Leaf area index determination method, apparatus, device, medium, and product

Abstract

The present application provides a kind of leaf area index determination method, device, equipment, medium and product, it is related to agricultural remote sensing and near-ground optical measurement technical field, to solve the defects of lower accuracy and poor efficiency in determining the leaf area index of crop in the prior art, realize the accuracy and efficiency of improving the leaf area index of crop is determined.Imcluding: the target hemispherical image collected is carried out image correction processing, obtain the target hemispherical image after correction, target hemispherical image is obtained by vertical upward shooting for shooting device in plant row center position;Determine the normalized difference vegetation index NDVI histogram corresponding to plant in the target hemispherical image after correction;Based on NDVI histogram and preset zenith angle weight, determine the void fraction of plant;Based on void fraction, the leaf area index LAI of plant is inversed.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing and near-ground optical measurement technology, and in particular to a method, apparatus, equipment, medium and product for determining leaf area index. Background Technology

[0002] Leaf area index (LAI) is an important parameter for measuring crop canopy structure and photosynthetic potential, and can be used for growth monitoring, nitrogen water management, stress diagnosis, and model calibration. Current direct measurement methods (such as harvesting and weighing, specific leaf area conversion, etc.) are highly destructive to crops, labor-intensive, and difficult to implement at high frequencies. Indirect optical methods suffer from significant light dependence and scattering interference; uneven sky and high-contrast backgrounds amplify the accuracy of porosity estimation; they are insensitive to low light and complex backgrounds; fixed thresholds for image segmentation are difficult to standardize across different plots and batches; and the geometric properties of rows and the leaf clustering effect lead to large errors in the determined LAI. Therefore, current methods for determining crop leaf area index are characterized by low accuracy and poor efficiency. Summary of the Invention

[0003] This invention provides a method, apparatus, equipment, medium, and product for determining leaf area index, which addresses the shortcomings of low accuracy and poor efficiency in determining the leaf area index of crops in the prior art, thereby improving the accuracy and efficiency of determining the leaf area index of crops.

[0004] This invention provides a method for determining leaf area index, comprising the following steps.

[0005] The acquired target hemisphere image is processed to obtain a corrected target hemisphere image. The target hemisphere image is obtained by shooting vertically upward from the center of the plant row with the shooting device. Determine the normalized vegetation index (NDVI) histogram for the plants in the corrected target hemisphere image. The porosity of plants is determined based on the NDVI histogram and the preset zenith angle weights. Leaf area index (LAI) of plants is derived from porosity.

[0006] According to the leaf area index determination method provided by the present invention, image correction processing is performed on the acquired target hemisphere image to obtain a corrected target hemisphere image, including: Dark current correction is performed on the acquired target hemisphere image to obtain the target hemisphere image after dark current correction. Energy balancing is performed on the dark current-corrected target hemisphere image to obtain the corrected target hemisphere image.

[0007] According to the leaf area index determination method provided by the present invention, the target hemisphere image includes multiple channels, including: R channel, G channel, B channel and NIR channel; Dark current correction is performed on the acquired target hemispherical image to obtain a dark current-corrected target hemispherical image, including: Linearize the pixel values ​​of multiple channels in the target hemispherical image to determine the linear pixel value of each channel; Based on the preset dark current corresponding to each channel, the preset dark current is subtracted from the linear pixel value of each channel to obtain the corrected pixel value of each channel; Based on the corrected pixel value of each channel, the target hemisphere image after dark current correction is obtained.

[0008] According to the leaf area index determination method provided by the present invention, energy balancing processing is performed on a target hemisphere image after dark current correction to obtain a corrected target hemisphere image, including: Determine the sky reference region from the dark current corrected target hemisphere image; Based on the sky reference region, the scale factor of the R channel and the scale factor of the NIR channel in the target hemisphere image after dark current correction are determined. Based on the scale factors of the R channel and the NIR channel, the energy levels of the R channel and NIR channel are calibrated to obtain the corrected target hemisphere image.

[0009] According to a method for determining leaf area index provided by the present invention, based on an NDVI histogram and a preset zenith angle band weight, the porosity of a plant is determined, including: Determine the main peak in the NDVI histogram, and then determine the valley points on both sides of the main peak with the main peak as the center. The target threshold is determined based on the NDVI values ​​corresponding to the valley points on both sides of the main peak. Based on the target threshold, the pixels in the corrected target hemisphere image are divided into plant canopy pixels and sky background pixels; The porosity is determined based on the plant canopy pixels, sky background pixels, and preset zenith angle weights.

[0010] According to the present invention, a method for determining leaf area index (LAI) of plants is provided, which inverts the LAI based on porosity, including: The plant's LAI (Laminated Area Index) is retrieved based on porosity, cluster correction coefficient, and leaf tilt extinction coefficient.

[0011] The present invention also provides a leaf area index determination device, comprising the following modules: a processing module, a determination module, and an inversion module; The processing module is used to perform image correction processing on the acquired target hemisphere image to obtain the corrected target hemisphere image. The target hemisphere image is obtained by the shooting device being located at the center of the plant row and shooting vertically upwards. The determination module is used to determine the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemispherical image. The determination module is also used to determine the porosity of plants based on the NDVI histogram and preset zenith angle weights. The inversion module is used to invert the leaf area index (LAI) of plants based on porosity.

[0012] According to the present invention, a leaf area index determination device, including a processing module, is specifically used for: Dark current correction is performed on the acquired target hemisphere image to obtain the target hemisphere image after dark current correction. Energy balancing is performed on the dark current-corrected target hemisphere image to obtain the corrected target hemisphere image.

[0013] According to the leaf area index determination device provided by the present invention, the target hemisphere image includes multiple channels, including: R channel, G channel, B channel and NIR channel; The processing module is specifically used for: Linearize the pixel values ​​of multiple channels in the target hemispherical image to determine the linear pixel value of each channel; Based on the preset dark current corresponding to each channel, the preset dark current is subtracted from the linear pixel value of each channel to obtain the corrected pixel value of each channel; Based on the corrected pixel value of each channel, the target hemisphere image after dark current correction is obtained.

[0014] According to the present invention, a leaf area index determination device, including a processing module, is specifically used for: Determine the sky reference region from the dark current corrected target hemisphere image; Based on the sky reference region, the scale factor of the R channel and the scale factor of the NIR channel in the target hemisphere image after dark current correction are determined. Based on the scale factors of the R channel and the NIR channel, the energy levels of the R channel and NIR channel are calibrated to obtain the corrected target hemisphere image.

[0015] According to the present invention, a leaf area index determining device includes a determining module specifically used for: Determine the main peak in the NDVI histogram, and then determine the valley points on both sides of the main peak with the main peak as the center. The target threshold is determined based on the NDVI values ​​corresponding to the valley points on both sides of the main peak. Based on the target threshold, the pixels in the corrected target hemisphere image are divided into plant canopy pixels and sky background pixels; The porosity is determined based on the plant canopy pixels, sky background pixels, and preset zenith angle weights.

[0016] According to the present invention, a leaf area index determination device and an inversion module are provided, specifically used to invert the LAI of a plant based on porosity, cluster correction coefficient and leaf tilt angle extinction coefficient.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the leaf area index determination methods described above.

