Farmland disease and insect pest image recognition inspection robot

Abstract

The invention relates to the technical field of intelligent agriculture, and discloses a farmland pest image recognition inspection robot which comprises a moving and navigation platform, a wide-area sensing and screening system, a multifunctional diagnosis and marking mechanical arm and a data processing and control unit. The wide-area sensing and screening system of the robot is used for primarily screening images to lock suspicious target plants; an active physiological probe and an in-situ micro-marking nozzle module are integrated at the tail end of the multifunctional diagnosis and marking mechanical arm. The method comprises the following steps: performing contact measurement on a suspicious target plant to obtain multi-modal physiological information and performing preliminary diagnosis; when the diagnosis state is suspected, performing in-situ micro-labeling on the plant; and after a preset time interval, repositioning the plant and retesting. According to the method, wide-area image screening and contact type time sequence physiological diagnosis are combined, and the accuracy of disease and pest diagnosis, the early warning capability and the decision reliability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, specifically to a farmland pest and disease image recognition inspection robot. Background Technology

[0002] With the development of smart agriculture and robotics, the use of inspection robots or drones equipped with machine vision systems for automated monitoring of crop diseases and pests has become an important technological direction to replace traditional manual inspections. This type of technology analyzes visible light or multispectral images of the crop canopy using image recognition algorithms such as deep learning, aiming to achieve large-scale, high-efficiency automatic identification and location of diseases.

[0003] However, current inspection technologies that heavily rely on image recognition have inherent limitations. Their diagnostic basis is limited to external morphological characteristics of crops, such as macroscopic symptoms like leaf spots and yellowing. This means that diagnostic results are often only available after the disease has progressed to a certain stage, exhibiting a significant lag. Furthermore, early symptoms of various diseases, or physiological traits caused by environmental stress or nutrient imbalances, often appear visually very similar. This severely impacts the accuracy and reliability of diagnostic models that rely solely on image analysis when faced with atypical cases. Crucially, these technologies are completely incapable of detecting diseases in their latent stage, before any visible symptoms are observed.

[0004] There is an urgent need in this field for a novel technological solution that can break through the reliance on purely visual information and achieve deeper, earlier, and more reliable diagnosis of crop diseases, thereby providing stronger decision support for precision agricultural management. Therefore, this invention provides a farmland pest and disease image recognition inspection robot to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a farmland pest and disease image recognition inspection robot, which solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a farmland pest and disease image recognition inspection robot, comprising: a mobile and navigation platform for autonomous movement in the farmland environment; a wide-area perception and screening system for collecting image data of plants during inspection and preliminarily screening suspected target plants with diseases through image analysis; a multifunctional diagnostic and marking robotic arm, comprising a robotic arm body, the end of which is integrated with an integrated end-effector head, which integrates at least one active physiological probe and an in-situ micro-marking nozzle module, the active physiological probe being used for contact-based depth physiological information measurement of suspected target plants; and a data processing and control unit for coordinating and controlling the robot components and data processing. The processing and control unit first guides a multi-functional diagnostic and labeling robotic arm to probe suspicious target plants based on the screening results of the wide-area perception and screening system, in order to obtain multimodal physiological information inside them. Then, the unit integrates image data with the acquired multimodal physiological information to make a preliminary diagnosis. When the diagnosis result is unclear, i.e., the diagnosis status is "suspected", the unit will control the in-situ micro-labeling nozzle module to physically label the plant. After a preset time interval, the robot will return and acquire multimodal physiological information from the labeled plant a second time. Finally, the data processing and control unit compares and analyzes the data from the two measurements to reveal the changing trend of the plant's physiological state, thereby obtaining a high-confidence final diagnostic decision.

[0007] Through the above technical solutions, image recognition is positioned as an efficient wide-area screening tool, and the focus of diagnosis is placed on subsequent contact-based, multimodal physiological information acquisition and time-series tracking analysis. This establishes a new diagnostic paradigm from superficial screening to internal detection and then to disease course trend analysis, significantly improving the accuracy of diagnosis and early warning capabilities.

