Small seed precision sowing monitoring method and system

Abstract

The invention relates to the technical field of agricultural machinery intellectualization, in particular to a small seed precision sowing monitoring method and system. The method comprises the following steps: setting a time window, and acquiring first signal data passing through a surface laser sensor and second signal data passing through an infrared sensor based on the time window; respectively correcting the first signal data and the second signal data to obtain first optimized data and second optimized data; performing matching degree calculation according to a preset seed sample database by utilizing the first optimization data and the second optimization data to obtain a fusion matching degree score; and calculating the actual pulse frequency of the time window according to the fusion matching degree score, and determining miss-seeding and reseeding by using the actual pulse frequency. According to the invention, the problem that reseeding and miss-seeding are difficult to monitor due to the fact that small seeds are difficult to identify during mechanical seeding in the prior art is solved, and the identification capability of the small seeds is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent agricultural machinery, and in particular to a small-grain seed precision sowing monitoring method and system. Background Art

[0002] In the development of intelligent modern agriculture, the quality of precision seeding of small seeds (such as vegetable seeds, with a particle size of 0.5-1.5 mm) directly impacts crop yield and resource utilization. Real-time monitoring and anomaly detection during the seeding process have become key technical requirements. Currently, small seed sowing monitoring primarily relies on devices such as infrared sensors, which detect signal changes as seeds pass through a sensing area to count and identify anomalies. However, practical applications still face numerous technical bottlenecks.

[0003] First, due to their tiny size and chaotic falling patterns, small seeds generate weak and short-lived signals when passing through the sensing area, easily confounding background noise. This results in numerous blind spots and large errors in sensor detection, resulting in insufficient single-seed identification accuracy and making it difficult to meet the high-precision requirements of precision seeding. Second, the physical parameters of small seeds (geometry, density, surface characteristics, etc.) vary significantly. Existing monitoring systems are often designed for specific seed types, requiring manual adjustment of sensor thresholds and algorithm parameters, resulting in poor versatility. Changing seed types requires re-commissioning, increasing operational complexity and equipment costs. Furthermore, in field environments, dust adhering to the sensor surface reduces the sensitivity of the photosensor, while mechanical vibrations and light fluctuations can interfere with signal stability, increasing sensor misjudgment rates. Monitoring accuracy is particularly reduced in dusty and vibrating seeding environments. Consequently, missed and reseeded seeds are difficult to detect, requiring replanting after seedlings have established, significantly increasing seeding costs.

[0004] Therefore, a more intelligent and accurate seeding detection method is urgently needed to solve the problems of missed seeding and reseeding. Summary of the Invention

[0005] In response to the defects in the prior art, one aspect of the present invention provides a small-grain seed precision sowing monitoring method, which solves the problem in the prior art that it is difficult to accurately detect missed sowing and re-sowing of small-grain seeds.

[0006] In order to achieve the above-mentioned objectives, the present invention provides a small-grain seed precision sowing monitoring method, comprising: setting a time window, and acquiring first signal data of a surface laser sensor and second signal data of an infrared sensor based on the time window; respectively correcting the first signal data and the second signal data to obtain first optimized data and second optimized data; respectively using the first optimized data and the second optimized data to perform matching calculations based on a pre-set seed sample database to obtain a fusion matching score; calculating the actual pulse frequency of the time window based on the fusion matching score, and using the actual pulse frequency to determine missed sowing and replay.

[0007] The present invention adopts a time window to synchronously collect surface laser and infrared dual sensor signals to realize comprehensive capture of seed characteristics, avoid the limitations of a single sensor, filter and correct the data, improve data stability, and improve matching accuracy based on the multi-feature matching mechanism of the seed sample database. By comparing and analyzing the actual pulse frequency with the theoretical value, it accurately identifies missed and reseeded events, greatly improving the accuracy of fine sowing monitoring of small-grain seeds.

[0008] Optionally, the respectively correcting the first signal data and the second signal data to obtain the first optimized data and the second optimized data includes: sampling the first signal data and the second signal data with a predetermined time length to obtain the first signal segment and the second signal segment; sorting the sampling points in the first signal segment and the second signal segment respectively to obtain the first sorted signal and the second sorted signal; extracting the first intermediate size value and the second intermediate size value from the first sorted signal and the second sorted signal; using the first intermediate size value and the second intermediate size value to replace the corresponding intermediate position value of the first signal segment and the intermediate position value of the second signal segment respectively to obtain the first optimized segment and the second optimized segment; based on the first optimized segment and the second optimized segment, sliding sampling and iterative calculation are performed on the first signal data and the second signal data respectively with the predetermined time length to obtain the first optimized data and the second optimized data.

[0009] This method intercepts and sorts signal segments at predetermined time lengths, extracts intermediate values, and replaces them with the mid-segment values. This effectively filters out transient spike noise and vibration interference, preserving the core characteristics of the seed signal. Through continuous sliding sampling and iterative calculations, the raw data within the full time window is converted into smooth, optimized data, avoiding feature loss caused by single filtering and improving the stability and accuracy of the signal data.

[0010] Optionally, the correcting the first signal data and the second signal data respectively to obtain the first optimized data and the second optimized data further includes: extracting the surface laser signal mean and the infrared signal mean when no seed passes through in the first optimized data and the second optimized data respectively; calculating the deviation between the surface laser signal mean and the sampling point in the first optimized data and the deviation between the infrared signal mean and the sampling point in the second optimized data respectively to obtain a surface laser deviation set and an infrared deviation set; fitting the surface laser deviation set and the infrared deviation set respectively to obtain a surface laser deviation trend and an infrared deviation trend; using the surface laser deviation trend to correct the first optimized data, and using the infrared deviation trend to correct the second optimized data.

[0011] This method uses trend fitting to capture the temporal drift of surface laser and infrared signals, effectively identifying systematic errors caused by temperature changes, equipment aging, and other factors. A secondary correction of the optimized data based on deviation trends eliminates baseline drift caused by long-term environmental interference, bringing the signal closer to the actual characteristics and further improving data stability and accuracy.

