A rice plant growth state monitoring method and system based on image recognition
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GOLDEN RICE SEED CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-09
Smart Images
Figure CN122176506A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and specifically to a method and system for monitoring the growth status of rice plants based on image recognition. Background Technology
[0002] As an important food crop, rice requires crucial phenotypic screening during its breeding process. Among these, grain filling degree (i.e., the fullness of the panicle grains) and panicle morphology are key indicators for evaluating the yield potential and lodging resistance of varieties. In modern agricultural breeding, how to quickly, non-destructively, and accurately obtain these phenotypic data is a core challenge for accelerating the selection of superior varieties.
[0003] Current mainstream rice growth monitoring technologies primarily rely on computer vision and static image processing algorithms. These technologies typically utilize visible light cameras mounted on drones or ground robots to acquire field images, extracting target areas of rice panicles through color space conversion, threshold segmentation, or morphological operations. They then calculate the projected area, coverage, or relative chlorophyll content based on color indices to estimate crop maturity. While these two-dimensional static image-based methods have achieved some success in assessing leaf area index or biomass, they have significant limitations in monitoring the specific internal physiological indicator of rice panicle grain filling.
[0004] First, static image processing faces a serious challenge of visual confusion. In two-dimensional images under natural lighting conditions, unfertilized or poorly filled empty rice ears and full, ripe rice ears exhibit extremely high similarity in external morphology, projected area, and surface color characteristics. It is difficult to perceive the internal starch filling and weight differences through the husk solely based on external geometric features or color information. This overlap of phenotypic features leads to low accuracy of traditional algorithms in distinguishing between empty husks and ripe grains, making misjudgments highly likely.
[0005] Secondly, existing image monitoring technologies lack effective utilization and physical modeling of complex field environmental factors. Field environments are often accompanied by wind fields of varying strengths, causing rice plants to constantly sway dynamically. Traditional approaches generally treat this wind-induced plant movement as interference or noise, tending to avoid motion blur through complex image stabilization algorithms or selecting windless times for shooting, aiming to obtain clear, still images. However, this approach ignores the important physical mechanism behind object motion: objects of different masses produce different inertial responses under the same wind excitation. Full-filled rice ears, due to their large mass and inertia, exhibit significantly different swaying frequencies and amplitudes compared to lighter, empty rice ears. Existing technologies fail to capture and utilize this crucial dynamic characteristic, resulting in the loss of the possibility of sensing the weight dimension of rice ears non-contactly. Summary of the Invention
[0006] The purpose of this invention is to propose a method and system for monitoring the growth status of rice plants based on image recognition, in order to solve the technical problems in the prior art that rely on static images and cannot distinguish between empty rice husks and full grains, and that are easily affected by natural wind interference, resulting in inaccurate monitoring results.
[0007] To address the above problems, the present invention proposes a technical solution for monitoring the growth status of rice plants based on image recognition: A method for monitoring the growth status of rice plants based on image recognition includes: A time-series image sequence of rice canopy was obtained, and each frame in the time-series image sequence was subjected to environment-adaptive segmentation and deblurring to obtain a clear target region of rice ears. The morphological skeleton of the target region of the rice panicle is extracted, and the dynamic characteristics of the rice panicle under wind excitation are calculated by tracking the motion trajectory of the morphological skeleton using a curve fitting algorithm. The target region of the rice ear is subjected to gray-level compression, and the micro-texture features of the rice ear surface are calculated based on the gray-level co-occurrence matrix. An environmental wind speed reference value is obtained and the weights are adaptively adjusted based on the environmental wind speed reference value. A multi-dimensional coupled evaluation model is constructed by combining the dynamic features and the micro-texture features, and the comprehensive score of rice grain filling maturity is calculated.
[0008] Beneficial effects: This invention overcomes the problem of single-frame images being easily interfered with by environmental noise and motion blur by acquiring time-series image sequences and combining them with environmental adaptive segmentation and deblurring processing, thus ensuring the reliability of the feature extraction source data. In addition, by extracting the dynamic characteristics of rice ears under wind excitation, this invention utilizes the physical difference in mass inertia between full grains and empty husks, solving the technical problem that traditional static image analysis cannot distinguish rice ears that look similar but have different internal fullness. At the same time, this invention combines gray-level co-occurrence matrix to calculate micro-texture features and adaptively adjusts the weights according to the environmental wind speed reference value, realizing stable monitoring under different wind conditions and effectively improving the accuracy of the comprehensive score of rice grain filling maturity.