[0008] A method for diagnosing farmland pests and diseases using an image recognition and inspection robot for farmland pests and diseases includes the following steps: First, the wide-area perception and screening system is controlled to perform autonomous inspection, collect image data and screen out suspicious target plants. Once a suspicious target is found, the multi-functional diagnostic and marking robotic arm is controlled to approach the plant and use the active physiological probe on it to obtain the multimodal physiological information of the plant. This information may specifically include the maximum photochemical quantum efficiency obtained by the pulse-modulated chlorophyll fluorometer module, the acoustic emission signal sequence obtained by the bio-ultrasonic acoustic sensor module, and the dielectric spectrum data obtained by the high-frequency microwave dielectric spectrometer module. Subsequently, the wide-area perception and screening system integrates image data and acquired multimodal physiological information to construct a comprehensive state vector. The system then uses a preliminary diagnostic model to determine the vector as one of three states: "healthy," "confirmed," or "suspected." For plants determined to be "suspected," the system immediately initiates a time-series tracking process, controls the in-situ micro-labeling nozzle module to label the plant, for example, by spraying liquid containing fluorescent markers, and creates a corresponding time-series tracking record in the database. After a preset time interval, the robot will revisit the area and excite the marker with an ultraviolet light source. It will then use its vision system to capture the fluorescence signal to accurately relocate to the same plant that was previously marked. The robot will repeat the process of acquiring multimodal physiological information for the plant.

[0009] The wide-area perception and screening system will conduct time-series comparative analysis of the multimodal physiological information obtained from the first measurement and the current retest, calculate the time-series physiological change vector that can quantify changes in physiological state, and finally generate the diagnostic decision directly based on the deterioration, stabilization or recovery trend of the plant's physiological state represented by the vector.

[0010] After diagnosis, information on all confirmed plants can be compiled to generate a comprehensive digital agricultural map and management report containing precise geographic coordinates, disease types, and development trends, providing decision support for subsequent precision agriculture management.

[0011] This invention provides a farmland pest and disease image recognition inspection robot. It has the following beneficial effects: 1. This invention is not limited to image recognition of external visual symptoms of plants. Instead, it uses an active physiological probe on a multifunctional diagnostic and labeling robotic arm to perform contact measurements on suspected target plants, obtaining multimodal physiological information such as photochemical quantum efficiency, acoustic emission signals, and dielectric spectra. This surface-to-interior diagnostic approach can capture internal physiological abnormalities in the early stages of disease occurrence when visual symptoms are not yet obvious, thereby achieving accurate diagnosis of latent or early-stage diseases. This is significantly superior to techniques that rely solely on image analysis, improving diagnostic accuracy and early warning capabilities.

[0012] 2. This invention introduces a unique time-series tracking diagnostic mechanism for uncertain cases with an initial diagnostic status of "suspected". By performing in-situ micro-labeling on the target plant and retesting after a preset time interval, the system can obtain dynamic data on the evolution of the plant's physiological state over time. The final diagnostic decision is based on the time-series physiological change vector generated by comparing and analyzing the two measurement data. It is based on the development trend of the disease course rather than a static snapshot, thereby effectively eliminating the interference of temporary factors such as environmental stress, making the diagnostic conclusion more reliable, significantly improving the reliability of diagnostic decisions, and solving the uncertainty problem of single diagnosis.

[0013] 3. This invention employs a phased intelligent diagnostic strategy. First, it utilizes a wide-area perception and screening system to perform large-scale and rapid initial image screening to efficiently identify suspicious targets. Subsequently, it initiates a relatively resource-intensive contact-based deep diagnostic and time-series tracking process only for these few suspicious targets. This hierarchical operation mode avoids unnecessary precision detection of all plants, greatly optimizing the robot's operational efficiency and energy consumption. This allows the robot to achieve both breadth and precision in practical applications in large-scale farmland, realizing high efficiency and intelligence in the diagnostic process, and balancing inspection range and diagnostic depth. Attached Figure Description

[0014] Figure 1 This is a three-dimensional view of the robot structure of the present invention; Figure 2 This is a diagram of the robot framework architecture of the present invention; Figure 3 This is a modular architecture diagram of the integrated end-effector of the present invention; Figure 4 This is a system architecture diagram of the diagnostic inspection robot of the present invention; Figure 5 This is a flowchart of the farmland disease diagnosis method of the present invention.

[0015] Among them, 100 is the mobile and navigation platform; 200 is the wide-area perception and screening system; 300 is the multifunctional diagnostic and labeling robotic arm; 310 is the robotic arm body; 320 is the integrated end effector head; 330 is the active physiological probe; 331 is the pulse-modulated chlorophyll fluorometer module; 332 is the bio-ultrasonic acoustic sensor module; 333 is the high-frequency microwave dielectric spectrometer module; 340 is the in-situ micro-labeling nozzle module; 350 is the near-field visual guidance system; and 400 is the data processing and control unit. Detailed Implementation

[0016] The technical solutions in 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.

