Intelligent agricultural information collection method and system

Abstract

This application provides a smart agriculture information collection method and system, relating to the field of agricultural information collection technology. The method involves: acquiring a data stream from fixed field sensors using a first acquisition mode, and acquiring a mobile platform traversal scanning data stream using a second acquisition mode; generating a synchronization trigger signal when the mobile platform enters a preset key location; extracting time window data associated with the key location from the fixed sensor data stream based on the synchronization trigger signal, and establishing a spatiotemporal correlation pair with the scanning data of the mobile platform at that location; performing data alignment and fusion within the spatiotemporal correlation pair, and outputting a site fusion snapshot. Using the scheme of this application, proactive spatiotemporal correlation of multi-source field sensing data can be achieved, thereby improving the accuracy of the agricultural condition characterization of the site fusion snapshot.

Description

Technical Field

[0001] This application relates to the field of agricultural information collection technology, and more specifically, to a smart agricultural information collection method and system. Background Technology

[0002] With the rapid development of precision agriculture and intelligent agricultural machinery, smart agriculture information collection has gradually evolved from simple manual inspections to a three-dimensional perception mode that combines fixed sensor networks with mobile platforms. By using sensors such as soil moisture and weather sensors fixed in the field, continuous time-series monitoring of the crop growth environment is achieved. At the same time, high-resolution spatial information is obtained by using mobile platforms such as drones and unmanned ground vehicles equipped with multispectral and lidar payloads. The fusion of the two can form a spatiotemporally complementary multimodal dataset of farmland, providing data support for decision-making in operations such as variable irrigation and precision fertilization.

[0003] However, fixed sensors and mobile platforms operate as two independent systems. The former continuously collects temporal parameters at a preset cycle, while the latter collects spatial data traversally along a planned path. These systems inherently lack synchronization mechanisms in terms of timing and spatial reference. In practice, the mobile platform relies on real-time dynamic positioning or near-field beacons to reach each monitoring site. However, existing methods simply perform approximate matching in the cloud using their respective collection timestamps, without establishing proactive triggering associations at the collection end. This results in unquantifiable and cumulative spatiotemporal mismatches between fixed and mobile data due to clock drift, inconsistent sampling intervals, and path execution deviations. Consequently, the environmental time-series data at the same monitoring site cannot accurately correspond to the spatial representation of the crop canopy, severely weakening the ability of the fused snapshot to reconstruct the true agricultural condition at that site. Therefore, how to achieve proactive spatiotemporal correlation of multi-source sensing data in the field to improve the accuracy of agricultural condition representation in site fusion snapshots has become a challenge for the industry. Summary of the Invention

[0004] This application provides a smart agriculture information collection method and system, which can realize active spatiotemporal correlation of multi-source sensing data in the field, thereby improving the accuracy of agricultural condition characterization of site fusion snapshots.

[0005] In a first aspect, this application provides a smart agriculture information collection method, comprising the following steps: The first acquisition mode acquires the data stream from fixed field sensors, and the second acquisition mode acquires the data stream from the mobile platform's traversal scanning. When the mobile platform is detected to have entered a preset key location, a synchronization trigger signal is generated; Based on the synchronization trigger signal, time window data associated with the key site is extracted from the fixed sensor data stream, and a spatiotemporal correlation pair is established with the scanning data of the mobile platform at that site; Within the spatiotemporal correlation pair, data alignment and fusion are performed, and a site fusion snapshot is output.

[0006] Preferably, based on the synchronization trigger signal, time window data associated with the key location is extracted from the fixed sensor data stream, and a spatiotemporal correlation pair is established with the scanning data of the mobile platform at that location, specifically including: The generation time of the aforementioned synchronization trigger signal is taken as the reference time; In the fixed sensor data stream, data within a time window centered on the reference time and within a preset duration is extracted; The spatial coordinates of the key site are bound to the scanning data collected by the mobile platform at that site; Using the spatial coordinates of the key site as the primary key, the time window data and the bound scan data are associated as the spatiotemporal association pair.

[0007] Preferably, the execution data alignment and fusion specifically includes: For the fixed sensor data stream within the spatiotemporal correlation pair, perform missing value compensation and outlier removal to obtain preprocessed continuous time-series data; For the mobile platform scanning data within the spatiotemporal correlation pair, geographic coordinate registration and noise filtering are performed to obtain preprocessed spatial distribution data; The preprocessed continuous time-series data and the preprocessed spatial distribution data are used together as inputs for data alignment and fusion.

[0008] Preferably, performing data alignment and fusion also includes: In preprocessed continuous time series data, identify time series trend characteristics within a time window; In the preprocessed spatial distribution data, the spatial distribution features corresponding to key sites are extracted; Based on the time series trend features and the spatial distribution features, dynamic time warping is performed to align the two types of features on the time axis.

[0009] Preferably, dynamic time warping is performed, specifically including: Construct a matching cost matrix between time series trend features and spatial distribution features with respect to the collection time. Search the matching cost matrix for the normalized path with the minimum cumulative cost to obtain the optimal matching relationship at each acquisition time. Based on the optimal matching relationship, the fixed sensor data stream and the mobile platform scanning data are resampled and aligned.