[0009] Furthermore, the environmental adaptive segmentation and deblurring process for each frame of the temporal image sequence includes: Gaussian filtering is used to denoise a single frame image, removing random noise generated by the image sensor; The optimal segmentation threshold for the current frame is calculated using the maximum inter-class variance method, and the image is binarized to separate the rice ears foreground from the field ridges background. The image edge gradient intensity is detected. When the edge gradient intensity is lower than a preset sharpness threshold, Wiener filtering is used to restore the image based on the estimated point spread function to eliminate motion blur caused by wind.
[0010] Furthermore, the step of using a curve fitting algorithm to track the motion trajectory of the morphological skeleton and calculate the dynamic characteristics of the rice spike under wind excitation includes: Morphological thinning is performed on the binarized image to extract the central skeleton of the rice ear with a single pixel width; The bottom end of the rice panicle central skeleton is selected as the starting point and the top end as the ending point. The point with the largest curvature is selected as the control point, and the third-order Bézier curve is used to fit the rice panicle central skeleton. By tracking the angle between the top tangent of the fitted curve and the vertical direction in each frame, an angle time series is constructed. Perform a fast Fourier transform on the angle time series to obtain the frequency with the highest power spectral density as the main oscillation frequency, and calculate the variance of the angle time series as the oscillation variance. The dynamic characteristics include the principal oscillation frequency and the oscillation variance.
[0011] Beneficial effects: This scheme uses morphological refinement to extract the central skeleton of the rice spike and fits it with a third-order Bézier curve, which solves the problem of the difficulty in accurately tracking the non-rigid deformation of the rice spike as a flexible object in the wind; by tracking the angle between the tangent at the top of the fitted curve and the vertical direction to construct an angular time series and performing a fast Fourier transform, the main oscillation frequency and oscillation variance can be accurately calculated, thereby transforming the complex visual motion into dynamic parameters that reflect the physical inertia of the rice spike.
[0012] Furthermore, the calculation of the micro-texture features of the rice ear surface based on the gray-level co-occurrence matrix includes: Convert the target area of rice ears into a grayscale image and compress the grayscale levels to 16 levels; Construct four gray-level co-occurrence matrices with a step size of 1 pixel and directions of 0 degrees, 45 degrees, 90 degrees, and 135 degrees, and calculate their mean values to obtain the comprehensive matrix; Contrast and homogeneity are calculated based on the comprehensive matrix, where contrast represents the depth of texture grooves and homogeneity represents the uniformity of texture distribution. The microtexture features include the contrast and the homogeneity.
[0013] Beneficial effects: This scheme eliminates the influence of different rice panicle growth directions on texture calculation by constructing gray-level co-occurrence matrices in multiple directions and taking the average value to obtain a comprehensive matrix; based on the comprehensive matrix, the contrast and homogeneity can be calculated, which can reflect the depth of the grooves and the density of the arrangement on the surface of the rice panicle at the micro level, providing a basis for judging the grain filling fullness in addition to dynamic characteristics.
[0014] Furthermore, before calculating the comprehensive score for rice grain-filling maturity, the calculation also includes calculating the gravitational inertial damping index, the formula of which is:
[0015] In the formula, This represents the gravitational inertial damping index; This represents the normalized environmental wind speed proxy value; Indicates the main oscillation frequency; Indicates the variance of the swing; To prevent odd outliers; It is the basic noise constant.
[0016] Beneficial effects: This solution utilizes the principle of physics that, under the same wind speed excitation, objects with greater mass exhibit lower frequencies and smaller variances. By calculating the differences in values, it effectively distinguishes between empty husks and full grains, achieving non-contact perception of the weight attributes of rice ears.
[0017] Furthermore, before calculating the comprehensive score of rice grain-filling maturity, the textural density factor is also calculated, and its calculation formula is as follows:
[0018] In the formula, Indicates the texture density factor; This represents the contrast of the normalized gray-level co-occurrence matrix; This indicates the homogeneity of the normalized gray-level co-occurrence matrix; This represents the average gray value of the target area of rice ears; This is the color correction factor.