[0017] Please see the appendix Figure 1 - Appendix Figure 5 The present invention provides a diagnostic inspection robot configured to autonomously perform early diagnosis and disease tracking tasks of pests and diseases in a farmland environment.

[0018] The diagnostic inspection robot includes: a mobile and navigation platform 100, a wide-area perception and screening system 200, a multi-functional diagnostic and marking robotic arm 300, and a data processing and control unit 400. The data processing and control unit 400 is electrically connected to and controls the mobile and navigation platform 100, the wide-area perception and screening system 200, and the multi-functional diagnostic and marking robotic arm 300.

[0019] The mobile and navigation platform 100 is the mobile carrier of the entire system. The mobile and navigation platform 100 adopts a tracked chassis structure to adapt to unstructured terrain in the field and integrates a real-time dynamic differential positioning module, an inertial measurement unit, and a lidar. The real-time dynamic differential positioning module is used to obtain high-precision absolute geographical coordinates; the inertial measurement unit is used to sense the attitude and angular velocity of the platform; and the lidar is used for 3D mapping of the environment and real-time obstacle detection. Multiple modules work together to support the platform's autonomous navigation and precise docking on a preset path.

[0020] The wide-area sensing and screening system 200 is installed on the mast of the mobile and navigation platform 100 and is used to conduct a non-contact survey of the appearance of crops over a large area. The wide-area sensing and screening system 200 includes a visible light camera and a multispectral camera. The visible light camera is used to acquire high-resolution texture and color information of crops, and the multispectral camera is used to acquire reflectance data of crops in specific bands such as near-infrared and red light.

[0021] The multi-functional diagnostic and marking robotic arm 300 is the core working unit for performing contact-based deep diagnostics and marking. The multi-functional diagnostic and marking robotic arm 300 includes a robotic arm body 310 and an integrated end-effector 320. The robotic arm body 310 is a six-degree-of-freedom collaborative robotic arm. Its high precision and flexible motion capabilities ensure that the end-effector can approach and contact specific tissue parts of the plant with precise posture. The integrated end-effector head 320 compactly integrates an active physiological probe 330, an in-situ micro-label nozzle module 340, and a near-field visual guidance system 350; the active physiological probe 330 is a composite sensor unit, which includes a pulse-modulated chlorophyll fluorometer module 331, a bio-ultrasonic acoustic sensor module 332, and a high-frequency microwave dielectric spectrometer module 333. The pulse-modulated chlorophyll fluorometer module 331 is used to measure the health status of plant photosynthetic systems. The pulse-modulated chlorophyll fluorometer module 331 consists of a measuring optical fiber, an LED light source group, a photodiode detector, and a dark-adapted leaf clip. The LED light source group can emit low-intensity measuring light, continuous photochemical light, and high-intensity saturated pulse light according to programmed instructions. The pulse-modulated chlorophyll fluorometer module 331 evaluates the activity of the photosystem by measuring chlorophyll fluorescence parameters, such as maximum photochemical quantum efficiency. The bio-ultrasonic acoustic sensor module 332 is used to detect structural damage signals inside plant stems. The core component of the bio-ultrasonic acoustic sensor module 332 is a wideband piezoelectric transducer, and it is equipped with a preamplifier and a bandpass filter. During measurement, a flexible clamping structure and an acoustic coupling agent are used to ensure good acoustic contact between the sensor and the surface of the plant stem, so as to effectively capture acoustic emission signals generated by vascular bundle embolism or tissue rupture. The high-frequency microwave dielectric spectrometer module 333 is used to measure the dielectric properties of local plant tissues. It employs an open coaxial probe, which is connected to a vector network analyzer or a dedicated integrated radio frequency circuit. When the probe contacts a plant leaf or fruit, it emits a microwave signal and detects its reflection parameters, thereby calculating the complex dielectric constant of the target tissue within a specific frequency range. ,in: ; The imaginary unit, Where is the dielectric constant. It is the dielectric loss factor; The in-situ micro-labeling nozzle module 340 is used to label plants to be tracked. The in-situ micro-labeling nozzle module 340 is a piezoelectrically driven inkjet printhead. Its ink cartridge contains a biodegradable fluorescent marker that is harmless to plants. The marker is colorless and transparent under natural light, but emits visible fluorescence under ultraviolet light of a specific wavelength. The near-field vision guidance system 350 is an RGB-D camera used to achieve precise positioning of the end-effector. The near-field vision guidance system 350 can simultaneously acquire high-resolution color images and pixel-level depth information of the target area, providing input for the visual servo control algorithm, and guiding the robotic arm body 310 to drive the end-effector 320 to accurately reach the predetermined measurement point of the plant leaf or stem. The data processing and control unit 400 is the central processing and control core of the entire robot system. The data processing and control unit 400 adopts a high-performance embedded industrial computer and integrates the data acquisition interfaces of all sensors, the robotic arm controller interface, and the motor drive interface of the mobile platform. The data processing and control unit 400 is responsible for running navigation and path planning algorithms and performing real-time analysis of wide-area perception data.