[0010] Preferably, after resampling and aligning the fixed sensor data stream and the mobile platform scanning data according to the optimal matching relationship, the method further includes: Calculate the local similarity at each matching time point for the aligned fixed sensor data and the mobile platform scan data; Data points corresponding to matching times with local similarity below a preset threshold are filtered out to obtain a refined set of matching pairs.

[0011] Preferably, the output site fusion snapshot specifically includes: A set of multidimensional agricultural feature vectors is generated for each of the key sites. The multidimensional agricultural feature vectors include soil parameter statistics, meteorological environment statistics, and crop growth characterization. The geographic coordinates of the key sites, the multidimensional agricultural feature vectors, and the corresponding timestamps are encapsulated into a structured site fusion snapshot. According to the traversal path order of the mobile platform, the multiple site fusion snapshots are stitched together into a farmland fusion snapshot sequence and output.

[0012] Secondly, this application provides a smart agricultural information collection system, comprising: The data acquisition module is used to acquire field fixed sensor data streams in a first acquisition mode and mobile platform traversal scanning data streams in a second acquisition mode. The processing module is used to generate a synchronization trigger signal when the mobile platform is detected to have entered a preset key location; The processing module is also used to extract time window data associated with the key site from the fixed sensor data stream according to the synchronization trigger signal, and establish a spatiotemporal correlation pair with the scanning data of the mobile platform at the site; The execution module is used to perform data alignment and fusion within the spatiotemporal correlation pair and output a site fusion snapshot.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described smart agriculture information collection method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described smart agriculture information collection method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application constructs an active spatiotemporal correlation mechanism using key locations as spatial anchors and synchronization trigger signals as temporal references. Combined with feature-level dynamic time warping and point-by-point quality refinement, it achieves high-precision fusion snapshot output of multi-source sensing data in the field. First, by pre-setting key locations in the field, continuous temporal acquisition from fixed sensors and traversal spatial scanning by a mobile platform are unified into the same spatial reference system, providing a common anchoring reference for the two types of heterogeneous data. Second, when the mobile platform enters a key location, a synchronization signal is generated based on this spatial event. Using this signal as a reference time, time window data is extracted from the fixed sensor data stream, transforming spatial position changes into precise temporal extraction instructions, eliminating temporal mismatches caused by clock drift and sampling rhythm differences at the source of acquisition. Then, using the spatial coordinates of the key locations as the primary key, the time window data and the mobile platform scanning data are associated as spatiotemporal correlation pairs, ensuring that the temporal environmental parameters of the same monitoring location are consistent. A one-to-one pairing data unit is formed with the spatial crop representation. Based on this, preprocessing is performed on the two types of data within the spatiotemporal correlation pair, and dynamic time warping is performed between the time series trend features and spatial distribution features. By constructing a matching cost matrix, the warping path with the minimum cumulative cost is searched to obtain the optimal matching relationship and complete resampling alignment. Then, low reliability matching points are screened out by local similarity to obtain a refined set of matching pairs. Finally, multidimensional agricultural feature vectors are generated based on the refined matching pairs, and encapsulated with geographic coordinates and timestamps into structured site fusion snapshots. These snapshots are then spliced ​​together in the order of the traversal path to form a farmland fusion snapshot sequence and output. In summary, this application realizes active spatiotemporal correlation of multi-source sensing data in the field through the full-link collaboration of unified spatial anchor points, event-driven triggering, primary key association pairing, feature-level alignment refinement, and structured snapshot output, thereby improving the accuracy of site fusion snapshots in restoring the true state of agricultural conditions. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of a smart agriculture information collection method according to some embodiments of this application; Figure 2 This is an exemplary flowchart of a smart agriculture information collection method according to some embodiments of this application; Figure 3 This is a schematic diagram of the process for establishing spatiotemporal association pairs according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a smart agriculture information collection system according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a smart agriculture information collection method according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] refer to Figure 1 This figure is a schematic diagram of an application scenario of a smart agriculture information collection method according to some embodiments of this application. The figure includes two main components: a collection end and a server. The collection end is used to acquire field fixed sensor data streams in a first collection mode and mobile platform traversal scanning data streams in a second collection mode. When the mobile platform is detected to enter a preset key site, a synchronization trigger signal is generated, and the fixed sensor data stream, mobile platform scanning data stream, and synchronization trigger signal are sent to the server through a communication network. The server is used to extract time window data associated with the key site from the fixed sensor data stream according to the synchronization trigger signal, establish a spatiotemporal correlation pair with the mobile platform scanning data at that site, perform data alignment and fusion within the spatiotemporal correlation pair, and output a site fusion snapshot.

[0019] refer to Figure 2 The figure is an exemplary flowchart of a smart agriculture information collection method according to some embodiments of this application. The smart agriculture information collection method mainly includes the following steps: In step 101, a field fixed sensor data stream is acquired using a first acquisition mode, and a mobile platform traversal scan data stream is acquired using a second acquisition mode.

