Paddy rice mature period judgment method based on image recognition

Abstract

The invention discloses a rice mature period judgment method based on image recognition, and relates to the technical field of computer vision and agricultural informationization, and the method comprises the following steps: carrying out the continuous shooting of a rice field at a set initial image capture frequency through a high-definition visible light camera carried by an unmanned plane, and obtaining the original image data of rice ears; according to the method, a dynamic regulation and control mechanism is constructed based on an image recognition and machine learning fusion technology, and intelligent sensing of the rapid swinging state of the rice ears and self-adaptive adjustment of image acquisition parameters are achieved. The space and time dimension features are extracted through feature engineering, the model evaluation result is combined, the image capture frequency and the shutter speed are dynamically optimized, gale interference is effectively dealt with, the image definition and the recognition accuracy are improved, high-quality data support is provided for rice mature period judgment, and the system environment adaptability is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of computer vision and agricultural information technology, and in particular to a rice maturity judgment method based on image recognition. Background Art

[0002] Image recognition-based rice maturity judgment uses computer vision technology to analyze images of rice plants to automatically identify their growth status and thus determine whether they are mature. This method usually uses drones to capture images of rice in the field and combines deep learning algorithms (such as convolutional neural networks (CNNs)) to extract image features such as leaf color changes, rice ear morphology, and grain fullness, thereby achieving intelligent recognition and classification of rice maturity stages. Compared with traditional methods that rely on manual observation, this technology has the advantages of being efficient, objective, and can be promoted on a large scale. It is of great significance to improving the accuracy of crop harvesting timing and the level of agricultural intelligence.

[0003] Existing technologies have the following shortcomings: They typically use drones equipped with high-definition visible light cameras (sometimes also equipped with multispectral or super-resolution cameras) to conduct low-altitude, close-range flights over rice fields, clearly capturing the morphological characteristics of rice ears from vertical or oblique angles, comprehensively capturing their appearance. The drones fly automatically along a preset route and continuously capture images, monitoring image acquisition in real time. They continuously acquire ground images at a set frequency, forming a dataset with wide coverage and high image overlap, providing high-quality raw data for subsequent image recognition and analysis.

[0004] However, in windy weather conditions, rice ears swing violently due to the influence of the wind, and the swing frequency increases significantly. If the preset image capture frequency and original camera parameters (such as shutter speed and exposure time) are still used for image acquisition in this situation without any optimization and adjustment, a series of serious consequences will occur. First, strong winds cause the rice ears to shake violently at the moment of shooting, which can easily cause image motion blur. In the blurred image, the outline of the rice ear is blurred, the color is distorted, and the edge of the grain is unclear. Key features such as the ear structure, grain plumpness, and curvature will be difficult to accurately extract, which will seriously affect the accuracy and reliability of subsequent analysis results such as image recognition and maturity judgment.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The present invention aims to provide a method for determining rice maturity based on image recognition. By integrating image recognition with machine learning, a dynamic control mechanism is constructed to intelligently perceive the rapid swaying of rice ears and adaptively adjust image acquisition parameters. This method extracts spatial and temporal features through feature engineering, and, combined with model evaluation results, dynamically optimizes image capture frequency and shutter speed. This method effectively addresses wind interference, improves image clarity and recognition accuracy, provides high-quality data support for rice maturity determination, and enhances the system's environmental adaptability, thereby addressing the aforementioned issues.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for determining rice maturity based on image recognition, comprising the following steps:

[0008] The high-definition visible light camera carried by the drone continuously photographs the rice fields at a set initial image capture frequency to obtain the original image data of the rice ears;

[0009] After obtaining the original rice ear images, the obtained raw data is preprocessed, and the valid rice ear images after preprocessing are uniformly managed and labeled to construct an organized and labeled data set;

[0010] Based on the data set, feature engineering technology is used to extract feature vectors reflecting the rapid swaying of rice ears in the current area from the rice ear images, and the obtained feature vectors are further analyzed and processed.

[0011] The analyzed and processed feature vector indicators are input into a pre-trained machine learning model, which then performs an intelligent assessment of the degree of rice ear swaying to determine whether the rice ears in the current collection area are swaying rapidly due to strong winds.

[0012] When the machine learning model determines that the rice ears in the current area are in a rapid swinging state, it triggers the image capture parameter adjustment mechanism, adaptively increases the image capture frequency, increases the number of images captured per unit area, and dynamically increases the camera's shutter speed to adapt to the rapid movement of the rice ears.

[0013] Preferably, the specific steps of continuously photographing the rice field with a high-definition visible light camera carried by the drone at a set initial image capture frequency to obtain the original image data of the rice ears include:

[0014] First, the UAV's flight route, altitude, speed, and image capture frequency parameters are set in the ground control system based on the rice field area and operational requirements.

[0015] The drone then takes off autonomously according to a preset route and conducts close-range flight operations at low altitude. At the same time, it automatically captures images of the rice fields at a set frequency using its onboard camera, continuously collecting image information above the rice ears.

[0016] During the flight, the ground operating system can monitor the image acquisition status in real time to ensure complete image coverage and stable quality;

[0017] After the flight is completed, the acquired raw image data will be downloaded or synchronized to the ground equipment for subsequent image preprocessing and analysis.