[0022] Specifically, unit 400 calculates a suspicion score function for each crop based on data collected by the wide-area sensing and screening system 200. ; ; in, The color feature vector of the target plant; The reference color feature vector for healthy plants; The normalized vegetation index is the target plant. The baseline NDVI value for healthy plants; and , which is the preset weighting coefficient.

[0023] when When the threshold is exceeded, the data processing and control unit 400 identifies the target plant as a suspicious target and controls the subsequent actions of the robot system. At the same time, the data processing and control unit 400 is also responsible for executing the fusion algorithm of multimodal physiological data, the diagnostic decision model and the time series data analysis, and synchronizing data and interacting with the cloud server through the wireless communication module.

[0024] This invention also provides a method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot. This method is executed by a data processing and control unit deployed on the diagnostic inspection robot system and includes the following steps: S100 performs autonomous inspections and uses the wide-area sensing and screening system 200 to collect crop appearance image data. The image data is analyzed to screen out suspicious target plants. First, autonomous inspections and preliminary screening of farmland crops are performed. The goal of this stage is to efficiently identify any individual plant whose appearance deviates from the healthy baseline from a large area of ​​farmland through non-contact sensing methods, and record its precise location to provide target guidance for subsequent contact-based in-depth diagnosis.

[0025] S101, the data processing and control unit 400 plans an inspection path that can traverse the crop rows in the target area based on the inspection area set by the user or the preset farmland digital map, using the navigation module of the mobile and navigation platform 100. The mobile and navigation platform 100 then begins to move autonomously along the path. S102, during the movement, the data processing and control unit 400 activates the wide-area perception and filtering system 200, which synchronously acquires visible light images and multispectral images at a preset frequency. Each acquisition action is strictly bound to the geographical location information and timestamp obtained by the RTK-GPS module of the mobile and navigation platform 100 to form a raw image dataset with geographical and time labels. S103, the data processing and control unit 400 performs real-time preprocessing on the acquired raw image data to eliminate environmental interference and ensure data consistency. First, it performs spatial registration on the visible light image and multispectral image acquired at the same time to ensure that the corresponding pixels in the two images represent crop tissues in the same geographical location. Then, it performs radiometric correction on the multispectral image to convert the dimensionless digital values ​​recorded by the sensor into surface reflectance with clear physical meaning, effectively reducing the impact of changes in light intensity and atmospheric conditions on the authenticity of the data. S104, for the preprocessed image data, the data processing and control unit 400 segments the individual crops or crop plots in the image and extracts visual features to quantify their apparent health status. From the corrected multispectral image, the normalized vegetation index is calculated. At the same time, the registered visible light image is converted from the RGB color space to the HSV color space, and the average hue (H) and saturation (S) values ​​of the canopy region are extracted to form a color feature vector. S105, the data processing and control unit 400 uses a suspicion scoring function. This function assesses the degree of health deviation of each target plant and filters plants based on this score. It integrates the deviation between color features and vegetation index features. ; in, It is the average normalized vegetation index calculated from the canopy area of ​​the target plant, and its calculation formula is: ; in and These represent the reflectance in the near-infrared band and the red band, respectively. It is a color feature vector extracted from the canopy region of the target plant. ; and These are a pre-defined reference color feature vector and a reference NDVI baseline value representing the health status of the crop at the current growth stage. The baseline value can be obtained statistically from selected healthy samples in the field or retrieved from a historical database. The Euclidean distance between the color characteristics of the target plant and the healthy reference characteristics quantifies the degree of color abnormality. This item quantifies the degree of vegetation index decline. This form ensures that a penalty score is only applied when the NDVI value of the target plant falls below the health baseline. and These are preset weighting coefficients used to adjust the relative importance of color anomalies and growth decline in the overall score; S106, the data processing and control unit 400 will calculate the... With a preset filtering threshold If a comparison is made, If the target plant is identified as a "suspicious target plant", the system will record the precise geographic coordinates of the suspicious target plant in the queue of targets to be diagnosed for use in subsequent stages. After completing the screening of a region, the system will continue to inspect along the path, or start to execute the next stage of diagnosis task based on the target location in the queue.