[0020] It should be noted that this application uses orchard agricultural information collection as an example for illustration, but this example is only used to clearly illustrate the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. In the orchard scenario, due to the significant shading effect of the canopy layer, it is difficult to accurately correlate the data between the ground sensors in the rows and the canopy layer in space, and the problem of spatiotemporal synchronization of multimodal data is particularly prominent. Selecting this scenario can fully demonstrate the technical advantages of the present invention in triggering synchronization by key points and establishing spatiotemporal correlation. Those skilled in the art should understand that, based on the same inventive concept, the technical solution of this application can be equally applied to other smart agriculture scenarios such as field crops and greenhouses. The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] In some embodiments, acquiring field fixed sensor data streams in a first acquisition mode and acquiring mobile platform traversal scanning data streams in a second acquisition mode specifically includes: Within the orchard planting area, data streams uploaded by fixed soil moisture sensor arrays and canopy microclimate monitoring nodes deployed in a grid pattern are acquired using the first acquisition mode. During the preset period from fruit enlargement to maturity, the second acquisition mode is used to acquire the canopy spectral data stream and point cloud data stream obtained by the unmanned ground vehicle equipped with a multispectral camera and lidar, which scans along the fruit tree rows in a preset path. The preset key sites correspond to the soil sampling points and canopy monitoring node deployment locations between each row of fruit trees.

[0022] In practice, firstly, within the orchard planting area, the area is pre-divided into multiple preset grids based on the row spacing and plant spacing of the fruit trees. Soil sampling points are determined at the center point or diagonal intersection of each preset grid, and a fixed soil moisture sensor array is buried at each sampling point. This fixed soil moisture sensor array includes multiple soil moisture and temperature sensor probes arranged in layers along the soil profile. The burial depth of each probe at least covers the main root distribution layer. These multiple soil moisture and temperature sensor probes are connected to a data acquisition terminal located on the ground via cables. The data acquisition terminal has a built-in wireless communication module and a waterproof casing. Next, in the middle of the canopy of each row or at predetermined intervals, canopy monitoring node deployment positions are selected. At these positions, the main trunk or branches of the fruit trees... A canopy microclimate monitoring node is installed on the support frame. The canopy microclimate monitoring node includes an air temperature and humidity sensor and a light sensor integrated into a radiation shield. The fixed soil moisture sensor array and the canopy microclimate monitoring node continuously collect soil moisture content and temperature parameters at different depths, as well as air temperature, relative humidity, and light intensity parameters at the canopy height, according to a preset first sampling period. The first sampling period ranges from 5 minutes to 30 minutes. The collected parameters are packaged into fixed sensor data packets according to the collection timestamp and the corresponding device identifier, and uploaded via a wireless communication gateway deployed in the orchard. The wireless communication gateway is deployed at the intersection of each preset grid and equipped with a high-gain antenna to form the fixed sensor data stream.

[0023] Then, the period from fruit enlargement to maturity was selected as the monitoring period. Within this period, a scanning cycle was defined as one day or every preset interval of 3 to 7 days. In each scanning cycle, an unmanned ground vehicle equipped with a multispectral camera and lidar, featuring a real-time dynamic differential positioning module and an inertial navigation unit, autonomously navigated between rows of fruit trees according to a pre-planned traversal path. This pre-planned traversal path was generated by a data fusion processing unit based on the coordinate sequence of each key location and the passable area between rows of fruit trees, and then sent to the path planning module of the unmanned ground vehicle. The traversal path covered all the soil sampling points and the canopy monitoring nodes located in the inter-row passageways. During the journey, the multispectral camera... A camera continuously acquires canopy spectral images including red and near-infrared bands at a preset frame rate. The lens of the multispectral camera is mounted vertically or tilted upwards. The preset frame rate ranges from 1 to 5 frames per second, forming a canopy spectral data stream. A lidar synchronously acquires three-dimensional point clouds of the fruit tree canopy at a preset scanning frequency. The lidar is mounted on the top of the unmanned ground vehicle, and the scanning frequency ranges from 10 Hz to 20 Hz, forming a point cloud data stream. The locations of each soil sampling point and canopy monitoring node are pre-marked as preset key points by recording their spatial coordinates in the global navigation satellite system coordinate system and storing them in the path planning module of the unmanned ground vehicle. This unifies the spatial reference of the fixed sensor data stream acquisition source and the mobile platform scanning data stream into the same key point system, so as to trigger synchronization and association in the future.

[0024] In step 102, when the mobile platform is detected to have entered a preset key location, a synchronization trigger signal is generated.

[0025] In this embodiment, the real-time dynamic differential positioning module of the unmanned ground vehicle outputs the current positioning coordinates of the unmanned ground vehicle at a preset frequency. The path planning module of the unmanned ground vehicle pre-stores the spatial coordinates of each key point and delineates an electronic fence area with a preset radius centered on the spatial coordinates of each key point. The preset radius ranges from 0.5 meters to 2 meters. When the unmanned ground vehicle travels along the traversal path, if the current positioning coordinates output by the real-time dynamic differential positioning module fall within the electronic fence area corresponding to any key point, it is determined that the unmanned ground vehicle has entered that key point. At this time, the vehicle-mounted control unit of the unmanned ground vehicle generates the synchronization trigger signal. The synchronization trigger signal carries the identification information of the key point and the trigger timestamp, and is sent to the data fusion processing unit through the wireless communication gateway deployed in the orchard to trigger subsequent time window data interception and the establishment of spatiotemporal correlation pairs.

