Wheat drilling operation quality post-evaluation method based on unmanned aerial vehicle

Abstract

The invention discloses a wheat drilling operation quality post-evaluation method based on an unmanned aerial vehicle, and relates to the technical field of agricultural informatization processing, and the method comprises the steps: obtaining an unmanned aerial vehicle multiband remote sensing image after wheat seedling emergence in a target land parcel, and carrying out the preprocessing of the remote sensing image, and obtaining an orthoimage map; inputting the orthoimage into a pre-trained seedling strip recognition model, recognizing and outputting a semantic segmentation result of a seedling strip, the seedling strip recognition model being constructed based on an SEANet architecture; post-processing the semantic segmentation result to generate seedling strip data with vector boundaries, and extracting vector data of each seedling strip based on the vector boundaries; according to the vector data, calculating a core index for evaluating the quality of the drilling operation; and comparing the core index with an operation quality standard to perform grading evaluation on the drilling operation quality of the target land parcel, so that accurate and efficient quantitative evaluation can be performed on the wheat drilling operation quality.

Description

Technical Field

[0001] This application relates to the technical field of agricultural information processing, and in particular to a method for post-evaluation of the quality of wheat strip seeding operations based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Currently, the quality of wheat row sowing is fundamental to constructing a reasonable plant population structure, directly determining the uniformity of emergence and ultimately affecting the realization of crop yield potential. The heterogeneity of seedling strip spatial distribution leads to unbalanced competition among plants for light, water, and nutrients: localized overcrowding hinders individual development, increasing the risk of lodging and disease. Many complex factors influence emergence uniformity, including mechanical operating parameters such as the stability of the sowing machine's seeding and the consistency of row spacing.

[0003] In related technologies, the evaluation of wheat row sowing quality in the field mainly relies on manual sampling surveys. This method involves setting up quadrats in a limited representative area, manually counting seedlings, measuring broken strip length and row spacing, and then calculating statistical indicators such as the coefficient of variation. However, this method has significant drawbacks: such as low efficiency and high cost, it is difficult to meet the evaluation needs of large-scale planting. Due to the limitations of the sampling space, it cannot fully capture the spatial variation characteristics of the operation quality within the field, resulting in insufficient representativeness of the evaluation results. In addition, this method is highly dependent on human visual interpretation and manual measurement, and the evaluation results are easily affected by the operator's subjective experience, making it difficult to guarantee repeatability and objectivity. In recent years, UAV remote sensing technology has been widely used in agricultural remote sensing monitoring due to its advantages such as high spatiotemporal resolution, flexibility, efficiency, and low data acquisition cost. However, related methods are concentrated on crop type identification or growth monitoring, and a specialized and automated remote sensing evaluation system for wheat row sowing quality—especially core indicators such as sowing uniformity, broken strip rate, and adjacent row spacing qualification rate—has not yet been formed.

[0004] Therefore, there is an urgent need in this field for a method to achieve accurate, efficient and quantitative evaluation of key operational indicators using UAV remote sensing technology. Summary of the Invention

[0005] The purpose of this application is to provide a method for post-evaluation of the quality of wheat strip seeding operations based on unmanned aerial vehicles (UAVs), which can provide accurate and efficient quantitative evaluation of the quality of wheat strip seeding operations.

[0006] To achieve the above objectives, this application provides the following solution: This application provides a method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs), comprising: acquiring multi-band remote sensing images of wheat after emergence in a target plot using UAVs; preprocessing the remote sensing images to obtain orthophoto maps; inputting the orthophoto maps into a pre-trained seedling strip recognition model to identify and output semantic segmentation results of the seedling strips, wherein the seedling strip recognition model is constructed based on the SEANet architecture; post-processing the semantic segmentation results to generate seedling strip data with vector boundaries, and extracting vector data for each seedling strip based on the vector boundaries; calculating core indicators for evaluating the quality of strip seeding operations based on the vector data, wherein the core indicators include at least one of sowing uniformity, strip breakage rate, and adjacent row spacing qualification rate; and comparing the core indicators with preset operation quality standards to classify and evaluate the quality of strip seeding operations in the target plot.

