An adaptive monitoring system for crop growth status

Abstract

This invention relates to the field of agricultural monitoring technology, and more particularly to an adaptive monitoring system for crop growth status, comprising: a spectral acquisition unit with a wide-angle field-of-view probe and a narrow-angle field-of-view probe, wherein the observation area of ​​the narrow-angle probe is located within the observation area of ​​the wide-angle probe; a height adjustment mechanism for adjusting the height of the spectral acquisition unit; a local processor for generating a difference feature vector based on the spectral differences between the two fields of view, determining the canopy coverage, and controlling the height adjustment mechanism to perform height adjustment according to the changing trend of the canopy coverage; a communicator and a cloud server, wherein the cloud server receives the difference feature vector and outputs the growth status prediction result, which is fed back to the local processor to update the adjustment strategy. This invention utilizes the spectral differences between the two field-of-view probes to achieve automatic and accurate determination of crop growth status, unaffected by changes in light intensity, suitable for long-term field monitoring. Furthermore, through adaptive adjustment of the observation height and collaborative prediction with the cloud, it achieves dynamic and accurate monitoring of the crop growth process.

Description

Technical Field

[0001] This invention relates to the field of agricultural monitoring technology, and in particular to an adaptive monitoring system for crop growth status. Background Technology

[0002] Accurate monitoring of crop growth status is crucial for guiding agricultural production activities such as irrigation, fertilization, and harvesting. Traditionally, the assessment of crop growth status has relied primarily on manual visual observation and experience. However, due to the complex variations in morphological characteristics among different crops, varieties, and growing environments, manual assessment is not only inefficient and highly subjective, but also difficult to achieve continuous, objective, and automated monitoring.

[0003] Currently, spectral monitoring methods have been applied in agriculture. Existing technologies typically employ fixed-height spectral sensors to collect crop reflectance spectral data from a single field of view, assessing crop growth parameters by extracting spectral features or calculating vegetation indices. For example, some schemes use canopy reflectance to calculate the normalized vegetation index to estimate leaf area index or biomass. However, because crop height and canopy structure change continuously during growth, fixed-height sensors are prone to misalignment between the observation field of view and the target organ, resulting in spectral data that cannot consistently reflect the growth information of key crop parts. Furthermore, traditional schemes often only perform simple threshold judgments or static model analyses locally, lacking the ability to mine trends from multi-time period monitoring data and perform cloud-based collaborative optimization. Existing technologies struggle to achieve dynamic adaptive tracking and accurate prediction of crop growth status.

[0004] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an adaptive monitoring system for crop growth status.

[0006] This invention provides an adaptive monitoring system for crop growth status, the technical solution of which is as follows:

[0007] The crop growth status adaptive monitoring system includes:

[0008] A spectral acquisition device is used to acquire the reflected light signal of the target plant, including a wide-angle field-of-view probe and a narrow-angle field-of-view probe, wherein the observation area of ​​the narrow-angle field-of-view probe is located within the observation area of ​​the wide-angle field-of-view probe;

[0009] A height adjustment mechanism is used to adjust the height of the spectral acquisition device relative to the target plant;

[0010] The local processor is used to generate a difference feature vector characterizing the spectral difference between the two field of view based on the reflected light signal, determine the canopy coverage of the target plant, and determine an adaptive observation and adjustment strategy based on the changing trend of the canopy coverage over multiple acquisition cycles and control the lift regulator to perform height adjustment operations.

[0011] The communicator is used to establish a communication connection between the local processor and the cloud server, upload the differential feature vector to the cloud server, and receive the growth status prediction results sent by the cloud server.

[0012] The cloud server communicates with the local processor via a communicator to receive the differential feature vector, output the growth status prediction result of the target plant based on the pre-trained classification model, and feed the growth status prediction result back to the local processor to update the adaptive observation and adjustment strategy.

[0013] The beneficial effects of the crop growth status adaptive monitoring system of the present invention are as follows:

[0014] The system of this invention utilizes the spectral differences of dual-field-of-view probes to achieve automatic and accurate determination of crop growth status, unaffected by changes in light intensity, and suitable for long-term field monitoring. At the same time, through adaptive adjustment of observation height and collaborative prediction with the cloud, it achieves dynamic and accurate monitoring of the crop growth process.

[0015] Based on the above scheme, the crop growth status adaptive monitoring system of the present invention can be further improved as follows.

[0016] In one alternative approach, the observation area of ​​the narrow-angle field-of-view probe is pointed towards the top region of the target plant to acquire top feature reflection information of the target plant, which includes spectral signals of the top leaves, flowers, or fruits of the target plant; the observation area of ​​the wide-angle field-of-view probe covers the entire canopy of the target plant and the surrounding background ground to acquire overall reflection information of the target plant and its surrounding area, which includes mixed spectral signals of the canopy, stems, and background ground of the target plant.

[0017] Among the above-mentioned optional methods, by clearly defining the differences in the observed objects of the dual-field-of-view probes, focusing on top characteristic organs (such as flowers and fruits) with a narrow angle, and covering the whole and background with a wide angle, the spectral differences between the two can accurately reflect the changes in organ composition at different growth stages of crops, providing highly sensitive characteristic data for determining the growth period.

[0018] In one alternative embodiment, the wide-angle field-of-view probe includes a cosine receiver to achieve a 180-degree field of view, and the narrow-angle field-of-view probe includes a focusing lens to achieve a 25-degree field of view; the ratio between the mounting height of the spectral acquisition unit relative to the ground and the observation diameter of the narrow-angle field-of-view probe is 2:1, and the ratio between the mounting height of the spectral acquisition unit relative to the ground and the effective observation diameter of the wide-angle field-of-view probe is 1:3.

[0019] In the above-mentioned optional methods, by limiting the specific field of view angle parameters and height-diameter ratio, it is ensured that the narrow field of view can accurately focus on the core area of ​​the crop, while the wide field of view can cover sufficient background information, thus ensuring the geometric consistency of spectral data acquisition and improving the generalization ability of the model.

[0020] In one alternative embodiment, the height adjuster includes a drive motor, a telescopic rod, and a distance sensor. One end of the telescopic rod is fixedly connected to the spectral collector. The distance sensor is used to measure the real-time distance between the spectral collector and the target plant to provide a feedback control reference for height adjustment. The drive motor is used to drive the telescopic rod to extend and retract to adjust the height of the spectral collector relative to the target plant.

[0021] Among the above-mentioned optional methods, a specific mechanical structure for the lifting adjuster is provided. Height adjustment is achieved by driving the telescopic rod with a motor. The structure is simple and reliable, and easy to deploy and maintain in the field environment.

[0022] In one alternative approach, the local processor is specifically used to perform the height adjustment operation, including:

[0023] The canopy coverage of the target plant is calculated based on the reflected light signal corresponding to the narrow-angle field-of-view probe.

[0024] The current height of the target plant is estimated based on the difference feature vector, or the current distance between the spectral acquisition device and the target plant is determined based on the real-time distance obtained by the ranging sensor, and the target observation height is calculated based on the current height or the current distance and the field of view angle of the narrow-angle field of view probe; the target observation height is the height required for the observation area of ​​the narrow-angle field of view probe to cover the core area of ​​the canopy of the target plant.

[0025] If the canopy coverage is lower than a preset coverage threshold, a descent control command is generated. The descent control command is used to control the drive motor to drive the telescopic rod to shorten, so as to lower the spectral acquisition device to the target observation height.

[0026] If the canopy coverage is higher than a preset coverage threshold, an ascent control command is generated. The ascent control command is used to control the drive motor to extend the telescopic rod to raise the spectral acquisition device to the target observation height.

[0027] If the canopy coverage continues to decrease over multiple preset acquisition cycles, and the canopy coverage has not yet fallen below the preset coverage threshold, a predictive descent command is generated to lower the spectral acquisition device to the target observation height in advance.

[0028] A minimum observation height threshold is set to control the height of the spectral acquisition device to be no lower than the minimum observation height threshold, so as to prevent the spectral acquisition device from physically colliding with the target plant and to ensure the observation integrity of the wide-angle field-of-view probe.

[0029] Among the aforementioned optional methods, a control logic combining canopy coverage threshold determination and trend prediction enables a dynamic response to changes in crop growth height. In particular, the introduction of predictive descent commands allows for adjustments before the coverage falls below the threshold, avoiding monitoring blind spots; the setting of a minimum observation height threshold ensures equipment safety and the integrity of the wide-angle field of view, greatly enhancing the system's intelligence level.

[0030] In one alternative approach, the local processor is specifically used for:

[0031] Based on the reflected light signal, obtain the first reflectivity data corresponding to the wide-angle field-of-view probe and the second reflectivity data corresponding to the narrow-angle field-of-view probe;

[0032] The first reflectance data and the second reflectance data are preprocessed by Savitzky-Golay filtering to obtain the first preprocessed data and the second preprocessed data.

[0033] Based on the first preprocessed data and the second preprocessed data, at least one of the reflectance difference, reflectance ratio, red edge feature difference, and vegetation index difference within the preset band range is calculated to form the difference feature vector.

[0034] The calculation method for the red-edge feature difference value includes: performing linear fitting on the first preprocessed data and the second preprocessed data in the red-edge band region from 690nm to 750nm respectively to obtain the first red-edge feature parameter and the second red-edge feature parameter, and determining the red-edge feature difference value based on the difference between the first red-edge feature parameter and the second red-edge feature parameter; the red-edge feature parameter includes at least one of red-edge position, red-edge slope and red-edge fit goodness;

[0035] The method for calculating the vegetation index difference value includes: calculating the normalized vegetation index based on the first preprocessed data and the second preprocessed data respectively to obtain the first vegetation index and the second vegetation index, and determining the vegetation index difference value based on the difference between the first vegetation index and the second vegetation index.

[0036] In the aforementioned optional methods, a feature vector representing the differences between the two fields of view is constructed from the original spectral data through specific preprocessing and feature extraction procedures. The introduction of differences in red-edge position and vegetation index effectively captures the physiological changes of crops at different growth stages, improving the effectiveness of feature data in determining the growth period.

[0037] In an alternative approach, the differential feature vector further includes moisture-sensitive differential features and nutrient-sensitive differential features;

[0038] The water-sensitive differential feature is used to characterize the difference in water content between the canopy and the understory of the target plant in order to assess the water stress status of the target plant.