[0026] In other embodiments, near-field communication (NFC) beacons are pre-deployed at each key location. These NFC beacons are at least one of RFID tags or Bluetooth Low Energy (BLE) beacons, and each NFC beacon stores unique identification information corresponding to the key location. The unmanned ground vehicle (UAV) is equipped with a reader or scanning module matched to the NFC beacon. When the UAV travels along the traversal path and enters the effective communication range of a NFC beacon deployed at a key location, the reader or scanning module on the UAV reads the unique identification information stored in the NFC beacon and detects the received signal strength. When the received signal strength exceeds a preset signal strength threshold, it is determined that the UAV has entered the key location. At this time, the onboard control unit of the UAV packages the unique identification information of the key location with the current timestamp to generate the synchronization trigger signal, which is then sent to the data fusion processing unit via a wireless communication gateway deployed in the orchard.

[0027] In step 103, based on the synchronization trigger signal, time window data associated with the key site is extracted from the fixed sensor data stream, and a spatiotemporal correlation pair is established with the scanning data of the mobile platform at that site.

[0028] In some embodiments, reference Figure 3 As shown in the figure, this is a schematic flowchart illustrating the establishment of spatiotemporal correlation pairs according to some embodiments of this application. In this embodiment, based on the synchronization trigger signal, time window data associated with the key site is extracted from the fixed sensor data stream, and a spatiotemporal correlation pair is established with the scanning data of the mobile platform at that site. Specifically, this includes: In step 1031, the generation time of the synchronization trigger signal is used as the reference time; In step 1032, data within a time window centered on the reference time and within a preset duration is extracted from the fixed sensor data stream. In step 1033, the spatial coordinates of the key site are bound to the scanning data collected by the mobile platform at that site; In step 1034, the spatial coordinates of the key site are used as the primary key to associate the time window data with the bound scan data to form the spatiotemporal association pair.

[0029] It should be noted that, in this application, the spatiotemporal correlation pair refers to a paired data unit formed by associating fixed sensor temporal dimension monitoring data and mobile platform spatial dimension scanning data under the same key location, with the spatial coordinates of the key location as the primary key.

[0030] It should also be noted that this application, through the aforementioned spatiotemporal correlation pair establishment mechanism using key locations as spatial anchors and the generation time of the synchronization trigger signal as the time series reference, achieves active spatiotemporal synchronization of two types of heterogeneous data—fixed sensors and mobile platforms—at the data acquisition source. This avoids data mismatch caused by the accumulation of clock deviations and the inability to align sampling times, which is a problem in traditional post-event timestamp alignment methods. This enables time-dimensional monitoring data and spatial-dimensional scanning data to be correlated and fused under high-precision spatiotemporal registration, significantly improving the accuracy and reliability of multimodal agricultural data fusion. It provides a highly reliable data foundation for precise agricultural diagnosis and variable-based operational decision-making based on fused snapshots.

[0031] In specific implementation, the data fusion processing unit receives the synchronization trigger signal, extracts the identification information of the key point and the trigger timestamp carried in the synchronization trigger signal, and determines the time marked by the trigger timestamp as the reference time. Centered on the reference time, a continuous data segment with a time span equal to a preset duration is extracted from the received and cached or stored fixed sensor data stream; this continuous data segment is the time window data. The preset duration is determined based on the first sampling period of the fixed sensor and the rate of change of agricultural parameters at the key location. The preset duration ranges from 5 minutes to 30 minutes, and in this embodiment, it is set to 10 minutes. That is, with the reference time as the midpoint, 5 minutes of continuous data collection from the fixed sensor are taken before and after the reference time, which together constitute the time window data. At the same time, the spatial coordinates corresponding to the identification information of the key location are extracted from the synchronization trigger signal, or the spatial coordinates of the key location are retrieved from the pre-stored list of key location coordinates based on the identification information of the key location. The spatial coordinates of the key location are then bound one by one with each frame of spectral image in the canopy spectral data stream and each frame of point cloud in the point cloud data stream collected by the mobile platform at the key location. The spatial coordinates of the key location are attached to each frame of spectral image and each frame of point cloud as a spatial reference label, thereby converting the scanning data collected by the mobile platform at the key location into the bound scanning data with spatial location tags. Using the spatial coordinates of the key site as the primary key, the time window data corresponding to the key site and the bound scan data corresponding to the key site are associated and stored together to form a spatiotemporal association pair in which the time dimension data and the spatial dimension data of the key site correspond one-to-one, so that subsequent data alignment and fusion operations can be performed within the spatiotemporal association pair.

[0032] In step 104, within the spatiotemporal correlation pair, data alignment and fusion are performed, and a site fusion snapshot is output.