[0018] Preferably, based on the data set, feature engineering technology is used to extract feature vectors reflecting the rapid swing of rice ears in the current area from the rice ear image. The extracted features include the image structure at different resolutions extracted by the multi-scale image pyramid and the inter-frame consistency in the image sequence. The image structure at different resolutions extracted by the multi-scale image pyramid and the inter-frame consistency in the image sequence are analyzed under the detection window to generate multi-scale fluctuation reference values and grayscale reconstruction error reference values, respectively. The dynamic instability of the texture structure of the rice ear image at different spatial scales is quantified by the multi-scale fluctuation reference value, and the degree of destruction of the dynamic consistency between frames of the rice ear image in the time series is quantified by the grayscale reconstruction error reference value.

[0019] Preferably, the specific steps of analyzing the image structures at different resolutions extracted by the multi-scale image pyramid in the detection window to generate the multi-scale fluctuation reference value are as follows:

[0020] First, a multi-scale image pyramid is constructed for the rice ear image to generate multiple image levels at different resolutions. i ={I1, I2, ..., I n}, where I i is the image of the i-th layer in the image pyramid, n is the total number of layers in the image pyramid, and each layer corresponds to a scale. Then, the structural change rate index is calculated in the local window area for the two adjacent layers of images to measure the degree of difference in image structure between scales. The calculation expression is as follows:

[0021]

[0022] , where SVR i is the structural change rate indicator between the i-th layer and the i+1-th layer image, (x, y)∈Ω is the set of pixel coordinate points in the local image window area, Ω is the local image window area, is the gradient value of the i-th layer image at coordinate (x, y), is the gradient value of the i+1th layer image at coordinate (x, y), and ε is a smoothing term to prevent the denominator from being zero;

[0023] Completing the structural change rate indicator SVR i After extraction, in order to further enhance the sensitivity of the structure's "anti-interference direction", the texture direction divergence in the image is calculated. The texture direction divergence index calculation expression is as follows:

[0024]

[0025] , where TOD i is the texture direction divergence index generated by the i-th layer image in the multi-scale image pyramid, θ i (x, y) is the main texture direction angle at pixel coordinate (x, y) in the i-th level of the image pyramid, Θ i is the main texture direction set of all pixels in the monitoring window Ω of the i-th layer of the image pyramid, median(Θ i ) is the set Θ i The median direction angle, γ is the direction deviation sensitivity coefficient;

[0026] Finally, the structural change rate index SVR at multiple scales i and texture direction divergence index TOD i Fusion generates a multi-scale volatility index reference value. The generation formula is as follows:

[0027]

[0028] , where MSF is the multi-scale fluctuation reference value, α is the weight coefficient of the structure change rate index, and β is the weight coefficient of the texture direction divergence index.

[0029] Preferably, the specific steps of analyzing the inter-frame consistency in the image sequence under the detection window to generate the grayscale reconstruction error reference value are as follows:

[0030] Any frame in the detection window is selected as the current analysis frame, and its estimated reconstructed image is generated by the image reconstruction algorithm using several frames before and after it. Then, the image is divided into N×N local area blocks of the same size. In each area block, the maximum absolute difference of the pixel grayscale values in the local area is calculated. The calculation expression is as follows:

[0031]

[0032] , where δ q is the maximum grayscale reconstruction error of the qth local area block of the image, Indicates that in area block B q In the process, all pixel points (x, y) are traversed to extract the maximum value of the grayscale difference in the current area, B qis the qth local image region block, H(x, y) is the grayscale value of the current frame image at position (x, y), and R(x, y) is the grayscale value of the corresponding pixel in the reconstructed image;

[0033] The grayscale differences of the divided regions are integrated using the response weight function to construct the final grayscale reconstruction error reference value. The generation formula is as follows:

[0034]

[0035] , where GLRE is the grayscale reconstruction error reference value, e is the natural base, ω is the slope control parameter of the response function, and θ is the trigger threshold.

[0036] Preferably, the analyzed and processed multi-scale fluctuation reference value and grayscale reconstruction error reference value are input into a pre-trained machine learning model, and the rice ear swing amplitude coefficient is generated by the machine learning model. The rice ear swing amplitude coefficient is used to intelligently evaluate the rice ear swing degree to determine whether the rice ears in the current collection area are in a rapid swinging state under the influence of strong winds.

[0037] Preferably, the rice ear swing amplitude coefficient generated when the rice ear swing degree is intelligently evaluated by a pre-trained machine learning model is compared with a preset reference threshold value of the rice ear swing amplitude coefficient to determine whether the rice ears in the current collection area are in a rapid swinging state under the influence of strong wind. The judgment process is as follows:

[0038] If the rice ear swing amplitude coefficient is greater than the rice ear swing amplitude coefficient reference threshold, it is judged that the rice ears in the current collection area are in a rapid swinging state under the influence of strong wind; if the rice ear swing amplitude coefficient is less than or equal to the rice ear swing amplitude coefficient reference threshold, it is judged that the rice ears in the current collection area are not in a rapid swinging state under the influence of strong wind.