[0026] S200, the driving mobile and navigation platform 100 approaches the suspected target plant and controls the multi-functional diagnostic and labeling robotic arm 300 to perform contact-based multimodal physiological information acquisition on predetermined parts of the plant. In this stage, the working unit on the robot system is used to perform close-up, contact-based deep physiological state detection on the screened suspected target plants to obtain internal health indicators that cannot be reflected by the external visual system, including the following steps: S201, the data processing and control unit 400 extracts the geographic coordinates of the suspicious target plants from the target queue to be diagnosed. Based on the coordinates, the data processing and control unit 400 instructs the movement and navigation platform 100 to move from the current position to a preset work point next to the target plant. The position of the work point is optimized to ensure that the multi-functional diagnostic and marking robotic arm 300 has sufficient working space while avoiding disturbance to the surrounding crops. S202, after the mobile and navigation platform 100 is stably docked, the data processing and control unit 400 activates the near-field visual guidance system 350 at the end of the multi-functional diagnostic and marking robotic arm 300 to collect three-dimensional point cloud data and high-resolution color images of the target plant and construct a local three-dimensional digital model of the plant. S203, the data processing and control unit 400 analyzes the three-dimensional digital model to automatically identify and locate specific parts suitable for physiological detection. Through point cloud segmentation and morphological analysis algorithms, it identifies the main stem of the plant and the leaves that meet the measurement conditions from the model. The system calculates the precise three-dimensional coordinates and surface normal vectors of these target parts in the robot coordinate system. S204, based on the pose information of the target part, the data processing and control unit 400 plans a collision-free motion trajectory for the robotic arm body 310 from the current position to the target contact point. Subsequently, the data processing and control unit 400 drives the robotic arm body 310 so that the integrated end-effector 320 at its end moves precisely along the planned trajectory until the corresponding sensor module of the active physiological probe 330 reaches the test point of the target part. This process adopts a visual servo control strategy, that is, it uses the real-time feedback of the near-field visual guidance system 350 to dynamically correct the movement of the robotic arm to cope with the slight displacement of the plant caused by environmental factors such as wind, and ensure the accuracy of positioning and the flexibility of contact. S205, after the active physiological probe 330 is precisely positioned, the data processing and control unit 400 triggers each sensor module to acquire data sequentially according to a preset programmed timing sequence. First, the dark adaptation leaf clamp of the pulse-modulated chlorophyll fluorometer module 331 closes, subjecting the selected leaf area to dark processing for a period of time. After dark adaptation, the module automatically executes the measurement sequence to obtain the initial fluorescence yield. and maximum fluorescence yield Based on this, the key physiological parameter reflecting the activity of the photosynthetic system, namely the maximum photochemical quantum efficiency, was calculated. ; in, The maximum fluorescence yield under saturated pulsed light irradiation; The initial fluorescence yield after dark adaptation.

[0027] S206, After completing the fluorescence measurement, the robotic arm automatically adjusts its posture to ensure stable contact between the probe of the bio-ultrasonic acoustic sensor module 332 and the surface of the plant's main stem. The data processing and control unit 400 then collects and records the time sequence of acoustic emission signals from inside the stem within a preset time window. ,in These are discrete-time sampling points; S207, Subsequently, the robotic arm body 310 moves again, gently attaching the probe of the high-frequency microwave dielectric spectrometer module 333 to the surface of the target leaf or fruit. The network analyzer or radio frequency circuit built into the module performs frequency sweep measurement within a preset frequency range, acquiring the frequency points at each frequency. The corresponding complex reflection coefficient is used to calculate the complex permittivity. This forms a dielectric spectrum data; S208, after all physiological information acquisition tasks are completed, the robotic arm body 310 and the data processing and control unit 400 will process all the acquired physiological data (including parameters) into a single file. Acoustic signal sequence Data such as dielectric spectrum are bound to the target plant's identity, geographic coordinates, and current timestamp to form a complete multimodal physiological information dataset, preparing for the next stage of data fusion and diagnostic analysis.

