Tobacco variety identification method and system based on visible light image and near-infrared spectrum fusion

Abstract

The present application relates to the technical field of tobacco intelligent identification, and particularly relates to a tobacco variety identification method and system based on visible light image and near-infrared spectrum fusion. The method comprises the following steps: obtaining visible light image and near-infrared spectrum data of tobacco leaves; performing light interference analysis according to the visible light image of the tobacco leaves, and recording brightness offset parameters; generating brightness compensation data according to the brightness offset parameters; applying the brightness compensation data to the visible light image of the tobacco leaves, and detecting leaf aging characteristics; performing spectrum reflection characteristic analysis according to the near-infrared spectrum data, and recording characteristic band reflection parameters; and determining tobacco spectrum characteristics based on the characteristic band reflection parameters. The present application effectively reduces the influence of light interference on the identification result based on the tobacco intelligent identification technology, and improves the accuracy, stability and classification reliability of the tobacco variety identification under complex environments.

Description

Technical Field

[0001] This invention relates to the field of tobacco intelligent identification technology, and in particular to a method and system for identifying tobacco varieties based on the fusion of visible light images and near-infrared spectra. Background Technology

[0002] Current tobacco variety identification technologies typically rely on manual observation of leaf color, shape, and vein characteristics. This method is susceptible to variations in worker experience, resulting in low efficiency, significant subjective errors, and difficulties in large-scale testing. With the development of artificial intelligence, some solutions have begun using visible light image recognition for automatic tobacco leaf classification. However, existing image recognition technologies are easily affected by factors such as lighting changes, shooting angle, leaf aging, and pests, leading to insufficient recognition stability. On the other hand, near-infrared spectroscopy can detect physiological changes such as chlorophyll and water content within tobacco leaves. However, most existing spectral recognition schemes only analyze single biochemical indicators, lacking the ability to integrate these with tobacco appearance features. Furthermore, current image and spectral recognition methods often employ independent processing, lacking a unified fusion analysis mechanism. This makes it difficult to accurately determine tobacco varieties by comprehensively considering both appearance and physiological characteristics, resulting in low accuracy in complex environments. Summary of the Invention

[0003] Therefore, the present invention needs to provide a method and system for identifying tobacco varieties based on the fusion of visible light images and near-infrared spectra, in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for tobacco variety identification based on the fusion of visible light images and near-infrared spectra includes the following steps: Step S1: Acquire visible light images and near-infrared spectral data of tobacco leaves; perform illumination interference analysis based on the visible light images of tobacco leaves and record the brightness shift parameters; Step S2: Generate brightness compensation data based on brightness offset parameters; apply the brightness compensation data to the visible light image of tobacco leaves and detect leaf aging characteristics; Step S3: Perform spectral reflectance characteristic analysis based on near-infrared spectral data and record the reflectance parameters of characteristic bands; determine the spectral characteristics of tobacco based on the reflectance parameters of characteristic bands; Step S4: Perform feature fusion of leaf aging characteristics and tobacco spectral characteristics to generate fusion recognition results; determine the tobacco variety based on the fusion recognition results.

[0005] Preferably, this specification also provides a tobacco variety identification system based on the fusion of visible light images and near-infrared spectra, used to perform the tobacco variety identification method based on the fusion of visible light images and near-infrared spectra as described above. This tobacco variety identification system based on the fusion of visible light images and near-infrared spectra includes: The data acquisition module is used to acquire visible light images and near-infrared spectral data of tobacco leaves; it performs light interference analysis based on the visible light images of tobacco leaves and records brightness shift parameters. The leaf aging detection module is used to generate brightness compensation data based on brightness offset parameters; apply the brightness compensation data to visible light images of tobacco leaves, and detect leaf aging characteristics; The spectral analysis module is used to perform spectral reflectance characteristic analysis based on near-infrared spectral data, record reflectance parameters of characteristic bands, and determine the spectral characteristics of tobacco based on the reflectance parameters of characteristic bands. The variety identification module is used to fuse leaf aging characteristics and tobacco spectral characteristics to generate a fused identification result; the tobacco variety is determined based on the fused identification result.