[0033] In some embodiments, performing data alignment and fusion specifically includes: For the fixed sensor data stream within the spatiotemporal correlation pair, perform missing value compensation and outlier removal to obtain preprocessed continuous time-series data; For the mobile platform scanning data within the spatiotemporal correlation pair, geographic coordinate registration and noise filtering are performed to obtain preprocessed spatial distribution data; The preprocessed continuous time-series data and the preprocessed spatial distribution data are used together as inputs for data alignment and fusion.

[0034] In specific implementation, firstly, the data fusion processing unit performs missing value compensation and outlier removal on the fixed sensor data streams within the established spatiotemporal correlation pair. It iterates through the soil moisture content, soil temperature, air temperature, relative humidity, and light intensity parameter sequences in each fixed sensor data packet contained in the fixed sensor data stream, detecting missing moments caused by temporary sensor power outages or wireless communication packet loss. For missing moments with a continuous missing duration less than a first preset missing tolerance threshold, a linear interpolation method based on adjacent valid data before and after the missing moment is used to fill in the missing parameter values, obtaining the compensated parameters. The sequence, wherein the first preset missing tolerance threshold ranges from 2 to 5 first sampling periods; simultaneously, the sliding window median method is used to identify abnormal jump values ​​in each parameter sequence that exceed the reasonable range of preset parameters or deviate from the median of the sliding window by more than a preset multiple of the median absolute deviation, and the identified abnormal jump values ​​are replaced with the median value of the data within the sliding window to obtain the parameter sequence after outlier removal. The parameter sequences after the above missing value compensation and outlier removal processing are rearranged according to the original acquisition timestamp to form the preprocessed continuous time series data that is continuous in time and has no abnormal jumps.

[0035] Then, the data fusion processing unit performs geographic coordinate registration and noise filtering on the mobile platform scanning data within the spatiotemporal correlation pair. This mobile platform scanning data includes canopy spectral data streams and point cloud data streams. From the motion trajectory data recorded by the unmanned ground vehicle's path planning module or real-time dynamic differential positioning module, the unit obtains the real-time positioning coordinates and attitude angles of the unmanned ground vehicle when acquiring each frame of spectral image and each frame of point cloud at the key location. Using the spatial coordinates of the key location as a reference, the unit performs spatial position correction on each frame of spectral image and each frame of point cloud, and integrates the local coordinates of each frame of spectral image and each frame of point cloud relative to the unmanned ground vehicle's body coordinate system. The coordinates are converted to global coordinates with the spatial coordinates of the key site as the origin, thus completing geographic coordinate registration. At the same time, Gaussian smoothing filtering is performed on each frame of spectral image after geographic coordinate registration to reduce dark current noise and salt-and-pepper noise of the image sensor. For each frame of point cloud after geographic coordinate registration, statistical outlier removal is performed. Points whose average distance from a preset number of neighboring points exceeds a preset standard deviation are removed as noise points, thus completing noise filtering. The canopy spectral image and point cloud after geographic coordinate registration and noise filtering are rearranged according to their respective acquisition timestamps to form the preprocessed spatial distribution data.

[0036] Finally, the data fusion processing unit uses the preprocessed continuous time-series data and the preprocessed spatial distribution data together as inputs for data alignment and fusion.

[0037] Preferably, in some embodiments, performing data alignment and fusion further includes: In preprocessed continuous time series data, identify time series trend characteristics within a time window; In the preprocessed spatial distribution data, the spatial distribution features corresponding to key sites are extracted; Based on the time series trend features and the spatial distribution features, dynamic time warping is performed to align the two types of features on the time axis.

[0038] In specific implementation, the identification of time series trend features within a time window can be achieved in the following way: The data fusion processing unit identifies the time series trend features within the time window from the obtained preprocessed continuous time series data. The time window is divided into multiple consecutive preset sub-windows. The duration of each preset sub-window is determined according to the first sampling period and the rate of change of agricultural parameters. Within each preset sub-window, the mean, standard deviation, slope of change, and cumulative change of the soil moisture content sequence, soil temperature sequence, air temperature sequence, relative humidity sequence, and light intensity sequence in the preprocessed continuous time series data are calculated respectively. The slope of change is obtained by performing a sliding window linear fitting on the data points within the preset sub-window. The mean, standard deviation, slope of change, and cumulative change calculated within each preset sub-window are concatenated to form the time series statistical feature sub-vector corresponding to the preset sub-window. The time series statistical feature sub-vectors of all preset sub-windows within the time window are combined in chronological order to form the time series trend feature that characterizes the overall change direction and evolution law of each agricultural parameter within the time window.

[0039] In specific implementation, the extraction of spatial distribution features corresponding to key sites can be achieved in the following way: Extract the spatial distribution features corresponding to the key sites from the obtained preprocessed spatial distribution data. Delineate a preset spatial neighborhood range centered on the spatial coordinates of the key sites. The radius of this spatial neighborhood range is determined based on the canopy width of the fruit trees and the planting row spacing. Extract multispectral reflectance data from the canopy spectral image in the preprocessed spatial distribution data within this spatial neighborhood range. Calculate the average reflectance in the red band and the average reflectance in the near-infrared band, and calculate the normalized vegetation index based on these two values. Simultaneously, calculate the spatial variation coefficient of the normalized vegetation index for each pixel within this spatial neighborhood range. On the point cloud in the preprocessed spatial distribution data, the point cloud height quantile, point cloud density, and canopy volume estimate within the spatial neighborhood are extracted as canopy three-dimensional geometric structure features. The normalized vegetation index, the spatial variation coefficient of the normalized vegetation index, the point cloud height quantile, the point cloud density, and the canopy volume estimate are then concatenated to form the spatial distribution feature that characterizes the spatial distribution differences of the crop canopy spectral reflectance characteristics and canopy three-dimensional geometric structure within the spatial neighborhood corresponding to the key site.