[0012] Optionally, the use of the first optimization data and the second optimization data to perform matching calculations based on a preset seed sample database to obtain a fusion matching score includes: extracting sample contour signal features, sample amplitude signal features, and sample width signal features corresponding to the seeds to be sown in the preset seed sample database; extracting contour features from the first optimization data, and performing matching calculations using the contour features and the sample contour signal features to obtain a first matching score; extracting actual amplitude and actual width from the second optimization data, and performing matching calculations using the actual amplitude, the actual width, the sample amplitude signal features, and the sample width signal features to obtain a second matching score; and fusing the first matching score and the second matching score to obtain a fusion matching score.

[0013] This method extracts the standard outline, amplitude, and width signal characteristics of the seed to be sown as a benchmark, providing a reliable reference for matching. Contour features are extracted from the laser-optimized data and compared with the sample outline. A first score is generated through morphological feature matching, accurately capturing differences in seed geometry. The actual amplitude and width are extracted from the infrared-optimized data and quantitatively compared with the sample amplitude and width features to generate a second score reflecting the matching degree between signal fluctuations and temporal characteristics. Finally, the dual-dimensional scores are integrated, combining the advantages of optical morphology and signal timing characteristics to improve the recognition of small seeds.

[0014] Optionally, the matching calculation using the contour feature and the sample contour signal feature to obtain a first matching score includes: binarizing the contour feature to obtain a contour binary image; calculating the Zernike moment eigenvector of the contour binary image, and normalizing the Zernike moment eigenvector to obtain a first normalized eigenvector; extracting the normalized Zernike moment eigenvector from the sample contour signal feature to obtain a second normalized eigenvector; and calculating a first matching score based on the first normalized eigenvector and the second normalized eigenvector.

[0015] The present invention performs binarization processing on the contour features, converting the continuous signal into a contour binary image, and clearly distinguishing the seed contour from the background noise. By calculating the Zernike moment eigenvector, the geometric shape, edge curvature and other key features of the seed contour are effectively captured, and its rotational invariance is used to eliminate the interference caused by rotation and displacement when the seed passes. The eigenvector is normalized to ensure that the features at different scales are comparable. The measured normalized vector is directly compared with the standard vector in the sample library to calculate the matching degree, realizing the precise quantitative verification of the contour features and improving the accuracy and robustness of the surface laser scoring.

[0016] Optionally, calculating the first matching score based on the first normalized eigenvector and the second normalized eigenvector includes: obtaining multiple sets of dust concentrations and optimal adjustment coefficients; constructing a coefficient prediction model using the dust concentrations and the optimal adjustment coefficients; obtaining the current dust concentration, and inputting the current dust concentration into the coefficient prediction model to obtain a dynamic adjustment coefficient; and calculating the first matching score using the dynamic adjustment coefficient in combination with the first normalized eigenvector and the second normalized eigenvector.

[0017] This method pre-acquires multiple sets of dust concentrations and optimal adjustment coefficients, constructs a coefficient prediction model, and achieves accurate modeling of different dust environments. During actual monitoring, the current dust concentration is collected in real time and input into the model, rapidly generating an adaptive dynamic adjustment coefficient to perform targeted compensation on the matching calculations of the first and second normalized eigenvectors. This dynamic adjustment mechanism effectively counteracts the effects of dust-induced contour signal attenuation or distortion, avoids feature matching deviations in high-dust environments, and further improves the accuracy and robustness of surface laser scoring.

[0018] Optionally, the sample amplitude signal characteristics include a maximum amplitude and a minimum amplitude, and the sample width signal characteristics include a maximum width and a minimum width. The matching calculation using the actual amplitude, the actual width, the sample amplitude signal characteristics and the sample width signal characteristics to obtain a second matching score includes: obtaining a reference amplitude and a reference width of the seeds to be sown; calculating an amplitude score based on the reference amplitude, the maximum amplitude, the minimum amplitude and the actual amplitude; calculating a width score based on the reference width, the maximum width, the minimum width and the actual width; and calculating a second matching score based on the amplitude score and the width score.

[0019] The present invention provides a comprehensive reference for actual signal verification through the sample amplitude and width characteristics including maximum value, minimum value interval and benchmark value. Taking the benchmark value as the core, the amplitude score and width score are calculated respectively through the deviation ratio between the actual value and the benchmark and the interval constraint, and the matching degree of infrared light intensity fluctuation and the fit of signal duration with the benchmark characteristics are quantified, thereby improving the accuracy and robustness of infrared scores.

[0020] Optionally, the fusing of the first matching score and the second matching score to obtain a fused matching score includes: using the surface laser sensor and the infrared sensor to perform identification tests on the seeds to be sown, respectively, to obtain a surface laser omission rate and an infrared omission rate; determining the surface laser weight and infrared weight of the surface laser sensor based on the surface laser omission rate and the infrared omission rate; and using the first matching score and the second matching score combined with the surface laser weight and the infrared weight to perform weighted summation to obtain a fused matching score.

[0021] This method uses recognition tests to determine the missed detection rates of both surface laser and infrared sensors, quantifying the differences in recognition performance between the two types of sensors in real-world scenarios. Weights are assigned inversely proportional to the missed detection rates, with sensors with lower missed detection rates receiving higher weights. This allows the characteristics of sensors with better performance to dominate the fusion process, avoiding the limitations of a single sensor. The fusion matching score combines the contour morphology advantages of the surface laser with the signal fluctuation characteristics of the infrared, effectively balancing the anti-interference capabilities of the different sensors and improving the reliability and accuracy of seed recognition.

[0022] Optionally, the small-grain seed precision sowing monitoring method further includes: based on the missed sowing and the reseeding, obtaining the row number updated in real time by the trigger pulse signal of the row number sensor; determining the sowing starting point of the row number, and calculating the distance between the seeder and the sowing starting point within the time window using the walking wheel encoder data obtained in real time; and determining the positions corresponding to the missed sowing and the reseeding according to the row number and the distance.