[0039] The nutrient sensitivity difference feature is used to characterize the difference in chlorophyll or nitrogen nutrition between the canopy and the understory of the target plant in order to assess the nutrient deficiency status of the target plant.

[0040] Among the above-mentioned optional methods, based on the determination of the growth period, the dimension of the feature vector is further expanded by introducing water and nutrient sensitive features, so that the system can not only determine the growth period, but also simultaneously monitor the physiological stress status of crops, providing multi-dimensional data support for precision agricultural management.

[0041] In one alternative approach, the cloud server is specifically used for:

[0042] The differential feature vector is input into the pre-trained classification model or empirical model;

[0043] The growth stage label of the target plant, output by the classification model or empirical model, is obtained as the growth state prediction result.

[0044] Among the above-mentioned optional methods, deploying a high-performance classification model on a cloud server and utilizing cloud computing power to process complex feature vector mapping relationships ensures high accuracy and real-time performance in determining the growth period.

[0045] In an optional manner, the growth status prediction result further includes the water stress status and nutrient deficiency status of the target plant; the water stress status is determined based on the water-sensitive differential features in the differential feature vector, and the nutrient deficiency status is determined based on the nutrient-sensitive differential features in the differential feature vector.

[0046] The cloud server is also specifically used for: generating early warning information when the water stress state or the nutrient deficiency state is abnormal, and generating agricultural decision-making suggestions based on the water stress state and the nutrient deficiency state; and sending the early warning information and the agricultural decision-making suggestions to the local processor or user terminal through a communicator.

[0047] The agricultural decision-making recommendations include at least one of irrigation timing recommendations, fertilizer application recommendations, and harvesting time recommendations; the irrigation timing recommendations determine the appropriate irrigation time window based on the severity of the water stress state, and the fertilizer application recommendations determine the recommended fertilizer type and amount based on the type and severity of the nutrient deficiency state.

[0048] Among the above-mentioned optional methods, a leap from "monitoring" to "decision-making" has been achieved. Based on the state of water and nutrient stress, agricultural suggestions are automatically generated to directly guide agricultural production, significantly improving the practical value and economic benefits of the system.

[0049] In one alternative approach, cloud servers are also used for:

[0050] Based on the changing trend of the differential feature vector over multiple consecutive acquisition cycles, the growth state transition node of the target plant is identified; the growth state transition node is the moment when the change amplitude of the differential feature vector between adjacent acquisition cycles exceeds a preset change threshold.

[0051] When the growth state transition node is identified, a dynamic observation strategy adjustment instruction is generated; the dynamic observation strategy adjustment instruction includes at least one of the following adjustments: increasing the data acquisition frequency of the spectral acquisition device to capture spectral details during the rapid change period of growth state; adjusting the height regulator to make the height of the spectral acquisition device reach the optimal observation height corresponding to the growth state transition node to ensure that the narrow-angle field-of-view probe focuses on the key organs of the current growth stage; and switching the feature extraction focus of the local processor to match the sensitive band and vegetation index type corresponding to the growth state transition node.

[0052] The local processor sends the dynamic observation strategy adjustment command to the local processor via the communicator, and the local processor updates the adaptive observation adjustment strategy based on the dynamic observation strategy adjustment command.

[0053] Among the aforementioned optional methods, by identifying growth state transition nodes and dynamically adjusting the observation strategy, precise capture of key crop growth stages is achieved. During periods of rapid growth change, the system automatically increases the acquisition frequency, adjusts the observation height, and focuses on feature extraction, ensuring the integrity and accuracy of key data and further enhancing the system's adaptability to complex growth processes.

[0054] This invention provides an adaptive monitoring system for crop growth status. Through the design of wide-angle and narrow-angle dual-field-of-view probes, and utilizing the inclusion relationship between their observation areas, it can simultaneously acquire spectral difference information between the entire crop and its top, thereby accurately characterizing the crop's vertical structural features and solving the problem of insufficient information dimension in single-view monitoring. Through the coordinated control of the height adjustment mechanism and the local processor, the system can dynamically adjust the observation height based on the changing trend of canopy coverage, achieving an adaptive response to changes in crop growth height and avoiding field drift and data distortion in long-term field monitoring. Furthermore, by constructing a differential feature vector containing water and nutrient-sensitive characteristics and combining it with a cloud-based classification model, the system can not only accurately determine the crop growth stage but also simultaneously monitor physiological stress states and generate agricultural decision-making suggestions. This achieves a closed-loop process from data acquisition and intelligent analysis to decision support, significantly improving the intelligence level and application value of agricultural monitoring.

[0055] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0056] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0057] Figure 1 This is a schematic diagram of an embodiment of the crop growth status adaptive monitoring system of the present invention;

[0058] Figure 2 This is a schematic diagram of the observation area of ​​the narrow-angle field-of-view probe of the present invention;

[0059] Figure 3 This is a schematic diagram of the observation area of ​​the wide-angle field-of-view probe of the present invention;

[0060] Figure 4 This is a comparison diagram of the dual-field-angle reflectance spectrum curves of summer maize at the milk stage according to the present invention;

[0061] Figure 5 This is a comparison of the dual-field-angle reflectance spectral curves of summer maize at maturity according to the present invention. Detailed Implementation

[0062] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0063] Example 1:

[0064] Figure 1 A schematic diagram of an embodiment of the adaptive monitoring system for crop growth status provided by the present invention is shown, as follows: Figure 1 As shown, the system includes:

[0065] The system includes a spectrum acquisition unit 1, a height adjustment unit 2, a local processor 3, a communicator 4, and a cloud server 5. The spectrum acquisition unit 1 is equipped with a wide-angle and a narrow-angle field-of-view probe. The height adjustment unit 2 consists of a drive motor and a telescopic rod. The spectrum acquisition unit 1, height adjustment unit 2, local processor 3, and communicator 4 together form the local terminal device. The spectrum acquisition unit 1 transmits the collected reflected light signals to the local processor 3. The local processor 3 sends height control commands to the height adjustment unit 2. The height adjustment unit 2 and the spectrum acquisition unit 1 are mechanically connected to adjust the physical position of the spectrum acquisition unit. The local processor 3 and the communicator 4 perform bidirectional data transmission. The communicator 4 then wirelessly connects with the cloud server 5 to achieve bidirectional data interaction, completing information exchange between the local terminal and the cloud. Specifically:

[0066] A spectral acquisition device is used to acquire reflected light signals from a target plant, including a wide-angle field-of-view probe and a narrow-angle field-of-view probe, wherein the observation area of ​​the narrow-angle field-of-view probe is located within the observation area of ​​the wide-angle field-of-view probe.

[0067] Specifically, the spectral acquisition unit is typically deployed above the target plant and fixed in place by a support. A wide-angle field-of-view probe is used to acquire a broad range of overall spectral information, while a narrow-angle field-of-view probe is used to focus on key organs at the top of the plant (such as leaves, flowers, and fruits). The observation areas of the two probes are overlapping, allowing the system to simultaneously acquire spectral data at different scales, providing a physical basis for subsequent extraction of features characterizing differences in the crop's vertical structure.

[0068] It should be understood that although the description uses the downward viewing posture of the probe as an example, in other embodiments, the probe angle can be finely adjusted according to the crop height and planting density, as long as the narrow field of view is within the wide field of view.

[0069] A height adjustment mechanism is used to adjust the height of the spectral acquisition device relative to the target plant.

[0070] Specifically, the height adjustment mechanism is mechanically connected to the spectral acquisition unit, allowing it to move vertically according to control commands. During crop growth, plant height constantly changes. If the probe height is fixed, the field of view coverage will drift, and narrow-angle probes may deviate from the core canopy area. The height adjustment mechanism enables the system to dynamically adjust the observation height, maintaining the optimal observation angle and solving the data distortion problem caused by crop growth in long-term field monitoring.

[0071] The local processor is used to generate a difference feature vector representing the spectral difference between the two field of view based on the reflected light signal, determine the canopy coverage of the target plant, and determine an adaptive observation and adjustment strategy based on the changing trend of the canopy coverage over multiple acquisition cycles, and control the lift regulator to perform height adjustment operations.

[0072] Specifically, the local processor is the core of the system's local control, with pre-installed data processing algorithms. It first preprocesses the raw reflected light signal acquired by the spectral acquisition unit (e.g., radiometric correction, noise reduction), then calculates the differences between wide-angle and narrow-angle field-of-view data (e.g., reflectance difference, vegetation index difference), constructing a difference feature vector. Simultaneously, the local processor uses the narrow-angle field-of-view data to calculate canopy coverage, serving as a direct basis for height adjustment. When an abnormal trend in coverage change is detected (e.g., a continuous decline), the local processor generates a corresponding adjustment strategy (e.g., a descent command), driving the height regulator to operate. This predictive adjustment based on time-series trends is more forward-looking than a single threshold judgment, effectively avoiding monitoring blind spots.

[0073] The communicator is used to establish a communication connection between the local processor and the cloud server, upload the differential feature vector to the cloud server, and receive the growth status prediction results sent by the cloud server.

[0074] Specifically, the communicator can be a 4G / 5G module, a LoRa module, or a Wi-Fi module, responsible for data interaction between the local machine and the cloud. It uploads the differential feature vectors generated by the local processor in real time or periodically, while receiving decision feedback from the cloud. This design offloads complex model computation tasks to the cloud, reducing local hardware costs, while leveraging the powerful computing capabilities of the cloud to achieve high-precision predictions.

[0075] The cloud server communicates with the local processor via a communicator to receive the differential feature vector, output the growth status prediction result of the target plant based on the pre-trained classification model, and feed the growth status prediction result back to the local processor to update the adaptive observation and adjustment strategy.

[0076] Specifically, the cloud server deploys machine learning models (such as support vector machines and random forests) trained on a large amount of historical data. Upon receiving differential feature vectors, the model classifies and identifies them, outputting the crop's current growth stage (such as seedling stage, jointing stage, tasseling stage, etc.). This prediction result is not only used for display but is also sent back to the local processor as a feedback signal. The local processor updates its observation strategy based on the prediction result, for example, increasing the sampling frequency during rapid growth periods or adjusting the observation height at specific growth stages.