[0018] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the leaf area index determination methods described above.

[0019] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the leaf area index determination methods described above.

[0020] This invention provides a method, apparatus, device, medium, and product for determining leaf area index (LAI). The method involves capturing a target hemisphere image vertically upwards from the center of the plant rows using an imaging device. The acquired target hemisphere image is then processed to obtain a corrected image. Next, the Normalized Difference Vegetation Index (NDVI) histogram corresponding to the plant in the corrected target hemisphere image is determined. The dominant peak of the NDVI histogram is used to set a restricted search interval, enabling rapid and robust threshold selection. Furthermore, by combining a preset zenith angle weight, the porosity of the plant is determined, allowing the LAI to be retrieved based on the porosity. This significantly improves the stability and processing efficiency of canopy segmentation under complex lighting and row geometry conditions, enhancing the accuracy and repeatability of effective LAI estimation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is one of the flowcharts illustrating the leaf area index determination method provided by the present invention.

[0023] Figure 2 This is the second flowchart of the leaf area index determination method provided by the present invention.

[0024] Figure 3 This is the third flowchart of the leaf area index determination method provided by the present invention.

[0025] Figure 4 This is the fourth flowchart of the leaf area index determination method provided by the present invention.

[0026] Figure 5 This is the fifth flowchart of the leaf area index determination method provided by the present invention.

[0027] Figure 6 This is the sixth flowchart of the leaf area index determination method provided by the present invention.

[0028] Figure 7 This is one of the structural schematic diagrams of the leaf area index determination system provided by the present invention.

[0029] Figure 8 This is the second schematic diagram of the leaf area index determination system provided by the present invention.

[0030] Figure 9 This is the third schematic diagram of the leaf area index determination system provided by the present invention.

[0031] Figure 10 This is a schematic diagram of the leaf area index determination device provided by the present invention.

[0032] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0034] Currently, in order to meet the needs of rapid, non-destructive and high-throughput determination of leaf area index in the field, indirect optical methods can be used to invert LAI through canopy porosity, transmittance or image segmentation, and efforts should be made to maintain the stability and comparability of results under different light and canopy geometry conditions.

[0035] Specifically, the near-hemispherical field of view can be decomposed into five concentric angular zones. Under narrow blue light bands, the transmittance ratio of the upper and lower canopies is obtained to calculate porosity, and the effective LAI and average leaf tilt angle are retrieved based on the radiative transfer model. To reduce the influence of the operator and distant targets, the instrument is equipped with a field-of-view shading cap and ring shielding, suitable for measurements under diffuse light conditions in forests or under crop canopies. This type of instrument has a mature application base in the field, but it is sensitive to light uniformity; direct sunlight in clear skies or sunspots through gaps in clouds can easily introduce systematic bias. The output primarily characterizes the effective LAI, requiring additional correction when encountering significant clusters or striped structures in rows of crops. The measurement format is biased towards integration, lacking pixel-level evidence to support on-site quality control and post-hoc error correction. Under high reflectivity or strong contrast backgrounds in bare soil between rows, angular zone statistics are prone to accumulating bias, affecting the stability and repeatability of the results.

[0036] Alternatively, an upward-facing hemispherical image can be recorded using a fisheye lens, and the sky and canopy can be distinguished by thresholding or learning segmentation. The porosity of different zenith angle zones can then be statistically analyzed and the LAI (Layer Area Index) can be retrieved. This approach preserves pixel-level information, facilitating visualization quality control, ring masking, and distant scene removal, and is suitable for structural anisotropy analysis in row crop fields. However, under low-light conditions such as canopy base, backlighting, or cloud gaps with sunspots, relying solely on RGB brightness and contrast is limited, and threshold selection is sensitive. The traditional global Otsu method exhaustively searches the entire grayscale range, resulting in a heavy computational burden and insufficient stability under multi-peak histograms. Desktop post-processing workflows have limited real-time performance, and parameter consistency relies on manual intervention. Without utilizing near-infrared information, there is a risk of confusion between the canopy and the dark background, making porosity estimation prone to errors.

[0037] Therefore, the following common problems still exist: First, the light dependence and scattering interference are significant, and the unevenness of the sky and the strong contrast background will amplify the accuracy of porosity estimation; Second, the segmentation robustness is limited. When using only RGB, it is not sensitive to low light and complex backgrounds, and the threshold for image segmentation is difficult to unify between different plots and batches; Third, the geometry of rows and the leaf clustering effect cause a deviation between the effective LAI and the true LAI, which needs to be corrected by combining masking, corner banding strategies or external parameters; Fourth, the efficiency of field work and the traceability of quality inspection are insufficient. Integrating equipment lacks pixel evidence, and offline software is difficult to provide timely feedback and correction.

[0038] Based on this, this application proposes a method for improving the accuracy of crop canopy LAI assessment based on modified NDVI and solid angle weighted inversion. After performing dark current correction and linearization on the end-side of the acquired RGB-NIR hemispherical image, a difference scale factor is cleverly calculated and global energy balance is performed based on the reflectance difference between the near-infrared and red bands of the selected pure sky, ensuring consistent spectral responses across different bands. This eliminates radiation inhomogeneities caused by variations in illumination or equipment characteristics without whiteboard correction, constructing a modified NDVI reflectance index. The modified NDVI histogram's main peak is used to set a restricted search interval, replacing the traditional full-grayscale exhaustive search, thus achieving rapid and robust threshold selection. Furthermore, by combining zenith angle band weighting and configurable ring shielding to suppress distant and strong contrast interference, solid angle weighted porosity statistics are performed within the hemispherical field of view. Porosity P and LAI are calculated in near real-time on an embedded platform, simultaneously generating quality control layers such as threshold, mask, corner band proportion, and effective sample size, ensuring consistent parameter caliber and high-throughput field operations. It can significantly improve the stability and processing efficiency of canopy segmentation under complex lighting and crop geometry conditions, improve the accuracy and repeatability of effective LAI estimation, and provide a reliable data foundation for subsequent cluster correction and model docking.

[0039] Thus, this application utilizes RGB-IR spectral imaging technology to achieve rapid and non-destructive monitoring of crop canopy leaf area index (LAI) and physiological parameters, providing support for the development of precision agriculture. To improve the robustness of segmentation under open ambient lighting and background interference, addressing the instability of RGB brightness segmentation caused by non-uniform skies, high-contrast backgrounds, and low illumination at the canopy base, a "radiological preprocessing + corrected NDVI" method is adopted. In the candidate sky region S, a representative area (i.e., the sky reference area) is first selected as a reference. This area is typically an open sky region used to estimate the uniformity of the light source. Next, the energy difference between red and near-infrared light in the reference area is calculated, resulting in a difference scaling coefficient k. Then, based on this coefficient, the global spectral response is adjusted to align the near-infrared and red light energy levels, thereby obtaining a more accurate NDVI value that better reflects the true spectral characteristics of the crop.