[0023] The present invention uses missed and replayed events to trigger a row number sensor pulse signal, updating the current row number in real time and clearly identifying the row range where the anomaly occurs. Using the row number starting point as a reference, combined with data from the wheel encoder, the distance between the planter and the starting point is accurately calculated, mapping the anomaly within the time window to a specific spatial location. This two-dimensional coordinate combination of "row number + distance" allows for quantitative location of missed and replayed events, enhancing the practicality of the present invention.

[0024] Another aspect of the present invention provides a small-grain seed precision sowing monitoring system, comprising: a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute a small-grain seed precision sowing monitoring method as described in any one of the previous aspects of the present invention.

[0025] The small-grain seed precision sowing monitoring system of the present invention has a compact structure, stable performance, high integration and simple composition. It can stably execute the small-grain seed precision sowing monitoring method provided in the previous aspect of the present invention, further improving the overall applicability and practical application capabilities of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a small seed precision sowing monitoring method according to an embodiment of the present invention; Figure 2 The figure is a structural diagram of a small seed precision sowing monitoring system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] Specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the present invention. In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that these specific details are not necessarily required to practice the present invention. In other instances, well-known circuits, software, or methods are not specifically described to avoid obscuring the present invention.

[0028] Throughout this specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the present invention. Therefore, appearances of the phrases "in one embodiment," "in an embodiment," "an example," or "an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combinations and / or subcombinations in one or more embodiments or examples. Furthermore, those of ordinary skill in the art will appreciate that the figures provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0029] See Figure 1 In an alternative embodiment, as Figure 1 The small seed precision sowing monitoring method shown includes the following steps: Step S1: setting a time window, and acquiring first signal data of a passing area laser sensor and second signal data of an infrared sensor based on the time window.

[0030] In this embodiment, when setting the time window, a time window of fixed length is pre-set based on the typical falling speed of small seeds and the actual sowing density. The set time window must ensure that multiple seeds pass through within the time window. This is used as the basic unit for single signal acquisition. The surface laser sensor and the infrared sensor are synchronously triggered by a high-precision clock module to ensure that the two sensors start data acquisition within the same time window. The surface laser sensor collects the three-dimensional contour reflection signal when the seeds pass through the detection area at a preset sampling frequency to form the first signal data. The infrared sensor synchronously collects the infrared light intensity change signal within the corresponding time period to form the second signal data. The time window adopts a continuous sliding mode, and the next window is automatically started when the previous window ends. A certain overlap rate (such as 20%) can be retained between windows to avoid the interruption of continuous seed signals. Finally, the original signal data of the two sensors are intercepted and stored in each time window, providing a basic data segment with a unified time dimension for subsequent signal correction and feature extraction.

[0031] Step S2: Correct the first signal data and the second signal data respectively to obtain first optimized data and second optimized data.

[0032] The steps of respectively correcting the first signal data and the second signal data to obtain first optimized data and second optimized data specifically include the following steps: Step S201: sampling the first signal data and the second signal data at a predetermined time length to obtain a first signal segment and a second signal segment.

[0033] In this embodiment, the predetermined time length is set to be smaller than the above-mentioned time window. The time of 3 to 5 sampling points can be selected as the preset time length, and the first signal data and the second signal data continuously collected within the time window are initially sampled: in the complete signal data stream covered by the time window, a short-term signal matching the predetermined time length is intercepted starting from the start moment of the window to form a first signal segment (surface laser local signal segment) and a second signal segment (infrared local signal segment) containing all sampling points within the local time period. The segment length is recommended to be 1 / 5-1 / 10 of the time window, providing a basic local data unit for subsequent sliding window iterative calculations (such as median filtering). Subsequent sliding sampling will use this predetermined time length as a step to progressively intercept signal segments within the entire time window.

[0034] Step S202: Sort the sampling points in the first signal segment and the second signal segment respectively to obtain a first sorted signal and a second sorted signal.

[0035] In this embodiment, for the first signal segment and the second signal segment obtained by sampling for a predetermined time length, the signal values ​​of all sampling points in the segment are respectively extracted (the surface laser signal segment includes the contour reflection intensity sampling value, and the infrared signal segment includes the light intensity change sampling value). The sampling point data in each segment are independently sorted in ascending or descending order of values. The surface laser sampling points in the first signal segment are arranged from small to large (or from large to small) according to the signal intensity values ​​to form an ordered first sorted signal. The infrared sampling points in the second signal segment are simultaneously sorted according to the light intensity values ​​using the same rule to obtain a corresponding second sorted signal. During the sorting process, the integrity of the sampling points in a single segment is maintained, and data is not mixed across segments.

[0036] Step S203: extracting a first intermediate size value and a second intermediate size value from the first sorting signal and the second sorting signal.

[0037] In this embodiment, in the first sorted signal (ordered arrangement of sampling points of the local signal segment of the surface laser) and the second sorted signal (ordered arrangement of sampling points of the local signal segment of the infrared laser), the signal value at the middle position is extracted as the first middle size value and the second middle size value respectively: for the first sorted signal, the total number of sampling points contained therein is counted. If the total number is an odd number, the value of the sampling point at the middle position of the sorted sequence is directly selected as the first middle size value; if the total number is an even number, the average value of the sampling points at the two middle positions is taken as the first middle size value. The second sorted signal is processed using the same rule, that is, according to the parity of the total number of sampling points, the value at the middle position or the average value of the two middle values ​​is selected as the second middle size value, to ensure that the extracted middle size value can represent the typical characteristics of the local signal segment.

[0038] Step S204: Use the first intermediate size value and the second intermediate size value to replace the corresponding intermediate position value of the first signal segment and the intermediate position value of the second signal segment respectively, to obtain a first optimized segment and a second optimized segment.

[0039] In this embodiment, the extracted first intermediate size value (the intermediate eigenvalue after the surface laser signal segments are sorted) and the second intermediate size value (the intermediate eigenvalue after the infrared signal segments are sorted) are used to correct the corresponding first signal segments and the second signal segments, respectively: in the first signal segment, the original sampling point corresponding to the middle position on the time axis of the segment is located (that is, the middle point of all sampling points in the segment arranged in order of acquisition time), and the signal value of the original sampling point is replaced with the first intermediate size value. The same operation is performed on the second signal segment to find the original sampling point at the middle position of its time axis, and the signal value of the point is replaced with the second intermediate size value. After the replacement, the first signal segment and the second signal segment eliminate the interference of local extreme noise (especially high-frequency noise generated by vibration) on the middle position, forming a smoothed first optimized segment and a smoothed second optimized segment.