[0077] Thus, the system constructs a complete closed loop of "sensing (spectral acquisition) - adjustment (lifting and lowering control) - decision-making (cloud prediction) - feedback (strategy update)," realizing adaptive and high-precision monitoring of crop growth status.

[0078] It should be noted that the aforementioned communication device working in conjunction with the cloud server is a preferred implementation of this system. In another usage mode, the communication device can also be replaced with a data interface (such as RS485, USB, or Ethernet interface) to directly connect to the local data acquisition unit. In this mode, after the local processor completes the differential feature vector extraction and canopy coverage calculation, it transmits the data to the data acquisition unit through the data interface for storage or offline analysis, for subsequent manual processing or import into other analysis systems. This mode does not involve real-time feedback and decision updates from the cloud server and is a passive data export method, suitable for application scenarios with limited network conditions or where real-time decision-making is not required. Those skilled in the art will understand that the above two usage modes can be flexibly selected according to the actual deployment environment, and both fall within the protection scope of this invention.

[0079] Example 2:

[0080] Based on Example 1, this embodiment provides a detailed description of the specific physical structure and observation logic of the dual field-of-view probe in the spectral acquisition device.

[0081] The observation area of ​​the narrow-angle field-of-view probe is pointed to the top area of ​​the target plant, and is used to directionally acquire the top feature reflection information of the target plant, which includes the spectral signals of the top leaves, flowers or fruits of the target plant.

[0082] The observation area of ​​the wide-angle field-of-view probe covers the entire canopy of the target plant and the surrounding background ground, and is used to obtain the overall reflection information of the target plant and its surrounding area. The overall reflection information includes the mixed spectral signals of the canopy, stem and background ground of the target plant.

[0083] Specifically, the narrow-angle field-of-view probe is designed for directional observation. Its narrow field of view allows it to penetrate gaps in the crop canopy surface and precisely target key organs at the top of the crop. For example, during the flowering or fruiting stages of a crop, the narrow-angle field-of-view probe primarily captures the specific spectral reflectance signals of flowers or fruits, which often carry crucial physiological information about the crop's reproductive growth stage.

[0084] In contrast, wide-angle field-of-view probes are designed for extensive coverage, with a larger field of view capable of simultaneously covering the main canopy, lower stems, and exposed soil or background ground between rows of crops. This observation logic is based on the fact that the vertical structure of crops (such as the number of leaf layers, stem height, and flower / fruit position) changes significantly at different growth stages. For example, in the seedling stage, the top of the crop is mostly composed of young leaves, while the soil background accounts for a high proportion of the overall field of view; in the mature stage, the top of the crop may contain fruit or older leaves, while the proportion of stems and lower leaves increases in the overall field of view. By comparing the spectral differences between narrow-angle and wide-angle fields of view, this change in vertical structure can be accurately captured, thus providing highly sensitive feature data for determining growth status.

[0085] Furthermore, the wide-angle field-of-view probe includes a cosine receiver to achieve a 180-degree field of view, and the narrow-angle field-of-view probe includes a focusing lens to achieve a 25-degree field of view; the ratio between the installation height of the spectral acquisition device relative to the ground and the observation diameter of the narrow-angle field-of-view probe is 2:1, and the ratio between the installation height of the spectral acquisition device relative to the ground and the effective observation diameter of the wide-angle field-of-view probe is 1:3.

[0086] Specifically, the wide-angle field-of-view probe uses a cosine receiver with an effective field of view of up to 180 degrees. This is an optical receiving device that conforms to Lambert's cosine law. It collects light energy at different incident angles according to the cosine value of the angle, which can restore the overall reflection characteristics of ground objects under natural lighting conditions to the greatest extent. Its effective observation area is approximately a circular area centered on the probe.

[0087] The narrow-angle field-of-view probe uses a 25-degree focusing lens, and its optical path design is similar to that of a telescope system. It can image distant targets onto the detector and has extremely high spatial resolution. Its observation area is approximately a small circular spot of light.

[0088] To ensure that the observation area of ​​the narrow-angle field-of-view probe is always within the observation area of ​​the wide-angle field-of-view probe, thereby ensuring that both are observing the same crop or the same crop population, this embodiment strictly limits the ratio of installation height to observation diameter.

[0089] Assuming the spectral acquisition device is installed at a height of H relative to the ground, according to the principles of geometric optics, the observation diameter D of the narrow-angle field-of-view probe... narrow ≈2Htan(25° / 2)≈0.5H, meaning the ratio of installation height to observation diameter is approximately 2:1; the effective observation diameter D of the wide-angle field-of-view probe. wide(Based on the effective response range of the cosine receiver, typically effective data within a 144-degree range) ≈ 3H, meaning the ratio of installation height to observation diameter is approximately 1:3. This parameter setting ensures that the observation area of ​​the narrow-angle field-of-view probe (approximately 0.5H in diameter) falls entirely within the observation area of ​​the wide-angle field-of-view probe (approximately 3H in diameter), forming a concentric circle observation structure. This structural design not only ensures the consistency of the observed target, but more importantly, it makes the "top information" observed by the narrow-angle field-of-view probe a subset of the "overall information" observed by the wide-angle field-of-view probe. By calculating the spectral differences between the two, "background information" (such as stems and soil) can be extracted from the "overall information," thereby accurately extracting characteristic signals representing changes in the top state of the crop.

[0090] It should be understood that the above-mentioned field-of-view parameters (180 degrees and 25 degrees) and height-to-diameter ratios (2:1 and 1:3) are only preferred embodiments. In other embodiments, the field-of-view angle and ratio can be adaptively adjusted according to different crop types (such as tall or short crops) and planting densities, as long as the condition that the narrow-angle observation area is located within the wide-angle observation area is met. For example, the wide-angle field-of-view angle can be 144 degrees, and the narrow-angle field-of-view angle can be 15 degrees. As long as there is a significant difference between the two and the observation area of ​​the narrow-angle field of view is completely located within the observation area of ​​the wide-angle field of view, the technical effect of the present invention can also be achieved.

[0091] It should be noted that the monitoring system of this invention is applied to regional monitoring of crop populations. A narrow-field-of-view probe targets several uniformly growing plants (e.g., 2-8 plants) within the population, while a wide-field-of-view probe covers a larger area including multiple plants and the soil between rows. In this case, the observation area of ​​the narrow-field-of-view is always located within the wide-field-of-view; therefore, the dual-field-of-view spectral difference analysis method and adaptive adjustment strategy described in this invention are directly applicable without requiring changes to the algorithm logic.

[0092] To intuitively understand dual field of view (wide-angle field of view probe and narrow-angle field of view probe). Figure 2 This is a schematic diagram of the field of view of a narrow-angle field-of-view probe. Figure 3 This is a schematic diagram of the field of view of a wide-angle field-of-view probe. A 25° narrow field of view has a smaller detection area, enabling precise focusing on key organs of the crop canopy (such as the top leaf or tassel), with minimal interference from background soil. In contrast, a 180° wide field of view has a larger detection area, covering the entire crop canopy and its surrounding environment. This complementary observation mode of "focusing on local features" and "perceiving the overall environment" lays the physical foundation for subsequent extraction of spectral difference features.

[0093] Example 3:

[0094] This embodiment, based on Embodiment 1, provides a detailed description of the specific structure of the lifting regulator and its adaptive control logic.

[0095] The height adjustment device includes a drive motor, a telescopic rod, and a distance sensor. One end of the telescopic rod is fixedly connected to the spectral collector. The distance sensor is used to measure the real-time distance between the spectral collector and the target plant to provide a feedback control reference for height adjustment. The drive motor is used to drive the telescopic rod to extend and retract to adjust the height of the spectral collector relative to the target plant.

[0096] Specifically, the telescopic rod can be either an electric push rod or a hydraulic rod, with its fixed end mounted on a bracket and its movable end connected to the spectral collector via a flange or clip. The drive motor receives control signals from a local processor and drives the telescopic rod to extend or retract via forward and reverse rotation, thereby raising or lowering the spectral collector. A distance sensor is mounted on the spectral collector or the telescopic rod, with its measurement direction facing the top of the target plant. It is used to collect the distance between the spectral collector and the top of the plant in real time, and this distance serves as the feedback control reference for the local processor to adjust the height. This mechanical and sensor-integrated structural design ensures the reliability of the lifting and lowering action, provides real-time feedback data for closed-loop control, adapts to complex field conditions, and has low maintenance costs.

[0097] It should be understood that in this embodiment, the telescopic pole is installed perpendicular to the ground. However, in other embodiments, depending on the terrain or crop planting pattern, the telescopic pole can also be installed at an angle, as long as the height of the spectral collector relative to the target plant can be adjusted.

[0098] The local processor is specifically used to execute the height adjustment operation, and its control logic mainly includes the following key steps:

[0099] Step S301: Calculate the canopy coverage of the target plant based on the reflected light signal corresponding to the narrow-angle field-of-view probe.

[0100] Specifically, the local processor processes the spectral data acquired by the narrow-angle field-of-view probe. By calculating vegetation indices (such as NDVI) and setting thresholds, it distinguishes between vegetation pixels and background pixels, and then calculates the proportion of vegetation pixels within the narrow-angle field-of-view observation area, which is the canopy coverage. This coverage directly reflects the canopy coverage of the crop by the narrow-angle field-of-view probe at the current height of the spectral acquisition device.

[0101] Step S302: Estimate the current height of the target plant based on the difference feature vector, or determine the current distance between the spectral acquisition device and the target plant based on the real-time distance obtained by the ranging sensor, and calculate the target observation height based on the current height or the current distance and the field of view of the narrow-angle field of view probe;

[0102] Wherein, the target observation height is the height required for the observation area of ​​the narrow-angle field-of-view probe to cover the core area of ​​the canopy of the target plant;

[0103] Specifically, the difference feature vector contains spectral difference information between wide-angle and narrow-angle fields of view, which is closely related to the crop's vertical structure (such as plant height and leaf area index). The local processor has a pre-built calibration model trained on historical data, capable of retrieving the crop's current height from the difference feature vector. Based on the current height and the field of view angle (25 degrees) of the narrow-angle field-of-view probe, combined with geometric optics principles, the target observation height required to cover the core canopy area (e.g., a certain range downwards from the top of the plant) can be calculated. For example, if the estimated plant height is H and the optimal observation distance of the narrow-angle field-of-view probe is D, then the target observation height can be set to H+D.