[0007] For example, the acquisition of multi-band remote sensing images of wheat after emergence in the target plot includes: operating a drone equipped with a multi-band sensor to fly and photograph the target plot under clear and windless weather conditions 7-15 days after wheat emergence and when the outline of the seedlings is clear, so as to obtain a sequence of remote sensing images covering the target plot; the flight altitude of the drone is set to 10 meters, the forward overlap rate is not less than 80%, and the lateral overlap rate is not less than 60%.

[0008] For example, the preprocessing of the remote sensing image includes: performing image stitching and orthorectification processing on the acquired remote sensing image sequence to generate a high-precision orthorectified image map.

[0009] For example, the vector data includes the centerline and geometric parameters of each seedling strip, the geometric parameters including at least the strip length, the average strip width, and the strip orientation angle.

[0010] For example, the post-processing of the semantic segmentation results to generate seedling strip data with vector boundaries, and the extraction of vector data for each seedling strip based on the vector boundaries, includes: performing a morphological opening operation on the semantic segmentation results to remove isolated noise patches with an area smaller than a preset threshold; identifying and repairing broken parts in the seedling strips, and when the gap between breaks is within a preset pixel distance, using linear interpolation to connect the broken endpoints to restore the connectivity of the seedling strips; performing smoothing and polygon approximation processing on the repaired seedling strip boundaries to generate smooth vector boundaries; using a skeletonization algorithm to extract the centerline representing the direction of the seedling strip's center based on the vector boundaries; and calculating the geometric parameters of each seedling strip based on the centerline and the vector boundaries, specifically including: calculating the seedling strip length based on the centerline coordinates, calculating the average width of the seedling strip based on the distance between vector boundaries on the direction perpendicular to the centerline, and calculating the angle between the centerline and the reference direction as the seedling strip's direction angle.

[0011] For example, calculating core indicators based on the vector data specifically includes: dividing the seedling strip into calculation units along its length; extracting the multispectral reflectance of each calculation unit to calculate the normalized vegetation index; statistically analyzing the normalized vegetation index of all calculation units and calculating the coefficient of variation as an evaluation indicator, where the coefficient of variation characterizes the sowing uniformity; identifying broken segments in each seedling strip where the length of consecutive seedlings without seedlings exceeds a set length threshold; calculating the total length of broken segments in all seedling strips based on the broken segments; calculating the broken segment rate based on the total length of broken segments, the total length of the target plot, and the total number of seedling strips; uniformly setting measuring points along the length direction between all adjacent seedling strips; measuring the actual adjacent row spacing at each measuring point; and statistically analyzing the proportion of measuring points where the deviation between the actual adjacent row spacing and the specified row spacing is within the allowable error range as the adjacent row spacing qualification rate.

[0012] For example, the preset operation quality standards include: seeding uniformity variation coefficient ≤ 40%, breakage rate ≤ 2%, and adjacent row spacing qualification rate ≥ 80%.

[0013] For example, the quality of strip seeding operations on the target plot is evaluated in a graded manner, including: if all core indicators meet the operation quality standards, the evaluation is excellent; if some core indicators meet the operation quality standards and the unmet core indicators do not exceed the set tolerance limit, the evaluation is qualified; if the number of indicators that meet the operation quality standards is less than two, or at least some core indicators exceed the tolerance limit, the evaluation is unqualified.

[0014] For example, the pre-trained seedling strip recognition model is obtained by: constructing a deep learning training dataset, which includes sub-images cropped from orthophotos of various scenes and labeled with wheat seedling strip regions; and training the model using the training dataset based on the SEANet model architecture until the model's seedling strip recognition accuracy on the validation set reaches a preset value.