[0040] Step S205: Based on the first optimization segment and the second optimization segment, sliding sampling is performed on the first signal data and the second signal data for the predetermined time length, and iterative calculation is performed to obtain first optimization data and second optimization data.

[0041] In this embodiment, based on the generated first and second optimized segments, continuous sliding sampling is performed on the first and second signal data within the time window with a preset step size (e.g., the time interval corresponding to a single sampling point is the sliding step size). Starting from the starting position of the initial signal segment, after each calculation of sorting the sampling points → extracting the intermediate size value → replacing the intermediate position value, the sampling window is shifted backward along the time axis by one sampling point, and the local signal segment containing the new sampling point is intercepted as the next set of first and second signal segments to be processed. The above iterative calculation process is repeated for the new segments to generate subsequent optimized segments. Ultimately, through a point-by-point sliding and intensive iteration method, all the original signal data within the time window is converted into continuous optimized segments. After splicing, the smoother first and second optimized data are obtained, effectively suppressing the transient spike noise in the signal while preserving the signal detail characteristics when the seed passes to the greatest extent.

[0042] Step S206 , extracting the surface laser signal mean and the infrared signal mean when no seed passes through from the first optimization data and the second optimization data respectively.

[0043] In this embodiment, a preset seed signal identification threshold (eg, a signal amplitude exceeding a certain percentage of a baseline is considered to indicate seed passage) is used to filter out time periods without seed passage in the first optimization data and the second optimization data.

[0044] In the first optimized data, all continuous signal segments where no seed contour features were detected (stable areas without obvious amplitude jumps) were located, and the signal numerical averages of all sampling points in the area were calculated to obtain the mean value of the surface laser signal when no seeds passed through. Simultaneously, in the second optimized data, stable signal segments where no seeds blocked the infrared light path (light intensity fluctuations were within the background noise range) were identified, and the light intensity numerical averages of all sampling points in this period were calculated to obtain the mean value of the infrared signal when no seeds passed through.

[0045] Step S207 , respectively calculating the deviation between the surface laser signal mean and the sampling points in the first optimized data and the deviation between the infrared signal mean and the sampling points in the second optimized data to obtain a surface laser deviation set and an infrared deviation set.

[0046] In this embodiment, for each sampling point in the first optimized data (the complete data stream of the surface laser after median filtering), the difference between the signal value at that sampling point and the mean of the surface laser signal is calculated one by one. All differences are aggregated to form a surface laser deviation set, which reflects the degree of fluctuation of each sampling point in the first optimized data relative to the seedless baseline. Similarly, for each sampling point in the second optimized data (the complete data stream of the infrared after median filtering), the difference between its signal value and the mean of the infrared signal is calculated. These differences are then stored collectively to form an infrared deviation set, which quantifies the deviation of each sampling point in the second optimized data from the infrared baseline. The deviation calculation process covers all sampling points in both optimized data streams over the entire time period, ensuring that both the surface laser deviation set and the infrared deviation set fully contain all fluctuation information of the signal in the time dimension.

[0047] Step S208 : fitting the surface laser deviation set and the infrared deviation set respectively to obtain a surface laser deviation trend and an infrared deviation trend.

[0048] In this embodiment, a time series fitting method is used to perform a systematic analysis. With the time axis as the horizontal axis and the deviation value as the vertical axis, all difference data in the surface laser deviation set are arranged in order of sampling time. The data distribution trend is fitted by linear regression to obtain a surface laser deviation trend (such as linear drift, periodic fluctuation or slow attenuation trend) that can characterize the change law of the surface laser deviation over time. Simultaneously, the difference data in the infrared deviation set are arranged according to the same time dimension, and the same fitting algorithm is used to analyze its change characteristics over time to obtain an infrared deviation trend that reflects the overall change law of the infrared deviation. The full amount of data in the deviation set is covered during the fitting process to ensure that the trend model can accurately capture the systematic deviation characteristics caused by environmental interference (such as temperature changes and sensor aging).

[0049] Step S209: correcting the first optimization data using the surface laser deviation trend, and correcting the second optimization data using the infrared deviation trend.

[0050] In this embodiment, for each sampling point in the first optimized data, the trend deviation value corresponding to that time point is extracted from the surface laser deviation trend model. This trend deviation value is then subtracted from the original optimized signal value at that sampling point to eliminate systematic offsets caused by environmental factors (such as temperature drift and device aging). Similarly, for each sampling point in the second optimized data, the trend deviation value corresponding to that time point is obtained based on the infrared deviation trend model. This infrared trend deviation value is then subtracted from the optimized signal value at that sampling point to compensate for baseline drift of the infrared sensor caused by long-term operation or environmental interference.

[0051] Step S3: Use the first optimized data and the second optimized data to perform matching calculations based on a preset seed sample database to obtain a fusion matching score.

[0052] The steps of performing matching calculation using the first optimized data and the second optimized data according to a preset seed sample database to obtain a fusion matching score specifically include the following steps: Step S301: extracting sample contour signal features, sample amplitude signal features, and sample width signal features corresponding to the seeds to be sown from a preset seed sample database.

[0053] In this embodiment, sample contour signal features of seeds to be sown detected by a surface laser sensor in a standard interference-free environment are extracted from a pre-set seed sample database, and sample amplitude signal features recorded when the seeds pass through the infrared sensor are simultaneously extracted (covering the maximum value, minimum value and reference amplitude of the signal amplitude when multiple seeds of the same type with differences block the infrared light path, reflecting the range of infrared light intensity variation) and sample width signal features (including the maximum value, minimum value and reference width corresponding to the signal duration when multiple seeds of the same type with differences pass through the detection area, characterizing the signal features of the physical size of the seeds). The extraction process ensures that the sample features strictly correspond to the variety attributes of the seeds to be sown, and all sample features are constructed based on the statistical average value of multiple repeated experiments.