[0040] Preferably, in some embodiments, performing dynamic time warping specifically includes: Construct a matching cost matrix between time series trend features and spatial distribution features with respect to the collection time. Search the matching cost matrix for the normalized path with the minimum cumulative cost to obtain the optimal matching relationship at each acquisition time. Based on the optimal matching relationship, the fixed sensor data stream and the mobile platform scanning data are resampled and aligned.

[0041] It should be noted that the matching cost matrix in this application is used to measure the temporal deviation cost between the temporal statistical feature sub-vector of each preset sub-window in the time series trend feature and the spatial feature vector of each collection time in the spatial distribution feature; the optimal matching relationship refers to the point-to-point correspondence sequence of the time series trend feature and the spatial distribution feature on the time axis established by the regular path with the minimum cumulative cost from the starting point to the ending point in the matching cost matrix.

[0042] In specific implementation, the matching cost matrix between the time series trend features and the spatial distribution features with respect to the acquisition time can be constructed in the following way: the data fusion processing unit constructs the matching cost matrix between the time series trend features and the spatial distribution features with respect to the acquisition time, arranges the time series statistical feature sub-vectors of each preset sub-window contained in the time series trend features in the chronological order of the preset sub-windows as the first feature sequence, and arranges the spatial feature vectors of each acquisition time contained in the spatial distribution features in the chronological order of the acquisition time as the second feature sequence. Calculate the Euclidean or Manhattan distance between the temporal statistical feature vector of each preset sub-window in the first feature sequence and the spatial feature vector of each acquisition time in the second feature sequence. Use this distance as the matching cost between the preset sub-window and the acquisition time. Arrange the pairwise matching costs between all preset sub-windows and all acquisition times into a two-dimensional matrix according to the sequence number of the first feature sequence and the sequence number of the second feature sequence. The number of rows in the two-dimensional matrix is ​​equal to the total number of preset sub-windows in the first feature sequence, and the number of columns in the two-dimensional matrix is ​​equal to the total number of acquisition times in the second feature sequence. Each element in the two-dimensional matrix is ​​the matching cost between the corresponding preset sub-window and the corresponding acquisition time, thus forming the matching cost matrix.

[0043] In specific implementation, searching for the regularized path with the minimum cumulative cost in the matching cost matrix to obtain the optimal matching relationship at each acquisition time can be achieved as follows: Starting from the starting point of the matching cost matrix and ending at its endpoint, a dynamic programming algorithm is used to calculate the cumulative cost of the path traversed from the starting point to each element in the matching cost matrix, and the predecessor elements traversed to reach that cumulative cost are recorded. The path with the minimum cumulative cost to reach the endpoint of the matching cost matrix is ​​determined as the regularized path. Each step on this regularized path corresponds to a matching pair between a preset sub-window in the first feature sequence and a certain acquisition time in the second feature sequence. The regularized path contains a complete set of matching relationships between all preset sub-windows in the first feature sequence and all acquisition times in the second feature sequence under the condition of minimum cumulative cost. Each matching pair contained in this regularized path is taken as the optimal matching relationship at each acquisition time.

[0044] In specific implementation, the resampling and alignment of the fixed sensor data stream and the mobile platform scanning data based on the optimal matching relationship can be achieved in the following way: Based on the obtained optimal matching relationship, the fixed sensor data stream and the mobile platform scanning data are resampled and aligned. Each matching pair in the optimal matching relationship is traversed, and the fixed sensor acquisition time range corresponding to the preset sub-window and the mobile platform acquisition time corresponding to the spatial feature vector are obtained for each matching pair. For each fixed sensor data point within the fixed sensor acquisition time range, based on the time deviation between the mobile platform acquisition time and the fixed sensor acquisition time range in the matching pair, the corresponding fixed sensor parameter estimate is obtained by interpolation or extrapolation at the mobile platform acquisition time using a linear interpolation method. Simultaneously, for the mobile platform scanning data point at the mobile platform acquisition time, based on the same matching relationship, the corresponding mobile platform parameter value is obtained at each acquisition time within the fixed sensor acquisition time range using a nearest neighbor interpolation method. Therefore, the fixed sensor data stream and the mobile platform scanning data, which were originally collected at inconsistent times, are converted into aligned data with the same timestamp at each collection time under a unified time reference, and the fixed sensor data stream and the mobile platform scanning data are resampled and aligned.

[0045] Preferably, in some embodiments, after resampling and aligning the fixed sensor data stream and the mobile platform scanning data according to the optimal matching relationship, the method further includes: Calculate the local similarity at each matching time point for the aligned fixed sensor data and the mobile platform scan data; Data points corresponding to matching times with local similarity below a preset threshold are filtered out to obtain a refined set of matching pairs.