[0019] Beneficial effects: This scheme integrates the contrast, homogeneity, and average gray value of the gray-level co-occurrence matrix and introduces a color correction coefficient, enabling the model to not only evaluate the roughness of the texture but also take into account the color change characteristics of mature rice ears, thus improving the comprehensiveness and accuracy of the description of the physical properties of rice ear surfaces.
[0020] Furthermore, the formula for calculating the comprehensive score of rice grain-filling maturity is as follows:
[0021] In the formula, This represents the overall score for the grain-filling maturity of rice. Basic score; The gravitational inertial damping index is mentioned above. Texture density factor; For dynamically adaptive weighting coefficients, and , For adjustment coefficients, The wind speed threshold, This is the normalized environmental wind speed proxy value.
[0022] Beneficial effects: This solution can automatically adjust the weight distribution of the gravitational inertial damping index and texture density factor based on the real-time acquired environmental wind speed proxy value. When the wind speed is high, it emphasizes dynamic characteristics, and when the wind speed is low or there is no wind, it emphasizes texture characteristics, thus ensuring the adaptability of the monitoring system in all-weather environments and the stability of the evaluation results.
[0023] Furthermore, obtaining the ambient wind speed reference value includes: Select the leaf tip region in the image background as the reference region; The average amplitude of the optical flow field in the reference area is calculated using the optical flow method, and the average amplitude of the optical flow field is normalized to obtain an environmental wind speed proxy value with a value between 0 and 1, which is used as the environmental wind speed reference value.
[0024] Furthermore, the method for acquiring the time-series image sequence is as follows: using an industrial-grade camera set up in the field, shooting from a 45-degree angle (overhead or side view), and acquiring a video stream with a frame rate of 60 frames per second and a duration of 5 seconds.
[0025] The technical solution of the rice plant growth status monitoring system based on image recognition proposed in this invention is as follows: A rice plant growth status monitoring system based on image recognition includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the image recognition-based rice plant growth status monitoring method described in any of the above technical solutions is implemented.
[0026] The beneficial effects of this invention are as follows: This invention enables precise monitoring of the growth status of rice plants in complex field environments, effectively solving the technical problem of traditional static image technology, which struggles to distinguish between empty husks and full grains due to their similar appearance. The solution of this invention can fully utilize the natural wind field environment to accurately capture the dynamic characteristics and surface microtexture features of rice panicles under wind excitation, thereby truly reflecting the fullness and physical properties of the rice panicles. Simultaneously, this invention can automatically balance the weights of dynamic characteristics and microtexture features under different wind speeds, ensuring the stability and reliability of the evaluation results under windy or calm conditions. Attached Figure Description
[0027] Figure 1 This is a flowchart of the rice plant growth status monitoring method based on image recognition provided in an embodiment of the present invention; Figure 2 This is a graph showing the inertial damping response characteristics under wind field excitation established in an embodiment of the present invention. Figure 3 This is a multidimensional clustering distribution map based on gravity inertia and texture features in an embodiment of the present invention; Figure 4 This is a comparison chart of the accuracy of monitoring rice plant growth status under different working conditions between the present invention and existing technologies. Detailed Implementation
[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0029] Specific embodiments of the image recognition-based rice plant growth status monitoring method proposed in this invention: like Figure 1 As shown, the image recognition-based rice plant growth status monitoring method in this embodiment includes the following steps: S1. Obtain a time-series image sequence of the rice canopy, and perform environment-adaptive segmentation and deblurring on each frame of the time-series image sequence to obtain a clear target region of the rice ear.
[0030] Specifically, to capture and evaluate the dynamic response characteristics of rice under wind excitation, it is necessary to acquire a continuous time-series image sequence. In this embodiment, an industrial-grade CMOS camera installed in the field is used as the acquisition device, with the camera resolution set to 1920 x 1080 pixels and the frame rate set to 60 frames per second. The camera is installed at a 45-degree angle between the lens optical axis and the horizontal plane, capturing the rice canopy from a top-down or side-view perspective. This angle minimizes leaf shading and highlights the morphology of the rice ears. The acquisition strategy is set to automatically trigger an acquisition task every 30 minutes, acquiring a 5-second video stream each time, thus obtaining a time-series image sequence containing 300 frames.