[0040] Furthermore, addressing the issue of LAI bias caused by canopy heterogeneity and clustering effects, since row crops exhibit strong anisotropy and leaf clustering, the clustering effect typically causes crop leaves to aggregate in certain areas, leading to an overestimation or underestimation of the LAI. Traditional integral measurements struggle to explicitly characterize this spatial structure, easily resulting in effective LAI bias. Therefore, this application introduces a clustering correction coefficient during the LAI inversion process. This is used to adjust the actual structure of the canopy. By parametrically correcting for clusters, this effect is quantified and eliminated, ensuring that the final LAI estimation results are closer to the true value, reducing the systematic impact of cluster and row structure on the inversion results, and improving the comparability between different plots.

[0041] Furthermore, this application constructs an integrated "acquisition-processing-quality inspection" workflow, employing coaxial co-registered RGB-NIR fisheye imaging and an embedded RK3588 platform to achieve real-time data processing, effectively improving field measurement efficiency and ensuring caliber consistency. Under complex lighting and background interference conditions, an NDVI vegetation mask constructed using NIR channels and red light is used, combined with histogram peak settings to define a restricted search interval, improving segmentation efficiency. Through parallelization and buffered processing chains, near real-time porosity and LAI estimation results are output, generating a quality inspection layer to support on-site verification, meeting the needs of large-scale batch operations, and significantly improving the efficiency and stability of field operations.

[0042] The following is combined with Figures 1 to 11 This invention describes the leaf area index determination method, apparatus, equipment, medium, and product provided by the present invention.

[0043] Figure 1 This is one of the flowcharts illustrating the leaf area index determination method provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Perform image correction processing on the acquired target hemisphere image to obtain the corrected target hemisphere image.

[0044] The target hemisphere image was obtained by taking a vertical upward shot from the center of the plant rows using the imaging device.

[0045] Step 102: Determine the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image.

[0046] Step 103: Determine the porosity of the plants based on the NDVI histogram and the preset zenith angle weight.

[0047] Step 104: Invert the leaf area index (LAI) of the plant based on porosity.

[0048] In one possible implementation, to ensure the stability of the upward hemisphere imaging when acquiring the target hemisphere image, the imaging device (e.g., an RGB-NIR fisheye camera) can be placed at the center between the crop rows, with the lens pointing vertically upward (to avoid tilting that could cause the horizon to shift or the canopy to become uneven) to acquire the target hemisphere image.

[0049] In one possible implementation, when acquiring the target hemispherical image, the RGB and NIR channels can be triggered simultaneously to ensure that the acquisition time and field of view of each channel are completely consistent (no time difference, no registration deviation). Furthermore, after image acquisition is complete, a central circular ROI (Region of Interest, a circular target area with a specified radius centered on the geometric center of the image, often used to extract central region features, shield edge interference, and focus on core target analysis) can be established on the target hemispherical image plane. An outer ring shielding template can be enabled based on the horizon and background conditions to reduce the influence of non-target areas.

[0050] Furthermore, the acquired target hemisphere image undergoes image correction processing to obtain a corrected target hemisphere image. The Normalized Difference Vegetation Index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image is then determined. Based on the NDVI histogram and preset zenith angle weights, the porosity of the plants is determined. Finally, the leaf area index (LAI) of the plants is retrieved based on the porosity. A detailed description of this process can be found below and will not be repeated here.

[0051] In this embodiment, a target hemisphere image is captured vertically upwards from the center of the plant rows using an imaging device. The acquired target hemisphere image is then processed to obtain a corrected target hemisphere image. Next, the Normalized Difference Vegetation Index (NDVI) histogram corresponding to the plant in the corrected target hemisphere image is determined. The main peak of the NDVI histogram of the corrected target hemisphere image is used to set a restricted search interval, enabling rapid and robust threshold selection. Furthermore, by combining a preset zenith angle weight, the porosity of the plant is determined, allowing the leaf area index (LAI) to be retrieved based on the porosity. This significantly improves the stability and processing efficiency of canopy segmentation under complex lighting and row geometry conditions, enhancing the accuracy and repeatability of effective LAI estimation.

[0052] Figure 2 This is a second schematic flowchart of the leaf area index determination method provided by the present invention, as shown below. Figure 2 As shown, the method includes the following: Step 201: Perform dark current correction on the acquired target hemisphere image to obtain the target hemisphere image after dark current correction.

[0053] Step 202: Perform energy balancing processing on the target hemisphere image after dark current correction to obtain the corrected target hemisphere image.

[0054] Step 203: Determine the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image.

[0055] Step 204: Determine the porosity of the plant based on the NDVI histogram and the preset zenith angle weight.

[0056] Step 205: Invert the leaf area index (LAI) of the plant based on porosity.

[0057] In one possible implementation, to improve the stability of NDVI, this application adds a camera dark current correction and a white-plate-free energy balancing process before calculating NDVI. Dark current correction is used to subtract the dark current (black level) of the camera sensor under no-light conditions and correct the sensor's nonlinear response to restore the true radiance values ​​of each channel. Energy balancing does not require a white plate (standard reflector); it only uses the clean sky area of ​​the image itself (i.e., the sky reference area) as a reference to calculate the scale factors of the red (R) and near-infrared (NIR) channels, eliminating cross-channel energy level drift caused by illumination variations / device differences.

[0058] It should be noted that a detailed description of dark current correction and energy balancing can be found below, and will not be repeated here.

[0059] In one possible implementation, the target hemispherical image includes multiple channels, including: R channel, G channel, B channel and NIR channel.

[0060] Figure 3 This is the third flowchart of the leaf area index determination method provided by the present invention, as shown below. Figure 3 As shown, "Step 201, performing dark current correction on the acquired target hemispherical image to obtain a dark current-corrected target hemispherical image" specifically includes the following: Step 301: Linearize the pixel values ​​of multiple channels in the target hemispherical image to determine the linear pixel value of each channel.

[0061] Step 302: Based on the preset dark current corresponding to each channel, subtract the preset dark current from the linear pixel value of each channel to obtain the corrected pixel value of each channel.

[0062] Step 303: Based on the corrected pixel values ​​of each channel, obtain the target hemisphere image after dark current correction.

[0063] In one possible implementation, the pixel values ​​output by the camera often have a gamma or non-linear response, so they can be obtained through the inverse function of the response function. For each channel raw pixel values The pixel value (current) is restored to linear light intensity (i.e., the original pixel value is linearized, restoring the nonlinear output of the sensor to a linear value proportional to the incident light intensity), thus obtaining the restored linear pixel value. As shown in Formula 1.

[0064] Formula 1 in, Indicates the channel identifier.

[0065] Furthermore, the restored linear pixel values ​​are then... Based on the preset dark current d corresponding to each channel c From the linear pixel values ​​of each channel Subtracting the preset dark current d c This yields the radiation value free of dark noise, which is equivalent to obtaining the corrected pixel value for each channel. As shown in Formula 2.

[0066] Formula 2 Among them, the preset dark current d c The constant obtained for dark field / black level calibration can be obtained by pre-collecting the black level of each channel in a dark box and taking the average value as the preset dark current d for that channel. c And it is embedded into the equipment parameters.

[0067] Furthermore, based on the corrected pixel value of each channel The target hemisphere image after dark current correction was determined.