[0015] For example, before inputting the orthophoto image into the seedling strip recognition model, the method further includes: dividing the orthophoto image into multiple image blocks of fixed size, and maintaining a preset proportion of overlapping area between adjacent image blocks; the identification and output of the semantic segmentation result of the seedling strip specifically involves: inputting each image block into the seedling strip recognition model for inference, and fusing the prediction results of the overlapping area of ​​adjacent image blocks to generate a semantic segmentation result of the complete plot.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method for post-evaluation of wheat row sowing quality based on unmanned aerial vehicles (UAVs). It automatically identifies seedling strips using a pre-trained deep learning model and calculates core indicators based on the model's output, replacing traditional methods that rely on manual sampling, visual interpretation, and manual measurement. This avoids errors introduced by human subjectivity, ensuring the repeatability and objectivity of the evaluation results while effectively improving evaluation efficiency. Using multi-band UAV imagery as the data source, it ultimately outputs three core agronomic indicators directly affecting crop growth: sowing uniformity, strip breakage rate, and adjacent row spacing qualification rate. This achieves a multi-dimensional and quantitative comprehensive evaluation of row sowing quality, providing direct and accurate data support for agronomic management. Utilizing multi-band remote sensing data, it directly correlates spectral information with sowing quality, which is more accurate than relying solely on morphology. By post-processing the model's identification results and extracting precise geometric parameters such as seedling strip vector boundaries and centerlines, the reliability of the overall evaluation results is improved. Therefore, it enables accurate, efficient, and reliable post-evaluation of wheat row sowing quality. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the post-evaluation method for wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) in an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the wheat strip seedling identification results in an embodiment of this application.

[0020] Figure 3 This is a schematic diagram of the division of the sowing uniformity calculation unit in the embodiments of this application.

[0021] Figure 4 This is a schematic diagram of the breakage rate identification and length measurement in an embodiment of this application.

[0022] Figure 5 This is a schematic diagram illustrating the setting of adjacent row spacing measurement points and the statistics of pass rate in the embodiments of this application.

[0023] Figure 6 This is a flowchart illustrating the post-evaluation method for wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) in this application embodiment. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] like Figure 1 As shown in the figure, this application provides a method for post-evaluation of the quality of wheat strip seeding operations based on unmanned aerial vehicles (UAVs), including the following steps: S110. Acquire multi-band remote sensing images of wheat after emergence in the target plot using unmanned aerial vehicles (UAVs), preprocess the remote sensing images, and obtain orthophoto maps.

[0027] S120. Input the orthophoto image into the pre-trained seedling strip recognition model, identify and output the semantic segmentation result of the seedling strip, wherein the seedling strip recognition model is built based on the SEANet architecture.

[0028] S130. Post-process the semantic segmentation results to generate seedling strip data with vector boundaries, and extract vector data for each seedling strip based on the vector boundaries.

[0029] S140. Calculate the core indicators for evaluating the quality of row seeding operations based on vector data. The core indicators include at least one of seeding uniformity, row breakage rate, and adjacent row spacing qualification rate.

[0030] S150. Based on the core indicators and the preset operation quality standards, the quality of the strip seeding operation of the target plot is graded and evaluated.

[0031] This application provides a method for post-evaluation of wheat row sowing quality based on unmanned aerial vehicles (UAVs). It employs a pre-trained deep learning model to automatically identify seedling strips and calculates core indicators based on the model's output. This replaces traditional methods relying on manual sampling, visual interpretation, and manual measurement, avoiding errors introduced by human subjectivity and ensuring the repeatability and objectivity of the evaluation results, while effectively improving evaluation efficiency. Using multi-band UAV imagery as the data source, it ultimately outputs three core agronomic indicators directly affecting crop growth: sowing uniformity, strip breakage rate, and adjacent row spacing qualification rate. This achieves a multi-dimensional and quantitative comprehensive evaluation of row sowing quality, providing direct and accurate data support for agronomic management. Utilizing multi-band remote sensing data, it directly correlates spectral information with sowing quality, which is more accurate than relying solely on morphology. By post-processing the model's identification results and extracting precise geometric parameters such as the seedling strip vector boundary and centerline, the reliability of the overall evaluation results is improved. Therefore, it enables accurate, efficient, and reliable post-evaluation of wheat row sowing quality.