[0104] The range sensor can directly measure the real-time distance between the spectral acquisition device and the top of the target plant. Based on this distance and the field of view of the narrow-angle field-of-view probe, the target observation height required to cover the core area of ​​the canopy can be calculated. For example, if it is desired to maintain a constant distance L between the spectral acquisition device and the top of the plant, the target observation height is the current height of the spectral acquisition device (if the plant height remains unchanged), or it can be adjusted to a height equal to L through closed-loop control. In practical applications, a desired distance value is usually set, and the height adjustment mechanism is controlled to make the reading of the range sensor approach this desired value.

[0105] Step S303 (Coverage Threshold Determination and Height Adjustment): If the canopy coverage is lower than a preset coverage threshold, a descent control command is generated. The descent control command is used to control the drive motor to drive the telescopic rod to shorten, thereby lowering the spectral collector to the target observation height. If the canopy coverage is higher than the preset coverage threshold, an ascent control command is generated. The ascent control command is used to control the drive motor to drive the telescopic rod to extend, thereby raising the spectral collector to the target observation height.

[0106] Specifically, the preset coverage threshold is an empirical value set based on crop type and growth stage. For example, for corn, it can be set to 80% during the jointing stage. When the coverage is below the threshold, it indicates that the spectral acquisition device is positioned too high, and the narrow-angle field-of-view probe does not observe enough vegetation. In this case, the system automatically lowers the probe to increase the coverage. Conversely, when the coverage is above the threshold, it indicates that the probe is positioned too low, and too much ground background or non-target area may be observed. In this case, the system automatically raises the probe.

[0107] This closed-loop control mechanism ensures that the spectral acquisition device is always in the optimal observation position, avoiding field-of-view drift caused by crop growth or plant lodging.

[0108] Step S304 (Generation of Predictive Descent Instruction): If the canopy coverage continues to decrease over multiple preset acquisition cycles, and the canopy coverage has not yet fallen below the preset coverage threshold, a predictive descent instruction is generated to lower the spectral acquisition device to the target observation height in advance.

[0109] Specifically, crop growth is a continuous process, with plant height constantly increasing. If only thresholds are used for judgment, adjustments may be made only when the coverage falls below the threshold, potentially creating a monitoring blind spot for a period. This embodiment introduces a trend prediction mechanism. The local processor records coverage data for multiple consecutive collection periods (e.g., the past 3 days). If a continuously decreasing trend in coverage is detected (e.g., a 2% decrease per day), even if the current coverage is still above the threshold, the system will determine that the crop is growing rapidly and generate a reduction instruction in advance. This predictive adjustment strategy demonstrates the system's intelligence level, effectively eliminates monitoring blind spots, and ensures data continuity and integrity.

[0110] Step S305 (Minimum observation height limit): Set a minimum observation height threshold to control the height of the spectral acquisition device to be no lower than the minimum observation height threshold, so as to prevent the spectral acquisition device from physically colliding with the target plant and to ensure the observation integrity of the wide-angle field of view probe.

[0111] Specifically, the minimum observation height threshold is a safety distance set based on the maximum possible height of the crop and the physical dimensions of the spectral acquisition device. For example, if the maximum height of corn can reach 2.5 meters and the length of the spectral acquisition device is 0.3 meters, then the minimum observation height threshold can be set to 3 meters. When the target observation height calculated by the system is lower than this threshold, the height will be forcibly locked at the minimum observation height to prevent the probe from colliding with the top of the crop and causing equipment damage. At the same time, this threshold also ensures that the wide-angle field-of-view probe has a sufficient field of view to cover a certain area around the plant, ensuring the integrity of the background information.

[0112] It should be noted that, in addition to providing the current distance for target observation height calculation in step S302, the ranging sensor can also serve as a closed-loop feedback reference for height adjustment. During the system's execution of the height adjustment command, the local processor can read the distance value from the ranging sensor in real time, compare it with the desired distance, and precisely adjust the position of the telescopic rod through control algorithms such as PID control, ensuring a stable relative distance between the spectral acquisition device and the top of the plant. This feedback control based on hardware ranging has a higher response speed and stronger anti-interference capability compared to pure software estimation.

[0113] Furthermore, in the above steps, steps S301 (calculating canopy coverage) and S302 (estimating height and target height) are fixed steps that must be executed in each height adjustment cycle. Steps S303 (comparing canopy coverage thresholds and issuing rise / fall commands) and S304 (predictive descent commands) are conditional branches that are selectively executed based on real-time monitoring results. The system automatically selects and executes the corresponding adjustment action based on the relationship and trend between canopy coverage and preset thresholds. Step S305 (minimum observation height limit) is executed only when a descent command is generated to ensure that the spectral acquisition device does not collide with the target plant after descent. If an rise command is generated or no adjustment is required, this safety verification step is skipped. This flexible control logic ensures both the system's response speed and improves energy efficiency and equipment safety.

[0114] Through the control logic described above, this embodiment constructs a complete "perception-decision-execution" closed loop. The system can not only passively respond to changes in coverage, but also proactively predict crop growth trends and make adjustments in advance, truly achieving adaptive monitoring of crop growth status.

[0115] Example 4:

[0116] This embodiment, based on Embodiment 1, provides a detailed explanation of the specific data processing flow for constructing differential feature vectors using the local processor. The differential feature vectors are the core data foundation for this system's growth state determination and stress prediction; their construction quality directly determines the accuracy of subsequent model predictions.

[0117] The local processor is specifically used to perform the following data processing steps:

[0118] Step S401: Based on the reflected light signal, obtain the first reflectivity data corresponding to the wide-angle field-of-view probe and the second reflectivity data corresponding to the narrow-angle field-of-view probe.

[0119] Specifically, the raw signal acquired by the spectral acquisition device is the light intensity value (DN value). The local processor first uses the incident light signal observed by the wide-angle field-of-view probe (usually obtained through standard whiteboard calibration or a dedicated incident light probe) to perform reflectance inversion on the raw DN value, obtaining dimensionless reflectance data. The first reflectance data characterizes the mixed spectral characteristics of the overall crop canopy and background environment, while the second reflectance data characterizes the fine spectral characteristics of key organs at the top of the crop.

[0120] Step S402: Perform Savitzky-Golay filtering preprocessing on the first reflectivity data and the second reflectivity data respectively to obtain the first preprocessed data and the second preprocessed data.

[0121] Specifically, spectral data from field environments often contain random noise (such as instantaneous changes in illumination caused by wind swaying, electronic noise, etc.). Savitzky-Golay filtering (SG filtering for short) is a convolutional smoothing algorithm based on the least squares method. It smooths the data by fitting a polynomial within a moving window, effectively removing high-frequency noise while preserving the peak and valley features of the spectral curve (such as the position of the red edge and the depth of the absorption valley). In this embodiment, a 3rd-order polynomial with a window width of 9 is preferably used for SG filtering parameters to ensure that the slope information of the red edge region is not lost while smoothing noise.

[0122] Understandably, to further improve the robustness of the differential feature vectors under different environmental conditions, after completing the SG filtering preprocessing, the local processor can also dynamically correct or label the reflectance data based on the temporal field information at the time of spectral acquisition and the real-time weather identification results. Specifically:

[0123] Regarding temporal factors: The solar altitude angle and light intensity vary significantly across different time periods, directly affecting the spectral reflectance characteristics of the crop canopy, especially the top leaves focused by narrow-angle field-of-view probes. Their specular reflectance is enhanced under strong midday sunlight, while it is relatively weaker in the morning or afternoon. To avoid characteristic fluctuations within the same growth stage due to different sampling times, the system can identify the current time period:

[0124] Morning period (9:00-11:00): The sunlight is oblique and the proportion of shadows is high, which is suitable for obtaining information on the side structure of the canopy;

[0125] Midday (11:00-13:00): Direct sunlight, uniform light exposure at the top of the canopy, vegetation index tends to saturate;

[0126] Afternoon (13:00-15:00): The sunlight shines at an angle again, symmetrical to the morning.

[0127] The local processor can normalize reflectance data to the level of a reference time period (e.g., 10:00 AM) based on preset time-period correction coefficients. These correction coefficients can be obtained through offline calibration experiments, such as measuring the reflectance variation curves of the same standard reference plate at different time periods and fitting a correction function.

[0128] Regarding weather factors: The system can identify weather types by analyzing the stability of incident irradiance. Under clear weather, incident irradiance is stable and relatively high; under cloudy weather, irradiance fluctuates frequently but maintains a certain overall intensity; under overcast weather, irradiance is low and the proportion of blue light in the spectral composition increases. The signal-to-noise ratio of the dual-field spectral difference features differs under different weather conditions. For example, the diffuse light environment under cloudy weather can reduce canopy shading, allowing the difference between the narrow-angle and wide-angle fields of view to more accurately reflect the internal structure of crops, thus allowing for higher confidence weights on the data; while strong midday sunlight may introduce specular reflection noise, which can be mitigated by reducing the weights. The system can adjust the weights of feature vectors in subsequent model inputs based on weather type, or model data for different weather conditions separately during training.

[0129] By introducing dynamic corrections for time fields and weather factors, this invention further enhances the environmental adaptability of differential feature vectors, ensuring consistency of data collected under different conditions at the same growth stage, thereby improving the generalization ability and prediction accuracy of the classification model.

[0130] Step S403: Based on the first preprocessed data and the second preprocessed data, calculate at least one of the reflectance difference, reflectance ratio, red edge feature difference value and vegetation index difference value within the preset band range to form the difference feature vector.

[0131] Specifically, the construction of the differential feature vector is one of the core innovations of this invention. Traditional single-field-of-view monitoring can only obtain absolute reflectance, while this invention, by comparing the data differences between wide-angle and narrow-angle fields of view, can remove background interference and extract features characterizing changes in the vertical structure of crops. Methods for constructing the differential feature vector include, but are not limited to, the following:

[0132] (1) Reflectance difference and ratio: Directly calculate the difference (ΔR=R180-R25) or ratio (Ratio=R25 / R180) between the first preprocessed data and the second preprocessed data in a specific band (such as 550nm in the green band, 670nm in the red band, and 800nm ​​in the near-infrared band). Wherein, R180 represents the reflectance value collected by the wide-angle field-of-view probe (field of view 180°) after SG filtering preprocessing, i.e., the first preprocessed data; R25 represents the reflectance value collected by the narrow-angle field-of-view probe (field of view 25°) after SG filtering preprocessing, i.e., the second preprocessed data; ΔR is the reflectance difference, which represents the absolute difference between the reflectance of the wide-angle field of view and the narrow-angle field of view at the same wavelength; Ratio is the reflectance ratio, which represents the relative ratio between the reflectance of the narrow-angle field of view and the wide-angle field of view at the same wavelength.