[0068] Figure 4 This is the fourth flowchart of the leaf area index determination method provided by the present invention, as shown below. Figure 4 As shown, "Step 202, performing energy balancing processing on the dark current-corrected target hemisphere image to obtain the corrected target hemisphere image" specifically includes the following: Step 401: Determine the sky reference region from the dark current corrected target hemisphere image.

[0069] Step 402: Based on the sky reference region, determine the scale factor of the R channel and the scale factor of the NIR channel in the target hemisphere image after dark current correction.

[0070] Step 403: Based on the scale factor of the R channel and the scale factor of the NIR channel, calibrate the energy levels of the R channel and the NIR channel to obtain the corrected target hemisphere image.

[0071] In one possible implementation, a sky reference region S can be determined from a dark current-corrected target hemispherical image, and the sky reference region S can be used as a reference irradiation source to estimate the cross-channel scale factor and align the energy levels.

[0072] First, it is necessary to construct the brightness. The RGB values ​​are combined to form a "luminance map," which is used later to determine where there are fewer edges and smoother changes (the sky is usually smoother), as shown in Formula 3.

[0073] Formula 3 in, This represents the corrected pixel value of the R channel. This represents the corrected pixel value of the G channel. The value represents the corrected pixel value of the B channel. 0.299, 0.587, and 0.114 are weighting coefficients, which represent the perceptual weight of the human eye for the three RGB channels. Formula 3 is used to synthesize the brightness value of the RGB channels, which represents the overall brightness of the image.

[0074] Furthermore, it is necessary to determine the smoothing gradient. First, Gaussian smoothing is performed, and then the gradient magnitude ∇ is taken. This is because the gradient in the sky region is generally small (with less texture), as shown in Formula 4.

[0075] Formula 4 in, Formula 4 represents Gaussian smoothing of the brightness L, where σ represents the smoothing kernel radius and ∇ represents the gradient operator (calculating the gradient in the x / y directions). Based on Formula 4, the brightness gradient can be calculated; the smaller the gradient, the more uniform the region (gradient ≈ 0 in the sky region, large gradient in the canopy region).

[0076] Then, the sky reference region is determined based on color / spectral features, which requires two discriminants: one is the proportion of blue (B channel). (The sky is usually bluish, and the B channel accounts for a higher proportion of the total RGB energy), as shown in Formula 5. Another factor is the NIR / R ratio. As shown in Formula 6, since the NIR of the sky region is much greater than R, and the R of the canopy is relatively high, the relative relationship between near-infrared and red light is used to assist in the judgment (especially when there are clouds or the sky is not pure blue, relying solely on the "blue proportion" is not stable).

[0077] Formula 5 Formula Six Where ε is a preset positive number (a very small constant greater than 0) used to prevent the denominator from being 0. This represents the corrected pixel value for the R channel.

[0078] In one possible implementation, the sky reference area is determined ( When this is the case, it is necessary to determine the regions where pixels meet the following conditions: few textures ( Like the sky ( or Brightness is within a reasonable range. Therefore, all pixels that meet the conditions constitute the sky candidate region S, that is, the determined sky reference region S satisfies... .otherwise, Pixel regions that are not selected as sky reference regions cannot be used. If the number of pixels in the selected sky candidate region S is less than 5% of the total pixels of the target hemisphere image, the sky candidate region is deemed insufficient, triggering a warning (prompting the user to adjust the acquisition point).

[0079] Where, τ g τ represents the gradient threshold (e.g., 5). BN τ represents the threshold for the proportion of channel B (e.g., 0.4). NR The threshold representing the NIR / R ratio (e.g., 3), τ bright τ represents the minimum brightness threshold (e.g., 50). sat This represents the saturation threshold (e.g., 240). τ g τ BN τ NR The value can be adapted to different crops / lighting scenarios (e.g., τ on sunny days). BN (Take 0.4 for cloudy days and 0.3 for overcast days), and you can preset the parameter set.

[0080] In one possible implementation, the scale factor of the R channel and the scale factor of the NIR channel are determined in a defined sky reference region S. That is, in a region like the sky reference region S where the spectral relationship should be relatively stable, the overall magnification difference between NIR and Red is estimated. The median is used to avoid being affected by a small number of outliers, as shown in Equation 7.

[0081] Formula 7 Among them, K R represents the scaling factor, and median represents the median (to resist outlier interference). Indicates taking Positive values ​​are used (negative numbers are avoided). Thus, using the sky reference area S as a reference, the scale factor of the NIR / R channel is calculated (the median is more robust than the mean, avoiding the influence of a small number of noise points).

[0082] Furthermore, by aligning the energy levels of the two channels and stretching / compressing the red channel as a whole, the relative scales of the R channel and the NIR channel become more consistent under different shooting times and exposures / lighting conditions, as shown in Formula 8.

[0083] Formula 8 Thus, through the scaling factor K R Calibrate the R channel to align the energy levels of the red / near-infrared channels, eliminating energy level drift caused by illumination / devices.

[0084] In one possible implementation, NDVI can be calculated based on the calibrated red / near-infrared channels to ensure consistency of NDVI under different lighting / device conditions, as shown in Formula 9.

[0085] Formula Nine in, This indicates the modified NDVI, with a value range of [-1, 1], and its stability is much higher than that of the traditional NDVI.

[0086] Thus, based on the above formula, each pixel is transformed into a "comparable value" that is closer to the true radiation intensity, avoiding the subsequent NDVI and threshold segmentation being skewed by camera nonlinearity and black level drift.

[0087] Figure 5 This is the fifth flowchart illustrating the leaf area index determination method provided by the present invention, as shown below. Figure 5 As shown, "Step 103, determining the porosity of the plant based on the NDVI histogram and preset zenith angle weights" specifically includes the following: Step 501: Determine the main peak in the NDVI histogram, and determine the valley points on both sides of the main peak with the main peak as the center.

[0088] Step 502: Determine the target threshold based on the NDVI values ​​corresponding to the valley points on both sides of the main peak.

[0089] Step 503: Based on the target threshold, divide the pixels in the corrected target hemisphere image into plant canopy pixels and sky background pixels.

[0090] Step 504: Determine the porosity based on the plant canopy pixels, sky background pixels, and preset zenith angle weights.

[0091] One possible implementation is based on the restricted search method of Otsu, which finds a target threshold among all candidate thresholds that maximizes the inter-class variance between the two classes on either side of the target threshold. Specifically, an NDVI histogram can be plotted within the ROI to automatically identify the main peak and local valleys (i.e., the valleys on either side of the main peak), thereby determining the candidate threshold interval. The candidate threshold interval is searched using the Otsu objective function to obtain the optimal threshold (i.e., the target threshold). ).

[0092] Furthermore, with A vegetation mask M(x,y)∈{0,1} is generated, where 1 represents crop canopy leaves. Opening and closing operations are used to refine the boundaries and remove isolated noise points. This method significantly reduces the computational burden of exhaustive full grayscale enumeration while maintaining equivalent segmentation accuracy, thus improving the real-time performance of embedded systems.