[0032] For example, the process of acquiring remote sensing images by manipulating a drone to take pictures, specifically acquiring multi-band remote sensing images of wheat after emergence in a target plot, in this application includes: 7-15 days after wheat emergence, when the seedling outline is clear (and there is no severe shading), a drone equipped with an RGB-nir multi-band sensor (a sensor that simultaneously acquires images in the visible light (red, green, and blue) and near-infrared bands) is used to conduct aerial photography of the target plot under clear, windless weather conditions to obtain a sequence of remote sensing images covering the target plot. The drone's flight altitude is set at 10 meters to ensure that the image resolution meets the requirements for seedling identification and indicator measurement, with a forward overlap rate of no less than 80% and a lateral overlap rate of no less than 60%. Multi-band remote sensing images are obtained through the above photography method, and after preprocessing, high-precision orthophoto maps (such as...) can be obtained. Figure 2 (As shown).

[0033] The preprocessing process is as follows: The acquired remote sensing image sequence undergoes image stitching and orthorectification to generate a high-precision orthophoto map. Specifically, the acquired UAV original images are a series of overlapping single images, and geometric distortions exist due to factors such as flight attitude and terrain undulations. Image stitching refers to using computer vision algorithms to automatically align and merge these single images with overlapping areas, generating a seamless, large-scale image covering the entire target area. The orthorectification process further eliminates image distortions (displacement, stretching, etc.) caused by sensor tilt and terrain elevation differences. This process typically requires Digital Elevation Model (DEM) data as input, using geometric correction algorithms to project each pixel in the image to the correct planar coordinate position. The final high-precision orthophoto map has a unified coordinate system, a consistent scale, and extremely high geometric accuracy, ensuring that the spatial relationships of ground features such as distances and angles between any two points in the image are consistent with those in the actual terrain, providing a necessary foundation for the accurate measurement of subsequent seedbed geometric parameters.

[0034] After obtaining the orthophoto image, a pre-trained seedling strip recognition model can be used to identify and output the semantic segmentation results of the seedling strips. This model is built upon the Semantic Edge Aware Network (SEANet) architecture. The training process for the seedling strip recognition model is as follows: A deep learning training dataset is constructed, comprising cropped sub-images of wheat seedling belt regions from orthophotos of various scenes, each with annotated. Specifically, these are cropped sub-images containing different seedling belt densities, soil backgrounds, and scenes with light straw cover. The annotated seedling belt regions include seedling belt boundaries and sowing row positions. A validation and test set can also be constructed using this method. The training dataset contains at least 1000 images and covers wheat field scenes with different soil types and sowing methods.

[0035] Based on the SEANet model architecture, the network parameters were set as follows: learning rate 0.001, number of iterations 50, and batch size 16. The model was trained using the training dataset until the seedling identification accuracy on the validation set reached a preset value. In some embodiments, the preset value was set to 95%, and training was stopped when the validation set accuracy was consistently higher than 95%.

[0036] SEANet is a convolutional neural network (CNN) architecture specifically designed for semantic segmentation tasks. Its core lies in the introduction of an edge-aware module. Traditional semantic segmentation models sometimes produce blurry or inaccurate predictions at object boundaries. SEANet addresses this problem by processing two branches in parallel: one branch focuses on standard semantic feature extraction and classification (semantic flow), while the other branch specifically learns edge information of objects in the image (edge ​​flow). Deep within the network, the features from these two branches are adaptively fused, enabling the model to predict the category of each pixel not only based on its appearance and contextual semantics but also by enhancing its perception of object boundary positions. This makes SEANet perform exceptionally well in applications requiring precise object shape delineation (such as seedling strip recognition in this application), producing clearer and more accurate seedling strip segmentation maps, laying a reliable foundation for subsequent vectorization and geometric parameter extraction. In some embodiments, the publicly available SEANet can be used as the basic architecture, and transfer learning or retraining can be performed using the labeled dataset constructed in this application to adapt it to the specific task of wheat seedling strip recognition.