[0054] Step S302: extracting contour features from the first optimized data, and performing a matching calculation using the contour features and the sample contour signal features to obtain a first matching score.

[0055] In this embodiment, a signal threshold segmentation algorithm is used (such as setting the area where the contour signal amplitude is 20% higher than the baseline as the effective contour area) to separate the characteristic signal segment when the seed passes through the detection area from the continuous surface laser signal. The waveform of the characteristic signal segment is analyzed to extract the key parameters reflecting the geometric shape of the seed, including the coordinates of the start and end points of the contour edge, the change curve of the signal amplitude, the distribution of the peak points of the contour, etc., to form contour features.

[0056] The step of performing matching calculation using the contour feature and the sample contour signal feature to obtain a first matching score specifically includes the following sub-steps: Step S30201, binarize the contour features to obtain a contour binary image.

[0057] In this embodiment, the OTSU adaptive threshold method is used to perform binarization processing on the contour features to obtain a contour binary image.

[0058] The algorithm automatically analyzes the signal grayscale distribution of the contour feature (converting the contour signal amplitude into a grayscale value scale) and calculates the optimal threshold that maximizes the inter-class variance between the foreground (seed contour area) and the background (non-contour area). Using this threshold as the dividing line, the area in the contour feature with a grayscale value higher than the threshold is determined as the seed contour valid area and assigned a value of 1, and the background area with a grayscale value lower than the threshold is assigned a value of 0. Finally, the continuous contour feature signal is converted into a contour binary image containing only two pixel values ​​0 and 1.

[0059] Step S30202: Calculate the Zernike moment eigenvector of the contour binary image, and normalize the Zernike moment eigenvector to obtain a first normalized eigenvector.

[0060] In this embodiment, Zernike moments, a rotationally invariant image feature extraction method, can effectively capture geometric features such as the overall morphology, edge curvature, and concave-convex details of seeds in binary contour images. By selecting a Zernike polynomial of a preset order (e.g., an 8th-order polynomial, which can be flexibly selected based on the complexity of the seed contour), a mathematical model is constructed for the pixel distribution of the seed contour region in the image. The numerical values ​​of the Zernike moments of each order are then solved through integral calculations to form a feature vector containing multidimensional shape information. The core significance of this process lies in converting the two-dimensional seed contour image into a quantifiable mathematical vector, which not only preserves the key morphological features of the seed but also, through the orthogonality and rotational invariance of the Zernike moments, eliminates interference with feature extraction caused by rotation or slight displacement of the seed as it passes through the detection area.

[0061] After completing the construction of the feature vector, in order to eliminate the influence of numerical differences in different dimensions and unify the feature scale, the vector modulus normalization method (i.e., L2 normalization) is adopted. First, the Euclidean modulus of the Zernike moment feature vector is calculated (obtained by taking the square root of the sum of the squares of each element of the vector), and then each element in the vector is divided by the obtained modulus so that the normalized vector modulus is 1. Finally, the first normalized feature vector is generated, laying the foundation for subsequent seed feature matching and other operations. With the help of the orthogonality and rotation invariance of the Zernike moment, the feature extraction interference caused by rotation and slight displacement when the seed passes through the detection area is effectively reduced.

[0062] The first normalized eigenvector satisfies the following formula: in, is the first normalized eigenvector, is the Zernike moment eigenvector of the contour binary image.

[0063] Step S30203: extracting a normalized Zernike moment eigenvector from the sample profile signal features to obtain a second normalized eigenvector.

[0064] In this embodiment, the sample contour signal characteristics of the seed to be sown are deeply analyzed to accurately extract a normalized Zernike moment feature vector. This vector is a typical morphological feature vector formed by calculating the Zernike moments of the contour binary images of a large number of identical seed samples and then normalizing them. It contains quantitative information about the key geometric attributes of the seed contour (such as overall shape, edge details, and concave-convex distribution), and the values ​​of each dimension are all within a uniform scale range (such as the interval [0,1]). The extraction process ensures that this vector is completely consistent with the first normalized feature vector obtained from the actual detection signal in terms of order, dimension, and normalization standard, ultimately resulting in a second normalized feature vector that can be directly used for matching calculations.

[0065] The second normalized eigenvector satisfies the following formula: in, is the second normalized eigenvector, is the Zernike moment eigenvector of the sample profile signal feature.

[0066] Step S30204: Calculate a first matching score based on the first normalized feature vector and the second normalized feature vector.

[0067] Calculating a first matching score based on the first normalized feature vector and the second normalized feature vector specifically includes the following sub-steps: Step S3020401, obtain multiple sets of dust concentrations and optimal adjustment coefficients.

[0068] In this embodiment, based on a dust simulation test, a test species contour signal of the test species at different dust concentrations is obtained; the test species is binary labeled to generate a binary label; an adjustment coefficient is introduced, and the test score of the test species contour signal is calculated based on the adjustment coefficient; a matching judgment threshold is set, and the matching judgment threshold is compared with the test score to obtain a comparison result; the accuracy is calculated using the binary label and the comparison result; the adjustment coefficient is adjusted, and the optimal adjustment coefficient is determined based on the adjustment result and the accuracy.

[0069] To set the matching judgment threshold, first select a standard sample of seeds to be sown (typical seeds without dust interference), collect more than 100 sets of pure contour signals through the surface laser sensor, calculate the matching score of the first normalized eigenvector and the second normalized eigenvector, and form a sample benchmark score set. Perform statistical analysis on this set and obtain the lower limit value of the distribution (such as the minimum value of the 95% confidence interval) as the initial threshold. Then, a preliminary test is performed to verify the rationality of the threshold: in an environment without dust interference (concentration = 0g / m 3 ), use the initial threshold to judge 100 groups of actual seed signals, and calculate the missed detection rate (the proportion of actual seeds mistakenly identified as noise) and the false detection rate (the proportion of noise mistakenly identified as seeds). If the missed detection rate is greater than 5%, lower the threshold (for example, to 0.65); if the false detection rate is greater than 5%, increase the threshold (for example, to 0.75), and repeat the verification until the missed detection rate and false detection rate are both ≤5%. The threshold at this time is the matching threshold for this seed type. .