[0093] In one possible implementation, while maintaining the central ROI, an outer ring shield is activated based on factors such as distant bright objects and high contrast at the horizon; if necessary, a fan-shaped shielding template is introduced to shield the operator's orientation. The pixel weights of the shielded areas are reset to zero, and they do not participate in subsequent gap statistics and inversion, ensuring the stability and repeatability of the estimation.

[0094] It is understandable that histogram statistics are performed only on the central circular ROI region of the preprocessed corrected NDVI image (eliminating invalid edge regions), resulting in a histogram H(v), where v is the NDVI value (normalized to 0~255 gray levels, or retaining the original [-1,1] floating-point value). This focuses on the effective analysis area and avoids interference introduced by full-image statistics. Furthermore, the main peak and local valleys are automatically identified, restricted intervals are defined, and the gray level v with the highest peak value in the histogram H(v) is found. peak (Corresponding to the pixel type with the largest proportion in the image, such as the sky background or the canopy subject); Using the main peak as the center, search for local minima (valleys) of the histogram to both sides, denoted as v. min (Left valley point) and v max (Right-side valley point). This defines the restricted interval (candidate threshold interval) as follows: (The range is usually only 10% to 30% of the full grayscale).

[0095] Thus, the valley points of the histogram serve as a natural boundary between the "background (sky)" and the "foreground (canopy)." Defining the interval using these valleys allows for precise targeting of the threshold search range, avoiding unnecessary traversal. Within this constrained interval, the Otsu algorithm is executed to solve for the optimal target threshold τ. Pixels within the ROI are divided into two categories: background... and canopy .

[0096] Furthermore, based on the target threshold τ, the binarized NDVI image is used to generate an initial mask, as shown in Formula 10.

[0097] Formula 10 In one possible implementation, when determining the porosity based on plant canopy pixels, sky background pixels, and a preset zenith angle weight, direct pixel counting can lead to inaccuracies due to field-of-view distortion in the hemispherical projection of a fisheye camera (pixels at different zenith angles correspond to different solid angles). Therefore, by weighting each pixel with solid angle weights, the physical meaning of the "hemispherical solid angle ratio" is restored, ensuring the physical accuracy of the porosity.

[0098] Specifically, the polar radius-zenith angle mapping can be provided by the factory calibration. This establishes a one-to-one correspondence between "image pixel polar radius" and "true zenith angle". For arbitrary circular symmetric projections... The Jacobian determinant of the solid angle of a pixel is shown in Formula 11. It can describe the mapping relationship between "pixel area" and "true solid angle" under projection transformation. The larger the Jacobian, the larger the true solid angle corresponding to the pixel.

[0099] Formula Eleven Where θ represents the zenith angle corresponding to the pixel (0° represents the zenith, 90° represents the horizon), and ρ represents the polar radius (pixel distance) from the pixel to the center of the image. This represents the factory-calibrated polar radius-zenith angle mapping function, where J(x,y) represents the solid angle Jacobian determinant of the pixel (a correction factor for projection distortion), ρ'(θ) represents the derivative of the polar radius with respect to the zenith angle, and sinθ represents the sine of the zenith angle (sin0°=0 at the zenith, sin90°=1 at the horizon). The larger the zenith angle and the closer to the horizon, the larger the sinθ and the higher the weight.

[0100] Thus, through the factory-calibrated polar radius-zenith angle mapping The algorithm is embedded into the device using a "lookup table + interpolation" method (e.g., storing ρ values ​​at 1° zenith angle intervals) to avoid real-time calculation of complex functions. Furthermore, ω(x,y) is pre-calculated for all pixels in the image and stored as a weight template, which is directly called during acquisition (significantly reducing the real-time computation load on the edge). The weights ω(x,y) are normalized to the [0,1] interval for easier subsequent weighted statistics.

[0101] It should be noted that it is necessary to calculate the "aperture rate within the hemispherical field of view." Physically, the poerture rate corresponds to the transmission probability per unit solid angle of the hemisphere. The natural area element of a hemisphere is the solid angle dΩ = sinθdθdφ, but fisheye images count pixels. The area element in polar coordinates is dA = ρdρdφ. Therefore, to transform the "pixel summation" into "hemispherical integration," a transformation coefficient (Jacobi) must be multiplied, as shown in Formula XII.

[0102] Formula 12 Therefore, the solid angle weight of each pixel can be written as shown in Formula 13 (ignoring the constant pixel area ΔA).

[0103] Formula Thirteen The equivalent notation of Formula Thirteen (using ρ=Ψ(θ)) is: .

[0104] Furthermore, the void binary is directly given by the vegetation mask segmentation result of the restricted search Otsu method. The "void / non-void" state of each pixel is directly obtained from the canopy mask: G(x,y)=1-M(x,y). M(x,y) represents the vegetation mask, and G(x,y) represents the void binary, which is the basis for porosity statistics. M=1 for vegetation mask represents leaves, G=1 for void represents sky pores, and G=0 for canopy.

[0105] Therefore, the hemispherical weighted porosity P is shown in Formula 14. This is the porosity ratio weighted by solid angle. The larger the P, the sparser the area.

[0106] Formula Fourteen in, ω(x,y) represents the "weighted sum" of all void pixels within the ROI (reflecting the void ratio under the true solid angle), where ω(x,y) represents the solid angle weight. This represents the "weighted sum" of all pixels within the ROI (reflecting the total solid angle corresponding to the ROI).

[0107] It should be noted that segmentation of the band can also be performed for quality inspection purposes. Let the j-th band be... When there are effective gaps in a certain band (saturation / occlusion makes the available weights insufficient), the gap ratio m is used. j Perform band-level corrections, as shown in Formula 15.

[0108] Formula Fifteen Among them, Ω j This represents the j-th zenith angle zone (divided into segments based on zenith angle to accommodate the geometric characteristics of canopies at different heights), m j m represents the "gap ratio" (the proportion of insufficient effective weights) in the j-th band. j =0 indicates that the band has no gaps (all weights are valid), m j The larger the value, the more severe the gap (e.g., occlusion / saturation leading to fewer effective weights), P j The porosity of the j-th band after correction is represented by the gap ratio m. j The portion of the effective weight that is insufficient is compensated for, and the true porosity of the band is restored.

[0109] Furthermore, error variance quantification is performed based on Formula 16 to quantify the statistical error (variance) of the porosity P, reflecting the reliability of the results. 2 The smaller (P) is, the more reliable P is, and the more uniform the weight distribution is. The smaller the value of u, the smaller the error. 2 If (P) exceeds the preset threshold (e.g., 0.01), a quality inspection alarm will be triggered.

[0110] Formula Sixteen It should be noted that, for engineering simplification, the default is to use P with full hemispherical aggregation; if needed, piecewise statistics with finite angular bands can be enabled to enhance the adaptation to extreme geometries.

[0111] Figure 6 This is the sixth flowchart of the leaf area index determination method provided by the present invention, as shown below. Figure 6 As shown, the method includes the following: Step 601: Perform image correction processing on the acquired target hemisphere image to obtain the corrected target hemisphere image.

[0112] Step 602: Determine the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image.

[0113] Step 603: Determine the porosity of the plant based on the NDVI histogram and the preset zenith angle weight.

[0114] Step 604: Based on porosity, cluster correction coefficient, and leaf tilt extinction coefficient, retrieve the plant's LAI.