[0039] Preferably, when the machine learning model determines that the rice ears in the current area are in a rapid swaying state, the image capture frequency is adaptively increased, the number of images captured per unit area is increased, and the shutter speed of the camera is dynamically increased to adapt to the rapid movement of the rice ears. The specific steps are as follows:

[0040] When the rice ear swing amplitude coefficient (POA) given by the machine learning model is greater than the reference threshold of the rice ear swing amplitude coefficient, it indicates that the rice ears are currently in a rapid swinging state caused by strong winds. At this time, the adaptive adjustment mechanism of the image capture parameters is triggered. First, the image capture frequency of the drone is adaptively increased. The calculation expression for the adaptive adjustment of the image capture frequency is as follows;

[0041]

[0042] , where F adaptis the adaptively adjusted image capture frequency, F init is the initial set image capture frequency, μ is the adjustment coefficient used to control the sensitivity of the frequency increase, POA ref is the reference threshold of the rice ear swing amplitude coefficient, ε is a smoothing term to prevent the denominator from being zero, cos -1 It is an inverse cosine function, which ensures that the output value is monotonically increasing;

[0043] To adapt to the high risk of motion blur caused by the rapid swaying of rice ears, when the rice ear sway amplitude coefficient (POA) exceeds the reference threshold, the shutter speed of the drone's camera should be adaptively increased to more clearly capture the rapid movement of the rice ears. The dynamic shutter speed adjustment formula is as follows:

[0044]

[0045] , where S adapt is the shutter speed after dynamic adjustment, S init is the default shutter speed, It is the shutter adjustment index parameter, which is used to control the shutter change amplitude.

[0046] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0047] By introducing a dynamic control mechanism based on the combination of image recognition and machine learning, this invention achieves intelligent perception of the rapid swaying state of rice ears and adaptive adjustment of image acquisition parameters, effectively overcoming problems such as image blur and reduced recognition accuracy caused by the violent swaying of rice ears in windy weather. By extracting dynamic features in spatial and temporal dimensions through feature engineering and combining it with a machine learning model to conduct real-time assessment of the degree of rice ear sway, the system can actively adjust the image capture frequency and shutter speed under unstable environmental conditions, thereby improving image quality and feature retention. This provides stable and reliable image data support for the accurate identification of rice maturity, significantly enhancing the system's environmental adaptability and recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0049] Figure 1 The figure is a flow chart of the method for determining rice maturity based on image recognition according to the present invention. DETAILED DESCRIPTION

[0050] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0051] The present invention provides Figure 1 The rice maturity determination method based on image recognition shown includes the following steps:

[0052] The high-definition visible light camera carried by the drone continuously photographs the rice fields at a set initial image capture frequency to obtain the original image data of the rice ears;

[0053] The drone operated according to a pre-planned route and altitude, using a gimbal stabilization system to maintain a stable image during flight, ensuring the initial image quality and coverage. The primary goal of this step was to obtain a sufficient number of rice ear images with sufficient coverage to provide raw material for subsequent data analysis and feature extraction.

[0054] The high-definition visible light camera carried by the drone continuously photographs the rice field at a set initial image capture frequency. The specific steps for obtaining the original image data of rice ears include: first, according to the scope of the rice field area and operation requirements, the drone's flight route, flight altitude, speed and image capture frequency parameters are set in the ground control system; then, the drone takes off autonomously according to the preset route and performs close-range flight operations at low altitude. At the same time, the onboard camera automatically captures rice field images at a set frequency, continuously collecting image information above the rice ears; during the flight, the ground operating system can monitor the image acquisition status in real time to ensure complete image coverage and stable quality; after the flight is completed, the acquired original image data is downloaded or synchronized to the ground equipment for subsequent image preprocessing and analysis.

[0055] After obtaining the original rice ear images, the obtained raw data is preprocessed, and the valid rice ear images after preprocessing are uniformly managed and labeled, thereby constructing an organized and labeled data set (such as shooting time, shooting location, environmental conditions, etc.);

[0056] After obtaining the original rice ear images, the image data needs to be preprocessed to improve the image quality and the accuracy of subsequent analysis. Preprocessing operations usually include image denoising, color correction, brightness / contrast adjustment, cropping, resolution unification and other steps, aiming to eliminate interference factors during the shooting process and make the image clearer and more standardized. Subsequently, images that meet the quality standards are screened from the preprocessed images and uniformly managed and labeled, that is, relevant metadata information is added to each image, such as shooting time, shooting location (GPS coordinates), current weather conditions (wind speed, light), drone parameters, etc. These steps together construct a structured, standardized, and labeled data set (Dataset), providing a high-quality data foundation for subsequent feature extraction, model training and evaluation.

[0057] The dataset will serve as a crucial support for subsequent feature engineering and model evaluation. Once established, it will facilitate batch processing, feature extraction, and integration with machine learning models. It will also provide a unified standard for analyzing rice ear status at different times and using different behavioral indicators.

[0058] Based on the data set, feature engineering technology is used to extract feature vectors reflecting the rapid swaying of rice ears in the current area from the rice ear images, and the obtained feature vectors are further analyzed and processed.

[0059] Based on the data set, feature engineering technology is used to extract feature vectors reflecting the rapid swing of rice ears in the current area from rice ear images. The extracted features include image structures at different resolutions extracted by the multi-scale image pyramid and inter-frame consistency in the image sequence. The image structures at different resolutions extracted by the multi-scale image pyramid and the inter-frame consistency in the image sequence are analyzed under the detection window to generate multi-scale fluctuation reference values and grayscale reconstruction error reference values, respectively. The multi-scale fluctuation reference values are used to quantify the dynamic instability of the texture structure of rice ear images at different spatial scales, and the grayscale reconstruction error reference values are used to quantify the degree of destruction of the dynamic consistency between frames of rice ear images in the time series.