[0006] The beneficial effects of this invention are as follows: (1) By performing illumination interference analysis on visible light images of tobacco leaves and generating brightness compensation data based on brightness offset parameters, adaptive correction of tobacco leaf images under complex lighting conditions was achieved, reducing the impact of strong light, shadows and changes in shooting angle on tobacco variety identification and improving the stability and accuracy of tobacco leaf image recognition.

[0007] (2) By combining the yellowing area of ​​the leaves, the vein texture and the curling degree of the leaf edge to analyze the aging characteristics of the leaves, and using near-infrared spectroscopy to detect the chlorophyll content, moisture content and tobacco maturity, the collaborative identification of the appearance characteristics and internal physiological characteristics of tobacco leaves was realized, which improved the reliability and anti-interference ability of tobacco variety identification.

[0008] (3) By integrating leaf aging characteristics with tobacco spectral characteristics, a dual-modal recognition mechanism of visible light image and near-infrared spectrum is constructed, realizing the joint judgment of external morphological information and internal biochemical information of tobacco leaves. This can effectively reduce the misjudgment rate of single image recognition or single spectrum recognition, and improve the intelligent level and recognition accuracy of tobacco variety recognition in complex environments. Attached Figure Description

[0009] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps of the tobacco variety identification method based on the fusion of visible light images and near-infrared spectra of the present invention; Figure 2 This is a schematic diagram of the tobacco variety identification system based on the fusion of visible light images and near-infrared spectra of the present invention. Figure 3 This is a schematic diagram of the near-infrared spectral curve in this invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0010] The technical method 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. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0011] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0013] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for identifying tobacco varieties based on the fusion of visible light images and near-infrared spectra, the method comprising the following steps: Step S1: Acquire visible light images and near-infrared spectral data of tobacco leaves; perform illumination interference analysis based on the visible light images of tobacco leaves and record the brightness shift parameters; In one embodiment, visible light images of tobacco leaves are acquired using an industrial RGB camera (1920×1080 resolution), and leaf reflectance spectral data in the 700nm-1100nm wavelength range are acquired using a near-infrared spectral sensor. During the acquisition process, stable illumination is provided by a ring-shaped LED supplementary lighting device, and the camera exposure parameters are fixedly controlled. Subsequently, brightness analysis is performed on the visible light images of the tobacco leaves. Leaf brightness data is extracted using the L channel in the LAB color space, and the average brightness value of the local area is calculated using a 15×15 pixel window. When the brightness value of a local area is higher than a preset high-brightness threshold, the area is marked as a high-brightness area; when the brightness value of a local area is lower than a preset shadow threshold, the area is marked as a shadow area. The brightness difference between the high-brightness area and the shadow area is calculated, and a brightness offset parameter is generated based on the brightness difference.

[0014] In another embodiment, a 12-megapixel industrial camera was used to acquire images of 300 samples of Yunyan 87, Honghua Dajinyuan, and inferior tobacco leaves. The image acquisition distance was controlled between 25cm and 35cm. The sampling interval of the near-infrared spectral sensor was set to 2nm, and a total of 201 band points were acquired. During the image brightness analysis, the high-brightness threshold was set to 210, and the shadow threshold was set to 65. In a certain tobacco leaf image, a total of 42 bright areas and 37 shadow areas were detected. The average brightness value of the bright areas was 228, and the average brightness value of the shadow areas was 51. The final brightness difference was calculated to be 177, which was recorded as the brightness offset parameter.

[0015] Step S2: Generate brightness compensation data based on brightness offset parameters; apply the brightness compensation data to the visible light image of tobacco leaves and detect leaf aging characteristics; In one embodiment, a brightness compensation curve is constructed based on brightness offset parameters, and the CLAHE local brightness equalization algorithm is used to correct the brightness of the tobacco leaf image, generating a corrected tobacco leaf image. Subsequently, leaf color parameters are extracted based on the HSV color space, and yellowing areas are identified based on changes in the H and S channels. Simultaneously, vein texture extraction is performed on the corrected tobacco leaf image, and the Frangi filtering algorithm is used to enhance the structure of the main and lateral veins. Then, the leaf edge curvature change value is calculated based on the vein texture to identify curled areas at the leaf edges. The leaf aging level is assessed by combining the proportion of yellowed areas with the degree of leaf edge curling, and leaf aging characteristics are determined based on the leaf aging level.