[0115] In one possible implementation, for discrete inversions that include cluster / leaf tilt angle corrections, a dual correction for cluster and leaf tilt angles is introduced to achieve physically accurate LAI calculations. Aggregate inversions achieve near real-time solutions through extremely simplified formulas and rely on stage calibration to achieve equivalence with the reference caliber, perfectly adapting to the engineering needs of high-throughput field operations.

[0116] Specifically, within the Poisson penetration framework of canopy light radiation, the porosity P of different zenith angle zones was analyzed. j We perform weighted fusion and introduce lightweight parameterization of the cluster correction coefficient β(θ) and the leaf tilt extinction coefficient G(θ), as shown in Equation 17.

[0117] Formula 17 in, Through small sample control or using P j To estimate the shape minimization residual β(θ) j ) represents the cluster correction factor for the j-th zenith angle zone (related to the zenith angle); G(θ) j ) represents the extinction coefficient of the leaf tilt angle distribution in the j-th zenith angle zone (related to the zenith angle); P j This represents the porosity after correction for the notch in the j-th zenith angle zone. It is the core output of the segmented statistics for the zenith angle zone, reflecting the actual light transmittance at each zenith angle; w j The weighting coefficient for the j-th zenith angle zone is determined by the solid angle proportion of the zenith angle zone; θ j Represents the central zenith angle of the j-th zenith angle zone; θ represents the angular interval of the zenith angle zone.

[0118] It should be noted that the clustering effect refers to the fact that crop leaves are not randomly distributed, but grow in clusters / rows, causing the light penetration probability to deviate from the Poisson distribution, so it is corrected by β(θ); the leaf tilt angle distribution refers to the fact that the leaf tilt angles of different crops (such as wheat being upright and soybeans being flat) are different, and the light extinction characteristics are different, so it is corrected by G(θ) (extinction coefficient).

[0119] Specifically, by introducing the cluster correction coefficient β(θ) and the leaf tilt angle extinction coefficient G(θ), along a certain observation direction (zenith angle θ), the canopy's shading of light can be abstracted as the interception of light beams by randomly distributed leaf elements. The result is that transmittance (i.e., porosity) decays exponentially, as shown in Equation 18.

[0120] Formula 18 Wherein, P(θ) represents the porosity in the direction of the zenith angle θ (P obtained by angular zone statistics). j (It is its discrete approximation); LAI represents leaf area index (one-sided leaf area per unit surface area), which is the quantity to be inverted in the model, and is the one-sided leaf area density in the vertical direction of the surface (dimensionless); G(θ) represents the leaf projection function (determined by the leaf tilt angle distribution), which characterizes the influence of the leaf tilt angle distribution on the "projected area of ​​the leaf in the observation direction"; Ω(θ) represents the cluster / aggregation correction coefficient (rows and clusters will increase or decrease the "porosity under the same LAI"), which corrects the transmittance deviation caused by the "non-random distribution of leaves (clustered / rowed)"; cosθ is the zenith angle geometric correction term, which converts the "path length along the line of sight" into the positive path length perpendicular to the surface (the path is longer when viewed at an angle).

[0121] Furthermore, by taking the natural logarithm of both sides of the exponential decay formula (Formula 18), the nonlinear relationship is transformed into a linear relationship, and the LAI inversion formula in the direction of a single zenith angle θ is directly solved, as shown in Formula 19.

[0122] Formula 19 However, a single angle is easily affected by anomalies, so the hemispherical integral is usually discretized into multiple angular zones for summation. The continuous range of hemispherical zenith angles is discretized into multiple angular zones, and the inversion results of each angular zone are summed using solid angle weighting. Multi-directional data fusion is used to suppress unidirectional anomalies and improve the robustness and accuracy of LAI inversion, as shown in Equation 20.

[0123] Formula 20 Among them, P j θ represents the porosity of the j-th angular band. j The angle band represents an angle (such as the central angle), w jβ(θ) represents the integral weight (reflecting how much of the angular zone is a "solid angle" on the hemisphere). j G(θ) represents the cluster and geometric correction term. j ) represents the leaf tilt angle projection function, w j This indicates the weight of the solid angle of the angular band.

[0124] Furthermore, to ensure real-time performance, a convergent fast inversion can be used to provide the effective leaf area index (LAI), as shown in Equation 21. Integrating over the entire hemispherical zenith angle simplifies the discrete weighted summation to a constant 2 (the integral result of the hemispherical solid angle); and ignoring the zonal corrections for cluster and leaf tilt angles, the convergent porosity P directly reflects the overall penetration characteristics. This determines the effective leaf area index (LAI). eff This reflects the actual light interception capability of the canopy and is the core output of real-time edge-side calculation.

[0125] Formula 21 Thus, it only involves logarithmic operations and one multiplication, and the single-frame processing time on the embedded end is less than 100ms, achieving near real-time operation; and the input is only the hemispherical weighted void ratio P, without the need for corner strip segmented data.

[0126] Due to LAI eff The model has been simplified, and its alignment with the reference aperture LAI (LAI) needs to be established through staged linear calibration. ref This involves establishing a correspondence between the results of discrete inversion (e.g., the true value of the quadrat harvesting method) and the reference caliber, eliminating systematic errors caused by simplification, and achieving rapid calculation and accurate equivalence. Therefore, it is necessary to use stage calibration to establish a piecewise equivalence with the reference caliber, as shown in Equation 22.

[0127] Formula 22 Among them, LAI ref This represents the reference aperture leaf area index, where a1 and b1 represent the lower interval LAI. eff The slope and intercept of the linear segment ≤τ, where a2 and b2 represent the high interval LAI. eff >τ represents the slope and intercept of the linear segment, where τ is the segment threshold / inflection point, determined by the abrupt change point of the fitting residual of the sample data.

[0128] In one possible implementation, quality control and anomaly removal can be performed on the results. During processing, a quality control layer is generated simultaneously, including: threshold τ, NDVI histogram, mask overlay, ROI coverage, saturated pixel ratio, and processing latency. When any indicator exceeds the threshold (e.g., insufficient ROI coverage or excessively high saturation ratio), the system prompts for re-sampling or automatically adjusts the candidate threshold range and masking template. All quality control results are archived along with the original images for easy post-processing review and method reproduction.

[0129] That is, by using multi-dimensional quality inspection layers, the end-side processing process can be visualized and monitored and anomalies can be automatically removed. Furthermore, by relying on synchronous calibration with reference equipment, consistency of LAI results across time periods and plots can be established. Finally, the calibration parameters and quality inspection thresholds are solidified into the project configuration, ensuring the reliability, consistency of standards, and reproducibility of the results of batch field operations from a technical perspective.

[0130] In one possible implementation, to address the inconsistencies in LAI results caused by differences in caliber between end-side equipment and industry reference equipment, as well as changes in environmental / crop characteristics across time periods / plots, a linear / segmented calibration relationship is established through synchronous point sampling calibration. The calibration parameters and quality inspection thresholds are then solidified into the project configuration, enabling cross-device calibration and cross-temporal consistency maintenance of LAI results. This ensures uniformity of caliber during batch field operations, allowing end-side equipment results to be benchmarked against industry standard reference equipment (LAI-2200 / 2200C).

