

[0051] Calculate the mask area (reflecting size) and the overall average brightness.

[0052] The detailed feature vector sets of leaves and fruits are statistically aggregated separately to form a fixed-length "feature set" vector.

[0053] Leaf feature set Depend on The 10-dimensional detailed features of each leaf are calculated by taking the mean and standard deviation for each dimension and then stitching them together.

[0054] ; In the formula, This represents the mean. The standard deviation represents the average disease severity across the entire tree's leaves. The mean represents the average disease severity across all leaves, while the standard deviation reflects the uniformity of lesion distribution (early lesions may be sparsely distributed, resulting in a large standard deviation). This design allows the model to perceive the overall condition of the plant population, rather than individual leaves, making it more suitable for comprehensive field assessment.

[0055] Fruit feature set In the same way The mean and standard deviation of the 8-dimensional features of each fruit are aggregated.

[0056] The diagnostic process is conducted independently in two pathways: leaf and fruit. A predefined baseline template is used under standard laboratory conditions (e.g., uniform light, controlled temperature and humidity). For leaves, a healthy template is available. Early black star disease template Deficiency syndrome template For fruits, there are green fruit templates. Color change period template Mature fruit template These templates are related to feature sets. Vectors of the same dimension.

[0057] Environmental parameters Mapped to modulation parameters via a lightweight coding network (such as a multilayer perceptron). : ; here and These represent scaling and offset instructions for each dimension of the feature. Subsequently, affine modulation is performed on each feature in the benchmark library to generate the simulation environment. The following characteristics: , , ; in This represents element-wise multiplication. The modulated feature set is then compared with the environment. Aligned feature space.

[0058] In the aligned feature space, the cosine similarity between the feature set and the aligned baseline templates for each category is calculated. Taking a blade as an example: , , ; The similarity scores are transformed into a probability distribution using the Softmax function: ; In the formula, It is a predictor of blade condition. This represents the probability of being diagnosed with early-stage black spot disease. Similarly, the probability of a healthy person being diagnosed is calculated. and nutrient deficiency The probability of.

[0059] The final output includes two diagnostic results and their confidence levels: If it is If so, an alarm will be triggered and the probability will be given.

[0060] Meanwhile, the standard deviation of the fruit characteristic set can be used to assess orchard maturity uniformity.

[0061] The aforementioned scheme decomposes the complex task of field visual monitoring into a series of automatically executable sub-modules. Starting with raw images and environmental data, it obtains objects through instance segmentation, ensures data reliability through quality screening, and then forms robust population characteristics through statistical aggregation. Furthermore, it utilizes a learned environmental adapter to achieve feature alignment, and finally completes diagnosis by matching similarity with a modulated standard template. This effectively solves the most critical generalization problem in the implementation of agricultural vision systems, providing a concrete and feasible technical path for early warning of black spot disease and harvest period determination.

[0062] In some embodiments, the remaining features are filtered at least once more, and after the final filtering is completed, the remaining features belonging to the same type are integrated into a feature set; See Figure 3 The steps for further filtering the remaining features at least once more include: 3.1 Determine screening criteria based on type and environmental parameters; 3.2. Remove features that do not meet the screening criteria.

[0063] For example, suppose we have segmented 100 complete apple leaves from an image and extracted the "lesion contrast" feature for each leaf (the higher the feature value, the greater the theoretical probability of black spot disease lesions). The shooting environment was cloudy (weak light). Under low light, even healthy green leaves will have reduced overall contrast; and the contrast difference between early black spot disease lesions (small and dark in color) and the leaf will become very inconspicuous.

[0064] A secondary screening is performed, inputting: the set of contrast features of all remaining leaf lesions, and the current environmental parameters (the key here is light intensity l). A built-in concept states that light intensity l is positively correlated with the reliable lesion contrast feature threshold. That is, the stronger the light, the higher the reliable lesion feature threshold should be (because the signal is clearer); the weaker the light, the lower the threshold should be (to avoid missing weak signals).

[0065] Filtering: A low filtering threshold is dynamically calculated based on the currently measured cloudy day light intensity. This low threshold is then used to determine the characteristics of the 100 leaves. Leaves with extremely low contrast (e.g., close to 0, which may indicate a blurred image or severely shadowed areas) are filtered out because they do not provide useful information under the current conditions; while most leaves with medium to low contrast values ​​near the low filtering threshold are retained.

[0066] This screening process did not attempt to directly diagnose diseases. Instead, it filtered out feature data that was inherently unreliable under the current environmental conditions. The retained leaf features were those that were relatively informative under cloudy conditions. Subsequently, these leaf features were integrated into a feature set representing the state of the apple tree canopy leaves under the current cloudy conditions, and then fed into the subsequent diagnostic model. This ensured that the final feature set submitted for diagnosis consisted of instances most likely to carry valid information under the current specific environment, avoiding systematic misjudgments or omissions caused by environmental changes.

[0067] Finally, regarding step S1.3, outputting the prediction results in at least one form includes: With only a single prediction result, its advantage lies in direct decision-making and clear action instructions, eliminating the need for secondary human interpretation and significantly lowering the operational threshold. It can seamlessly integrate with automated execution equipment (such as precision sprayers and automatic sorting machines) to form a closed loop of perception, decision-making, and execution, offering the fastest response time. It is best suited for high-frequency, standardized operational scenarios requiring immediate automated response. For example, when the system is integrated into the real-time spraying module of a drone, once the probability of disease exceeds an absolute threshold, a spraying command must be output immediately and unconditionally.