[0060] When both high- and low-frequency texture fluctuations increase in images of different resolutions extracted through a multiscale image pyramid, it generally indicates that rice ears in the current region are experiencing rapid swaying due to strong winds. This is because when the ears are normally stationary or gently swaying, their texture structure at all scales typically maintains good continuity and stability. High-frequency layers exhibit clear details and sharp edges, while low-frequency layers exhibit balanced transitions between light and dark, as well as texture. However, when the ears sway violently due to strong winds, significant spatial misalignment, texture stretching, edge ghosting, and blurring appear in the image. These changes manifest simultaneously at both high and low frequency scales, leading to increased overall texture fluctuations and decreased stability. In particular, in higher-frequency layers, detailed texture jitter is exacerbated by motion blur, while in low-frequency layers, the overall structure is disturbed, resulting in abnormal variations in light and dark. This increased cross-scale consistent fluctuation reflects the disturbance characteristics of the rice ears in multidimensional space and can serve as strong evidence for determining whether the rice ears are experiencing rapid swaying due to strong winds.

[0061] The specific steps for analyzing the image structures at different resolutions extracted by the multi-scale image pyramid under the detection window to generate the multi-scale fluctuation reference value are as follows:

[0062] First, a multi-scale image pyramid is constructed for the rice ear image to generate multiple image levels at different resolutions. i ={I1, I2, ..., I n}, where I i is the image of the i-th layer in the image pyramid, n is the total number of layers in the image pyramid, and each layer corresponds to a scale. Then, the structural change rate index is calculated in the local window area for the two adjacent layers of images to measure the degree of difference in image structure between scales. The calculation expression is as follows:

[0063]

[0064] , where SVR i is the structural change rate indicator between the i-th layer and the i+1-th layer image, (x, y)∈Ω is the set of pixel coordinate points in the local image window area, Ω is the local image window area, is the gradient value of the i-th layer image at coordinate (x, y), is the gradient value of the i+1th layer image at coordinate (x, y), ε is a smoothing term to prevent the denominator from being zero, and is generally set to a minimum value (such as 1e-5);

[0065] The above steps measure the "relative change rate" of image structures of different scales in local areas. The more drastic the change, the larger the SVR, which means that the texture distortion caused by strong winds is more obvious.

[0066] Completing the structural change rate indicator SVR iAfter extraction, in order to further enhance the sensitivity to the structure's "anti-interference direction", the texture direction divergence in the image is calculated. That is, the main texture direction field is extracted at each scale and its inconsistency is measured. The texture direction divergence index is calculated as follows:

[0067]

[0068] , where TOD i is the texture direction divergence index generated by the i-th layer image in the multi-scale image pyramid, θ i (x, y) is the main texture direction angle at pixel coordinate (x, y) in the i-th level of the image pyramid, Θ i is the main texture direction set of all pixels in the monitoring window Ω of the i-th layer of the image pyramid, median(Θ i ) is the set Θ i The median direction angle is the one in the middle of all local directions. γ is the direction deviation sensitivity coefficient, which controls the sensitivity to small deviations.

[0069] The above steps measure the degree of deviation of the local texture direction from the median direction of the layer, that is, the "directional instability" of the texture. Strong wind disturbances can cause chaotic directional distribution and significantly increase TOD.

[0070] Finally, the structural change rate index SVR at multiple scales i and texture direction divergence index TOD i Fusion generates a multi-scale volatility index reference value. The generation formula is as follows:

[0071]

[0072] , where MSF is the multi-scale fluctuation reference value, α is the weight coefficient of the structure change rate index, and β is the weight coefficient of the texture direction divergence index.

[0073] The method uses a weighted fusion of the structural change rate and texture directional divergence extracted from rice ear images at different scales to construct a unified disturbance intensity index. By generating multi-scale fluctuation reference values, it comprehensively quantifies the instability of the image's spatial structure and directional distribution, useful for determining whether the rice ear is experiencing rapid oscillation.

[0074] The larger the multiscale fluctuation reference value, generated by analyzing the image structures at different resolutions extracted from the multiscale image pyramid within the detection window, the more intense the texture disturbance and structural instability in the rice ear image at multiple scale levels in the current region. This phenomenon is often caused by the violent swaying of the rice ears due to strong winds. Rapid swaying can lead to noticeable edge blurring and detail jumps in the high-frequency layers of the image, while in the low-frequency layers, it manifests as overall texture distortion and regional grayscale inconsistency. This causes synchronous disturbances in the multiscale features, increasing the multiscale fluctuation reference value. Therefore, the numerical value of the multiscale fluctuation reference value can serve as a quantitative basis for determining whether the rice ears are experiencing rapid swaying caused by strong winds. Conversely, a smaller reference value indicates that the image structure remains stable at different scales and the rice ears are not experiencing significant swaying.

[0075] When the position and morphology of rice ears vary significantly between frames in an image sequence, it typically indicates that the rice ears in the current area are experiencing rapid swaying, largely due to strong winds. Under normal climatic conditions, rice ears exhibit minimal movement, with smooth positional and morphological changes between frames. However, in windy conditions, the wind exerts irregular and intense disturbances on the rice ears, causing them to sway and distort significantly within a short period of time. This is manifested as increased displacement, significant shape and contour changes, and a loss of texture continuity between consecutive frames in the image sequence. This dramatic inter-frame variation reflects the high temporal instability of the rice ears. Combined with the image characteristics of wind-induced dynamics, it exhibits distinct external wind interference signatures, making it an important indicator for determining whether the rice ears are affected by strong winds and experiencing rapid swaying. Quantifying these inter-frame variations, such as using a grayscale reconstruction error reference or a boundary drift reference, can further improve the accuracy and stability of the judgment.