[0133] Taking the red light band as an example, during the seedling stage of rice crops, the wide-angle field of view includes a large amount of soil background, which has a high red light reflectance, while the narrow-angle field of view focuses on a small number of seedlings, which has a low red light reflectance (smaller R25). Therefore, the difference ΔR is positive and significant, and the ratio is less than 1. However, during the crop canopy closure stage, the observation targets of the two tend to be the same, the difference ΔR approaches zero, and the ratio approaches 1.

[0134] It should be noted that the sign of the difference may differ in different bands (for example, in the near-infrared band, the reflectivity of healthy vegetation is higher than that of soil, so R25>R180, ΔR is negative, and the ratio is greater than 1). However, the absolute value of the difference, the sign of the difference reflecting the high and low reflectivity of the two fields of view, and the degree and direction of the ratio deviating from 1 can all serve as effective features reflecting changes in crop coverage and vertical structure.

[0135] (2) Red edge feature difference value: Red edge features refer to parameters that reflect the physiological state of vegetation extracted from the red edge region (usually 680nm-750nm) of the vegetation spectrum, such as red edge position (REP), red edge slope, and red edge goodness of fit (R). 2 The red edge position is the wavelength corresponding to the maximum value of the first derivative of reflectance with wavelength, i.e., the inflection point where reflectance rises sharply from the red band to the near-infrared band, which is closely related to chlorophyll content; the red edge slope reflects the steepness of the red edge rise, which is related to leaf area index and canopy structure; the red edge goodness of fit characterizes the degree of agreement between the spectral data and the ideal red edge model, and can be used as an indicator of data quality.

[0136] In this embodiment, the calculation method of the red-edge feature difference value includes: performing linear fitting on the first preprocessed data and the second preprocessed data in the red-edge band region of 690nm to 750nm respectively to obtain the first red-edge feature parameter and the second red-edge feature parameter, and determining the red-edge feature difference value based on the difference between the first red-edge feature parameter and the second red-edge feature parameter; the red-edge feature parameter includes at least one of red-edge position, red-edge slope and red-edge fit goodness;

[0137] Specifically, as a concrete example of a red-edge feature parameter, the calculation method is illustrated below using the red-edge position as an example. The red-edge position is usually determined by the derivative method or the linear four-point interpolation method, and its calculation formula is as follows:

[0138] REP=700+40×((R670+R780) / 2-R700) / (R740-R700);

[0139] REP stands for red edge position, which physically means the inflection point wavelength where the reflectance of vegetation spectrum rises sharply from the red band to the near-infrared band, and the unit is nanometers (nm).

[0140] R represents reflectivity. R670, R700, R740, and R780 represent the reflectivity values ​​at wavelengths of 670nm, 700nm, 740nm, and 780nm, respectively.

[0141] 700nm is the starting reference wavelength of the red edge region, which can also be called the lower boundary wavelength of the red edge region. It is the position where reflectivity begins to rise rapidly.

[0142] 40 represents the wavelength interval, also known as the interval width (i.e., the difference between 740nm and 700nm).

[0143] The formula uses the reflectivity of four characteristic bands—the red absorption valley (670nm), the red edge initiation point (700nm), the red edge midpoint (740nm), and the near-infrared plateau (780nm)—to estimate the red edge position through linear interpolation.

[0144] It should be noted that 40 and 700 are calibration parameters for the formula. 700 is the reference wavelength for interpolation (usually taken near the center of the red-edge region, such as 700nm), and 40 is the wavelength interval (740nm-700nm in this embodiment). These parameters can be adjusted according to the specific wavelength selected.

[0145] The position of the red edge observed in a wide-angle field of view is often affected by the lower leaves and soil background, and may exhibit a red shift or blue shift. In contrast, the position of the red edge observed in a narrow-angle field of view primarily reflects the physiological state of the top leaves. The difference between the two (ΔREP = REP180 - REP25) can sensitively indicate the difference between the health status of the top of the crop canopy and the lower background. When the top leaves are under stress (such as drought, pests, or diseases), the position of the red edge in the narrow-angle field of view will exhibit a blue shift, leading to an increased difference value.

[0146] In this invention, when the red edge position is selected as the red edge feature parameter, ΔREP is the red edge feature difference value. Its physical meaning is the absolute difference between the red edge position in the wide-angle field of view and the red edge position in the narrow-angle field of view at the same time and location. It is one of the core components of the difference feature vector. It can be understood that in other cases, when the red edge slope is selected, the corresponding difference value is the slope difference; when the goodness of fit is selected, the difference value is the goodness of fit difference. All of the above difference values ​​can be used as components of the difference feature vector. In practical applications, the most sensitive feature parameter can be selected according to the crop type and monitoring needs.

[0147] REP180 is the first red-edge feature parameter (e.g., red-edge position, red-edge slope, or red-edge goodness of fit) calculated based on the first preprocessed data, and REP25 is the second red-edge feature parameter calculated based on the second preprocessed data.

[0148] For example, when the red edge position is selected as the red edge feature parameter, REP180 is the red edge position of the 180° wide-angle field of view, that is, the first red edge feature parameter. Its physical meaning is the red edge position calculated based on the first preprocessed data (the data of the reflected light signal collected by the 180° wide-angle field of view probe after SG filtering); REP25 is the red edge position of the 25° narrow-angle field of view, that is, the second red edge feature parameter. Its physical meaning is the red edge position calculated based on the second preprocessed data (the data of the reflected light signal collected by the 25° narrow-angle field of view probe after SG filtering).

[0149] It should be understood that when the red-edge slope is chosen as the red-edge characteristic parameter, linear regression can be performed on the reflectance data within the red-edge band. The red-edge slope is calculated as follows: linear regression is performed on the reflectance data within the red-edge band (e.g., 690nm-750nm), yielding the regression equation R=k·λ+b, where k is the red-edge slope. A larger slope indicates a steeper red edge and more vigorous vegetation growth. In other words, the slope of the resulting regression line is the red-edge slope, and its difference (Δslope = slope180 - slope25) reflects the change in the steepness of the red edge between two viewing angles, which is related to the vertical distribution of chlorophyll in the canopy. When the red-edge goodness-of-fit is chosen as the characteristic parameter, its difference can be used to assess the difference in the fitting quality between the spectral data at two viewing angles and the standard red-edge model, serving as a reference indicator for data reliability; the red-edge goodness-of-fit R... 2 The calculation method is as follows: after performing linear regression on the red-edge band data, the coefficient of determination R is calculated. 2 , used to evaluate the strength of the linear relationship between reflectivity and wavelength. R 2 The closer to 1, the more obvious the red edge feature and the higher the data quality.

[0150] It should be noted that the red shift and blue shift mentioned above refer to the shift of the red edge position towards longer wavelengths (such as towards 750nm), which is usually associated with increased chlorophyll content or vigorous vegetation growth; the blue shift refers to the shift of the red edge position towards shorter wavelengths (such as towards 690nm), which is usually associated with chlorophyll degradation, leaf stress or senescence.

[0151] (3) Vegetation index difference value: The vegetation index is an indicator that enhances vegetation information by combining the reflectance of different bands.

[0152] In this embodiment, the vegetation index difference value is calculated by: calculating the normalized vegetation index based on the first preprocessed data and the second preprocessed data respectively, obtaining the first vegetation index and the second vegetation index, and determining the vegetation index difference value based on the difference between the first vegetation index and the second vegetation index.

[0153] The formula for calculating the Normalized Difference Vegetation Index (NDVI) is as follows:

[0154] NDVI = (NIR - Red) / (NIR + Red);

[0155] Wherein, NIR represents the reflectance in the near-infrared band; and Red represents the reflectance in the red band.

[0156] The NDVI difference value (ΔNDVI = NDVI 180 - NDVI 25) can eliminate some of the influence of light conditions and more stably reflect the difference in greenness of crops in the vertical direction.

[0157] It is understood that this embodiment uses the Normalized Difference Vegetation Index (NDVI) as an example, but the vegetation index difference value described in this invention is not limited to NDVI. It can also be other vegetation indices that can reflect vegetation greenness or growth status, such as Enhanced Vegetation Index (EVI), Photochemical Reflectance Index (PRI), Soil-Regulated Vegetation Index (SAVI), Green Band Normalized Difference Vegetation Index (GNDVI), etc. It is only necessary to calculate the index values ​​under the two viewing angles and find the difference.

[0158] Furthermore, in order to enhance the system's ability to perceive crop physiological stress, the differential feature vector also includes water-sensitive differential features and nutrient-sensitive differential features.

[0159] The water-sensitive difference feature is used to characterize the difference in water content between the canopy and the understory of the target plant in order to assess the water stress status of the target plant.

[0160] Specifically, this embodiment preferably uses the Normalized Differential Water Index (NDWI) to construct the water-sensitive differential characteristics. The formula for calculating NDWI is:

[0161] NDWI=(Green-NIR) / (Green+NIR);

[0162] Wherein, Green represents the reflectivity of the green light band.

[0163] The water-sensitive differential characteristic (ΔNDWI = NDWI 180 - NDWI 25) reflects the difference in water content between the top of the crop canopy and the overall field of view. Under normal water supply conditions, the leaves at the top of the crop canopy have sufficient water, resulting in a higher NDWI. However, the overall field of view may contain some dry soil or older lower leaves, leading to a lower NDWI 180, resulting in a significant difference. When the crop is subjected to water stress, the top leaves preferentially lose water, causing NDWI 25 to decrease, thus reducing the difference. By monitoring the trend of ΔNDWI changes, the system can provide early warning of water stress.

[0164] The nutrient sensitivity difference feature is used to characterize the difference in chlorophyll or nitrogen nutrition between the canopy and the understory of the target plant in order to assess the nutrient deficiency status of the target plant.

