Straw coverage identification method based on machine vision

Abstract

The invention relates to the technical field of image analysis, in particular to a straw coverage identification method based on machine vision, which comprises the following steps that a farmland polarization image group is acquired through polarization multi-angle imaging equipment, and the image group comprises near-infrared band images under multiple polarization angles; calculating a specular reflection suppression image based on the polarization image group, performing parallax value calculation on the suppressed image, and generating a straw-soil height difference mark graph; counting the coverage degree according to the height difference mark graph, wherein the coverage degree = the straw mark pixel number / the soil exposure area pixel number * 100%. According to the method, the final straw coverage calculation result has higher stability and field adaptability, errors are reduced under various interference scenes (weeds, water films and broken straw), and a credible basis is provided for intelligent agricultural machine seeding decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to a straw coverage recognition method based on machine vision. Background Art

[0002] Straw mulching is an important agronomic measure for conservation tillage and reducing soil wind and water erosion. Accurate monitoring of straw mulch is crucial for guiding no-till seeding and assessing operational quality. Traditional straw mulch identification methods rely primarily on the following technical approaches: Manual photography and image annotation: Operators use ordinary cameras to capture images of the farmland surface and manually annotate straw and soil areas to estimate coverage. This method is time-consuming, labor-intensive, and highly subjective, making it difficult to adapt to the automated inspection needs of large farmland areas.

[0003] RGB or multispectral imaging methods classify and identify straw and soil in images based on color or spectral differences. However, in conditions of strong lighting changes, weed cover, or straw decay and discoloration, color and spectral features can easily become confused, significantly reducing recognition accuracy.

[0004] In addition, some studies have attempted to use polarization imaging to detect the surface of farmland, but most of them use a rotating polarizer + single camera acquisition method. The image acquisition time difference is large (usually more than 200ms), resulting in serious image misalignment. Especially when acquiring on mobile agricultural machinery, it is very easy to produce blur and spatial offset, making it difficult to use for subsequent three-dimensional structure analysis.

[0005] In summary, existing methods generally have the following problems: temporal and spatial inconsistency in imaging, insufficient utilization of polarization data, difficulty in distinguishing weeds from straw structures, overly rough definition of soil areas, and lack of a highly robust coverage calculation model. There is an urgent need for a new straw cover identification method that integrates image consistency assurance, strong reflection suppression capability, accurate expression of structural features, and clear physical constraints of the calculation model. Summary of the Invention

[0006] The present invention provides a straw coverage recognition method based on machine vision.

[0007] A straw coverage recognition method based on machine vision comprises the following steps: S1, collecting a group of polarization images of farmland using a polarization multi-angle imaging device, wherein the group of images includes near-infrared band images at multiple polarization angles; S2, calculating a specular reflection suppressed image based on the polarization image group, calculating a disparity value of the suppressed image, and generating a straw-soil height difference marker map; S3, calculating coverage according to the height difference marking map, coverage = number of straw marking pixels / number of pixels in the soil exposed area × 100%; In S2, the area where the disparity value is greater than the preset difference threshold is marked as a straw area.

[0008] Optionally, the S1 specifically includes: S11, uses a beam splitter prism to split the incident light path into three channels, and each channel is equipped with a multi-polarization angle lower linear polarizer; S12, set a narrow band filter in the near infrared band; S13, driving the three CMOS sensors to collect data simultaneously through a synchronous exposure controller to generate a polarization image group aligned in spatial position.

[0009] Optionally, the band of the near-infrared band image , enhancing the reflection difference between the straw and the background soil.

[0010] Optionally, the multiple polarization angles include polarization angles of 0°, 45°, and 90°, so as to fully obtain the surface reflection characteristics and be used to calculate the polarization angle difference value.

[0011] A cube-shaped beam splitter is used to separate the incident light beam into three channels according to the direction of the optical path. Each channel is equipped with a linear polarizer, and the polarization directions are: 0° (parallel polarization direction); 45° (oblique polarization direction); 90° (vertical polarization direction).