[0068] Alternatively, multiple predictions ordered in a preset sequence offer the advantage of comprehensive information and strong decision support capabilities. They not only provide the most probable outcome but also reveal other possibilities and their weights through ranking, which helps human experts make comprehensive judgments in situations of high uncertainty. This is primarily used in scenarios involving generating analytical reports and assisting human decision-making. For example, generating daily orchard health reports for managers to use for resource planning and developing weekly agricultural plans. Providing multiple results when the model's confidence level in a particular sample is low is also a responsible approach, triggering more in-depth human review or consultation with other data.

[0069] In the second embodiment, to solidify the method flow of the first embodiment into an integrated hardware and software entity, this embodiment focuses on achieving a more stable, reliable, and user-friendly end-to-end service through specific system architecture and component collaboration. This embodiment provides an image recognition-based intelligent diagnostic system for fruit tree diseases and pests. (See [link to relevant documentation]). Figure 4 ,include: The image acquisition module is used to acquire at least one image of the target object and the corresponding environmental parameters of the image; The prediction result analysis module is used to analyze one of the images and the corresponding environmental parameters to obtain at least one form of prediction result; In addition, a prediction result output module outputs prediction results to the outside in at least one form.

[0070] The purpose of this embodiment is to hide complex imaging technology behind a simple user interface. Non-professional farmers only need to point and take a picture of the target, and the system can automatically complete the entire process from image formation to providing action suggestions, significantly lowering the technical barrier to entry. Furthermore, the system's modular architecture allows for upgrades to hardware (such as replacing with a higher-precision sensor) or software algorithms, ensuring the sustainable evolution of the technology. Therefore, this system embodiment transforms the method in the first embodiment into a stable, easy-to-use, and deployable substantive product from both the physical carrier and interaction perspectives.

[0071] In the third embodiment, a computer storage medium is provided. This disclosure also provides a computer storage medium storing a computer program thereon. When the program is executed by a processing device, it implements the image recognition intelligent diagnosis method for fruit tree diseases and pests provided in this disclosure.

[0072] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent diagnostic method for fruit tree diseases and pests based on image recognition, characterized in that, include: Acquire at least one image of the target object and the corresponding environmental parameters; The image and its corresponding environmental parameters are analyzed to obtain at least one form of prediction result; the environmental parameters are used to dynamically modulate the internal knowledge base to simulate its performance under the current environment; the knowledge base is a structured information storage module that provides a decision basis for the output of the prediction result. The prediction results are output in at least one form.

2. The intelligent diagnostic method for fruit tree diseases and pests based on image recognition according to claim 1, characterized in that, Obtaining at least one image and the corresponding environmental parameters for that image includes: Only one image and its corresponding environmental parameters are captured; or, The system acquires multiple images and their corresponding environmental parameters. Based on each acquired image and its corresponding environmental parameters, it provides correction parameters for the next data acquisition until it acquires an image and its corresponding environmental parameters that meet the expectations. The correction parameters are used to guide users to obtain images that meet expectations and the corresponding environmental parameters.

3. The intelligent diagnostic method for fruit tree diseases and pests based on image recognition according to claim 2, characterized in that, The environmental parameters corresponding to the image refer to: It can reflect the parameters of the environment in which the target object is located when the image is generated.

4. The intelligent diagnosis method for fruit tree diseases and pests based on image recognition according to claim 1, characterized in that, The analysis, based on one of the images and the corresponding environmental parameters, includes: The analysis is performed using the last acquired image and its corresponding environmental parameters. The steps are as follows: Extract at least one type of feature from the image; By filtering the features from all types, a feature set corresponding to each type is obtained; The feature set and environmental parameters are used as input features for analysis.

5. The intelligent diagnostic method for fruit tree diseases and pests based on image recognition according to claim 4, characterized in that, Filtering features across all types includes: Features are categorized according to type; Remove incomplete features; Integrate the remaining features belonging to the same type into a feature set; Alternatively, the remaining features can be filtered at least once more, and after the final filtering, the remaining features belonging to the same type can be integrated into a feature set.

6. The intelligent diagnosis method for fruit tree diseases and pests based on image recognition according to claim 5, characterized in that, The remaining features are then filtered at least once more, including: Screening criteria are determined based on type and environmental parameters; Features that do not meet the screening criteria will be removed.

7. The intelligent diagnosis method for fruit tree diseases and pests based on image recognition according to claim 4, characterized in that, The steps for analyzing images and their corresponding environmental parameters include: Obtain input features; Using environmental parameters as prior information, a correction weight vector with the same dimension as the environmental parameters is output. The correction weight is then used to modulate the template in the knowledge base element by element to obtain the aligned template. Based on the aligned template, aggregation matching is used to perform calculations to obtain the prediction results; Output the prediction results.

8. The intelligent diagnostic method for fruit tree diseases and pests based on image recognition according to claim 1, characterized in that, At least one form of prediction result includes having only one prediction result; Alternatively, multiple prediction results sorted in a preset order.

9. A method for intelligent diagnosis of fruit tree diseases and pests based on image recognition, characterized in that, include: The image acquisition module is used to acquire at least one image of the target object and the corresponding environmental parameters of the image; The prediction result analysis module is used to analyze one of the images and the corresponding environmental parameters to obtain at least one form of prediction result; as well as, The prediction result output module outputs the prediction results to the outside world in at least one form.

10. A computer storage medium having a computer program stored thereon, characterized in that, When executed by the processing device, the program implements the intelligent diagnosis method for fruit tree diseases and pests based on image recognition as described in any of claims 1-8.

Images