[0165] Specifically, this embodiment preferably uses the chlorophyll index (CI) or photochemical reflectance index (PRI) to construct the nutrient sensitivity difference characteristic. For example, in the chlorophyll index CI = (NIR / Green) - 1, the nutrient sensitivity difference characteristic (ΔCI = CI180 - CI 25) reflects the difference in chlorophyll content between the top of the crop canopy and the whole crop. When nitrogen is sufficient, the chlorophyll content of the young leaves at the top is high, and the CI 25 is relatively high, showing a significant difference. When the crop shows nitrogen deficiency symptoms, chlorophyll synthesis in the top leaves is inhibited, the CI 25 decreases, and the difference value changes.

[0166] This feature extraction method based on the difference between two fields of view can effectively eliminate the interference of environmental factors such as soil background and illumination angle compared with the absolute value monitoring of a single field of view, and significantly improve the sensitivity of the perception of crop physiological stress state.

[0167] Through the steps described above, the local processor transforms the raw spectral reflectance data into a differential feature vector containing multi-dimensional information such as differences in red-edge features, vegetation index, water sensitivity, and nutrient sensitivity. This vector not only represents the crop's growth stage but also implicitly reveals the crop's physiological stress state, providing high-quality data input for the cloud server's high-precision prediction.

[0168] Example 5:

[0169] This embodiment, based on Embodiment 4, provides a detailed explanation of the model prediction and agricultural decision generation process using a cloud server. The cloud server, as the "brain" of the system, undertakes the tasks of complex model computation and decision support.

[0170] The cloud server is specifically used to perform the following steps:

[0171] Step S501: Input the differential feature vector into the pre-trained classification model or empirical model.

[0172] Specifically, the differential feature vectors generated by the local processor are uploaded to the cloud server via a communicator. The cloud server contains a pre-trained classification model or empirical model. Understandably, depending on the specific circumstances, in another implementation, the classification model or empirical model can be deployed directly on the local processor without communicating with the cloud server. The local processor completes the entire process of spectral data acquisition, preprocessing, feature extraction, and growth stage prediction, making it suitable for remote farmland scenarios with poor network signals.

[0173] It's important to note that classification models or empirical models can take various forms. For example, Support Vector Machines (SVM) and Random Forests (RF) are two commonly used machine learning models. SVM models use kernel functions to map nonlinear samples in low-dimensional space to high-dimensional space, searching for the optimal classification hyperplane, making them particularly suitable for handling small-sample, high-dimensional classification problems. RF models, on the other hand, construct multiple decision trees and use voting, exhibiting good resistance to overfitting and strong feature importance assessment capabilities. Additionally, deep learning models such as Backpropagation (BP) neural networks, or empirical models based on statistical thresholds or expert rules (e.g., directly determining the growth stage based on the range of differences at red-edge positions) can also be used. In practical applications, the choice can be flexibly made based on the sample size and accuracy requirements.

[0174] It should be understood that the SVM, RF, BP neural networks, etc. mentioned above are merely examples, and the classification model or empirical model of this invention is not limited to these. Any model that can classify growth states based on differential feature vectors (including but not limited to various machine learning models, deep learning models, rule bases, or statistical empirical models) is applicable.

[0175] In this embodiment, a support vector machine (SVM) or random forest (RF) model is preferred because the difference feature vector usually contains multidimensional data (such as differences in red edge positions, differences in vegetation indices, water-sensitive features, etc.), and these features often have complex nonlinear relationships with crop growth stages.

[0176] The SVM model maps nonlinear samples in low-dimensional space to high-dimensional space through kernel functions to find the optimal classification hyperplane, making it particularly suitable for handling classification problems with small samples and high dimensions. The RF model, on the other hand, constructs multiple decision trees and performs voting, giving it good resistance to overfitting and the ability to evaluate feature importance.

[0177] It should be understood that although SVM or RF is preferred in this embodiment, in other embodiments, deep learning models (such as convolutional neural networks CNN) or gradient boosting trees (such as XGBoost) can also be used, as long as they can achieve the classification function of the growth stage.

[0178] Step S502: Obtain the growth stage label of the target plant output by the classification model, as the growth state prediction result.

[0179] Specifically, classification models, trained on a large number of historical samples, or empirical models based on expert knowledge or statistical rules, have learned the mapping relationship between differential feature vectors and different growth stages. For example, in a corn monitoring scenario, the model may output specific growth stage labels such as "seedling stage," "jointing stage," "tasseling stage," "silking stage," "grain-filling stage," and "maturity stage." These labels not only characterize the phenological stages of the crop but also implicitly contain its morphological and structural features. For instance, when the model outputs "tasseling stage," it means that the tassel features are significant in the spectral signal observed by the narrow-angle field-of-view probe, and the difference between the wide-angle and narrow-angle fields of view exhibits a pattern specific to this stage. The cloud server feeds this prediction result back to the local processor via a communicator, and the local processor updates the adaptive observation adjustment strategy accordingly, such as increasing the sampling frequency during the tasseling stage to capture the crucial process of pollen shedding from the tassel.

[0180] The growth status prediction results also include the water stress status and nutrient deficiency status of the target plant; the water stress status is determined based on the water-sensitive differential feature in the differential feature vector, and the nutrient deficiency status is determined based on the nutrient-sensitive differential feature in the differential feature vector.

[0181] Specifically, the growth status prediction results include not only growth stage labels but also diagnoses of crop physiological health. The cloud server has a pre-built stress assessment model or rule base. For water stress states, the system analyzes water-sensitive differential features (such as ΔNDWI) in the differential feature vector.

[0182] For example, a water stress threshold can be set. When ΔNDWI is lower than the lower limit of the normal range, it indicates that the top of the crop canopy is severely water deficient and is judged as "mild water stress"; when ΔNDWI continues to decline and is lower than the severe threshold, it is judged as "severe water stress".

[0183] Similarly, for nutrient deficiency states, the system analyzes nutrient sensitivity differences (such as ΔCI). When ΔCI shows abnormal fluctuations (such as a significant decrease or even reversal in the difference value), it indicates that chlorophyll synthesis in the top leaves of the crop is hindered, and it is judged as "nitrogen nutrient deficiency" or "micronutrient deficiency".

[0184] This stress determination method based on the difference between two field angles can effectively eliminate the interference of environmental factors such as soil background and changes in light intensity compared with the determination of the absolute value of a single field angle, and significantly improves the accuracy of stress diagnosis.

[0185] The cloud server is also specifically used to: generate early warning information when the water stress state or the nutrient deficiency state is abnormal, and generate agricultural decision-making suggestions based on the water stress state and the nutrient deficiency state; and send the early warning information and the agricultural decision-making suggestions to the local processor or user terminal through a communicator.

[0186] The agricultural decision-making recommendations include at least one of irrigation timing recommendations, fertilizer application recommendations, and harvesting time recommendations; the irrigation timing recommendations determine the appropriate irrigation time window based on the severity of the water stress state, and the fertilizer application recommendations determine the recommended fertilizer type and amount based on the type and severity of the nutrient deficiency state.

[0187] Specifically, regarding irrigation timing recommendations, the cloud server determines the appropriate irrigation time window based on the severity of the water stress condition. For example, if it is determined to be "mild water stress", the system recommends "applying appropriate amounts of irrigation within the next 48 hours"; if it is determined to be "severe water stress", the system recommends "starting the irrigation program immediately and increasing the irrigation amount by 20%".

[0188] Regarding fertilizer application recommendations, the cloud server determines the recommended fertilizer type and dosage based on the type and severity of the nutrient deficiency. For example, if the deficiency is determined to be "nitrogen deficiency," the system recommends "applying 10 kg / mu of urea as a top dressing"; if the deficiency is determined to be "micronutrient deficiency," the system recommends "spraying foliar fertilizer."

[0189] Regarding harvesting time recommendations, the system combines growth stage tags and crop maturity characteristics (such as the significance of fruit spectral features in the differential feature vector) to predict the optimal harvesting window. For example, "It is expected that the optimal harvesting period will begin in 7 days, and it is recommended to prepare for harvesting in advance."

[0190] Finally, the cloud server sends early warning information and agricultural decision-making suggestions to local processors or user terminals (such as farmers' mobile apps or computer clients) via a communicator. Farmers can view this information in real time and carry out precise field management.

[0191] This embodiment achieves a leap from "data monitoring" to "agricultural guidance" through cloud-based intelligent decision-making, greatly improving the level of intelligence in agricultural production.

[0192] Example 6:

[0193] This embodiment, based on Embodiment 1, provides a detailed description of the system's dynamic observation strategy adjustment function. Throughout the crop's growth cycle, the transitions between different growth stages are often accompanied by drastic changes in physiological characteristics, which typically occur within a short period. To accurately capture data from these critical transition points, the system employs a dynamic observation strategy adjustment mechanism.

[0194] The cloud server is also used to: identify the growth state transition node of the target plant based on the changing trend of the differential feature vector over multiple consecutive acquisition cycles; the growth state transition node is the moment when the change amplitude of the differential feature vector between adjacent acquisition cycles exceeds a preset change threshold.

[0195] Specifically, when crops transition from one growth stage to another (e.g., from jointing to tasseling), their vertical structure (such as plant height and leaf extension angle) and physiological state (such as chlorophyll content and water status) undergo abrupt changes. These abrupt changes are directly reflected in the differential feature vector. For example, at the beginning of the tasseling stage, a narrow-angle field-of-view probe may suddenly capture the spectral signal of the tassel, causing a significant change in the red-edge feature difference value within a short period. The cloud server monitors the rate of change of each dimension in the differential feature vector in real time. Assuming a preset change threshold of 10%, if the change in the red-edge feature difference value exceeds 10% within two consecutive collection periods (e.g., with an interval of 1 day), or the change rate of the vegetation index difference value exceeds twice the historical average level, the cloud server can determine that the current moment is a growth state transition node. This trend-based identification method, compared to simple threshold determination, can more sensitively perceive qualitative changes in crop growth status.

[0196] Secondly, when the growth state transition node is identified, a dynamic observation strategy adjustment instruction is generated; the dynamic observation strategy adjustment instruction includes at least one of the following adjustments: increasing the data acquisition frequency of the spectral acquisition device to capture spectral details during the rapid change period of growth state; adjusting the height regulator to make the height of the spectral acquisition device reach the optimal observation height corresponding to the growth state transition node to ensure that the narrow-angle field-of-view probe focuses on the key organs of the current growth stage; and switching the feature extraction focus of the local processor to match the sensitive band and vegetation index type corresponding to the growth state transition node.