[0012] Each surface of the beam splitter prism is coated with a near-infrared band anti-reflection film, and the average transmittance satisfies: ; To avoid inconsistent brightness of multi-channel images and eliminate the field of view inconsistency and image misalignment problems caused by traditional rotating polarizers.

[0013] Polarization angles of 0°, 45°, and 90° comprehensively capture surface reflection characteristics. 0° and 90° represent two orthogonal directions of polarized light, helping to distinguish between specular and diffuse reflections. Adding a 45° angle further captures intermediate polarization characteristics caused by asymmetric surfaces or tilted structures (such as upright or curled portions of straw). The combination of these three angles can be used to calculate polarization angle differences (such as maximum, minimum, and contrast), improving the ability to distinguish between reflection patterns. Specular reflection is strongest at specific polarization angles and can typically be suppressed by taking the minimum value (min) or degree of polarization (DoP) of images at different polarization angles. Using only two angles, 0° and 90°, may not be sufficient to cover reflections from certain directions. Adding a 45° angle improves the ability to suppress irregular reflective areas, ensuring that "real structure" rather than "specular artifacts" can be effectively determined at different incident angles.

[0014] The near-infrared narrowband filter setting includes adding a central wavelength of nm narrowband filter, filtering range is: ; Half-wave width nm, used to suppress visible light interference and enhance the reflectivity difference between straw and weeds in the near-infrared region (the reflectivity of weeds rises sharply at 870 nm, and the interference is significant).

[0015] The sensor synchronous exposure control specifically includes driving three CMOS image sensors to expose simultaneously through an external synchronous exposure controller, and the exposure start time error satisfies: , to ensure that there is almost no time dislocation in the image acquisition process when the high-speed mobile platform (agricultural machinery) is in operation.

[0016] For example: vehicle speed km / h, in The displacement in ms is only: .

[0017] The wavelength range (850nm±15nm) is selected based on the following: 1. Avoiding visible light interference and improving imaging stability: The visible light band (approximately 400–700 nm) is strongly interfered with by changes in sunlight intensity, leaf reflection, and color differences. This results in poor robustness of traditional RGB or broadband imaging in straw and soil identification. The near-infrared band (NIR), especially around 850 nm, not only avoids these interferences but is also suitable for identifying texture and structural differences because of its relatively small light fluctuations and more uniform reflection under natural conditions.

[0018] 2. The characteristic window of the 850nm band in plant identification: In the spectral characteristics of plants, the main absorption bands of chlorophyll are located in the red and blue light regions (roughly 430nm and 660nm), while the reflection is significantly enhanced in the near-infrared region (700-900nm); however, straw, as dead plant tissue, has degraded cell structure and its reflectivity at 850nm is much lower than that of fresh vegetation, while dry soil has stable reflection in this band; the reflectivity of weeds (especially living leaves) rises sharply around 870nm (up to more than 30% reflection), and if the filter band is too wide, this type of interference signal will be introduced.

[0019] 3. Measured spectral data shows: Soil reflectance: stable at 800–900 nm, with a reflectance of approximately 20–30%; Reflectance of straw: significantly higher than soil at 850nm (difference of about 10–20%), but much lower than fresh green plants; Weed reflectance: Rapidly increases at 870–880nm (up to 30% compared to 850nm), which can be significantly misjudged as a high-contrast area of "straw."

[0020] The reflection difference between straw and background (soil) is enhanced while avoiding interference from the "climbing zone" (860-900nm) of living plant reflection. The high reflection zone of living weeds happens to appear at 870-900nm. The bandwidth of the present invention is controlled at 850±15nm, effectively avoiding the introduction of high reflection values above 870nm into the polarization image calculation, ensuring that the mirror suppression image will not be misjudged due to highly reflective weeds.

[0021] By limiting the wavelength to 850±15nm, the impact of non-structural reflections from different ground objects can be effectively reduced, the texture consistency between polarization angle images can be improved, and it can be ensured that the main source of reflection differences is the three-dimensional structure of the straw, rather than spectral disturbances.