[0046] The local similarity in this application is a quantitative indicator used to measure the degree of consistency in numerical change patterns between fixed sensor data and mobile platform scan data at the same matching time after resampling and alignment.

[0047] Specifically, the implementation involves the following steps: First, the data fusion processing unit calculates the local similarity at each matching time point for the resampled and aligned fixed sensor data and the mobile platform scan data. Iterating through each matching time point in the optimal matching relationship, at that matching time point, a data window from the resampled and aligned fixed sensor data and a data window from the mobile platform scan data are extracted. The window width of the data window is determined based on the first sampling period of the fixed sensor and the acquisition frame rate of the mobile platform scan data. The parameter sequences within the fixed sensor data window and the parameter sequences within the mobile platform scan data window are respectively subjected to mean normalization processing to eliminate dimensional differences between the two types of data. The Pearson correlation coefficient between the parameter sequences within the mean-normalized fixed sensor data window and the parameter sequences within the mobile platform scan data window is calculated, and the absolute value of this Pearson correlation coefficient is used as the local similarity at that matching time point. The local similarity value ranges from 0 to 1. The closer the local similarity value is to one, the more consistent the change patterns of the fixed sensor data and the mobile platform scanning data are at that matching time. Next, after calculating the local similarity of each matching time, the local similarity of each matching time is compared with a preset threshold, which ranges from 0.6 to 0.8. Matching times with local similarity lower than the preset threshold are marked as low-reliability matching times. The fixed sensor data points and mobile platform scanning data points corresponding to the low-reliability matching times are removed from the matching pair set established by the optimal matching relationship. The fixed sensor data points and mobile platform scanning data points corresponding to each matching time after removal are recombined into the refined matching pair set. The refined matching pair set is used as the data input for generating the site fusion snapshot.

[0048] In some embodiments, the output site fusion snapshot specifically includes: A set of multidimensional agricultural feature vectors is generated for each of the key sites. The multidimensional agricultural feature vectors include soil parameter statistics, meteorological environment statistics, and crop growth characterization. The geographic coordinates of the key sites, the multidimensional agricultural feature vectors, and the corresponding timestamps are encapsulated into a structured site fusion snapshot. According to the traversal path order of the mobile platform, the multiple site fusion snapshots are stitched together into a farmland fusion snapshot sequence and output.

[0049] Specifically, the implementation involves the following steps: First, the data fusion processing unit generates a set of multidimensional agricultural feature vectors for each key location. From the refined matching pair set, resampled and aligned fixed sensor data corresponding to the key location is extracted. The mean soil moisture content and mean soil temperature of the fixed sensor data within the time window are calculated as soil parameter statistics. The mean air temperature, mean relative humidity, and cumulative light intensity of the fixed sensor data within the time window are also calculated as meteorological environmental statistics. From the refined matching pair set, resampled and aligned mobile platform scanning data corresponding to the key location is extracted. Based on canopy spectral data, the mean normalized vegetation index (NVI) and the spatial variation coefficient of the NVI are calculated. Based on point cloud data, the mean canopy height, canopy point cloud density, and estimated canopy volume are calculated, collectively serving as crop growth characterization metrics. The aforementioned soil parameter statistics, meteorological environmental statistics, and crop growth characterization metrics are then concatenated according to a preset feature arrangement order to form a set of multidimensional agricultural feature vectors corresponding to the key location. Secondly, the geographic coordinates of the key location, the multidimensional agricultural condition feature vector, and the corresponding timestamp are encapsulated into a structured location fusion snapshot. Using the identification information of the key location as an index, the geographic coordinates of the key location as a spatial field, the multidimensional agricultural condition feature vector as a feature field, and the timestamp of the reference time as a time field, these are combined according to a preset data structure format to form the structured location fusion snapshot containing spatial, feature, and time fields. This structured location fusion snapshot serves as a standardized agricultural condition information recording unit for the key location at the reference time, and can be directly read and accessed by the farmland management decision-making system. Then, following the traversal path order of the mobile platform, multiple site fusion snapshots are stitched together into a farmland fusion snapshot sequence and output. The actual traversal path trajectory data of the unmanned ground vehicle within this scanning cycle is obtained. From this actual traversal path trajectory data, the passing order of each key site traversed by the unmanned ground vehicle is extracted. The structured site fusion snapshots corresponding to each key site are arranged sequentially according to this passing order, forming a time-ordered list of site fusion snapshots. This list of site fusion snapshots is serialized into the farmland fusion snapshot sequence. This farmland fusion snapshot sequence is output to a field management terminal or a cloud-based farmland management platform via a wireless communication gateway to generate a spatiotemporal distribution map of the overall farmland conditions corresponding to this scanning cycle.