[0031] After data acquisition, each frame of the video stream is preprocessed. Considering the thermal noise generated by the image sensor during long-term operation and in high-temperature outdoor environments, a Gaussian filtering algorithm is used to denoise each frame. By selecting an appropriate-sized Gaussian kernel and performing convolution with the image matrix, high-frequency random noise in the image is smoothed while preserving the main edge information. Subsequently, the optimal segmentation threshold for the current frame is automatically calculated using the maximum inter-class variance method. This algorithm searches for the gray value that maximizes the inter-class variance between the foreground target pixel set and the background pixel set as the optimal segmentation threshold by traversing all gray levels of the image. The denoised image is then binarized using this optimal segmentation threshold. Regions with pixel values higher than the optimal segmentation threshold are marked as rice ears foreground and assigned a pixel value of 1, while regions with pixel values lower than or equal to the optimal segmentation threshold are marked as field ridge background and assigned a pixel value of 0.
[0032] To address the issue of motion blur caused by rapid displacement of rice ears during camera exposure time in high wind speeds in the field, the system detects edge gradient intensity in each frame to determine motion blur. Specifically, an edge detection operator calculates the gradient magnitude of the rice ear edges in the binarized image. When the edge gradient intensity is below a preset sharpness threshold, the frame is considered to have motion blur. In this case, Wiener filtering is used to restore the image based on the estimated point spread function. The point spread function describes the imaging system's blur response to a point light source, while Wiener filtering, based on the minimum mean square error criterion and combined with the image's signal-to-noise ratio estimate, performs deconvolution on the degraded blurred image, thereby eliminating motion blur caused by wind and restoring the sharpness and detail of the rice ear edges.
[0033] Through the above-mentioned adaptive segmentation and deblurring process, this step can stably extract high-quality, well-defined rice ear target regions from field backgrounds with complex lighting changes and dynamic interference, eliminating the interference of environmental noise and motion blur on subsequent feature extraction.
[0034] S2. Extract the morphological skeleton of the target region of the rice panicle, and use a curve fitting algorithm to track the motion trajectory of the morphological skeleton to calculate the dynamic characteristics of the rice panicle under wind excitation.
[0035] Specifically, considering that rice ears are typical flexible objects, they exhibit complex non-rigid deformations in natural wind fields. Traditional centroid-based tracking methods are prone to failure due to leaf shading or the bending deformation of the rice ears themselves. Therefore, this step uses a combination of morphological skeleton extraction and curve fitting to accurately describe the motion state of the rice ears.
[0036] First, a morphological thinning operation is performed on the binarized image of the clear rice ear target region obtained in step S1. This operation iteratively strips the boundary pixels of the foreground region of the image, eroding the thick rice ear target region into a central skeleton of the rice ear with only a single pixel width while maintaining the topological structure of the object.
[0037] Subsequently, to evaluate the geometry of the skeleton and eliminate the influence of pixel-level jitter, a third-order Bézier curve was used to smoothly fit the extracted rice panicle central skeleton. During the fitting process, the bottom of the rice panicle central skeleton, i.e., the connection point between the stem and the stalk, was selected as the starting point. The apex of the central skeleton of the rice panicle was selected as the termination point. Meanwhile, at approximately one-third and two-thirds of the length of the central skeleton of the rice panicle, two points with the largest curvature were selected as the first control points. Second control point The fitted curve equation is constructed as follows:
[0038] Among them, parameters Let be a variable whose value ranges from 0 to 1. The curve equation is based on Bernstein basis functions, where ... The coefficient of the term changes with The increase and decrease, The term coefficient is The curve grows in three terms, with the middle two terms controlling the tangent direction and curvature of the curve at the beginning and end, respectively. Specifically, Indicates the corresponding parameter The coordinate vector of a point on the fitted curve at any given time. and The starting and ending points of the curve have been determined. and By influencing the derivative properties of the curve, the bending direction of the curve can be controlled, thus accurately approximating the physical shape of rice ears bending in the wind.
[0039] After curve fitting for each frame, the dynamic characteristics of the rice spike are calculated by tracking changes in the geometric features of the fitted curve. The specific calculation logic is as follows: assuming that in the first frame... In the frame image, the top of the fitted curve The coordinates are and bottom The coordinates are and The bottom coordinates remain fixed. The system calculates the angle between the tangent at the top of the fitted curve and the vertical direction in each frame, or simplifies the calculation by calculating the swing angle of the top relative to the bottom. The angle is obtained by calculating the arctangent function of the ratio of horizontal displacement to vertical distance. A continuous time-series image sequence is processed to construct an angle time series containing angle data from all frames.