[0070] Dust simulation test: by constructing 0g / m 3 -15g / m 3 Dust concentration gradient (according to 2g / m 3 Gradient points were set at intervals, and each concentration was maintained stable for 10 minutes. The small seeds to be monitored were controlled to continuously pass through the detection area at a typical sowing speed. The area laser sensor was triggered to collect 500 sets of seed contour signals and the actual seed passing events were synchronously recorded by a high-speed camera to generate binary labels (1 = seed passed effectively, 0 = invalid noise).

[0071] Construct a scoring calculation model: To score, is the Zernike moment normalized eigenvector of the measured target, is the Zernike moment normalized eigenvector of the comparison target, is the adjustment factor.

[0072] For each concentration point, the adjustment coefficient of the interval [0,1] is traversed in 0.01 steps, the score is calculated by the score calculation model, and the score is compared with the matching judgment threshold. Mark it as a "seed event", otherwise it is marked as a "non-seed event". Then, the seeds and non-seeds marked by the algorithm and the seeds and non-seeds marked manually are counted, and the accuracy is calculated based on the statistical results: in, is the accuracy, is the number of signal groups that the algorithm determines as seeds and manually marks as seeds, is the number of signal groups that the algorithm determines as non-seed and that are manually marked as non-seed, is the number of signal groups determined by the algorithm to be seeds but manually marked as non-seeds, The number of signal groups determined by the algorithm as non-seeds but manually marked as seeds.

[0073] According to the traversal iteration of the adjustment coefficient, if the accuracy increase is less than 1% after increasing the adjustment coefficient for 5 consecutive times, the current adjustment coefficient is recorded as the optimal adjustment coefficient. Finally, multiple sets of dust concentrations and optimal adjustment coefficients are obtained.

[0074] Step S3020402: construct a coefficient prediction model using the dust concentration and the optimal adjustment coefficient.

[0075] In this embodiment, dust concentration is used as the input dimension and the optimal adjustment coefficient is used as the output dimension. Based on the principle of least squares method, multiple groups of data are substituted into the linear regression model. (in is the slope, is the intercept) by calculating the minimum sum of squares of the errors between the actual data and the model prediction value and , complete the solution of model parameters, at the same time, use indicators such as determination coefficient and mean square error to divide the training set to verify the model fitting effect and the test set to test the generalization ability, and finally construct a model that can linearly predict the optimal adjustment coefficient based on dust concentration.

[0076] Step S3020403, obtain the current dust concentration, and input the current dust concentration into the coefficient prediction model to obtain a dynamic adjustment coefficient.

[0077] In this embodiment, relying on the dust concentration sensor, the current dust concentration data of the working environment is collected in real time. After pre-processing such as abnormal value filtering and format verification, it is input into a pre-built linear coefficient prediction model. The coefficient prediction model quickly calculates and outputs a dynamic adjustment coefficient adapted to the current environment based on the trained linear parameters.

[0078] Step S3020404: Calculate a first matching score using the dynamic adjustment coefficient in combination with the first normalized feature vector and the second normalized feature vector.

[0079] The first matching score satisfies the following formula: in, is the first matching score, is the first normalized eigenvector, is the second normalized eigenvector, is the dynamic adjustment coefficient.

[0080] The inner product of the first and second normalized eigenvectors is calculated by dynamically adjusting the coefficients Make compensation corrections and build The matching score calculation formula not only uses the dynamic adjustment coefficient to adapt to the dust concentration interference, but also avoids over-compensation through threshold limits, ultimately achieving accurate and robust calculation of seed contour feature matching.

[0081] The coefficient prediction model is the current dust concentration The predicted value of is the maximum adjustment threshold.

[0082] In the above formula, the maximum adjustment threshold is the maximum value of the 95% confidence interval among all the optimal adjustment coefficients obtained in step S3020401.

[0083] Step S303: extracting the actual amplitude and actual width from the second optimized data, and performing a matching calculation using the actual amplitude, the actual width, the sample amplitude signal feature, and the sample width signal feature to obtain a second matching score.

[0084] In this embodiment, the signal fluctuation segment when the seed passes through the infrared detection area is located by a signal peak detection algorithm, the maximum and minimum values ​​of the light intensity signal in the fluctuation segment are identified, and the difference between the two is calculated to obtain the actual amplitude reflecting the degree of seed occlusion; at the same time, based on the signal threshold (such as setting the moment when the light intensity changes by more than 10% of the baseline as the effective starting and ending points), the signal duration of the seed passing through the detection area is determined, and the time length is converted into the actual width representing the physical size of the seed in combination with the sampling frequency of the infrared sensor, to ensure that the extracted amplitude and width parameters can accurately reflect the infrared signal characteristics when the seed passes.

[0085] The step of performing a matching calculation using the actual amplitude, the actual width, the sample amplitude signal feature, and the sample width signal feature to obtain a second matching score includes: Step S30301, obtain the reference amplitude and reference width of the seeds to be sown.

[0086] In this embodiment, when obtaining the benchmark amplitude and benchmark width of the seeds to be sown, it is necessary to select more than 100 groups of valid signals of the seeds passing through the infrared sensor in a standard interference-free environment based on the statistical results of the infrared signal characteristics of the same seeds in the seed sample database, and extract the amplitude (the difference between the maximum and minimum light intensity) and width (the physical size converted from the signal duration) of each group of signals through signal peak detection; after removing outliers from these amplitude and width data, their arithmetic mean is calculated, which serves as the benchmark amplitude reflecting the typical infrared shielding intensity of the seeds and the benchmark width of the typical physical size, respectively. The determination of the benchmark value needs to cover the individual differences of the seeds to ensure that the subsequent calculation of the deviation between the actual amplitude and width and the benchmark value can effectively reflect the true characteristics of the seeds.