[0028] S300 integrates surface image data and multimodal physiological information to generate a preliminary diagnostic status. When the preliminary diagnostic status is "suspected," it triggers and executes in-situ micro-labeling of the plant. After acquiring multi-dimensional information on the surface and internal physiology of the target plant, it enters the data fusion, preliminary diagnosis, and classification labeling stage. By comprehensively analyzing all collected data, it makes a preliminary judgment on the plant's health status, including an assessment of uncertainty, and performs special labeling on individuals with unclear status, providing a basis for subsequent time-series tracking. This includes the following steps: S301, the data processing and control unit 400 first extracts features from the acquired raw physiological data, transforming it into standardized numerical features that can be used as model input. This includes the acoustic emission signal sequence acquired by the bio-ultrasound acoustic sensor module 332. Calculate its signal energy Ringing count , constitute acoustic feature vectors Signal energy The calculation formula is: ; in, This represents the total number of sampling points in the signal sequence.

[0029] Ringing count The calculation formula is: ; in, This is an indicator function; its value is 1 when the condition inside the parentheses is true, and 0 otherwise. This is the preset noise amplitude threshold.

[0030] For the dielectric spectrum data acquired by the high-frequency microwave dielectric spectrometer module 333, the dielectric constant at a specific sensitive frequency point is extracted. and dielectric loss factor The values ​​of these constitute the dielectric eigenvector: ; S302, the data processing and control unit 400 fuses the visual features acquired in stage S100 with the physiological features extracted in this stage, providing the current target plant at the measurement time. Construct a comprehensive integrated state vector The vector The structure is as follows: ; in, This is the visual feature vector of the plant, and its specific form is: ; The maximum photochemical quantum efficiency measured; For acoustic feature vectors; This is the dielectric eigenvector.

[0031] S303, based on the constructed comprehensive state vector The data processing and control unit 400 uses a preset preliminary diagnostic model. The plant's health status is categorized, and a preliminary diagnostic status is output. The model The specific implementation can be based on a multi-threshold decision logic in a multi-dimensional feature space, which divides the feature space into three regions, corresponding to the three states of "healthy", "confirmed" and "suspected". A state is considered "healthy" if and only if its state vector is... All feature components fall within their respective preset health threshold ranges, meaning that the plant's apparent traits and internal physiological indicators are normal. Determined as "confirmed": when the state vector Visual features exhibit typical signs (e.g., extremely low NDVI values ​​or severe color abnormalities), or multiple physiological features show significant, synergistically oriented abnormalities (e.g., chlorophyll fluorescence efficiency). While the acoustic signal energy decreased significantly, The abnormally high levels all point to vascular bundle diseases that obstruct water transport. The condition is classified as "suspected" in all other cases besides the two mentioned above. This state covers a variety of uncertain scenarios, such as a plant that appears normal but has an abnormal internal physiological indicator (in the latent period of the disease), or a plant that appears abnormal but has normal physiological indicators (due to nutritional imbalance or physical damage), or all indicators that are in a borderline state between health and disease. S304, Data processing and control unit 400 based on the output preliminary diagnostic status Implement classified treatment, if For a result of "healthy" or "diagnosed", the system simply records and archives the diagnosis along with all measurement data, and prepares to move on to the next target; S305, if the initial diagnosis status If the condition is "suspected", the data processing and control unit 400 activates the in-situ micro-labeling program and controls the multi-functional diagnostic and labeling robotic arm 300 to align the in-situ micro-labeling nozzle module 340 at its end with a stable and easily identifiable part of the plant (such as the base of the main stem or the petiole of a mature leaf). Then, the in-situ micro-labeling nozzle module 340 sprays a small drop of liquid containing a fluorescent marker in a non-contact manner, forming a micro-marker on the plant that is invisible to the naked eye but clearly identifiable under ultraviolet light. S306, while performing the tagging, the data processing and control unit 400 creates a unique time-series tracking record for the plant in its internal database. This record includes at least: a unique tag identifier (ID), the plant's precise geographic coordinates, and a timestamp of the tagging operation. And the complete integrated state vector that triggered this marking. This step completes the physical tagging and digital documentation of individuals with uncertain outcomes, laying the foundation for subsequent disease tracking and final diagnosis.