[0022] Optionally, the S2 specifically includes: S21, performing pixel-level registration on the near-infrared band images under the polarization angle, establishing a spatial coordinate mapping relationship, and ensuring that the images at each polarization angle reflect the same physical position at the same coordinate; S22, calculating the polarization angle difference value of each pixel; S23, constructing a specular reflection suppression image: taking the minimum value of the polarization angle difference values of all pixels as a reference, and generating a normalized specular reflection suppression image: S24, performing binocular disparity calculation on the mirror reflection suppressed image to generate a disparity map; S25, converting the disparity map into a height difference mark map; S26, marking the continuous area with a height difference greater than 0.5 mm in the height difference marking map as a straw area.

[0023] Optionally, the binocular disparity calculation includes taking the 45° polarization image as the reference view and using a semi-global matching algorithm to respectively and The images are stereo matched and two disparity maps are output, which are averaged and fused to generate the final disparity map.

[0024] Optionally, the polarization angle difference value is calculated as: ;in, The polarization angle is Image at coordinates The gray value at Represents the polarization angle difference value map, which is used to characterize the reflection angle anisotropy.

[0025] Optionally, the generation of the height difference marker map includes converting the disparity value into a real physical height difference, and the conversion is expressed as: ,in, is the baseline distance, i.e. the center distance between the two imaging channels, is the focal length of the imaging lens, is the physical height difference at the coordinate (x, y) in the height difference map, is the disparity value at coordinate (x,y).

[0026] Optionally, the S3 specifically includes: S31, performing a morphological closing operation on the height difference mark map, where the structural element of the closing operation is a circular core that connects the broken straw areas; S32: Extract connected areas with height differences greater than 0.5 mm as straw areas and count the total number of pixels. ; S33, extracting continuous areas with height differences ≤ 0.2 mm as soil exposure areas; S34, count the total number of pixels in the soil exposure area ; S35, calculate coverage: .

[0027] Optionally, the soil exposure area also meets the following requirements: Area ≥100cm 2 , converted to image pixels ≥ 500 pixels; The standard deviation of the height difference within the area is ≤0.05mm.

[0028] Beneficial effects of the present invention: This invention utilizes a beam splitter prism and three-channel synchronous exposure architecture, coupled with fixed 0° / 45° / 90° polarizers and a narrowband filter with a central wavelength of 850nm±15nm. This effectively avoids the image misalignment problem associated with rotating polarizers in traditional single-sensor sensors, while also filtering out spectral noise caused by the surge in reflections from living weeds at 870nm. Furthermore, the specular reflection suppression image constructed based on polarization angle differences enhances the optical structural differences between straw and soil, allowing for significant structural and texture differences in the straw region without contact, providing ideal input for subsequent height analysis.

[0029] The present invention applies a semi-global matching algorithm guided by a polarization map to farmland straw identification. By using a 45° polarization map as a reference view and combining it with 0° and 90° views to perform differential parallax calculations, effective straw parallax information can still be extracted in low-texture, partially occluded, or overlapping areas. Combined with the actual calibrated baseline distance and focal length parameters, the generated height difference marker map has significant distinguishing capabilities at the thickness level of rotten straw, and its structural texture performance is superior to traditional RGB or NIR depth mapping methods.

[0030] The present invention designs a soil exposure area identification mechanism based on the triple physical constraints of height difference threshold, area and flatness standard deviation, which effectively avoids the problem of traditional methods misidentifying stones, ruts and shadows as exposed soil. 2 Areas with a standard deviation of ≤0.05mm are screened, ensuring greater surface consistency and physical plausibility in the extracted soil areas. The final straw cover calculation results are more stable and adaptable to field conditions, reducing errors in various interference scenarios (weeds, water film, and broken straw), providing a reliable basis for seeding decisions made by intelligent agricultural machinery. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention; Figure 2 Schematic diagram of the coverage calculation process of an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art may also implement some known technologies in other alternative ways. The accompanying drawings are only for describing the embodiments in more detail and are not intended to limit the present invention in any specific way.

