A method for estimating daily aboveground biomass of crops based on phenological camera images

Abstract

This invention provides a method for estimating daily aboveground biomass of crops based on phenological camera images, belonging to the field of smart agriculture technology. This invention continuously and dynamically monitors crop canopy images daily, providing high timeliness. After analyzing the greenness color index, it performs cumulative analysis over time to obtain cumulative greenness features reflecting the cumulative change patterns of crop growth over time. It integrates temporal features and deep visual features, taking into account both the cumulative patterns of crop growth over time and the spatial structure characteristics of the canopy, avoiding the problem of insufficient representation ability of single features. It can extract highly robust deep features without requiring a large number of labeled samples, adapting to complex field scenarios with different light, weather, and growth stages. Compared with traditional single vegetation index models, it improves estimation accuracy and stability. It eliminates the need for high-cost equipment such as drones and satellites, and eliminates the need for manual field sampling, enabling all-weather unattended monitoring and making it suitable for large-scale field applications.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, specifically to a method for estimating daily aboveground biomass of crops based on phenological camera images. Background Technology

[0002] Crop aboveground biomass is a core agronomic indicator reflecting crop growth status, population vigor, and yield potential. It is also crucial foundational data for research on carbon cycling in farmland ecosystems and precision water and fertilizer management in the field. Current methods for acquiring and estimating crop aboveground biomass mainly fall into the following categories: Manual measurement methods obtain biomass through field sampling, drying, and weighing. While highly accurate, these methods are destructive, labor-intensive, and lack timeliness. They can only achieve discrete sampling during key crop growth stages and cannot obtain continuous daily dynamic data, failing to reflect the ongoing growth and changes in crops. Satellite remote sensing estimation methods have low temporal resolution, are easily obscured by cloud and rain, making it difficult to support high-frequency daily monitoring. Furthermore, their spatial resolution is limited, resulting in insufficient accuracy for small-plot crop monitoring and making them unsuitable for refined field monitoring scenarios. Unmanned aerial vehicle (UAV) remote sensing estimation methods require regular manual flight operations. Traditional methods suffer from high operating costs and poor continuity, making it impossible to achieve all-weather, daily automated dynamic monitoring and difficult to capture subtle changes in crop growth over short timescales. Fixed phenological camera monitoring, on the other hand, involves deploying RGB phenological cameras at fixed locations in the field to continuously and automatically collect canopy images, extract greenness vegetation indices for fitting, and has the advantages of being non-destructive, low-cost, and continuously and automatically collected. Deep learning-assisted estimation methods extract depth features from crop canopy images based on models such as convolutional neural networks and Transformers to construct biomass estimation models, which have stronger feature extraction capabilities compared to traditional vegetation indices.

[0003] Existing studies mostly use depth feature modeling of single-frame images or only combine daily vegetation indices, failing to make full use of the accumulated physiological information over time during the crop's growth period. They are easily affected by light intensity, weather changes, canopy tilt angle, and soil background, exhibiting extremely poor anti-interference ability. Furthermore, they are prone to estimation bias during critical growth transition stages such as jointing and heading, making it impossible to guarantee the continuity and accuracy of daily estimations.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for estimating daily aboveground biomass of crops based on phenological camera images, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for estimating daily aboveground biomass of crops based on phenological camera images, comprising the following steps:

[0008] S1. Deploy fixed phenological cameras to collect time-series RGB images of crops throughout their entire biological cycle in the target area. The time-series RGB images are red, green, and blue three-channel digital images continuously collected throughout the entire biological cycle of the crops.

[0009] S2. Taking the crop transplanting period as the starting point, the daily greenness color index is analyzed based on the time-series RGB image, and the cumulative greenness feature is generated. The cumulative greenness feature is used to simulate the physiological process of continuous accumulation of crop biomass with the growth process.

[0010] S3. Based on the pre-trained DINOv3 self-supervised vision basic large model, extract the global depth feature vector of the temporal RGB image and perform dimensionality reduction to obtain the canopy structure feature vector. The global depth feature vector is used to characterize the latent structural information of the crop canopy, including leaf tilt angle, texture, and coverage.

[0011] S4. Concatenate the cumulative greenness features with the canopy structure feature vector to construct a multi-source fusion feature set;

[0012] S5. Based on the multi-source fusion feature set and the corresponding measured aboveground biomass data, train a random forest regression model to obtain a daily aboveground biomass estimation model for crops.

[0013] S6. Input the time-series RGB image of the date to be estimated into the daily aboveground biomass estimation model of crops, and output the predicted daily aboveground biomass value of crops.

[0014] Furthermore, the phenological camera lens was pointed towards the crop canopy, and the sampling frequency was every 15 minutes.