[0032] S400, after a preset time interval, identifies and repositions the marked plant, and repeats the contact-based multimodal physiological information acquisition step to obtain a second set of multimodal physiological information; this stage aims to reveal the development trend of the disease by comparing and analyzing the physiological state of the same plant at different time points, thereby eliminating the uncertainty of a single measurement and making a high-confidence final diagnosis, including the following steps: S401, after a preset time interval, when planning a new inspection task, the data processing and control unit 400 will take all target points in the database containing valid time-series tracking records as high-priority inspection nodes, and the mobile and navigation platform 100 will be preferentially guided to the vicinity of the geographical coordinates of these marked plants. S402, after reaching the target area, the data processing and control unit 400 switches to the target relocation mode. The data processing and control unit 400 activates the ultraviolet light source installed on the robot system to illuminate the marked area of ​​the target plant. At the same time, the camera of the wide-area perception and screening system 200 or the near-field vision guidance system 350 switches to the high-sensitivity mode to capture the fluorescence of a specific wavelength emitted by the in-situ micro-markers. The system can accurately identify and re-lock the previously marked plant individuals from the environment by matching the morphology and code of the fluorescent markers through the image recognition algorithm. S403, after successfully relocating the target plant, the data processing and control unit 400 repeats the complete steps of S202 to S208. Through visual servo control, it drives the multifunctional diagnostic and labeling robotic arm 300 to perform pulse-modulated chlorophyll fluorescence, bio-ultrasound acoustics, and high-frequency microwave dielectric spectroscopy measurements on the same or adjacent parts of the plant, thereby obtaining the data at the current moment. The second set of multimodal physiological information was used to construct a new integrated state vector. ; S404, The data processing and control unit 400 retrieves the information about the plant at the time of its first marking from the database. ) Stored state vector and the newly acquired state vector To quantify changes in physiological state over time, a time-series comparative analysis was conducted, and the system calculated a time-series physiological change vector. This vector is composed of the differences between the core physiological characteristics measured in the two measurements: ; in, The change representing the maximum photochemical quantum efficiency; This represents the change in the energy of the acoustic emission signal. It represents the change in dielectric constant at a specific frequency.

[0033] S500, based on the initially acquired physiological information and the second set of multimodal physiological information, performs time-series change analysis to determine the disease progression trend and generate a final diagnostic decision, including the following steps: S501, based on the calculated temporal physiological change vector The data processing and control unit 400 uses the final diagnostic decision model. It makes a final judgment on the health status of the plant and outputs the final diagnostic status. The decision-making logic of this model is specifically implemented as follows: calculating the deterioration trend score. The rating is a vector of change. With the preset deterioration weight vector dot product: ; in, It is a three-dimensional weighted vector, and the signs of its components are set according to the correlation between physiological indicators and health status. For example, decreased photosynthetic efficiency, increased acoustic emission energy, and decreased dielectric constant usually indicate deteriorating health. The corresponding components are negative, positive, and negative, respectively; S502, calculate The final diagnostic status is output after comparing it with a preset decision threshold. : like (in If the threshold is positive, it indicates that the plant's physiological state is showing a clear and continuous deterioration trend. The system updates the plant's diagnostic status to "Confirmed - Developing Disease". like (in (If the threshold value is negative), it indicates that the plant's physiological indicators are trending toward restoring to a healthy baseline, and the system can update its status to "restored to health" or "low risk - continued observation". like If the plant is in a relatively stable condition with no obvious changes, then the "suspected - continued observation" status will be maintained. S503, after completing the final diagnosis, the data processing and control unit 400 will send the final diagnosis results and the complete timing data chain (including...) and The system updates the database record of the plant with the disease course analysis results. At the same time, the system summarizes the information of all diagnosed plants and generates a comprehensive digital agricultural map and management report that includes accurate geographical location, disease diagnosis type, and development trend assessment, providing users with decision-making basis for implementing precise pesticide application, diseased plant removal and other agricultural operations.

[0034] The method of this invention clearly positions image recognition technology as an efficient target screening entry point, and deeply couples and coordinates it with robotic contact physiological sensing, in-situ labeling and disease tracking technologies, thereby upgrading the conventional farmland pest and disease image recognition inspection robot into an intelligent diagnostic platform with three diagnostic dimensions: early, internal and development trend.