[0034] like Figure 1-Figure 2 As shown, a straw coverage recognition method based on machine vision includes the following steps: S1, collects a group of polarization images of farmland using polarization multi-angle imaging equipment, which includes near-infrared band images at multiple polarization angles; S2, calculating a specular reflection suppression image based on the polarization image group, calculating a disparity value of the suppressed image, and generating a straw-soil height difference marker map; S3, the coverage was calculated based on the height difference marking map, coverage = number of straw marking pixels / number of pixels in the soil exposed area × 100%; In S2, the area where the disparity value is greater than the preset difference threshold is marked as a straw area.

[0035] S1 specifically includes: S11, spectroscopic polarization imaging channel design: A cubic spectroscopic prism is used to separate the incident light beam into three channels according to the optical path direction. Each channel is equipped with a linear polarizer, and the polarization directions are: 0° (parallel polarization direction); 45° (oblique polarization direction); 90° (vertical polarization direction).

[0036] Each surface of the beam splitter prism is coated with a near-infrared band anti-reflection film, and the average transmittance satisfies: ; To avoid inconsistent brightness of multi-channel images and eliminate the field of view inconsistency and image misalignment problems caused by traditional rotating polarizers.

[0037] Polarization angles of 0°, 45°, and 90° comprehensively capture surface reflection characteristics. 0° and 90° represent two orthogonal directions of polarized light, helping to distinguish between specular and diffuse reflections. Adding a 45° angle further captures intermediate polarization characteristics caused by asymmetric surfaces or tilted structures (such as upright or curled portions of straw). The combination of these three angles can be used to calculate polarization angle differences (such as maximum, minimum, and contrast), improving the ability to distinguish between reflection patterns. Specular reflection is strongest at specific polarization angles and can typically be suppressed by taking the minimum value (min) or degree of polarization (DoP) of images at different polarization angles. Using only two angles, 0° and 90°, may not be sufficient to cover reflections from certain directions. Adding a 45° angle improves the ability to suppress irregular reflective areas, ensuring that "real structure" rather than "specular artifacts" can be effectively determined at different incident angles.

[0038] S12, near-infrared narrowband filter setting: add a central wavelength of nm narrowband filter, filtering range is: ; Half-wave width nm, used to suppress visible light interference and enhance the reflectivity difference between straw and weeds in the near-infrared region (the reflectivity of weeds rises sharply at 870 nm, and the interference is significant).

[0039] S13, sensor synchronous exposure control: an external synchronous exposure controller drives three CMOS image sensors to expose simultaneously, and the exposure start time error satisfies: , to ensure that there is almost no time dislocation in the image acquisition process when the high-speed mobile platform (agricultural machinery) is in operation.

[0040] For example: vehicle speed km / h, in The displacement in ms is only: .

[0041] The wavelength range (850nm±15nm) is selected based on the following: 1. Avoiding visible light interference and improving imaging stability: The visible light band (approximately 400–700 nm) is strongly interfered with by changes in sunlight intensity, leaf reflection, and color differences. This results in poor robustness of traditional RGB or broadband imaging in straw and soil identification. The near-infrared band (NIR), especially around 850 nm, not only avoids these interferences but is also suitable for identifying texture and structural differences because of its relatively small light fluctuations and more uniform reflection under natural conditions.

[0042] 2. The characteristic window of the 850nm band in plant identification: In the spectral characteristics of plants, the main absorption bands of chlorophyll are located in the red and blue light regions (roughly 430nm and 660nm), while the reflection is significantly enhanced in the near-infrared region (700-900nm); however, straw, as dead plant tissue, has degraded cell structure and its reflectivity at 850nm is much lower than that of fresh vegetation, while dry soil has stable reflection in this band; the reflectivity of weeds (especially living leaves) rises sharply around 870nm (up to more than 30% reflection), and if the filter band is too wide, this type of interference signal will be introduced.