[0087] Step S30302: Calculate the amplitude score based on the reference amplitude, the maximum amplitude, the minimum amplitude, and the actual amplitude.

[0088] The amplitude score satisfies the following formula: in, is the amplitude score, is the actual amplitude, is the reference amplitude, is the maximum amplitude, is the minimum amplitude value.

[0089] The above formula is based on the reference amplitude As the core reference, combined with the maximum amplitude and minimum value Constructing a valid interval When the actual amplitude When it is within this range, the absolute deviation between the actual amplitude and the reference amplitude is calculated as the span of the interval. The ratio of Get amplitude score The smaller the deviation, the closer the score is to 1, reflecting the degree of fit between the actual amplitude and the benchmark. If the actual amplitude exceeds the interval, the score is directly set to 0 to quantify the degree of match between the actual amplitude and the benchmark feature.

[0090] Step S30303: Calculate a width score based on the reference width, the maximum width, the minimum width, and the actual width.

[0091] The width score satisfies the following formula: in, Score for width, is the actual width, is the base width, is the maximum width, The minimum width.

[0092] Step S30304: Calculate a second matching score based on the amplitude score and the width score.

[0093] The second matching score satisfies the following formula: The above formula is used as the amplitude score and width score When both are not 0, it means that the actual amplitude and width are within the effective matching range. The average of the two is taken to obtain the second matching score, which reflects the degree of matching between the amplitude and width features. or If any one of them is 0, it indicates that the corresponding feature (amplitude or width) is not matched, and the second matching score is directly determined to be 0.

[0094] Step S304: Fusing the first matching score and the second matching score to obtain a fused matching score.

[0095] The step of fusing the first matching score and the second matching score to obtain a fused matching score includes: Step S30401: Use the area laser sensor and the infrared sensor to perform identification tests on the seeds to be sown, respectively, to obtain the area laser missed detection rate and the infrared missed detection rate.

[0096] In this embodiment, a known quantity (denoted as ) to be sown, control them to pass through the detection area of ​​the integrated area laser sensor and infrared sensor at a set speed and spacing, use the area laser sensor to collect seed contour signals, identify valid seed events through feature extraction and matching judgment processes, and count the actual number of identified events. , according to the surface laser missed detection rate formula Calculate, and similarly calculate the infrared missed detection rate .

[0097] Step S30402: Determine the surface laser weight and infrared weight of the surface laser sensor according to the surface laser missed detection rate and the infrared missed detection rate.

[0098] In this embodiment, the lower the missed detection rate, the better the sensor recognition performance, and the higher the corresponding weight. By constructing an inverse proportional mapping relationship between the missed detection rate and the weight, the surface laser weight , infrared weight .

[0099] Step S30403: Use the first matching score and the second matching score in combination with the surface laser weight and the infrared weight to perform weighted summation to obtain a fusion matching score.

[0100] In this embodiment, according to the multi-sensor fusion logic, a weighted summation model is constructed: the surface laser weight is multiplied by the first matching score to obtain the contribution value of the surface laser feature in the fusion result; the infrared weight is multiplied by the second matching score to obtain the contribution value of the infrared feature; finally, the two contribution values ​​are added together to obtain the fusion matching score. Satisfies the following formula: Step S4: Calculate the actual pulse frequency of the time window according to the fusion matching score, and use the actual pulse frequency to determine missed broadcasts and replays.

[0101] In this embodiment, a fusion score threshold is set, and signals with a fusion matching score greater than the threshold within the time window are determined to be valid seed events and counted; the sensor sampling frequency and window duration are combined to calculate the number of pulses of valid events per unit time to obtain the actual pulse frequency. At the same time, the theoretical pulse frequency is derived based on the seed disc rotation speed and the theoretical seed grain spacing. When the actual frequency is 0 for more than one consecutive time window, or is 50% lower than the theoretical frequency, it is determined to be a missed seed (depending on the accuracy requirements). When the actual frequency exceeds the range of ±30% of the theoretical frequency, it is determined to be a reseeding, thereby achieving real-time and accurate identification of sowing anomalies.

[0102] Step S5: Based on the missed broadcast and the rebroadcast, obtain the row number updated in real time by the trigger pulse signal of the row number sensor.

[0103] In this embodiment, after determining a missed broadcast or replay event, the system receives a trigger pulse signal from a row number sensor (such as a trigger sensor integrated in the seeding machine's sowing row) in real time, and each triggered pulse corresponds to a record of the current sowing row number; when a missed broadcast or replay event is identified, the latest row number synchronously updated by the trigger pulse signal is bound to the abnormal event, and the current row number is output.

[0104] Step S6, determining the sowing starting point of the row number, and calculating the distance between the sowing machine and the sowing starting point in the time window using the walking wheel encoder data obtained in real time.

[0105] In this implementation, when determining the sowing starting point of the row number, the position where the row number sensor first triggers the pulse signal is used as the reference starting point (the starting point after the row change is the zero distance point). During the sowing process, the pulse data of the walking wheel encoder is collected in real time. Combined with the encoder's preset pulse equivalent (such as each pulse corresponds to a walking distance of 0.5mm), the cumulative number of pulses from the starting point to the current time window is calculated, and the actual distance between the seeder and the reference starting point is obtained by converting "cumulative number of pulses × pulse equivalent". This distance data is bound to the time window and synchronously associated with the location information of missed and replayed events, providing accurate spatial coordinate support for sowing uniformity analysis and abnormality tracing.

[0106] Step S7: Determine the positions corresponding to the missed broadcast and the rebroadcast according to the row number and the distance.

[0107] In this embodiment, through the two-dimensional coordinate combination of "row number + distance", the missed broadcast or replay event is accurately mapped to a certain distance point of the specific broadcast line. The position information can also be synchronously transmitted to the alarm unit, so that the sound and light alarm is directly associated with the abnormal position of "X row - Y meters" displayed on the central control screen, thereby realizing the precise spatial positioning of the sowing abnormality.

[0108] like Figure 2 As shown, on the other hand, the present invention also provides a small-grain seed precision sowing monitoring system, including: a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute relevant steps of a relevant embodiment of a small-grain seed precision sowing monitoring method of the present invention.