[0131] Specifically, based on the sample pair (L) e Based on the distribution characteristics of L), univariate linear calibration or piecewise linear calibration is selected to establish a mathematical mapping relationship between the LAI of the end-side device and the LAI of the reference device, so as to achieve consistent maintenance across time periods and plots. The calibration parameters (a,b) and the quality inspection threshold are stored together and solidified into the project configuration to ensure the consistency of standards during batch field operations, as shown in Formula 23.

[0132] Formula 23 In one possible implementation, Figure 7 This is one of the structural schematic diagrams of the leaf area index determination system provided by the present invention, such as... Figure 7 As shown, the power supply module is the main control processing module, including a power supply and a power display module. The data acquisition module includes an RGBIR imaging module and a serializer. The main control processing module includes a display screen, a main control chip, and a deserializer. The main control processing module can use an RK3588 to complete image synchronous acquisition, NDVI calculation, restricted search Otsu segmentation, gap fraction estimation, and LAI inversion. The leaf area index data processing and visualization module includes an RGBIR image acquisition and preprocessing module, a vegetation index calculation and image segmentation module, and a leaf area index inversion module. The data acquisition module is a common RGBIR hemispherical imaging unit, using a 160° fisheye lens and coaxial confocal visible and near-infrared channels to ensure pixel-level spatial consistency. The display screen is used for on-site visualization and historical result review, and stores the original images, intermediate processing layers, and quality inspection indicators in a structured manner.

[0133] The software flow of this system includes, in sequence: system initialization, data acquisition, preprocessing and NDVI calculation, restricted threshold segmentation, gap fraction estimation, LAI inversion and quality control, local storage and visualization. Upon startup, calibration parameters such as projection and solid angle weights, channel black levels, etc., are loaded and hardware self-tests are completed to ensure the equipment is ready. Subsequently, a coaxially registered RGB-NIR fisheye camera acquires the upper hemisphere image and records the time. In the preprocessing stage, dark current correction and linearization are performed, and white-panel-free energy balancing is performed on the sky candidate region S to obtain the scale coefficient K. The corrected NDVI is calculated to eliminate energy level drift. In the segmentation stage, a restricted search interval is constructed based on the main peak of the NDVI histogram, and the threshold is calculated within it using the Otsu method. Geometric constraints are applied to the ROI and outer ring / fan-shaped shielding to suppress interference from the horizon and operator orientation, and a canopy mask is output. Then, based on the projection calibration, solid angle weights ω(x,y) are assigned to the pixels, and the hemispherical void fraction P is weighted and statistically calculated within the ROI. The LAI inversion is then completed according to the aggregated aperture or segmented weighted aperture, and quality inspection indicators are calculated. If the indicators are not met, a rollback strategy is triggered. Finally, the NDVI, mask, LAI, quality inspection layer and parameter set are stored locally and visualized in real time on the edge interface, supporting on-site verification and batch operation.

[0134] In one possible implementation, Figure 8 This is the second schematic diagram of the leaf area index determination system provided by the present invention, as shown below. Figure 8 As shown, the system includes a power supply 801, a main control processor 802 (RK3588), a data conversion module 803, an RGBIR imaging module 804, and a display / control terminal 805. The main control processor 802 and the RGBIR imaging module 804 are connected via a coaxial cable using a GMSL (Gigabit Multimedia Serial Link) high-speed serial link, achieving long-distance, low-error, and interference-resistant transmission of video and synchronization signals based on SerDes (serialization / deserialization) technology. The display / control terminal 805 is connected to the main control processor 802 and is used for parameter setting, result viewing, and historical playback. The main control processor 802 acquires RGBIR hemispherical image data through the GMSL high-speed link and completes processes such as NDVI calculation, restricted Otsu segmentation, porosity statistics, and LAI inversion. The final results are visualized in real-time and stored synchronously on the display terminal.

[0135] For example, in a complete embodiment, the power supply 801 is first turned on, and the main control processor 802 initiates system initialization, completing the self-test and readiness of the RGBIR imaging module 804 and the display / control terminal 805. Then, the device is vertically placed at the center of the crop rows, with the lens pointing upwards towards the canopy. Dual-channel synchronous acquisition is triggered via the touchscreen to acquire hemispherical images of the RGB and NIR bands. Subsequently, the main control processor 802 calculates the NDVI vegetation index in real time and performs normalization preprocessing on the NDVI values ​​to enhance spectral separability. Within a preset circular ROI, the main peak interval is automatically located based on the NDVI histogram, and the restricted search method (Otsu method) is executed to quickly determine the optimal segmentation threshold, generating a canopy vegetation mask. Depending on the site environment (such as strong reflective background or operator obstruction), a geometric masking template is dynamically activated, and the weights of distant or interfering areas are reset to zero. The weighted void fraction P within the masked ROI is calculated based on the zenith angle weight, and the effective LAI is inverted using a Poisson penetration model. Furthermore, it simultaneously generates quality inspection indicators: NDVI histogram, mask boundaries, saturated pixel ratio, and ROI coverage. If the quality inspection is abnormal (e.g., coverage <80%), the system prompts for re-sampled data or adaptively adjusts the threshold range. It outputs LAI results, spatialized mask layers, and quality inspection data to local storage and displays them in real-time on the touchscreen, supporting on-site verification.

[0136] In one possible implementation, Figure 9 This is the third schematic diagram of the leaf area index determination system provided by the present invention, as shown below. Figure 9 As shown, it includes an RGBIR imaging module 901, a device support rod 902, a touch screen 903, and a main control processor 904.

[0137] This application's embodiments employ a modular and lightweight design, facilitating portable operation and efficient data collection in the field. Utilizing natural sunlight as the illumination source avoids the complex calibration process of active lighting equipment. Embedded real-time processing synchronously acquires RGB-NIR dual-channel data and performs LAI inversion, significantly improving the efficiency and timeliness of field measurements. It solves the problem of unstable canopy segmentation caused by complex lighting and background interference. Through NDVI spectral enhancement and the restricted search Otsu method, combined with a dynamic geometric shielding strategy, it significantly improves the accuracy and efficiency of canopy-sky segmentation, ensuring the reliability of porosity statistics. Breaking through the spatial limitations of traditional point-based measurements, it achieves a visual representation of canopy porosity distribution. Through pixel-level masking and zenith angle weighted mapping, it outputs a two-dimensional porosity distribution map and spatialized LAI results, providing data support for crop row structural heterogeneity analysis and stress localization.

[0138] The leaf area index determination device provided by the present invention is described below. The leaf area index determination device described below can be referred to in correspondence with the leaf area index determination method described above.

[0139] Figure 10 This is a schematic diagram of the leaf area index determining device provided by the present invention, as shown below. Figure 10 As shown, the leaf area index determination device includes: a processing module 1001, a determination module 1002, and an inversion module 1003; Processing module 1001 is used to perform image correction processing on the acquired target hemisphere image to obtain a corrected target hemisphere image. The target hemisphere image is obtained by shooting vertically upward from the center position between the plant rows by the shooting device. The determination module 1002 is used to determine the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemispherical image. The determination module 1002 is also used to determine the porosity of plants based on the NDVI histogram and preset zenith angle weights; Inversion module 1003 is used to invert the leaf area index (LAI) of plants based on porosity.