[0043] 3. Measured spectral data shows: Soil reflectance: stable at 800–900 nm, with a reflectance of approximately 20–30%; Reflectance of straw: significantly higher than soil at 850nm (difference of about 10–20%), but much lower than fresh green plants; Weed reflectance: Rapidly increases at 870–880nm (up to 30% compared to 850nm), which can be significantly misjudged as a high-contrast area of "straw."

[0044] The reflection difference between straw and background (soil) is enhanced while avoiding interference from the "climbing zone" (860-900nm) of living plant reflection. The high reflection zone of living weeds happens to appear at 870-900nm. The bandwidth of the present invention is controlled at 850±15nm, effectively avoiding the introduction of high reflection values above 870nm into the polarization image calculation, ensuring that the mirror suppression image will not be misjudged due to highly reflective weeds.

[0045] By limiting the wavelength to 850±15nm, the impact of non-structural reflections from different ground objects can be effectively reduced, the texture consistency between polarization angle images can be improved, and it can be ensured that the main source of reflection differences is the three-dimensional structure of the straw, rather than spectral disturbances.

[0046] S2 specifically includes: S21, pixel-level registration processing: perform pixel-level registration on the near-infrared images at 0°, 45°, and 90° polarization angles to establish a unified spatial coordinate mapping relationship to ensure that the images at each polarization angle are at the same coordinate. reflect the same physical location.

[0047] S22, polarization angle difference calculation: For each pixel, calculate the polarization angle difference value: ;in, The polarization angle is Image at coordinates The gray value at Represents the polarization angle difference value map, which is used to characterize the reflection angle anisotropy.

[0048] S23, specular reflection suppression image construction: construct a normalized specular reflection suppression image based on the polarization angle difference value map: ; in, ; represents the minimum difference value of all pixels in the entire image; , Indicates the grayscale value of the mirror reflection suppression image, ranging from 0 to 255, dynamic range ; It is used to eliminate the interference of changes in light intensity under different shooting conditions on image differences and enhance the significance of straw structure.

[0049] S24, binocular disparity calculation: Polarization image as reference view; Use the Semi-Global Matching (SGM) algorithm to and Images are stereo matched; Output two disparity maps 、 , perform average fusion to generate the final disparity map .

[0050] S25, height difference marker map generation: convert the pixel disparity value into the real physical height difference using the following formula: ,in, is the baseline distance, i.e. the center distance between the two imaging channels, with a value range of 20–30 cm. is the focal length of the imaging lens, ranging from 8–12mm, is the physical height difference at the coordinate (x, y) in the height difference map, unit: mm, is the disparity value at the coordinate (x, y), in pixels.

[0051] S26, straw area marking: mark the continuous area in the height difference map that meets the following conditions as the straw area: mm Straw area; the height threshold of 0.5 mm is a robust segmentation standard set based on 382 groups of measured rotten straw thickness samples and the standard deviation (0.12 mm).

[0052] S3 specifically includes: S31, morphological closing operation processing: perform morphological closing operation on the height difference mark map to eliminate small holes and connect broken areas. The operation is as follows: The structural element kernel adopts a circular morphological kernel with a size of 5 × 5; The operation sequence is: dilation → erosion to achieve the connection of the broken straw area.

[0053] The size of the circular core is based on the measured average width of straw fracture of 4.2 mm, and the corresponding pixel under the adapted spatial resolution condition is approximately 5 pixels in diameter.

[0054] S32, extracting straw regions: extracting connected regions that meet the following conditions from the closing operation results: ; Count the total number of pixels in the area as: ;in, Represents the value of the pixel in the height difference map, represents the set of connected regions with a height difference greater than 0.5 mm, i.e., the straw region. Indicates the number of straw marker pixels.

[0055] S33, extract soil exposure area: define the judgment conditions for soil area extraction as follows: ; in, represents the candidate low-altitude region, Indicates the area of the region in pixels. Indicates area The standard deviation of the inner height differences.

[0056] S34, counting the total number of pixels in the soil exposure area: summing the total number of pixels in all areas that meet the above three conditions to obtain: ;in, Indicates that the height difference is met mm, area Pixels, standard deviation mm area set, Indicates the number of pixels in the soil exposed area.