[0076] The specific steps for analyzing the inter-frame consistency in the image sequence under the detection window to generate the grayscale reconstruction error reference value are as follows:

[0077] Any frame in the detection window is selected as the current analysis frame, and its estimated reconstructed image is generated by using several frames before and after it through image reconstruction algorithms (such as optical flow method or deformation network prediction); then, the image is divided into N×N local area blocks of the same size. In each area block, the maximum absolute difference of the pixel grayscale value in the local area is calculated. The calculation expression is as follows:

[0078]

[0079] , where δ q is the maximum grayscale reconstruction error of the qth local area block of the image, Indicates that in area block B qIn the process, all pixel points (x, y) are traversed to extract the maximum value of the grayscale difference in the current area, B q is the qth local image region block (image subregion), H(x, y) is the grayscale value of the current frame image at position (x, y), and R(x, y) is the grayscale value of the corresponding pixel in the reconstructed image;

[0080] The significance of this formula is to capture the extreme changes in the reconstruction error in local areas. It reflects the image structure mutation points more sensitively than the average error, especially the locations where rice ears undergo large displacement or bending under wind interference. These areas are often the parts where the image changes most significantly.

[0081] The grayscale differences of the divided regions are integrated using the response weight function to construct the final grayscale reconstruction error reference value. The generation formula is as follows:

[0082]

[0083] , where GLRE is the grayscale reconstruction error reference value, e is the natural base, ω is the slope control parameter of the response function, which is used to adjust the sensitivity of the index to changes, and θ is the trigger threshold, which adjusts the sensitivity of the model to identify swings. When δ q When <θ, the regional response will be weakened by the Sigmoid function, indicating that the region may be just noise or non-critical changes; when δ q When >θ, the response increases rapidly, indicating that the area is likely to be a wind disturbance area.

[0084] This step weights and amplifies regions of high-amplitude grayscale transitions in the image while ignoring minor noise interference, effectively focusing on key changes caused by strong winds. A larger GLRE value indicates more dramatic structural changes between image frames, indicating a higher likelihood of rapid rice ear movement.

[0085] The larger the grayscale reconstruction error reference value generated after analyzing the inter-frame consistency in the image sequence under the detection window, the more significant the difference between adjacent frames in the continuous image sequence, that is, there is an obvious inconsistency or deviation in the position, shape and grayscale information of the inter-frame images, which usually indicates that the rice ears in the current area have undergone drastic spatial displacement and morphological changes in a short period of time, reflecting the rapid swinging characteristics of the rice ears caused by strong winds; on the contrary, when the grayscale reconstruction error reference value is small, the consistency between continuous frames is high, the image sequence is stable, and the position and shape of the rice ears change little in a short period of time. It can be inferred that the rice ears in the current area are not affected by obvious wind disturbances and are in a normal state of relative stillness or slight swinging.

[0086] The analyzed and processed feature vector indicators are input into a pre-trained machine learning model, which then performs an intelligent assessment of the degree of rice ear swaying to determine whether the rice ears in the current collection area are swaying rapidly due to strong winds.

[0087] The analyzed and processed multi-scale fluctuation reference value and grayscale reconstruction error reference value are input into a pre-trained machine learning model, and the rice ear swing amplitude coefficient is generated by the machine learning model. The rice ear swing amplitude coefficient is used to intelligently evaluate the rice ear swing degree to determine whether the rice ears in the current collection area are in a rapid swinging state under the influence of strong winds.

[0088] A "pre-trained machine learning model" refers to a model that has been fully trained using existing sample data before the system is deployed, enabling it to accurately predict or classify under new circumstances or data. In this application scenario, the model is primarily used to output the current rice ear sway variation coefficient based on the input multi-scale fluctuation reference value and grayscale reconstruction error reference value, thereby determining whether the rice ear is in a state of rapid sway under strong winds. The training process typically includes the following key steps: First, a large amount of representative rice ear image data is collected. This data needs to cover different wind speed conditions, different sway states (such as static, slight sway, rapid sway, etc.), and changes in different environments. Second, these images are preprocessed and feature extracted to generate corresponding multi-scale fluctuation reference values and grayscale reconstruction error reference values, which are then associated with manually annotated sway state labels. Next, an appropriate machine learning model (such as support vector machine, random forest, gradient boosting tree, lightweight neural network, etc.) is selected and trained using these feature-label pairs, allowing the model to gradually learn the relationship between different reference value combinations and the actual sway state of the rice ear.

[0089] After training, the machine learning model is capable of automatically identifying and determining the swaying state of new image data. When the system captures a new rice ear image and calculates the corresponding multi-scale fluctuation reference value and grayscale reconstruction error reference value, these features are fed into the trained model. Based on the previously learned feature weight relationships and decision rules, the model quickly outputs a rice ear sway amplitude coefficient. This coefficient reflects the intensity of the rice ear sway and quantifies the swaying state. Based on a preset threshold (e.g., "variation coefficient > a certain value"), the system can determine whether the rice ears in the current acquisition area are experiencing rapid swaying due to strong winds. Because the model is trained on a large number of known samples, its judgment criteria are highly adaptable and generalizable, effectively handling complex variations in wind speed, plant density, and ambient lighting conditions. This provides a reliable and intelligent basis for dynamic adjustment of image acquisition parameters. This approach is more intelligent, stable, and practical than manual judgment or fixed threshold settings.