[0037] After obtaining the trained model, the orthophoto map is segmented into multiple image blocks of fixed size, with a preset overlap ratio maintained between adjacent image blocks. Each image block is then input into the seedling strip recognition model for inference, and the prediction results of the overlapping areas of adjacent image blocks are fused to generate a semantic segmentation result for the complete plot.

[0038] For example, when segmenting the image, considering computer memory limitations, the orthophoto image of the target plot is divided into 512×512 pixel blocks, with an overlap rate of 50% between adjacent blocks to avoid a decrease in the prediction accuracy of edge pixels. The segmented image is input into the trained and optimized SEANet model, which outputs the semantic segmentation results of each block, with encoding values ​​ranging from 0 to 3. When fusing overlapping regions, a voting method is used to decide on the pixels in the overlapping regions of adjacent blocks. The predicted encoding of the majority of blocks in the overlapping region is taken as the final result, eliminating stitching artifacts and generating a complete semantic segmentation map of the plot.

[0039] After obtaining the semantic segmentation structure through the model, step S130 is executed to post-process the semantic segmentation results, generating seedling strip data with vector boundaries, and extracting vector data for each seedling strip based on the vector boundaries. Specifically, the vector data includes the centerline and geometric parameters of each seedling strip, and the geometric parameters include at least the seedling strip length, the average width of the seedling strip, and the seedling strip orientation angle.

[0040] The post-processing process specifically includes the following steps: S131. Perform morphological opening operation on the semantic segmentation result to remove isolated noise patches with an area smaller than a preset threshold.

[0041] The above steps are for removing interference regions. Based on morphological opening operations (3×3 rectangular structuring elements), isolated small patches with an area ≤ 8 pixels in the semantic segmentation image (mostly noise interference) are removed, while continuous seedling strip regions are retained.

[0042] S132. Identify and repair broken parts in the seedling strip. When the gap between the breaks is within a preset pixel distance, use linear interpolation to connect the broken ends to restore the connectivity of the seedling strip.

[0043] This step aims to restore the connectivity of the seedling strip. Eight-neighbor connectivity analysis is used to identify broken segments in the seedling strip. If the distance between the broken points is ≤2cm (i.e., ≤2 pixels, based on a 1cm resolution), the broken endpoints are connected by linear interpolation to complete the broken boundary, thereby restoring the continuity of the seedling strip.

[0044] S133. The repaired seedling strip boundary is smoothed and approximated with polygons to generate a smooth vector boundary.

[0045] This step is to perform boundary smoothing optimization. Gaussian filtering (standard deviation σ=1.0) is applied to the seedling strip boundary, and then a polygon approximation algorithm (i.e. Douglas-Peucker algorithm, tolerance=0.5 pixels) is used to process it. While maintaining the overall shape of the seedling strip, boundary burrs are removed to generate a smooth seedling strip vector boundary.

[0046] S134. Based on vector boundaries, a skeletonization algorithm is used to extract the centerline representing the orientation of the seedling strip center.

[0047] Specifically, this step uses the Zhang-Suen thinning algorithm to extract the centerline from the optimized seedling strip vector boundary, ensuring that the centerline's overlap with the seedling strip center is ≤1 pixel. The centerline is part of the vector parameters and is mainly used to characterize the center direction of the seedling strip.

[0048] S135. Based on the centerline and vector boundary, calculate the geometric parameters of each seedling strip, specifically including: calculating the seedling strip length (L) based on the centerline coordinates, calculating the average width of the seedling strip (W) based on the distance between the vector boundaries on the direction line perpendicular to the centerline, and calculating the angle between the centerline and the reference direction (due north) as the seedling strip direction angle (θ, the angle with due north).