[0109] The present invention provides a small seed precision sowing monitoring system in which each functional component can be integrated into a single processing component, each component can exist physically separately, or two or more components can be integrated into a single component. The integrated components can be implemented in the form of hardware or software functions.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A small seed precision sowing monitoring method, characterized in that: The method comprises: Setting a time window, and acquiring first signal data of the surface laser sensor and second signal data of the infrared sensor based on the time window; Correcting the first signal data and the second signal data respectively to obtain first optimized data and second optimized data; Using the first optimized data and the second optimized data respectively to perform matching calculations according to a preset seed sample database to obtain a fusion matching score; An actual pulse frequency of the time window is calculated according to the fusion matching score, and missed broadcasts and rebroadcasts are determined using the actual pulse frequency.

2. A small seed precision sowing monitoring method according to claim 1, characterized in that: The respectively correcting the first signal data and the second signal data to obtain first optimized data and second optimized data includes: Sampling the first signal data and the second signal data at a predetermined time length to obtain a first signal segment and a second signal segment; sorting the sampling points in the first signal segment and the second signal segment respectively to obtain a first sorted signal and a second sorted signal; extracting a first intermediate magnitude value and a second intermediate magnitude value from the first sorted signal and the second sorted signal; Replacing the corresponding intermediate position value of the first signal segment and the intermediate position value of the second signal segment with the first intermediate size value and the second intermediate size value, respectively, to obtain a first optimized segment and a second optimized segment; Based on the first optimization segment and the second optimization segment, sliding sampling is performed on the first signal data and the second signal data respectively within the predetermined time length and iterative calculation is performed to obtain first optimization data and second optimization data.

3. A small seed precision sowing monitoring method according to claim 2, characterized in that: The respectively correcting the first signal data and the second signal data to obtain first optimized data and second optimized data further comprises: extracting the surface laser signal mean and the infrared signal mean when no seed passes through from the first optimized data and the second optimized data respectively; Calculating the deviations between the surface laser signal mean and the sampling points in the first optimization data and the deviations between the infrared signal mean and the sampling points in the second optimization data to obtain a surface laser deviation set and an infrared deviation set; Fitting the surface laser deviation set and the infrared deviation set respectively to obtain a surface laser deviation trend and an infrared deviation trend; The first optimization data is corrected using the surface laser deviation trend, and the second optimization data is corrected using the infrared deviation trend.

4. A small seed precision sowing monitoring method according to claim 1, characterized in that: The step of performing matching calculations using the first optimized data and the second optimized data according to a preset seed sample database to obtain a fusion matching score includes: Extracting sample contour signal features, sample amplitude signal features, and sample width signal features corresponding to the seeds to be sown from a pre-set seed sample database; Extracting contour features from the first optimized data, and performing a matching calculation using the contour features and the sample contour signal features to obtain a first matching score; Extracting an actual amplitude and an actual width from the second optimized data, and performing a matching calculation using the actual amplitude, the actual width, the sample amplitude signal feature, and the sample width signal feature to obtain a second matching score; The first matching score and the second matching score are fused to obtain a fused matching score.

5. A small seed precision sowing monitoring method according to claim 4, characterized in that: The performing matching calculation using the contour feature and the sample contour signal feature to obtain a first matching score includes: Binarizing the contour features to obtain a contour binary image; Calculating a Zernike moment eigenvector of the contour binary image, and normalizing the Zernike moment eigenvector to obtain a first normalized eigenvector; Extracting a normalized Zernike moment eigenvector from the sample profile signal feature to obtain a second normalized eigenvector; A first matching score is calculated according to the first normalized feature vector and the second normalized feature vector.

6. A small seed precision sowing monitoring method according to claim 5, characterized in that: Calculating a first matching score according to the first normalized feature vector and the second normalized feature vector includes: Obtain multiple sets of dust concentrations and optimal adjustment coefficients; Constructing a coefficient prediction model using the dust concentration and the optimal adjustment coefficient; Obtaining a current dust concentration, and inputting the current dust concentration into the coefficient prediction model to obtain a dynamic adjustment coefficient; A first matching score is calculated using the dynamic adjustment coefficient in combination with the first normalized feature vector and the second normalized feature vector.

7. A small seed precision sowing monitoring method according to claim 4, characterized in that: The sample amplitude signal feature includes a maximum amplitude and a minimum amplitude, the sample width signal feature includes a maximum width and a minimum width, and the performing a matching calculation using the actual amplitude, the actual width, the sample amplitude signal feature, and the sample width signal feature to obtain a second matching score includes: Obtaining the reference amplitude and reference width of the seeds to be sown; Calculating an amplitude score according to the reference amplitude, the maximum amplitude, the minimum amplitude, and the actual amplitude; Calculate a width score according to the reference width, the maximum width, the minimum width, and the actual width; A second matching score is calculated according to the amplitude score and the width score.

8. A small seed precision sowing monitoring method according to claim 4, characterized in that: The fusing the first matching score and the second matching score to obtain a fused matching score includes: The surface laser sensor and the infrared sensor are used to perform identification tests on seeds to be sown, respectively, to obtain a surface laser missed detection rate and an infrared missed detection rate; Determining a surface laser weight and an infrared weight of the surface laser sensor according to the surface laser missed detection rate and the infrared missed detection rate; The first matching score and the second matching score are combined with the surface laser weight and the infrared weight to perform weighted summation to obtain a fusion matching score.

9. The method for monitoring small seed precision sowing according to claim 1, characterized in that: The small seed precision sowing monitoring method further comprises: Based on the missed broadcast and the rebroadcast, obtaining a line number updated in real time by a trigger pulse signal of a line number sensor; Determine the sowing starting point of the row number, and calculate the distance between the seeder and the sowing starting point within the time window using the walking wheel encoder data obtained in real time; Positions corresponding to the missed broadcast and the rebroadcast are determined according to the row number and the distance.

10. A small seed precision sowing monitoring system, characterized in that: include: A processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute a small-grain seed precision sowing monitoring method as described in any one of claims 1 to 9.

Images