[0040] Next, a Fast Fourier Transform (FFT) was performed on the aforementioned angular time series to convert the oscillation signal in the time domain into a frequency domain signal. The frequency corresponding to the maximum power spectral density in the spectrum was then identified and determined as the principal oscillation frequency. Simultaneously, the mean of the sum of squares of the differences between each angle value and the average angle value in the angular time series was calculated using the statistical variance formula to obtain the oscillation variance. The dynamic characteristics of the rice spike include both the principal oscillation frequency and the oscillation variance. It should be noted that all data were dimensionless before the calculations.
[0041] Through the above-mentioned Bézier curve fitting and frequency domain analysis, this step successfully transformed the complex flexible deformation of rice ears into frequency and variance indices, accurately capturing the dynamic response characteristics of rice ears under wind excitation.
[0042] S3. Perform gray-level compression on the target area of the rice ear, and calculate the micro-texture features of the rice ear surface based on the gray-level co-occurrence matrix.
[0043] Fully filled rice ears, due to the ample filling of starch within, exhibit round and plump grains with a tight arrangement, resulting in deep, alternating light and dark textured grooves on the ear surface under light. In contrast, empty rice ears have relatively flat surfaces with indistinct texture features. To assess this microscopic visual difference, this embodiment employs gray-level co-occurrence matrix (GLCM) technology to extract texture features.
[0044] Specifically, the clear rice ear target area obtained in step S1 is converted from a color image to an 8-bit grayscale image, at which point the pixel grayscale level ranges from 0 to 255. To reduce the computational complexity of calculating the gray-level co-occurrence matrix and improve the algorithm's stability to minor changes in ambient lighting, the system performs grayscale compression processing on the image, compressing the original 256 grayscale levels to 16 grayscale levels. The calculation formula is defined as: the new pixel value equals the old pixel value divided by 16 and rounded down. This step effectively removes minor grayscale fluctuation noise caused by uneven lighting, making the texture features more significant and stable.
[0045] Next, the system constructs a gray-level co-occurrence matrix to describe the spatial correlation of gray values in the image. Considering the diversity of rice ears' growth directions in the field, texture statistics in a single direction cannot fully reflect its characteristics. Therefore, this embodiment selects a step size of 1 pixel and calculates four gray-level co-occurrence matrices for directions of 0 degrees, 45 degrees, 90 degrees, and 135 degrees, respectively. To obtain a comprehensive feature with rotation invariance, the gray-level co-occurrence matrices of the above four directions are superimposed and averaged to obtain a normalized comprehensive matrix. In this comprehensive matrix In, elements Represents grayscale level and grayscale The probability of adjacent occurrences in a rice ear image.
[0046] Based on this comprehensive matrix Further calculations are performed to determine the contrast, which reflects texture depth, and the homogeneity, which reflects texture uniformity.
[0047] Among them, contrast The calculation formula is:
[0048] This formula utilizes the concept of moment of inertia, and the term... As a weighting coefficient, the greater the difference in gray values between adjacent pixels in an image, i.e., the deeper the texture grooves and the clearer the edges, and The larger the difference, the more quadratically the weighting coefficients increase, significantly amplifying the contribution of the difference to the sum, resulting in a larger calculated contrast value. Conversely, if the image is flat, and If the values are similar, the contrast value is smaller. Therefore, this feature can effectively characterize the shadow depth caused by the gaps between grains on the rice panicle surface.
[0049] homogeneity The calculation formula is:
[0050] This formula uses the inverse moment method, and the denominator term... As a weighting adjustment factor, when the gray values of adjacent pixels are very close or even equal, the denominator term approaches 1, maximizing the contribution of that point to the sum; conversely, when the gray values of adjacent pixels differ significantly, the denominator term increases rapidly, causing the contribution of that term to decrease sharply. Therefore, a higher homogeneity value indicates a more regular distribution of image texture and more uniform local variations.
[0051] Through the above steps, by utilizing gray-level compression and multi-directional gray-level co-occurrence matrix analysis, this invention can extract two key micro-texture features, contrast and homogeneity, from the micro-level. This compensates for the lack of obvious features in macro-morphological skeleton analysis under calm or light wind conditions, and provides important static visual basis for the subsequent construction of a multi-dimensional coupled evaluation model.