[0049] After obtaining the key vector data, step S140 is executed to calculate the core indicators used to evaluate the quality of row seeding operations based on the vector data. These core indicators include seeding uniformity, row breakage rate, and adjacent row spacing compliance rate. The specific calculation process is shown below: S141, such as Figure 3As shown, calculation units are divided along the length of each seedling strip. The multispectral reflectance of each calculation unit is extracted to calculate the normalized vegetation index. The normalized vegetation index of all calculation units is statistically analyzed and the coefficient of variation is calculated as an evaluation index. The coefficient of variation characterizes the sowing uniformity.

[0050] During the calculation, RGB-nir multi-band reflectance data for each calculation unit are extracted, and the Normalized Difference Vegetation Index (NDVI) is calculated. The NDVI calculation formula is as follows: Where nir is the near-infrared reflectance and red is the red light reflectance. After the calculation is completed, the NDVI values ​​of all calculation units are counted to obtain the NDVI standard deviation and the NDVI average value, and then the coefficient of variation (CV) is calculated. This coefficient of variation is used as an evaluation index for sowing uniformity. The smaller the coefficient of variation, the better the sowing uniformity. The calculation method is: CV = (NDVI standard deviation / NDVI average value) × 100%.

[0051] S142, such as Figure 4 As shown, the method identifies segments in each seedling strip where the length of consecutive seedlings exceeds a set length threshold. The total length of broken segments in all seedling strips is calculated based on these segments. The breakage rate is then calculated based on the total length of broken segments, the total length of the target plot, and the total number of seedling strips.

[0052] Using the complete operational area of ​​the target plot as the evaluation scope, determine the total length of the target plot (L, in meters). For each identified seedling strip, traverse its entire length, and determine areas with no seedlings for more than 10cm consecutively as broken segments. Record the number of breaks (n) for each seedling strip and the length of each broken segment. (Unit: m, i = 1, 2, ..., n), used to calculate the total broken length D of a single seedling strip, calculated as follows: .

[0053] After calculating the total broken strip length D, the overall broken strip rate of the plot is calculated as follows: broken strip rate = (sum of total broken strip lengths of all seedling strips / (total length of target plot × number of sowing rows N)) × 100%, where N is the total number of sowing rows identified.

[0054] S143, such as Figure 5 As shown, measuring points are evenly set along the length of all adjacent seedling strips. The actual row spacing at each measuring point is measured, and the proportion of measuring points whose deviation from the specified row spacing is within the allowable error range is counted as the pass rate of adjacent row spacing.

[0055] Specifically, all adjacent planting row pairs are extracted from the identified seedling strip vector layer. For each pair of adjacent planting rows, measuring points are evenly set at 1m intervals, with no fewer than 20 measuring points per adjacent row pair, covering its entire length. Using these measuring points, the actual adjacent row spacing (Li, unit: cm) at each measuring point is measured and compared with the preset specified row spacing (L0). When the deviation |Li - L0| ≤ 4cm, the measuring point is considered qualified. The adjacent row spacing qualification rate H is then calculated as follows: H = (Number of qualified measuring points / Total number of measuring points) × 100%.

[0056] Through the above calculation process, three core indicators were obtained: sowing uniformity, breakage rate, and adjacent row spacing qualification rate. These core indicators were then compared with preset operational quality standards for evaluation. The preset operational quality standards included: sowing uniformity coefficient of variation ≤ 40%, breakage rate ≤ 2%, and adjacent row spacing qualification rate ≥ 80%.

[0057] When making comparisons and evaluations, the following criteria will be used: If all core indicators meet the work quality standards, the evaluation is excellent.

[0058] If some core indicators meet the operational quality standards, and the unmet core indicators do not exceed the set tolerance limits, the evaluation is considered qualified. The tolerance limits correspond to the following indicators: seeding uniformity coefficient of variation ≤ 50%, breakage rate ≤ 3%, and adjacent row spacing qualification rate ≥ 70%.

[0059] If the number of indicators that meet the work quality standards is less than two, or if at least some of the core indicators exceed the tolerance limit, the evaluation is unqualified.