[0015] Furthermore, in step S2, the temporal RGB image is masked to remove the soil background area, retain the crop canopy pixels, and calculate the instantaneous greenness color index. The formula used is:

[0016]

[0017] Wherein, R, G, and B are the average pixel values ​​of the red, green, and blue channels of the crop canopy pixels, respectively.

[0018] Furthermore, taking the crop transplanting period as the zero point, the instantaneous greenness color index is accumulated daily to generate a cumulative greenness characteristic, based on the following formula:

[0019]

[0020] in, The cumulative greenness characteristic on day t, Let i be the instantaneous greenness color index on day i. The date of commencement of the reproductive period.

[0021] Furthermore, in step S3, the temporal RGB image is input into the Transformer encoder, which outputs a 768-dimensional global depth feature vector. Principal component analysis was used to reduce the dimensionality of the 768-dimensional global depth feature vector, retaining the principal components, and generating a canopy structure feature vector. The variance contribution rate of the principal components is not less than 95%, and the projection transformation formula of principal component analysis is: W is the projection matrix for principal component analysis.

[0022] Furthermore, the random forest regression model outputs the average prediction results through an ensemble decision tree, providing the daily aboveground biomass prediction for the crop. The formula used is:

[0023]

[0024] Where n is the number of decision trees, Output values ​​for a single decision tree, where k is used to index the decision tree.

[0025] Furthermore, the daily aboveground biomass estimation model for crops is evaluated based on the coefficient of determination, using the following formula:

[0026]

[0027] Where m is used to index samples. To obtain the true value of aboveground biomass, This represents the average measured aboveground biomass. The aboveground biomass value predicted by the crop daily aboveground biomass estimation model.

[0028] Furthermore, by destructively sampling at sampling points in the target area, the dry weight data of crops per unit area obtained by the drying and weighing method were used as the measured aboveground biomass data.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] This invention continuously monitors crop canopy images daily, providing excellent timeliness. After analyzing the greenness color index, it performs cumulative analysis over time to obtain cumulative greenness features that reflect the cumulative changes in crop growth over time. It integrates temporal features and depth visual features, taking into account both the cumulative growth patterns of crops and the spatial structure of the canopy. This avoids the problem of insufficient representation capabilities of single features. It can extract highly robust depth features without a large number of labeled samples, adapting to complex field scenarios with different light, weather, and growth stages. Compared with traditional single vegetation index models, it improves estimation accuracy and stability. It does not require high-cost equipment such as drones and satellites, nor does it require manual field sampling, enabling all-weather unattended monitoring and making it suitable for large-scale field applications. Attached Figure Description

[0031] Figure 1 This is a scatter plot of the score of one of the principal components and biomass after dimensionality reduction of the fused feature set in this invention.

[0032] Figure 2 A scatter plot is used to verify the estimation effect of this invention on the test set;

[0033] Figure 3 This invention presents a time-series comparison diagram of daily biomass dynamic estimation throughout the entire growth period. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0035] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0036] Example:

[0037] Please see Figures 1-3 The present invention provides a technical solution:

[0038] A method for estimating daily aboveground biomass of crops based on phenological camera images, comprising the following steps:

[0039] Step 1: Deploy fixed phenological cameras to acquire time-series RGB images of crops throughout their entire biological cycle in the target area. These time-series RGB images are continuously acquired red, green, and blue three-channel digital images throughout the entire biological cycle of the crops. This captures the dynamic trajectory of the complete life cycle of crop growth, transforming discrete spatial information into continuous time-series data.

[0040] Fixed phenological cameras were deployed in rice fields, with lenses tilted at 30° towards the crop canopy at a height of 1.9m. These cameras automatically acquired temporal RGB images of the field throughout the day, from 11:00-13:00 and 17:45-19:30 daily, at 15-minute intervals. The resolution was 9216×5184 pixels, covering the entire crop growth period. The raw data typically contained a large number of redundant images from non-observation periods, such as black images at night or images showing equipment malfunctions. To eliminate noise interference from irrelevant data, a file traversal and date matching algorithm was used to filter and aggregate the raw temporal images acquired by the phenological cameras: the canopy image storage folder was located, and the acquisition date information was parsed based on the filenames of the temporal RGB images. Precise matching was performed according to preset key crop observation dates to select the temporal RGB images corresponding to those dates. The filtered temporal RGB images were then uniformly copied to a designated directory for aggregation, eliminating redundant images from irrelevant dates to ensure the temporal relevance and completeness of subsequent experimental data.

[0041] Step 2: Starting from the crop transplanting period, analyze the daily greenness color index based on time-series RGB images and accumulate it to generate a cumulative greenness feature. The cumulative greenness feature is used to simulate the physiological process of continuous accumulation of crop biomass as it grows. Crop transplanting marks the beginning of root regeneration and photosynthesis, at which point biomass is zero or at a baseline value, establishing the starting point for biomass accumulation. The instantaneous greenness color index decreases during the crop ripening and yellowing stage, but biomass is still increasing at this time. By accumulating and summing, the law of large numbers is utilized to improve the data's resistance to interference and long-term stability.