[0090] The machine learning model is not limited here. Any machine learning model that can perform a comprehensive analysis of the multi-scale fluctuation reference value MSF and the grayscale reconstruction error reference value GLRE to generate the rice ear swing amplitude coefficient POA can be used. In order to realize the technical solution of the present invention, the present invention provides a specific implementation method.

[0091] The formula for generating the rice ear swing amplitude coefficient POA is as follows: POA = f1·MSF+f2·GLRE, where f1 and f2 are the preset proportional coefficients of the multi-scale fluctuation reference value MSF and the grayscale reconstruction error reference value GLRE, respectively, and both f1 and f2 are greater than 0.

[0092] Preset scaling factors, denoted as f1 and f2, are weighting factors corresponding to the multiscale fluctuation reference value (MSF) and the grayscale reconstruction error reference value (GLRE) in the mathematical model for calculating the rice ear sway amplitude coefficient (POA). These factors adjust the influence of the two reference values in the overall calculation, controlling their respective contributions to the final POA. Because the MSF and GLRE have different physical units and potentially different magnitudes, their direct addition affects the interpretability and accuracy of the results. Therefore, a weighted fusion using these two scaling factors is necessary. These coefficients can be pre-set based on historical experimental data, model training results, or domain knowledge, and are typically set manually or automatically optimized during the training process. The paper indicates that when both f1 and f2 are greater than 0, it indicates that both indicators contribute positively to the POA and are not ignored. In short, the preset scaling factor serves as a weighting parameter used to balance and integrate the influence of different feature indicators on the final output, ensuring that the POA accurately reflects the intensity of rice ear sway.

[0093] It can be seen from the rice ear swing amplitude coefficient that the larger the multi-scale fluctuation reference value generated after analyzing the image structures at different resolutions extracted by the multi-scale image pyramid under the detection window, and the larger the grayscale reconstruction error reference value generated after analyzing the inter-frame consistency in the image sequence under the detection window, the larger the rice ear swing amplitude coefficient generated when the rice ear swing degree is intelligently evaluated by the pre-trained machine learning model, indicating that the rice ears in the current area are swinging rapidly due to strong winds. Otherwise, it indicates that the rice ears in the current area are not affected by obvious wind disturbances and are in a normal state of relative stillness or slight swinging.

[0094] The rice ear swing amplitude coefficient generated by the pre-trained machine learning model for intelligent evaluation of the rice ear swing degree is compared with the pre-set reference threshold of the rice ear swing amplitude coefficient to determine whether the rice ears in the current collection area are in a state of rapid swinging under the influence of strong winds. The judgment process is as follows:

[0095] If the rice ear swing amplitude coefficient is greater than the rice ear swing amplitude coefficient reference threshold, it is judged that the rice ears in the current collection area are in a rapid swinging state under the influence of strong wind; if the rice ear swing amplitude coefficient is less than or equal to the rice ear swing amplitude coefficient reference threshold, it is judged that the rice ears in the current collection area are not in a rapid swinging state under the influence of strong wind.

[0096] When the machine learning model determines that the rice ears in the current area are swaying rapidly, it triggers an adjustment mechanism for image capture parameters, adaptively increasing the image capture frequency and the number of images captured per unit area. It also dynamically increases the camera's shutter speed to adapt to the rapid movement of the rice ears.

[0097] When the machine learning model determines that the rice ears in the current area are swaying rapidly, the image capture parameter adjustment mechanism is triggered. Its main function is to adaptively adjust the image acquisition settings to adapt to the rapid movement of rice ears in a windy environment, thereby improving image quality and data availability. Specifically, adaptively increasing the image capture frequency means taking more images per unit time, and more key frames can be captured during the high-speed swaying of rice ears, increasing the probability of obtaining clear images and reducing data loss caused by blurring of a single frame. At the same time, increasing the number of images captured per unit area can form denser image coverage in the same area, providing a richer data foundation for subsequent image deblurring, image fusion, or key frame screening, and enhancing data redundancy and processing flexibility. In addition, dynamically increasing the camera's shutter speed (for example, from 1 / 500 second to 1 / 1000 second or faster) is to shorten the exposure time, allowing the camera to "freeze" the state of rice ears during high-speed movement, effectively reducing image motion blur caused by excessive target movement. By comprehensively adjusting these parameters, the stability and accuracy of image acquisition in harsh environmental conditions can be significantly improved without changing the drone's flight path, ensuring the smooth execution of subsequent analysis tasks such as image recognition, feature extraction, and maturity determination. Therefore, this step plays a critical feedback and control role in the entire system, serving as an important mechanism for achieving a dynamic balance between adaptive environmental perception and high-quality image acquisition.

[0098] When the machine learning model determines that the rice ears in the current area are swaying rapidly, it adaptively increases the image capture frequency, the number of images captured per unit area, and dynamically increases the camera's shutter speed to adapt to the rapid movement of the rice ears. The specific steps are as follows:

[0099] When the rice ear swing amplitude coefficient (POA) given by the machine learning model is greater than the reference threshold of the rice ear swing amplitude coefficient, it indicates that the rice ears are currently in a rapid swinging state caused by strong winds. At this time, the adaptive adjustment mechanism of the image capture parameters is triggered. First, the image capture frequency of the drone is adaptively increased. The calculation expression for the adaptive adjustment of the image capture frequency is as follows;

[0100]

[0101] , where F adapt is the adaptively adjusted image capture frequency, F init is the initial set image capture frequency, μ is the adjustment coefficient used to control the sensitivity of the frequency increase, POA ref is the reference threshold of the rice ear swing amplitude coefficient, ε is a smoothing term to prevent the denominator from being zero, and is generally set to a minimum value (such as 1e-5), cos -1 It is an inverse cosine function, which ensures that the output value is monotonically increasing;

[0102] The image capture frequency is dynamically adjusted based on the intensity of the rice ears' swaying. When the rice ears sway rapidly, the drone automatically increases the number of images per unit area, thereby increasing the probability of obtaining clear images. This mechanism enhances the spatiotemporal density of image data, improving the stability and effectiveness of subsequent recognition and analysis.