[0060] Figure 6 This is a flowchart illustrating the post-evaluation method for wheat strip seeding quality based on unmanned aerial vehicles (UAVs) in this application embodiment. The following is a summary of the process. Figure 6 The methods in the embodiments of this application are summarized and explained as follows: After the method begins, imagery is acquired using a drone, resulting in multi-band remote sensing images. These images are then processed to obtain orthophoto maps, which serve as the data source for post-evaluation. These orthophoto maps are then input into a seedling strip recognition model. This model is built using the SEANet architecture and trained on a deep learning dataset. The optimized model identifies the orthophoto maps, yielding semantic segmentation results for the seedling strips. After post-processing steps including block prediction, multi-stage post-processing optimization, and core parameter extraction, core indicators for wheat row sowing quality are calculated based on these parameters: sowing uniformity, row breakage rate, and adjacent row spacing compliance rate. These core indicators are compared with operational quality standards to achieve a comprehensive evaluation of operational quality. The evaluation results are displayed in a graded manner: excellent, qualified, and unqualified.

[0061] This application provides a method for post-evaluation of wheat row sowing quality based on unmanned aerial vehicles (UAVs). It automatically identifies seedling strips using a pre-trained deep learning model and calculates core indicators based on the model's output, replacing traditional methods that rely on manual sampling, visual interpretation, and manual measurement. This avoids errors introduced by human subjectivity, ensuring the repeatability and objectivity of the evaluation results while effectively improving evaluation efficiency. Using multi-band UAV imagery as the data source, it ultimately outputs three core agronomic indicators directly affecting crop growth: sowing uniformity, strip breakage rate, and adjacent row spacing qualification rate. This achieves a multi-dimensional and quantitative comprehensive evaluation of row sowing quality, providing direct and accurate data support for agronomic management. Utilizing multi-band remote sensing data, it directly correlates spectral information with sowing quality, which is more accurate than relying solely on morphology. By post-processing the model's identification results and extracting precise geometric parameters such as seedling strip vector boundaries and centerlines, the reliability of the overall evaluation results is improved. Therefore, it enables accurate, efficient, and reliable post-evaluation of wheat row sowing quality.

[0062] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0063] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0064] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0065] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0066] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0067] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0068] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0069] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for post-evaluation of the quality of wheat strip seeding operations based on unmanned aerial vehicles (UAVs), characterized in that, The method for post-evaluation of the quality of wheat strip seeding operations based on unmanned aerial vehicles (UAVs) includes: Acquire multi-band remote sensing images of wheat after emergence in the target plot using unmanned aerial vehicles (UAVs), and preprocess the remote sensing images to obtain orthophoto maps. The orthophoto image is input into a pre-trained seedling strip recognition model to identify and output the semantic segmentation result of the seedling strip, wherein the seedling strip recognition model is built based on the SEANet architecture; The semantic segmentation results are post-processed to generate seedling strip data with vector boundaries, and vector data of each seedling strip is extracted based on the vector boundaries. The core indicators for evaluating the quality of row seeding operations are calculated based on the vector data. The core indicators include at least one of seeding uniformity, row breakage rate, and adjacent row spacing qualification rate. The quality of strip seeding operations on the target plots is graded and evaluated by comparing the core indicators with the preset operation quality standards.

2. The method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The acquisition of multi-band remote sensing images of wheat after emergence in the target plot by unmanned aerial vehicles includes: 7-15 days after wheat emergence, when the seedling outline is clear, a drone equipped with a multi-band sensor is used to fly and photograph the target plot in clear, windless weather to obtain a sequence of remote sensing images covering the target plot; the drone's flight altitude is set to 10 meters, the forward overlap rate is not less than 80%, and the lateral overlap rate is not less than 60%.

3. The method for post-evaluation of wheat strip seeding quality based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, The preprocessing of the remote sensing images includes: performing image stitching and orthorectification on the acquired remote sensing image sequence to generate a high-precision orthorectified image map.