[0042] In step 2, the time-series RGB image is masked to remove the soil background area and retain the crop canopy pixels. Python code in PyCharm is used to read the image and calculate the Green Color Index (GCC), as well as the instantaneous Green Color Index. The formula used is:

[0043]

[0044] R, G, and B represent the mean values ​​of the red, green, and blue channels of the crop canopy pixels, respectively. The greenness color index reflects the proportion of green signal in all visible light signals within the crop canopy. Normalization eliminates the influence of light intensity variations, such as the difference in light intensity between morning and evening, on the absolute values ​​of RGB values, mapping images under different lighting conditions to the same dimension for comparison. The green channel is directly related to the chlorophyll content of the crop; as the number of leaves increases, green reflectance strengthens, and the G value rises. R and B values ​​are usually related to the soil background or aging leaves. The denominator serves as a normalization factor, offsetting the synchronous increase or decrease of each channel caused by changes in ambient light intensity.

[0045] Using the crop transplanting period as the zero point, the instantaneous greenness color index is accumulated daily to generate a cumulative greenness characteristic. The formula used is as follows:

[0046]

[0047] in, The cumulative greenness characteristic on day t, Let i be the instantaneous greenness color index on day i. The date of commencement of the reproductive period.

[0048] Step 3: Extract global depth feature vectors from time-series RGB images based on the pre-trained DINOv3 self-supervised visual foundation model, and perform dimensionality reduction to obtain canopy structure feature vectors. These global depth feature vectors represent the latent structural information of the crop canopy, including leaf tilt angle, texture, and coverage. The DINOv3 self-supervised visual foundation model is a fundamental visual model that employs self-supervised learning training, reducing training time and resource usage, making it suitable for small-sample agricultural scenarios. The pre-trained weights already contain rich low-level visual features and high-level semantic features; in small-sample agricultural scenarios, these weights are frozen and remain unchanged.

[0049] In step 3, the temporal RGB image is input into the Transformer encoder, which outputs a 768-dimensional global depth feature vector. The Transformer encoder calculates the relationships between image patches through a self-attention mechanism, aggregating them to generate a global depth feature vector containing spatial location information. Principal component analysis is then used to reduce the 768-dimensional global depth feature vector to 5 dimensions, retaining the principal components to generate a canopy structure feature vector. The variance contribution rate of the principal components is not less than 95%, and the projection transformation formula of principal component analysis is: W is the projection matrix for Principal Component Analysis (PCA). PCA projects features onto a new coordinate system through orthogonal transformation. The new coordinate axes sort the principal components by variance, retaining the top-ranked components to ensure their cumulative variance contribution reaches over 95%. This reduces the input dimensionality of subsequent random forest models, eliminating multicollinearity and improving inference speed and stability.

[0050] The global depth feature vector contains rich information such as the texture, edge, shape and spatial layout of the crop canopy. The projection matrix is ​​obtained by performing eigenvalue decomposition on the covariance matrix of the training set data. It consists of the feature vectors corresponding to the largest eigenvalues ​​in the sorted order. The role of the projection matrix is ​​to linearly transform the global depth feature vector to a new feature space, so that the new features, i.e. the principal components, are uncorrelated and arranged in descending order of variance.

[0051] Step 4: Concatenate the cumulative greenness features with the canopy structure feature vector to construct a multi-source fusion feature set;

[0052] The feature data of the time-series RGB images corresponding to the sampling date are matched and labeled with the measured aboveground biomass data to form training samples. The cumulative greenness features obtained in step 2 are matched and aligned with the canopy structure feature vector obtained in step 3. The shallow temporal growth features and deep visual structure features are fused through feature splicing to construct a fused feature set. The cumulative greenness features are used to characterize the cumulative change law of crop growth time dimension, and the canopy structure feature vector is used to characterize the structural growth features of crop canopy space dimension. After fusion, the core factors affecting aboveground biomass can be fully covered.

[0053] Step 5: Based on the multi-source fusion feature set and corresponding measured aboveground biomass data, train a random forest regression model to obtain a daily crop aboveground biomass estimation model. The crop dry weight per unit area, obtained through destructive sampling at sampling points in the target area and measured by drying and weighing, is used as the measured aboveground biomass data. A non-linear mapping between features and biomass is established by training the random forest regression model. Multiple decision trees are constructed, and their predictions are averaged to reduce the risk of overfitting that can easily occur with a single decision tree. Before training, a random partitioning algorithm is used to divide the dataset into training and test sets in an 8:2 ratio. (Refer to...) Figure 2 The scatter plot is used to verify the estimation effect of the constructed model on the test set.