[0103] To adapt to the high risk of motion blur caused by the rapid swaying of rice ears, when the rice ear sway amplitude coefficient (POA) exceeds the reference threshold, the shutter speed of the drone's camera should be adaptively increased to more clearly capture the rapid movement of the rice ears. The dynamic shutter speed adjustment formula is as follows:

[0104]

[0105] , where S adapt is the shutter speed after dynamic adjustment, S init is the default shutter speed, It is the shutter adjustment index parameter, which is used to control the shutter change amplitude.

[0106] A nonlinear adjustment mechanism increases the camera's shutter speed, enabling it to more quickly capture the instantaneous position of rice ears and reduce image blur caused by high-speed oscillation. This adjustment helps maintain image clarity in harsh environments and ensures that key feature information is accurately preserved.

[0107] By introducing a dynamic control mechanism based on the combination of image recognition and machine learning, this invention achieves intelligent perception of the rapid swaying state of rice ears and adaptive adjustment of image acquisition parameters, effectively overcoming problems such as image blur and reduced recognition accuracy caused by the violent swaying of rice ears in windy weather. By extracting dynamic features in spatial and temporal dimensions through feature engineering and combining it with a machine learning model to conduct real-time assessment of the degree of rice ear sway, the system can actively adjust the image capture frequency and shutter speed under unstable environmental conditions, thereby improving image quality and feature retention. This provides stable and reliable image data support for the accurate identification of rice maturity, significantly enhancing the system's environmental adaptability and recognition accuracy.

[0108] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0109] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0110] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0111] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0112] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0113] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0114] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0115] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0116] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0117] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A rice maturity determination method based on image recognition, characterized in that: The following steps are involved: The high-definition visible light camera carried by the drone continuously photographs the rice fields at a set initial image capture frequency to obtain the original image data of the rice ears; After obtaining the original rice ear images, the obtained raw data is preprocessed, and the valid rice ear images after preprocessing are uniformly managed and labeled to construct an organized and labeled data set; Based on the data set, feature engineering technology is used to extract feature vectors reflecting the rapid swaying of rice ears in the current area from the rice ear images, and the obtained feature vectors are further analyzed and processed. The analyzed and processed feature vector indicators are input into a pre-trained machine learning model, which then performs an intelligent assessment of the degree of rice ear swaying to determine whether the rice ears in the current collection area are swaying rapidly due to strong winds. When the machine learning model determines that the rice ears in the current area are in a rapid swinging state, it triggers the image capture parameter adjustment mechanism, adaptively increases the image capture frequency, increases the number of images captured per unit area, and dynamically increases the camera's shutter speed to adapt to the rapid movement of the rice ears.

2. The rice maturity determination method based on image recognition according to claim 1, characterized in that: The specific steps for obtaining the original image data of rice ears by continuously photographing the rice field with a high-definition visible light camera carried by the drone at a set initial image capture frequency include: First, the UAV's flight route, altitude, speed, and image capture frequency parameters are set in the ground control system based on the rice field area and operational requirements. The drone then takes off autonomously according to a preset route and conducts close-range flight operations at low altitude. At the same time, it automatically captures images of the rice fields at a set frequency using its onboard camera, continuously collecting image information above the rice ears. During the flight, the ground operating system can monitor the image acquisition status in real time to ensure complete image coverage and stable quality; After the flight is completed, the acquired raw image data will be downloaded or synchronized to the ground equipment for subsequent image preprocessing and analysis.

3. The rice maturity determination method based on image recognition according to claim 1, characterized in that: Based on the data set, feature engineering technology is used to extract feature vectors reflecting the rapid swing of rice ears in the current area from rice ear images. The extracted features include image structures at different resolutions extracted by the multi-scale image pyramid and inter-frame consistency in the image sequence. The image structures at different resolutions extracted by the multi-scale image pyramid and the inter-frame consistency in the image sequence are analyzed under the detection window to generate multi-scale fluctuation reference values and grayscale reconstruction error reference values, respectively. The multi-scale fluctuation reference values are used to quantify the dynamic instability of the texture structure of rice ear images at different spatial scales, and the grayscale reconstruction error reference values are used to quantify the degree of destruction of the dynamic consistency between frames of rice ear images in the time series.