[0057] S35, the final coverage calculation formula is as follows: .

[0058] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0059] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A straw coverage recognition method based on machine vision, characterized in that: The following steps are involved: S1, collecting a group of polarization images of farmland using a polarization multi-angle imaging device, wherein the group of images includes near-infrared band images at multiple polarization angles; S2, calculating a specular reflection suppressed image based on the polarization image group, calculating a disparity value of the suppressed image, and generating a straw-soil height difference marker map; S3, calculating coverage according to the height difference marking map, coverage = number of straw marking pixels / number of pixels in the soil exposed area × 100%; In S2, the area where the disparity value is greater than the preset difference threshold is marked as a straw area.

2. The method for identifying straw coverage based on machine vision according to claim 1, characterized in that: Said S1 specifically includes: S11, uses a beam splitter prism to split the incident light path into three channels, and each channel is equipped with a multi-polarization angle lower linear polarizer; S12, set a narrow band filter in the near infrared band; S13, driving the three CMOS sensors to collect data simultaneously through a synchronous exposure controller to generate a polarization image group aligned in spatial position.

3. The method for recognizing straw coverage based on machine vision according to claim 1, characterized in that: The band of the near-infrared band image , enhancing the reflection difference between the straw and the background soil.

4. The method for recognizing straw coverage based on machine vision according to claim 1, characterized in that: The multiple polarization angles include polarization angles of 0°, 45°, and 90°, so as to fully obtain the surface reflection characteristics and be used to calculate the polarization angle difference value.

5. The method for recognizing straw coverage based on machine vision according to claim 4, characterized in that: The S2 specifically includes: S21, performing pixel-level registration on the near-infrared band images under the polarization angle, establishing a spatial coordinate mapping relationship, and ensuring that the images at each polarization angle reflect the same physical position at the same coordinate; S22, calculating the polarization angle difference value of each pixel; S23, constructing a specular reflection suppression image: taking the minimum value of the polarization angle difference values of all pixels as a reference, and generating a normalized specular reflection suppression image: S24, performing binocular disparity calculation on the mirror reflection suppressed image to generate a disparity map; S25, converting the disparity map into a height difference mark map; S26, marking the continuous area with a height difference greater than 0.5 mm in the height difference marking map as a straw area.

6. The method for recognizing straw coverage based on machine vision according to claim 5, characterized in that: The binocular disparity calculation includes taking the 45° polarization image as the reference view and using the semi-global matching algorithm to respectively and The images are stereo matched and two disparity maps are output, which are averaged and fused to generate the final disparity map.

7. The method for identifying straw coverage based on machine vision according to claim 5, characterized in that: The polarization angle difference value is calculated as: ;in, The polarization angle is Image at coordinates The gray value at Represents the polarization angle difference value map, which is used to characterize the reflection angle anisotropy.

8. The method for identifying straw coverage based on machine vision according to claim 5, characterized in that: The generation of the height difference mark map includes converting the disparity value into the real physical height difference, and the conversion is expressed as: ,in, is the baseline distance, i.e. the center distance between the two imaging channels, is the focal length of the imaging lens, is the physical height difference at the coordinate (x, y) in the height difference map, is the disparity value at coordinate (x,y).

9. The method for identifying straw coverage based on machine vision according to claim 1, characterized in that: The S3 specifically includes: S31, performing a morphological closing operation on the height difference mark map, where the structural element of the closing operation is a circular core that connects the broken straw areas; S32: Extract connected areas with height differences greater than 0.5 mm as straw areas and count the total number of pixels. ; S33, extracting continuous areas with height differences ≤ 0.2 mm as soil exposure areas; S34, count the total number of pixels in the soil exposure area ; S35, calculate coverage: .

10. The method for recognizing straw coverage based on machine vision according to claim 9, characterized in that: The soil exposure area also meets the following requirements: Area ≥100cm 2 , converted to image pixels ≥ 500 pixels; The standard deviation of the height difference within the area is ≤0.05mm.

Images