[0197] Specifically, for the first adjustment strategy, the data acquisition frequency is increased. At the transition point of growth status, the morphological and physiological characteristics of crops change rapidly, and conventional low-frequency acquisition (such as once a day) may miss key information. The cloud server will issue instructions to the local processor to temporarily increase the acquisition frequency to once per hour or higher, thereby capturing the fine spectral change curves during the transition process and providing high temporal resolution data support for subsequent accurate modeling.

[0198] The second adjustment strategy involves adjusting the elevation and height to the optimal observation position. The location of key observation organs differs at different growth stages. For example, in the seedling stage, the key organ is the young leaf, located lower down; while in the grain-filling stage, the key organ is the fruit cluster, possibly located in the middle of the plant. The cloud server pre-stores optimal observation height models corresponding to different growth stages. When a growth state transition node is identified, the cloud server calculates the corresponding optimal observation height based on the predicted new growth stage and controls the elevation and height adjustment mechanism to adjust the spectral acquisition device to that height. This ensures that the narrow-angle field-of-view probe always focuses on the most representative key organ, avoiding data distortion caused by observation position deviations.

[0199] The third adjustment strategy involves switching the focus of feature extraction. The sensitive spectral bands and vegetation indices differ at different growth stages. For example, during the vegetative growth stage, chlorophyll and nitrogen content are key indicators, and the local processor should focus on extracting features from the green light band (550nm) and the red-edge band (720nm). During the reproductive growth stage, water status and fruit maturity become crucial, and the processor should switch to focusing on extracting water-sensitive bands (such as 970nm) or specific fruit pigment bands. The cloud server will issue instructions to update the feature extraction algorithm parameters of the local processor to match the current sensitive bands, thereby improving the signal-to-noise ratio of the feature vector and the accuracy of the prediction model.

[0200] Finally, the dynamic observation strategy adjustment instruction is sent to the local processor via the communicator, and the local processor updates the adaptive observation adjustment strategy based on the dynamic observation strategy adjustment instruction.

[0201] Specifically, upon receiving the instruction, the local processor immediately updates its internal configuration parameters, including the acquisition frequency, target observation height, and weight parameters of the feature extraction algorithm. This collaborative mechanism of "cloud recognition - local execution" constructs a complete "prediction-feedback-adjustment" closed loop, enabling the system to dynamically adjust observation methods based on crop growth status, much like a human expert, greatly improving the intelligence level of monitoring and the effectiveness of data.

[0202] It should be understood that the above three adjustment strategies can be executed individually or in combination, depending on the type and importance of the growth state transition node.

[0203] Example 7:

[0204] To more clearly illustrate the actual operating effect of the system of the present invention, a detailed description is given below using the monitoring scenario of summer maize in the milk-ripe stage as an example. In this application scenario, the monitoring object is summer maize plants in the milk-ripe stage, and the initial installation height of the spectral acquisition device relative to the ground is set to 2 meters.

[0205] After the system is started, the wide-angle field-of-view probe (180-degree cosine receiver) and the narrow-angle field-of-view probe (25-degree focusing lens) of the spectral acquisition device simultaneously acquire the reflected light signal of the target plant.

[0206] Specifically, at this stage, summer maize has entered the late reproductive growth phase. The top of the plant mainly grows tassels and upper leaves, while the ears (corn cobs) are located in the middle of the plant. A narrow-angle field-of-view probe focuses on the top area of ​​the plant, primarily capturing the spectral signals of the tassels and upper leaves; a wide-angle field-of-view probe covers the entire canopy and the surrounding ground, including a mixed spectral signal from the upper leaves, the middle ears, the lower stems, and the exposed soil between rows. Based on geometric optics principles, at an installation height of 2 meters, the narrow-angle field-of-view probe has an observation diameter of approximately 1 meter, while the wide-angle field-of-view probe has an effective observation diameter of approximately 6 meters. The narrow-angle field of view is entirely within the wide-angle field of view, ensuring the consistency of the observed target.

[0207] The local processor generates a difference feature vector representing the spectral differences between the two field-of-view angles based on the reflected light signal.

[0208] Specifically, the local processor first performs Savitzky-Golay filtering preprocessing on the acquired raw spectral data to remove environmental noise. Then, it calculates the red-edge feature difference values ​​(REP180-REP25).

[0209] like Figure 4 As shown in the figure, the blue curve represents the reflectance curve collected by the 25° narrow-angle field-of-view probe, the orange curve represents the reflectance curve collected by the 180° wide-angle field-of-view probe, and the gray curve is the reflectance difference curve between the two at the corresponding wavelengths (ΔR=R180-R25). The horizontal axis represents wavelength (unit: nm), covering the visible and near-infrared bands from 376nm to 880nm, and the vertical axis represents reflectance. It can be seen that during the milk-ripe stage, the reflectance of the 25° narrow-angle field-of-view probe is significantly higher than that of the 180° wide-angle field-of-view probe in the visible light band (approximately 376-700nm), with the blue curve generally above the orange curve; while in the near-infrared band (approximately 700-880nm), the two tend to be close. The difference curve shows a significant negative value throughout the visible light band, reaching a negative peak in the red-edge region (approximately 680-750nm), which is a unique spectral difference characteristic of the milk-ripe stage.

[0210] During the milk stage, the tassels at the top of summer maize begin to dry and turn yellow, and the chlorophyll content decreases, causing a blue shift (movement towards shorter wavelengths) in the red edge position observed in narrow-angle field of view. Meanwhile, in wide-angle field of view, the husks of the ears at the midsection still retain some green, and the spectral contributions of the lower stems and background soil remain relatively stable. Therefore, the calculated red edge feature difference values ​​exhibit numerical characteristics unique to the milk stage (e.g., REP180 is approximately 720 nm, REP25 is approximately 705 nm, and the difference is approximately 15 nm). This feature vector is uploaded to the cloud server.

[0211] The cloud server receives the differential feature vectors and outputs the prediction results of the growth status of the target plant based on the pre-trained classification model.

[0212] Specifically, the cloud server inputs the received feature vectors into a random forest model. Based on the red-edge position difference patterns specific to the milk-ripe stage in the training samples, the model outputs a growth stage label for "milk-ripe stage." Simultaneously, the cloud server analyzes the plant's water stress status based on water-sensitive difference features (such as NDWI differences) within the difference feature vectors. During the milk-ripe stage, if the water-sensitive difference features show a reduction in the water difference between the top of the canopy and the overall plant, it indicates that the plant may be under mild water stress. Based on this, the cloud server generates an agricultural decision suggestion of "appropriate irrigation within the next 48 hours" and sends it to the user terminal via a communicator.

[0213] When summer maize transitions from the milk stage to the maturity stage, the differences in the dual-field spectral density undergo further significant changes, such as... Figure 5 As shown in the diagram, at this point, the blue curve representing the 25° narrow field of view shows a flatter reflectance in the visible light band (approximately 376-700 nm), almost losing the typical red-edge rising characteristic of vegetation spectra, indicating that the top spikelets and leaves have completely dried and turned yellow. While the orange curve representing the 180° wide field of view also shows an overall increase in reflectance, it still retains certain vegetation spectral characteristics. This is because the 180° field of view allows observation of the stems and bracts in the lower and middle parts of the plant that are not yet completely dried. The difference curve between the two curves exhibits a larger negative value across the entire visible and near-infrared bands, a characteristic that becomes a key basis for the system to distinguish between the milky and mature stages.

[0214] During monitoring, as the summer maize plants continue to grow and their height gradually increases, the height of the spectral acquisition device relative to the plant canopy gradually decreases, causing changes in the canopy coverage observed by the narrow-angle field-of-view probe. The local processor calculates the canopy coverage of the target plant based on the reflected light signal corresponding to the narrow-angle field-of-view probe. Assuming that the calculated canopy coverage is 85%, 82%, and 79% in three consecutive acquisition cycles, showing a continuous decreasing trend, and not yet below the preset coverage threshold (e.g., 75%), the local processor recognizes this trend, generates a predictive descent command, and controls the drive motor in advance to shorten the telescopic rod, thereby lowering the spectral acquisition device to the target observation height.

[0215] This predictive adjustment mechanism avoids data loss that may occur if adjustments are made only after coverage falls below a threshold, ensuring the continuity and effectiveness of spectral data during periods of rapid growth and change.

[0216] If the canopy coverage is detected to be lower than the preset coverage threshold (e.g., due to plant lodging or rapid growth), the local processor generates a descent control command to control the drive motor to shorten the telescopic rod, thereby lowering the spectral acquisition device to the target observation height.

[0217] Conversely, if the coverage is too high, an ascent control command is generated. Through this adaptive height adjustment, the system always maintains the narrow-angle field-of-view probe focused on the core area of ​​the plant canopy, ensuring the accuracy of differential feature vector extraction and thus guaranteeing the reliability of the cloud server's prediction results.

[0218] It should be understood that the above-described monitoring scenario of summer maize at the milk stage is merely illustrative, and the system of the present invention is also applicable to monitoring different growth stages of other crops such as wheat and rice. By adjusting the installation height of the spectral acquisition device, the field of view parameters, and the classification model in the cloud server, the system can flexibly adapt to different application scenarios, demonstrating the high versatility and intelligence of the system of the present invention.

[0219] Example 8:

[0220] This embodiment, based on embodiments 1 to 7, further illustrates the adaptive monitoring method of the present invention for crops with different light adaptability.

[0221] Different crops have significantly different requirements for light intensity and duration. For example, corn and cotton are sun-loving plants that need strong direct sunlight to grow normally; while tea trees, Panax notoginseng, and ginseng are shade-loving plants and are prone to photoinhibition or sunscald under strong light. These physiological differences are directly reflected in the spectral characteristics of crops, especially the changing patterns of spectral differences between two viewing angles.