[0052] S4. Obtain the environmental wind speed reference value and adaptively adjust the weights based on the environmental wind speed reference value. Combine the dynamic features and the micro-texture features to construct a multi-dimensional coupled evaluation model and calculate the comprehensive score of rice grain filling maturity.
[0053] This step aims to establish a mapping relationship between physical stimuli, dynamic responses, and static textures, overcoming the limitations of single features in different environments through multi-source feature fusion. First, the system needs to acquire an environmental wind speed proxy value to characterize the current external stimulus intensity. During implementation, the leaf tip region, which is most sensitive to wind response in the image background, is selected as a reference region, and the average amplitude of the optical flow field at feature points within this region is calculated using optical flow. To eliminate the dimensional influence of different shooting distances or camera parameters, the calculated average amplitude is mapped to the interval between 0 and 1, resulting in a normalized environmental wind speed proxy value. For example, if the maximum wind speed calibrated by the system corresponds to an optical flow amplitude of 100 pixels per frame, when the detected average leaf tip movement speed is 50 pixels per frame, the environmental wind speed proxy value is 0.5.
[0054] Based on the acquired environmental wind speed proxy value and the dynamic characteristics calculated in step S2, a gravity inertial damping index is constructed. The design principle of this index is based on a damped vibration physical model, namely, under the same wind excitation, fully-filled rice ears, due to their large mass and inertia, exhibit a low oscillation frequency and small oscillation amplitude, while empty rice ears exhibit a high oscillation frequency and large oscillation amplitude. The formula for calculating the gravity inertial damping index is:
[0055] In the formula, This represents the gravitational inertial damping exponent; a larger value indicates a fuller rice ear. The numerator reflects the relationship between excitation and frequency. The normalized environmental wind speed proxy value represents the magnitude of the external stimulus force; The term represents the principal oscillation frequency extracted in step S2. Since the actual particle frequency is low, taking its reciprocal increases the value. Combined with the logarithmic function ln, this term monotonically increases as the frequency decreases. To prevent extremely small positive integers with a denominator of zero, a value of 0.01 is used. The denominator term reflects the system's response amplitude. For the variance of the swing, The base noise constant, with a value of 0.05, is used to stabilize the value under static conditions. The physical meaning of the entire formula lies in using the wind speed excitation multiplied by the inverse frequency response, and then divided by the oscillation amplitude, thereby mathematically transforming the invisible mass inertia into a visible dimensionless exponent.
[0056] Subsequently, combining the micro-texture features extracted in step S3 with the image grayscale information, the texture density factor is calculated. The texture density factor aims to assess the maturity of rice ears from a microscopic visual perspective, and its calculation formula is as follows:
[0057] In the formula, This represents the texture density factor. The first part of the formula uses the geometric mean to incorporate the contrast of the normalized gray-level co-occurrence matrix. and homogeneity This is used to comprehensively characterize the clarity and regularity of the texture. The latter part of the formula introduces a color correction term, where... This represents the average gray value of the target area of the rice ear; The color correction factor is set to 0.3. The color correction term is designed based on the biological characteristics of rice ripening, where its color changes from green to yellow and its brightness increases. When the rice ears turn yellow and mature, the average gray value... This increases the overall density factor, thereby achieving an organic combination of texture structure and color features.
[0058] To achieve stable monitoring around the clock, a multi-dimensional coupled evaluation model was constructed to calculate the comprehensive score of rice grain-filling maturity. The core of this model lies in introducing dynamic adaptive weighting coefficients, which automatically adjust the contribution ratios of dynamic and textural features based on real-time wind speed. The formula for calculating the comprehensive score of rice grain-filling maturity is as follows:
[0059] In the formula, This represents the final comprehensive score for rice grain filling maturity. Set the base score, for example, 20 points. The dynamic adaptive weighting coefficients are calculated following a modified version of the Sigmoid function:
[0060] In the formula, This is an adjustment coefficient used to control the sensitivity of weight changes; for example, a value of 10 is used. Let's say the wind speed threshold is 0.3. The principle behind this weighting formula is: when the ambient wind speed proxy value... Significantly greater than the wind speed threshold When the exponential term approaches 0, When the value approaches 1, the model primarily relies on the gravitational inertial damping exponent. An assessment should be conducted, making full use of the dynamic information provided by the wind field to distinguish between real and virtual phenomena; when the ambient wind speed is low or there is no wind, [further assessment is needed]. When the value approaches 0, the model automatically switches to primarily rely on the texture density factor. The system utilizes the exponential function exp to perform nonlinear mapping on texture features to improve discriminative power. Through this multidimensional coupling mechanism, the system can utilize physical inertial features when there is wind and microscopic texture features when there is no wind, thereby calculating a comprehensive score that truly reflects the grain-filling state of rice, achieving accurate and stable monitoring of the growth status of rice plants.