[0140] According to the leaf area index determination device provided by the present invention, the processing module 1001 is specifically used for: Dark current correction is performed on the acquired target hemisphere image to obtain the target hemisphere image after dark current correction. Energy balancing is performed on the dark current-corrected target hemisphere image to obtain the corrected target hemisphere image.

[0141] According to the leaf area index determination device provided by the present invention, the target hemisphere image includes multiple channels, including: R channel, G channel, B channel and NIR channel; Processing module 1001 is specifically used for: Linearize the pixel values ​​of multiple channels in the target hemispherical image to determine the linear pixel value of each channel; Based on the preset dark current corresponding to each channel, the preset dark current is subtracted from the linear pixel value of each channel to obtain the corrected pixel value of each channel; Based on the corrected pixel value of each channel, the target hemisphere image after dark current correction is obtained.

[0142] According to the leaf area index determination device provided by the present invention, the processing module 1001 is specifically used for: Determine the sky reference region from the dark current corrected target hemisphere image; Based on the sky reference region, the scale factor of the R channel and the scale factor of the NIR channel in the target hemisphere image after dark current correction are determined. Based on the scale factors of the R channel and the NIR channel, the energy levels of the R channel and NIR channel are calibrated to obtain the corrected target hemisphere image.

[0143] According to the present invention, a leaf area index determining device, wherein the determining module 1002 is specifically used for: Determine the main peak in the NDVI histogram, and then determine the valley points on both sides of the main peak with the main peak as the center. The target threshold is determined based on the NDVI values ​​corresponding to the valley points on both sides of the main peak. Based on the target threshold, the pixels in the corrected target hemisphere image are divided into plant canopy pixels and sky background pixels; The porosity is determined based on the plant canopy pixels, sky background pixels, and preset zenith angle weights.

[0144] According to the present invention, a leaf area index determination device, including an inversion module 1003, is specifically used to invert the LAI of a plant based on porosity, cluster correction coefficient, and leaf tilt angle extinction coefficient.

[0145] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11 As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a leaf area index determination method, which includes: performing image correction processing on the acquired target hemisphere image to obtain a corrected target hemisphere image, wherein the target hemisphere image is obtained by the imaging device being located vertically upward at the center position between the rows of plants; determining the normalized difference vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image; determining the porosity of the plants based on the NDVI histogram and a preset zenith angle band weight; and retrieving the leaf area index (LAI) of the plants based on the porosity.

[0146] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0147] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the leaf area index determination method provided by the above methods. The method includes: performing image correction processing on the acquired target hemisphere image to obtain a corrected target hemisphere image, wherein the target hemisphere image is obtained by shooting vertically upward from the center position between the rows of plants using an imaging device; determining the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image; determining the porosity of the plants based on the NDVI histogram and a preset zenith angle band weight; and retrieving the leaf area index (LAI) of the plants based on the porosity.

[0148] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the leaf area index determination method provided by the above methods. The method includes: performing image correction processing on a collected target hemisphere image to obtain a corrected target hemisphere image, wherein the target hemisphere image is obtained by shooting vertically upward from the center position between rows of plants using an imaging device; determining the normalized vegetation index (NDVI) histogram corresponding to the plants in the corrected target hemisphere image; determining the porosity of the plants based on the NDVI histogram and a preset zenith angle band weight; and retrieving the leaf area index (LAI) of the plants based on the porosity.

[0149] The device embodiments described above are merely illustrative. 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 the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining leaf area index, characterized in that, include: The acquired target hemisphere image is subjected to image correction processing to obtain a corrected target hemisphere image. The target hemisphere image is obtained by the imaging device being located at the center of the plant row and taken vertically upwards. Determine the normalized vegetation index (NDVI) histogram corresponding to the plant in the corrected target hemispherical image. The porosity of the plant is determined based on the NDVI histogram and the preset zenith angle weight. The leaf area index (LAI) of the plant is derived from the porosity.

2. The method for determining leaf area index according to claim 1, characterized in that, The step of performing image correction processing on the acquired target hemispherical image to obtain a corrected target hemispherical image includes: Dark current correction is performed on the acquired target hemisphere image to obtain the target hemisphere image after dark current correction. The target hemisphere image after dark current correction is subjected to energy balancing processing to obtain the corrected target hemisphere image.

3. The method for determining leaf area index according to claim 2, characterized in that, The target hemispherical image includes multiple channels, including: R channel, G channel, B channel and NIR channel; The step of performing dark current correction on the acquired target hemispherical image to obtain a dark current-corrected target hemispherical image includes: The pixel values ​​of the multiple channels included in the target hemispherical image are linearized to determine the linear pixel value of each channel; Based on the preset dark current corresponding to each channel, the preset dark current is subtracted from the linear pixel value of each channel to obtain the corrected pixel value of each channel; Based on the corrected pixel value of each channel, the target hemisphere image after dark current correction is obtained.

4. The method for determining leaf area index according to claim 2, characterized in that, The step of performing energy balancing processing on the dark current-corrected target hemisphere image to obtain the corrected target hemisphere image includes: Determine the sky reference region from the dark current corrected target hemisphere image; Based on the sky reference region, the scale factor of the R channel and the scale factor of the NIR channel in the dark current corrected target hemisphere image are determined. Based on the scale factor of the R channel and the scale factor of the NIR channel, the energy levels of the R channel and the NIR channel are calibrated to obtain the corrected target hemisphere image.

5. The method for determining leaf area index according to any one of claims 1-4, characterized in that, The determination of the porosity of the plant based on the NDVI histogram and a preset zenith angle weight includes: Determine the main peak in the NDVI histogram, and determine the valley points on both sides of the main peak with the main peak as the center; The target threshold is determined based on the NDVI values ​​corresponding to the valley points on both sides of the main peak; Based on the target threshold, the pixels in the corrected target hemisphere image are divided into plant canopy pixels and sky background pixels; The porosity is determined based on the plant canopy pixels, the sky background pixels, and the preset zenith angle weight.

6. The method for determining leaf area index according to any one of claims 1-4, characterized in that, The method of retrieving the leaf area index (LAI) of the plant based on the porosity includes: The LAI of the plant is inverted based on the porosity, cluster correction coefficient, and leaf tilt extinction coefficient.

7. A leaf area index determining device, characterized in that, include: Processing module, determination module, and inversion module; The processing module is used to perform image correction processing on the acquired target hemisphere image to obtain a corrected target hemisphere image. The target hemisphere image is obtained by the shooting device taking a vertical upward shot from the center of the plant row. The determining module is used to determine the normalized vegetation index (NDVI) histogram corresponding to the plant in the corrected target hemispherical image; The determining module is also used to determine the porosity of the plant based on the NDVI histogram and the preset zenith angle weight; The inversion module is used to invert the leaf area index (LAI) of the plant based on the porosity.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the leaf area index determination method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the leaf area index determination method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the leaf area index determination method as described in any one of claims 1 to 6.

Images