4. The method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The vector data includes the centerline and geometric parameters of each seedling strip, and the geometric parameters include at least the seedling strip length, the average width of the seedling strip, and the seedling strip orientation angle.

5. The method for post-evaluation of wheat strip seeding quality based on unmanned aerial vehicles (UAVs) according to claim 4, characterized in that, The post-processing of the semantic segmentation results to generate seedling strip data with vector boundaries, and the extraction of vector data for each seedling strip based on the vector boundaries, includes: A morphological opening operation is performed on the semantic segmentation result to remove isolated noise patches with an area smaller than a preset threshold. Identify and repair broken sections in the seedling strip. When the gap between the breaks is within a preset pixel distance, use linear interpolation to connect the broken ends to restore the connectivity of the seedling strip. The repaired seedling strip boundary is smoothed and approximated with polygons to generate a smooth vector boundary; Based on the vector boundary, a skeletonization algorithm is used to extract the centerline representing the orientation of the seedling strip center; Based on the centerline and the vector boundary, the geometric parameters of each seedling strip are calculated, specifically including: calculating the length of the seedling strip based on the centerline coordinates, calculating the average width of the seedling strip based on the distance between the vector boundaries on the direction line perpendicular to the centerline, and calculating the angle between the centerline and the reference direction as the seedling strip direction angle.

6. The method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The core indicators are calculated based on the vector data, specifically including: The calculation unit is divided along the length of each seedling strip. The multispectral reflectance of each calculation unit is extracted to calculate the normalized vegetation index. The normalized vegetation index of all calculation units is statistically analyzed and the coefficient of variation is calculated as an evaluation index. The coefficient of variation characterizes the sowing uniformity. Identify broken segments in each seedling strip where the length of consecutive seedless sections exceeds a set length threshold. Calculate the total length of broken sections in all seedling strips based on the broken segments. Calculate the breakage rate based on the total length of broken sections, the total length of the target plot, and the total number of seedling strips. Measurement points are evenly set along the length of all adjacent seedling strips. The actual adjacent row spacing at each measurement point is measured. The proportion of measurement points where the deviation between the actual adjacent row spacing and the specified row spacing is within the allowable error range is used as the adjacent row spacing qualification rate.

7. The method for post-evaluation of wheat strip seeding quality based on unmanned aerial vehicles (UAVs) according to claim 6, characterized in that, The preset operational quality standards include: sowing uniformity coefficient of variation ≤40%, breakage rate ≤2%, and adjacent row spacing qualification rate ≥80%.

8. The method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) according to claim 7, characterized in that, The quality of strip seeding operations on the target plots is evaluated and graded, including: If all core indicators meet the aforementioned work quality standards, the evaluation is excellent; If some core indicators meet the work quality standards, and the unmet core indicators do not exceed the set tolerance limit, the evaluation is qualified. If the number of indicators that meet the work quality standards is less than two, or if at least some of the core indicators exceed the tolerance limit, the evaluation is unqualified.

9. The method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) according to any one of claims 1-8, characterized in that, The pre-trained seedling strip recognition model is obtained through the following method: Construct a deep learning training dataset, which includes sub-images cropped from orthophoto maps of various scenes and labeled with wheat seedling belt regions; Based on the SEANet model architecture, the model is trained using the training dataset until the model's seedling identification accuracy on the validation set reaches a preset value.

10. The method for post-evaluation of wheat strip seeding operation quality based on unmanned aerial vehicles (UAVs) according to claim 9, characterized in that, Before inputting the orthophoto image into the seedling strip recognition model, the method further includes: The orthophoto image is divided into multiple image blocks of fixed size, and the adjacent image blocks are kept in an overlapping area of ​​a preset ratio. The process of identifying and outputting the semantic segmentation results of the seedling strips specifically involves: inputting each image block into the seedling strip identification model for inference, and fusing the prediction results of the overlapping areas of adjacent image blocks to generate the semantic segmentation results of the complete plot.

Images