[0061] The following combination Figures 2-4 The effects of the present invention will be further explained.
[0062] Figure 2 This study demonstrates the dynamic response characteristics of rice plants under wind excitation. The figures reveal that, under the same environmental wind speed, rice panicles at different developmental stages exhibit drastically different trajectories. The oscillation response curve of empty samples shows an extremely high oscillation frequency and a large oscillation amplitude, with its fluctuation trend closely following the changes in wind speed, indicating that the extremely light empty panicles possess almost no motion inertia. In stark contrast, the oscillation response curve of full samples exhibits a smooth and low-frequency characteristic, with significantly suppressed oscillation amplitude. This delayed and stable response fully reflects the gravitational inertia and damping effect brought about by starch filling in full rice panicles.
[0063] Figure 3 The figure presents a multidimensional clustering distribution based on gravitational inertia and texture features. Data points representing different growth stages are distributed throughout the figure. It can be observed that rice panicle samples from the early development stage, mid-filling stage, and maturity stage form independent, non-overlapping clusters in the feature space. Samples from the early development stage are concentrated in the lower left of the coordinate system, corresponding to a lower gravitational inertia damping index and a smaller texture density factor. As the grain filling progresses, the sample points continuously migrate to the upper right. Finally, mature stage samples cluster in a region with both a high gravitational inertia damping index and a high texture density factor. This obvious linear migration trend and good cluster separation indicate that the feature fusion model constructed in this invention can accurately assess the grain filling level of rice and achieve precise classification of growth states.
[0064] Figure 4 The bar chart visually illustrates the difference in monitoring accuracy between the present invention and existing technologies under different operating conditions. In strong wind environments, existing technologies suffer a significant drop in accuracy to around 65% due to their inability to handle motion blur and the influence of wind noise, while the present invention, leveraging its advantages in dynamic analysis, improves accuracy to 92%. In low-light environments and under complex background obstruction, the present invention, utilizing the anti-interference capabilities of micro-texture features, maintains a recognition level of over 85%, significantly outperforming traditional technologies. Data from the comprehensive average operating conditions shows that the monitoring accuracy of the present invention exceeds 90%, while existing technologies only maintain around 68%. This comparative result fully demonstrates that the multi-dimensional coupled evaluation model proposed in this invention has superior technical effects in improving monitoring accuracy, enhancing system stability, and coping with complex field environments.
[0065] Specific embodiments of the rice plant growth status monitoring system based on image recognition proposed in this invention: The image recognition-based rice plant growth status monitoring system includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the image recognition-based rice plant growth status monitoring method in the above embodiments is implemented.
[0066] The image recognition-based rice plant growth status monitoring system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.
[0067] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.
Claims
1. A method for monitoring the growth status of rice plants based on image recognition, characterized in that, include: A time-series image sequence of rice canopy was obtained, and each frame in the time-series image sequence was subjected to environment-adaptive segmentation and deblurring to obtain a clear target region of rice ears. The morphological skeleton of the target region of the rice panicle is extracted, and the dynamic characteristics of the rice panicle under wind excitation are calculated by tracking the motion trajectory of the morphological skeleton using a curve fitting algorithm. The target region of the rice ear is subjected to gray-level compression, and the micro-texture features of the rice ear surface are calculated based on the gray-level co-occurrence matrix. An environmental wind speed reference value is obtained and the weights are adaptively adjusted based on the environmental wind speed reference value. A multi-dimensional coupled evaluation model is constructed by combining the dynamic features and the micro-texture features, and the comprehensive score of rice grain filling maturity is calculated.