[0050] On the other hand, in some embodiments, this application provides a smart agricultural information collection system, with reference to... Figure 4The figure is a schematic diagram of the structure of a smart agriculture information collection system according to some embodiments of this application. The smart agriculture information collection system 400 includes: a collection module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire field fixed sensor data stream in the first acquisition mode and mobile platform traversal scanning data stream in the second acquisition mode. Processing module 402, in this application, is used to generate a synchronization trigger signal when the mobile platform is detected to have entered a preset key position; In this application, the processing module 402 is also used to extract time window data associated with the key site from the fixed sensor data stream according to the synchronization trigger signal, and establish a spatiotemporal correlation pair with the scanning data of the mobile platform at the site; The execution module 403 in this application is mainly used to perform data alignment and fusion within the spatiotemporal correlation pair and output a site fusion snapshot.

[0051] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described smart agriculture information collection method.

[0052] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a smart agriculture information collection method according to some embodiments of this application. The smart agriculture information collection method in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0053] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0054] The communication bus 502 can be used to transmit information between the aforementioned components.

[0055] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0056] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. In the above embodiments, the smart agriculture information acquisition method can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0057] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0058] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0059] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0060] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described smart agriculture information collection method.

[0061] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0062] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A smart agriculture information collection method, characterized in that, Includes the following steps: The first acquisition mode acquires the data stream from fixed field sensors, and the second acquisition mode acquires the data stream from the mobile platform's traversal scanning. When the mobile platform is detected to have entered a preset key location, a synchronization trigger signal is generated; Based on the synchronization trigger signal, time window data associated with the key site is extracted from the fixed sensor data stream, and a spatiotemporal correlation pair is established with the scanning data of the mobile platform at that site; Within the spatiotemporal correlation pair, data alignment and fusion are performed, and a site fusion snapshot is output.

2. The method as described in claim 1, characterized in that, Based on the synchronization trigger signal, time window data associated with the key location is extracted from the fixed sensor data stream, and a spatiotemporal correlation pair is established with the scanning data of the mobile platform at that location, specifically including: The generation time of the aforementioned synchronization trigger signal is taken as the reference time; In the fixed sensor data stream, data within a time window centered on the reference time and within a preset duration is extracted; The spatial coordinates of the key site are bound to the scanning data collected by the mobile platform at that site; Using the spatial coordinates of the key site as the primary key, the time window data and the bound scan data are associated as the spatiotemporal association pair.

3. The method as described in claim 1, characterized in that, The execution data alignment and fusion specifically includes: For the fixed sensor data stream within the spatiotemporal correlation pair, perform missing value compensation and outlier removal to obtain preprocessed continuous time-series data; For the mobile platform scanning data within the spatiotemporal correlation pair, geographic coordinate registration and noise filtering are performed to obtain preprocessed spatial distribution data; The preprocessed continuous time-series data and the preprocessed spatial distribution data are used together as inputs for data alignment and fusion.

4. The method as described in claim 3, characterized in that, Performing data alignment and fusion also includes: In preprocessed continuous time series data, identify time series trend characteristics within a time window; In the preprocessed spatial distribution data, the spatial distribution features corresponding to key sites are extracted; Based on the time series trend features and the spatial distribution features, dynamic time warping is performed to align the two types of features on the time axis.

5. The method as described in claim 4, characterized in that, Performing dynamic time warping specifically includes: Construct a matching cost matrix between time series trend features and spatial distribution features with respect to the collection time. Search the matching cost matrix for the normalized path with the minimum cumulative cost to obtain the optimal matching relationship at each acquisition time. Based on the optimal matching relationship, the fixed sensor data stream and the mobile platform scanning data are resampled and aligned.

6. The method as described in claim 5, characterized in that, Based on the optimal matching relationship, after resampling and aligning the fixed sensor data stream and the mobile platform scanning data, the method further includes: Calculate the local similarity at each matching time point for the aligned fixed sensor data and the mobile platform scan data; Data points corresponding to matching times with local similarity below a preset threshold are filtered out to obtain a refined set of matching pairs.

7. The method as described in claim 1, characterized in that, Output site fusion snapshots, specifically including: A set of multidimensional agricultural feature vectors is generated for each of the key sites. The multidimensional agricultural feature vectors include soil parameter statistics, meteorological environment statistics, and crop growth characterization. The geographic coordinates of the key sites, the multidimensional agricultural feature vectors, and the corresponding timestamps are encapsulated into a structured site fusion snapshot. According to the traversal path order of the mobile platform, the multiple site fusion snapshots are stitched together into a farmland fusion snapshot sequence and output.

8. A smart agricultural information collection system, characterized in that, include: The data acquisition module is used to acquire field fixed sensor data streams in a first acquisition mode and mobile platform traversal scanning data streams in a second acquisition mode. The processing module is used to generate a synchronization trigger signal when the mobile platform is detected to have entered a preset key location; The processing module is also used to extract time window data associated with the key site from the fixed sensor data stream according to the synchronization trigger signal, and establish a spatiotemporal correlation pair with the scanning data of the mobile platform at the site; The execution module is used to perform data alignment and fusion within the spatiotemporal correlation pair and output a site fusion snapshot.

9. A computer device comprising a memory and a processor, the memory storing code, characterized in that, The processor is configured to acquire the code and execute the smart agriculture information acquisition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the smart agriculture information collection method as described in any one of claims 1 to 7.

Images