Claims

1. A farmland pest and disease image recognition inspection robot, characterized in that, include: Mobile and navigation platform (100); A wide-area perception and screening system (200) installed on the mobile and navigation platform (100) is used to collect image data of plants during the inspection process and analyze the image data to screen out suspicious target plants. A multifunctional diagnostic and marking robotic arm (300) is mounted on the mobile and navigation platform (100). The multifunctional diagnostic and marking robotic arm (300) includes a robotic arm body (310) with an integrated end-effector head (320) at its end. The integrated end-effector head (320) integrates at least one active physiological probe (330) for contact measurement of suspected target plants, as well as an in-situ micro-marking nozzle module (340) and a near-field visual guidance system (350). The data processing and control unit (400) is electrically connected to the mobile and navigation platform (100), the wide-area perception and screening system (200), and the multi-functional diagnostic and marking robotic arm (300), respectively.

2. The farmland pest and disease image recognition inspection robot according to claim 1, characterized in that, The active physiological probe (330) includes: A pulse-modulated chlorophyll fluorometer module (331) is used to measure the health status of plant photosynthetic systems; A bio-ultrasonic acoustic sensor module (332) is used to detect structural damage signals inside plant stems; The high-frequency microwave dielectric spectrometer module (333) is used to measure the dielectric properties of local tissues in plants.

3. A method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot, applicable to any one of claims 1-2, characterized in that, Includes the following steps: (a) Control the wide-area sensing and screening system (200) to acquire image data of the target plants and screen out suspicious target plants; (b) Control the multifunctional diagnostic and labeling robotic arm (300) to approach the suspected target plant and acquire its multimodal physiological information using an active physiological probe (330); (c) The image data obtained in step (a) and the multimodal physiological information obtained in step (b) are fused to make a preliminary diagnosis of the suspected target plant; when the preliminary diagnosis status is "suspected", the in-situ micro-labeling nozzle module (340) is controlled to perform in-situ micro-labeling on the plant. (d) After a preset time interval, relocate to the marked plant and repeat step (b) to obtain the multimodal physiological information at the current moment; (e) Perform time-series comparative analysis on the two acquisitions of multimodal physiological information to generate a final diagnostic decision.

4. The method for diagnosing farmland diseases using an image recognition and inspection robot for farmland pests and diseases according to claim 3, characterized in that, In step (c), the preliminary diagnosis is based on image data and multimodal physiological information to construct a comprehensive state vector for the current target plant at the measurement time. Through a preset preliminary diagnosis model, the comprehensive state vector is determined to be one of three states: "healthy", "confirmed", or "suspected".

5. The method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot according to claim 3, characterized in that, When performing in-situ micro-labeling in step (c), the in-situ micro-labeling nozzle module (340) is controlled to spray liquid containing fluorescent markers onto the plant; at the same time, a time-series tracking record is created for the plant in the database, which includes the plant's unique identification code, geographic coordinates, labeling timestamp, and a comprehensive state vector that triggered the labeling.

6. The method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot according to claim 3, characterized in that, Obtaining multimodal physiological information in step (b) includes: The maximum photochemical quantum efficiency was obtained by using a pulse-modulated chlorophyll fluorometer module (331); Acoustic emission signal sequences are acquired using a bio-ultrasonic acoustic sensor module (332); Dielectric spectrum data are acquired using a high-frequency microwave dielectric spectrometer module (333).

7. The method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot according to claim 3, characterized in that, The time series comparison analysis in step (e) includes: Based on the core physiological characteristics obtained at the time of the first measurement and the time of the re-measurement, the temporal physiological change vector is calculated; the final diagnostic decision is generated based on the trend of plant physiological state change represented by the temporal physiological change vector.

8. The method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot according to claim 7, characterized in that, The generation of the final diagnostic decision includes the following steps: The deterioration trend score is obtained by performing a dot product operation between the time-series physiological change vector and the preset deterioration weight vector. The deterioration trend score is compared with a preset decision threshold to determine the disease development trend of the plant and output the final diagnostic decision.

9. The method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot according to claim 3, characterized in that, Relocation to the tagged plant in step (d) includes the following steps: The ultraviolet light source on the control robot system irradiates the target area; The fluorescence emitted by fluorescent markers is captured using a visual system; Image recognition algorithms are used to match the morphology and coding of fluorescence to locate the marked plants.

10. The method for diagnosing farmland diseases using a farmland pest and disease image recognition inspection robot according to claim 3, characterized in that, After generating the final diagnostic decision, the precise geographical location, disease diagnosis type, and development trend assessment of all diagnosed plants are summarized to generate a comprehensive digital agricultural map and management report.

Images