2. The method for monitoring the growth status of rice plants based on image recognition according to claim 1, characterized in that, The environmental adaptive segmentation and deblurring process for each frame of the time-series image sequence includes: Gaussian filtering is used to denoise a single frame image, removing random noise generated by the image sensor; The optimal segmentation threshold for the current frame is calculated using the maximum inter-class variance method, and the image is binarized to separate the rice ears foreground from the field ridges background. The image edge gradient intensity is detected. When the edge gradient intensity is lower than a preset sharpness threshold, Wiener filtering is used to restore the image based on the estimated point spread function to eliminate motion blur caused by wind.
3. The method for monitoring the growth status of rice plants based on image recognition according to claim 2, characterized in that, The calculation of the dynamic characteristics of rice ears under wind excitation by tracking the motion trajectory of the morphological skeleton using a curve fitting algorithm includes: Morphological thinning is performed on the binarized image to extract the central skeleton of the rice ear with a single pixel width; The bottom end of the rice panicle central skeleton is selected as the starting point and the top end as the ending point. The point with the largest curvature is selected as the control point, and the third-order Bézier curve is used to fit the rice panicle central skeleton. By tracking the angle between the top tangent of the fitted curve and the vertical direction in each frame, an angle time series is constructed. Perform a fast Fourier transform on the angle time series to obtain the frequency with the highest power spectral density as the main oscillation frequency, and calculate the variance of the angle time series as the oscillation variance. The dynamic characteristics include the principal oscillation frequency and the oscillation variance.
4. The method for monitoring the growth status of rice plants based on image recognition according to claim 3, characterized in that, The calculation of microtexture features on the surface of rice ears based on gray-level co-occurrence matrix includes: Convert the target area of rice ears into a grayscale image and compress the grayscale levels to 16 levels; Construct four gray-level co-occurrence matrices with a step size of 1 pixel and directions of 0 degrees, 45 degrees, 90 degrees, and 135 degrees, and calculate their mean values to obtain the comprehensive matrix; Contrast and homogeneity are calculated based on the comprehensive matrix, where contrast represents the depth of texture grooves and homogeneity represents the uniformity of texture distribution. The microtexture features include the contrast and the homogeneity.
5. The method for monitoring the growth status of rice plants based on image recognition according to claim 4, characterized in that, Before calculating the comprehensive score of rice grain-filling maturity, the calculation also includes the gravity inertial damping index, the formula of which is: In the formula, This represents the gravitational inertial damping index; This represents the normalized environmental wind speed proxy value; Indicates the main oscillation frequency; Indicates the variance of the swing; To prevent odd outliers; It is the basic noise constant.
6. The method for monitoring the growth status of rice plants based on image recognition according to claim 5, characterized in that, Before calculating the comprehensive score of rice grain-filling maturity, the textural density factor is also calculated, and its calculation formula is as follows: In the formula, Indicates the texture density factor; This represents the contrast of the normalized gray-level co-occurrence matrix; This indicates the homogeneity of the normalized gray-level co-occurrence matrix; This represents the average gray value of the target area of rice ears; This is the color correction factor.
7. The method for monitoring the growth status of rice plants based on image recognition according to claim 6, characterized in that, The formula for calculating the comprehensive score of rice grain-filling maturity is as follows: In the formula, This represents the overall score for the grain-filling maturity of rice. Basic score; The gravitational inertial damping index is mentioned above. Texture density factor; For dynamically adaptive weighting coefficients, and , For adjustment coefficients, The wind speed threshold, This is the normalized environmental wind speed proxy value.
8. The method for monitoring the growth status of rice plants based on image recognition according to claim 1, characterized in that, The acquisition of the environmental wind speed reference value includes: Select the leaf tip region in the image background as the reference region; The average amplitude of the optical flow field in the reference area is calculated using the optical flow method, and the average amplitude of the optical flow field is normalized to obtain an environmental wind speed proxy value with a value between 0 and 1, which is used as the environmental wind speed reference value.
9. The method for monitoring the growth status of rice plants based on image recognition according to claim 1, characterized in that, The method for acquiring the time-series image sequence is as follows: using an industrial-grade camera set up in the field, shooting from a 45-degree angle (overhead or side view), and acquiring a video stream with a frame rate of 60 frames per second and a duration of 5 seconds.
10. A rice plant growth status monitoring system based on image recognition, characterized in that, It includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the image recognition-based rice plant growth status monitoring method according to any one of claims 1-9.