4. The rice maturity determination method based on image recognition according to claim 3, characterized in that: The specific steps for analyzing the image structures at different resolutions extracted by the multi-scale image pyramid under the detection window to generate the multi-scale fluctuation reference value are as follows: First, a multi-scale image pyramid is constructed for the rice ear image to generate multiple image levels at different resolutions. i ={I1, I2, ..., I n }, where I i is the image of the i-th layer in the image pyramid, n is the total number of layers in the image pyramid, and each layer corresponds to a scale. Then, the structural change rate index is calculated in the local window area for the two adjacent layers of images to measure the degree of difference in image structure between scales. The calculation expression is as follows: ， Where, SVR i is the structural change rate indicator between the i-th layer and the i+1-th layer image, (x, y)∈Ω is the set of pixel coordinate points in the local image window area, Ω is the local image window area, is the gradient value of the i-th layer image at coordinate (x, y), is the gradient value of the i+1th layer image at coordinate (x, y), and ε is a smoothing term to prevent the denominator from being zero; Completing the structural change rate indicator SVR i After extraction, in order to further enhance the sensitivity of the structure's "anti-interference direction", the texture direction divergence in the image is calculated. The texture direction divergence index calculation expression is as follows: ， In the formula, TOD i is the texture direction divergence index generated by the i-th layer image in the multi-scale image pyramid, θ i (x, y) is the main texture direction angle at pixel coordinate (x, y) in the i-th level of the image pyramid, Θ i is the main texture direction set of all pixels in the monitoring window Ω of the i-th layer of the image pyramid, median(Θ i ) is the set Θ i The median direction angle, γ is the direction deviation sensitivity coefficient; Finally, the structural change rate index SVR at multiple scales i and texture direction divergence index TOD i Fusion generates a multi-scale volatility index reference value. The generation formula is as follows: ， Where MSF is the multi-scale fluctuation reference value, α is the weight coefficient of the structural change rate index, and β is the weight coefficient of the texture direction divergence index.

5. The rice maturity determination method based on image recognition according to claim 3, characterized in that: The specific steps for analyzing the inter-frame consistency in the image sequence under the detection window to generate the grayscale reconstruction error reference value are as follows: Any frame in the detection window is selected as the current analysis frame, and its estimated reconstructed image is generated by the image reconstruction algorithm using several frames before and after it. Then, the image is divided into N×N local area blocks of the same size. In each area block, the maximum absolute difference of the pixel grayscale values in the local area is calculated. The calculation expression is as follows: , where δ q is the maximum grayscale reconstruction error of the qth local area block of the image, Indicates that in area block B q In the process, all pixel points (x, y) are traversed to extract the maximum value of the grayscale difference in the current area, B q is the qth local image region block, H(x, y) is the grayscale value of the current frame image at position (x, y), and R(x, y) is the grayscale value of the corresponding pixel in the reconstructed image; The grayscale differences of the divided regions are integrated using the response weight function to construct the final grayscale reconstruction error reference value. The generation formula is as follows: ， Where GLRE is the grayscale reconstruction error reference value, e is the natural base, ω is the slope control parameter of the response function, and θ is the trigger threshold.

6. The rice maturity determination method based on image recognition according to claim 3, characterized in that: The analyzed and processed multi-scale fluctuation reference value and grayscale reconstruction error reference value are input into a pre-trained machine learning model, and the rice ear swing amplitude coefficient is generated by the machine learning model. The rice ear swing amplitude coefficient is used to intelligently evaluate the rice ear swing degree to determine whether the rice ears in the current collection area are in a rapid swinging state under the influence of strong winds.

7. The rice maturity determination method based on image recognition according to claim 6, characterized in that: The rice ear swing amplitude coefficient generated by the pre-trained machine learning model for intelligent evaluation of the rice ear swing degree is compared with the pre-set reference threshold of the rice ear swing amplitude coefficient to determine whether the rice ears in the current collection area are in a state of rapid swinging under the influence of strong winds. The judgment process is as follows: If the rice ear swing amplitude coefficient is greater than the rice ear swing amplitude coefficient reference threshold, it is judged that the rice ears in the current collection area are in a rapid swinging state under the influence of strong wind; if the rice ear swing amplitude coefficient is less than or equal to the rice ear swing amplitude coefficient reference threshold, it is judged that the rice ears in the current collection area are not in a rapid swinging state under the influence of strong wind.

8. The rice maturity determination method based on image recognition according to claim 7, characterized in that: When the machine learning model determines that the rice ears in the current area are swaying rapidly, it adaptively increases the image capture frequency, the number of images captured per unit area, and dynamically increases the camera's shutter speed to adapt to the rapid movement of the rice ears. The specific steps are as follows: When the rice ear swing amplitude coefficient (POA) given by the machine learning model is greater than the reference threshold of the rice ear swing amplitude coefficient, it indicates that the rice ears are currently in a rapid swinging state caused by strong winds. At this time, the adaptive adjustment mechanism of the image capture parameters is triggered. First, the image capture frequency of the drone is adaptively increased. The calculation expression for the adaptive adjustment of the image capture frequency is as follows; ， Where, F adapt is the adaptively adjusted image capture frequency, F init is the initial set image capture frequency, μ is the adjustment coefficient used to control the sensitivity of the frequency increase, POA ref is the reference threshold of the rice ear swing amplitude coefficient, ε is a smoothing term to prevent the denominator from being zero, cos -1 It is an inverse cosine function, which ensures that the output value is monotonically increasing; To adapt to the high risk of motion blur caused by the rapid swaying of rice ears, when the rice ear sway amplitude coefficient (POA) exceeds the reference threshold, the shutter speed of the drone's camera should be adaptively increased to more clearly capture the rapid movement of the rice ears. The dynamic shutter speed adjustment formula is as follows: ， Where S adapt is the shutter speed after dynamic adjustment, S init is the default shutter speed, It is the shutter adjustment index parameter, which is used to control the shutter change amplitude.

Images