[0222] For sun-loving plants (taking corn as an example): Sun-loving plants exhibit the most active photosynthesis under strong light, with fully open stomata and high chlorophyll content. The red edge position is typically located at a longer wavelength (e.g., 720-730nm). At this time, the difference between the spectrum of the top leaves acquired by a narrow-angle field-of-view probe and the overall spectrum acquired by a wide-angle field-of-view probe is most significant. Therefore, the system can be configured to automatically increase the data acquisition frequency during midday (11:00-13:00) and assign a higher confidence weight to the data from this period. During training, the classification model in the cloud server focuses on learning the differential feature patterns of sun-loving plants under strong light conditions; for example, the difference value of the red edge feature shows a rapid increasing trend during the tasseling stage.

[0223] For shade-loving plants (taking tea as an example): Under strong direct sunlight, the leaves of shade-loving plants are prone to photoinhibition, reduced chlorophyll content, and even sunscald spots. At this time, the spectrum of the top leaves collected by the narrow-angle field-of-view probe will show abnormalities (such as an increased blue shift at the red edge), leading to an abnormally large increase in the dual-field-of-view red edge feature difference value ΔREP. However, this does not indicate that the tea plant has entered a healthy growth stage; rather, it may be a stress signal. Therefore, the system should avoid collecting data under strong midday sunlight and instead prioritize collecting data in the morning (9:00-10:00) or evening (16:00-17:00), when the light is softer and the tea leaves are in a normal physiological state, allowing the difference feature vector to more accurately reflect the growth stage. The local processor can automatically adjust the dominant collection time according to the preset crop type, reducing the collection frequency or marking data as "low confidence" during midday.

[0224] During system initialization, the user terminal can receive the currently monitored crop type, such as "sun-loving" or "shade-loving." If no crop type is received, the system defaults to a general monitoring strategy applicable to most common crops. The cloud server then distributes corresponding dynamic observation strategies to the local processor based on the crop type, including:

[0225] Data collection time period configuration: Specify the time window for priority data collection during the day;

[0226] Weight configuration: Confidence weights for data from different time periods;

[0227] Feature extraction focuses on: for sun-loving plants, the focus is on monitoring differences in red edge position and vegetation index, while for shade-loving plants, the weight of differences in water sensitivity and nutrient sensitivity is increased, because shade-loving plants are more likely to show symptoms such as nitrogen deficiency under low light conditions.

[0228] Furthermore, the cloud server can train independent classification models for sun-loving and shade-loving plants respectively, to avoid model confusion caused by differences in the feature distribution of crops with different light adaptability. Through this embodiment, the system of the present invention can not only adaptively monitor general crops, but also provide customized monitoring strategies for different light-adaptive crops, significantly enhancing the system's universality and professionalism, and is especially suitable for precision agriculture scenarios involving mixed crop cultivation.

[0229] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive monitoring system for crop growth status, characterized in that, include: A spectral acquisition device is used to acquire the reflected light signal of the target plant, including a wide-angle field-of-view probe and a narrow-angle field-of-view probe, wherein the observation area of ​​the narrow-angle field-of-view probe is located within the observation area of ​​the wide-angle field-of-view probe; A height adjustment mechanism is used to adjust the height of the spectral acquisition device relative to the target plant; The local processor is used to generate a difference feature vector characterizing the spectral difference between the two field of view based on the reflected light signal, determine the canopy coverage of the target plant, and determine an adaptive observation and adjustment strategy based on the changing trend of the canopy coverage over multiple acquisition cycles and control the lift regulator to perform height adjustment operations. The communicator is used to establish a communication connection between the local processor and the cloud server, upload the differential feature vector to the cloud server, and receive the growth status prediction results sent by the cloud server. The cloud server communicates with the local processor via a communicator to receive the differential feature vector, output the growth status prediction result of the target plant based on the pre-trained classification model, and feed the growth status prediction result back to the local processor to update the adaptive observation and adjustment strategy.

2. The system according to claim 1, characterized in that, The narrow-angle field-of-view probe's observation area points towards the top region of the target plant, used to directionally acquire the top feature reflection information of the target plant, which includes the spectral signals of the top leaves, flowers, or fruits of the target plant; the wide-angle field-of-view probe's observation area covers the entire canopy of the target plant and the surrounding background ground, used to acquire the overall reflection information of the target plant and its surrounding area, which includes the mixed spectral signals of the canopy, stem, and background ground of the target plant.

3. The system according to claim 2, characterized in that, The wide-angle field-of-view probe includes a cosine receiver to achieve a 180-degree field of view, and the narrow-angle field-of-view probe includes a focusing lens to achieve a 25-degree field of view; the ratio between the installation height of the spectral acquisition unit relative to the ground and the observation diameter of the narrow-angle field-of-view probe is 2:1, and the ratio between the installation height of the spectral acquisition unit relative to the ground and the effective observation diameter of the wide-angle field-of-view probe is 1:

3.

4. The system according to claim 1, characterized in that, The height adjustment device includes a drive motor, a telescopic rod, and a distance sensor. One end of the telescopic rod is fixedly connected to the spectral collector. The distance sensor is used to measure the real-time distance between the spectral collector and the target plant to provide a feedback control reference for height adjustment. The drive motor is used to drive the telescopic rod to extend and retract to adjust the height of the spectral collector relative to the target plant.

5. The system according to claim 4, characterized in that, The local processor is specifically used to perform the height adjustment operation, including: The canopy coverage of the target plant is calculated based on the reflected light signal corresponding to the narrow-angle field-of-view probe. The current height of the target plant is estimated based on the difference feature vector, or the current distance between the spectral acquisition device and the target plant is determined based on the real-time distance obtained by the ranging sensor, and the target observation height is calculated based on the current height or the current distance and the field of view angle of the narrow-angle field of view probe; the target observation height is the height required for the observation area of ​​the narrow-angle field of view probe to cover the core area of ​​the canopy of the target plant. If the canopy coverage is lower than a preset coverage threshold, a descent control command is generated. The descent control command is used to control the drive motor to drive the telescopic rod to shorten, so as to lower the spectral acquisition device to the target observation height. If the canopy coverage is higher than a preset coverage threshold, an ascent control command is generated. The ascent control command is used to control the drive motor to extend the telescopic rod to raise the spectral acquisition device to the target observation height. If the canopy coverage continues to decrease over multiple preset acquisition cycles, and the canopy coverage has not yet fallen below the preset coverage threshold, a predictive descent command is generated to lower the spectral acquisition device to the target observation height in advance. A minimum observation height threshold is set to control the height of the spectral acquisition device to be no lower than the minimum observation height threshold, so as to prevent the spectral acquisition device from physically colliding with the target plant and to ensure the observation integrity of the wide-angle field-of-view probe.

6. The system according to claim 1, characterized in that, The local processor is specifically used for: Based on the reflected light signal, obtain the first reflectivity data corresponding to the wide-angle field-of-view probe and the second reflectivity data corresponding to the narrow-angle field-of-view probe; The first reflectance data and the second reflectance data are preprocessed by Savitzky-Golay filtering to obtain the first preprocessed data and the second preprocessed data. Based on the first preprocessed data and the second preprocessed data, at least one of the reflectance difference, reflectance ratio, red edge feature difference, and vegetation index difference within the preset band range is calculated to form the difference feature vector. The calculation method for the red-edge feature difference value includes: performing linear fitting on the first preprocessed data and the second preprocessed data in the red-edge band region from 690nm to 750nm respectively to obtain the first red-edge feature parameter and the second red-edge feature parameter, and determining the red-edge feature difference value based on the difference between the first red-edge feature parameter and the second red-edge feature parameter; the red-edge feature parameter includes at least one of red-edge position, red-edge slope and red-edge fit goodness; The method for calculating the vegetation index difference value includes: calculating the normalized vegetation index based on the first preprocessed data and the second preprocessed data respectively to obtain the first vegetation index and the second vegetation index, and determining the vegetation index difference value based on the difference between the first vegetation index and the second vegetation index.

7. The system according to claim 6, characterized in that, The differential feature vector also includes moisture-sensitive differential features and nutrient-sensitive differential features; The water-sensitive differential feature is used to characterize the difference in water content between the canopy and the understory of the target plant in order to assess the water stress status of the target plant. The nutrient sensitivity difference feature is used to characterize the difference in chlorophyll or nitrogen nutrition between the canopy and the understory of the target plant in order to assess the nutrient deficiency status of the target plant.

8. The system according to claim 7, characterized in that, Cloud servers are specifically used for: The differential feature vector is input into the pre-trained classification model or empirical model; The growth stage label of the target plant, output by the classification model or empirical model, is obtained as the growth state prediction result.

9. The system according to claim 8, characterized in that, The growth status prediction results also include the water stress status and nutrient deficiency status of the target plant; the water stress status is determined based on the water-sensitive differential feature in the differential feature vector, and the nutrient deficiency status is determined based on the nutrient-sensitive differential feature in the differential feature vector. The cloud server is also specifically used to: generate early warning information when the water stress state or the nutrient deficiency state is abnormal, and generate agricultural decision-making suggestions based on the water stress state and the nutrient deficiency state; The warning information and agricultural decision-making suggestions are sent to the local processor or user terminal via a communicator. The agricultural decision-making recommendations include at least one of irrigation timing recommendations, fertilizer application recommendations, and harvesting time recommendations; the irrigation timing recommendations determine the appropriate irrigation time window based on the severity of the water stress state, and the fertilizer application recommendations determine the recommended fertilizer type and amount based on the type and severity of the nutrient deficiency state.

10. The system according to claim 1, characterized in that, Cloud servers are also used for: Based on the changing trend of the differential feature vector over multiple consecutive acquisition cycles, the growth state transition node of the target plant is identified; the growth state transition node is the moment when the change amplitude of the differential feature vector between adjacent acquisition cycles exceeds a preset change threshold. When the growth state transition node is identified, a dynamic observation strategy adjustment instruction is generated; the dynamic observation strategy adjustment instruction includes at least one of the following adjustments: increasing the data acquisition frequency of the spectral acquisition device to capture spectral details during the rapid change period of growth state; adjusting the height regulator to make the height of the spectral acquisition device reach the optimal observation height corresponding to the growth state transition node to ensure that the narrow-angle field-of-view probe focuses on the key organs of the current growth stage; and switching the feature extraction focus of the local processor to match the sensitive band and vegetation index type corresponding to the growth state transition node. The local processor sends the dynamic observation strategy adjustment command to the local processor via the communicator, and the local processor updates the adaptive observation adjustment strategy based on the dynamic observation strategy adjustment command.

Images