[0054] The random forest regression model outputs the average prediction result through an ensemble decision tree, providing a daily forecast of crop aboveground biomass. The formula used is:

[0055]

[0056] Among them, the daily aboveground biomass forecast of crops This represents the daily aboveground biomass prediction of the crop, as output by the model, where n is the number of decision trees. The output value of a single decision tree represents the value of the k-th decision tree based on the input features. The judgment made, k is used to index the decision tree. It is the independent variable, representing the multi-source fusion feature set of the input model.

[0057] The evaluation of the daily aboveground biomass estimation model for crops based on the coefficient of determination is based on the following formula:

[0058]

[0059] Where m is used to index samples. To obtain the true value of aboveground biomass, This represents the average measured aboveground biomass. Coefficient of determination for aboveground biomass values ​​predicted by a daily crop aboveground biomass estimation model. It reflects the ability of the crop daily aboveground biomass estimation model to interpret the variation in biomass data. It is used to provide a standardized, dimensionless quantitative index to evaluate the overall estimation accuracy of the model throughout the entire growth period. The closer it is to 1, the smaller the residual, the better the model fit, and the higher the estimation accuracy.

[0060] Step 6: Input the time-series RGB image of the date to be estimated into the daily aboveground biomass estimation model of crops, and output the predicted daily aboveground biomass value of crops.

[0061] Reference Figure 3 This is a time-series comparison diagram of the dynamic estimation results of daily aboveground biomass of crops throughout the entire growth period according to the present invention.

[0062] This invention is particularly applicable to the monitoring and analysis of rice crop biomass.

[0063] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0064] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0066] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for estimating daily aboveground biomass of crops based on phenological camera images, characterized in that, The specific steps include: S1. Deploy fixed phenological cameras to collect time-series RGB images of crops throughout their entire biological cycle in the target area. The time-series RGB images are red, green, and blue three-channel digital images continuously collected throughout the entire biological cycle of the crops. S2. Taking the crop transplanting period as the starting point, the daily greenness color index is analyzed based on the time-series RGB image, and the cumulative greenness feature is generated. The cumulative greenness feature is used to simulate the physiological process of continuous accumulation of crop biomass with the growth process. S3. Based on the pre-trained DINOv3 self-supervised vision basic large model, extract the global depth feature vector of the temporal RGB image and perform dimensionality reduction to obtain the canopy structure feature vector. The global depth feature vector is used to characterize the latent structural information of the crop canopy, including leaf tilt angle, texture, and coverage. S4. Concatenate the cumulative greenness features with the canopy structure feature vector to construct a multi-source fusion feature set; S5. Based on the multi-source fusion feature set and the corresponding measured aboveground biomass data, train a random forest regression model to obtain a daily aboveground biomass estimation model for crops. S6. Input the time-series RGB image of the date to be estimated into the daily aboveground biomass estimation model of crops, and output the predicted daily aboveground biomass value of crops.

2. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 1, characterized in that: The phenology camera lens was pointed towards the crop canopy, and the sampling frequency was every 15 minutes.

3. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 1, characterized in that: In step S2, the temporal RGB image is masked to remove the soil background area, retain the crop canopy pixels, and the instantaneous greenness color index is calculated. The formula used is: Wherein, R, G, and B are the average pixel values ​​of the red, green, and blue channels of the crop canopy pixels, respectively.

4. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 3, characterized in that: Using the crop transplanting period as the zero point, the instantaneous greenness color index is accumulated daily to generate a cumulative greenness characteristic. The formula used is as follows: in, The cumulative greenness characteristic on day t, Let i be the instantaneous greenness color index on day i. The date of commencement of the reproductive period.

5. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 1, characterized in that: In step S3, the temporal RGB image is input into the Transformer encoder, and a 768-dimensional global depth feature vector is output. Principal component analysis was used to reduce the dimensionality of the 768-dimensional global depth feature vector, retaining the principal components, and generating a canopy structure feature vector. The variance contribution rate of the principal components is not less than 95%, and the projection transformation formula of principal component analysis is: W is the projection matrix for principal component analysis.

6. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 5, characterized in that: The random forest regression model outputs the average prediction result through an ensemble decision tree, providing a daily forecast of crop aboveground biomass. The formula used is: Where n is the number of decision trees, Output values ​​for a single decision tree, where k is used to index the decision tree.

7. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 6, characterized in that: The evaluation of the daily aboveground biomass estimation model for crops based on the coefficient of determination is based on the following formula: Where m is used to index samples. To obtain the true value of aboveground biomass, This represents the average measured aboveground biomass. The aboveground biomass value predicted by the crop daily aboveground biomass estimation model.

8. The method for estimating daily aboveground biomass of crops based on phenological camera images according to claim 1, characterized in that: The dry weight of crops per unit area obtained by destructive sampling at sampling points in the target area and by drying and weighing is used as the measured aboveground biomass data.

Images