[0016] In another embodiment, brightness compensation processing was performed on 200 groups of tobacco leaf samples, with the CLAHE contrast limit set to 2.5 and the grid size set to 8×8. After compensation processing, the average leaf brightness increased from the original 86 to 132, and the brightness fluctuation range in the shaded areas decreased. Reduce to Subsequent testing revealed that the yellowed area ranged from 12% to 38%, with the average yellowed area of ​​aged tobacco leaves reaching 31%. During vein texture analysis, one main vein and 24-37 lateral veins were identified; the maximum change in leaf edge curvature reached 0.72. Tobacco leaves with a yellowed area exceeding 30% and an edge curvature greater than 0.65 were classified as high-aging leaves.

[0017] Step S3: Perform spectral reflectance characteristic analysis based on near-infrared spectral data and record the reflectance parameters of characteristic bands; determine the spectral characteristics of tobacco based on the reflectance parameters of characteristic bands; In one embodiment, Savitzky-Golay smoothing and baseline correction were performed on the acquired near-infrared spectral data to reduce noise interference. Subsequently, spectral reflectance data in the chlorophyll-sensitive band (720nm-780nm) and the water absorption-sensitive band (900nm-980nm) were extracted, and the chlorophyll absorption migration state and water absorption peak variation characteristics were analyzed. Chlorophyll synthesis activity and chlorophyll degradation trends were assessed based on the chlorophyll absorption migration state, and the degree of water loss in leaf tissue was assessed based on the water absorption peak variation. The maturity of tobacco leaves was determined by combining chlorophyll content and leaf water content, and the corresponding standard spectral reflectance curve was matched to the tobacco leaf maturity. The reflectance intensity information of the corresponding bands was recorded to generate characteristic band reflectance parameters and determine the spectral characteristics of tobacco.

[0018] In another embodiment, analysis was performed on spectral data from 300 sets of tobacco leaves, with a spectral smoothing window length of 11 and a second-order polynomial fitting order of 2. At the 750 nm wavelength, the average reflectance of mature tobacco leaves was 0.63, while that of immature tobacco leaves was 0.48. At the 970 nm wavelength, the average absorption peak of high-moisture-content tobacco leaves reached 0.71, while that of low-moisture-content tobacco leaves was 0.39. The chlorophyll absorption edge was mainly distributed between 735 nm and 752 nm, with the absorption edge of aged tobacco leaves shifting to an average of 748 nm. Standard spectral reflectance curves were obtained based on different maturity stages, and the reflectance intensities at the 750 nm, 820 nm, 910 nm, and 970 nm wavelengths were recorded to construct the spectral characteristics of tobacco.

[0019] Step S4: Perform feature fusion of leaf aging characteristics and tobacco spectral characteristics to generate fusion recognition results; determine the tobacco variety based on the fusion recognition results.

[0020] In one embodiment, features such as the proportion of yellowed area, vein texture density, and leaf edge curling in leaf aging characteristics are normalized with chlorophyll absorption edge position, water absorption peak, and characteristic band reflectance intensity in tobacco spectral characteristics, and then input into a multimodal fusion recognition model. The fusion recognition model employs a dual-branch neural network structure, where the image feature branch uses MobileNetV2 to extract leaf appearance features, and the spectral feature branch uses an MLP network to extract spectral physiological features. The fusion recognition result is generated through a feature splicing layer and a Softmax classification layer, and the corresponding tobacco variety category is output.

[0021] In another embodiment, fusion recognition analysis was performed on 500 sets of samples, including Yunyan 87, Honghua Dajinyuan, and inferior tobacco leaves, with the training set accounting for 80% and the test set accounting for 20%. The image feature dimension was set to 128 dimensions, the spectral feature dimension was set to 64 dimensions, and the fused feature dimension was 192 dimensions. After 120 rounds of model training, the overall recognition accuracy of the fusion recognition model reached 96.8%, an improvement of 9.4% compared to the single image recognition model and 7.1% compared to the single spectral recognition model. Specifically, the recognition accuracy of Yunyan 87 reached 97.5%, the recognition accuracy of Honghua Dajinyuan reached 96.1%, and the recognition accuracy of inferior tobacco leaves reached 95.4%.

[0022] Preferably, in step S1, illumination interference analysis is performed based on the visible light image of the tobacco leaf, and the brightness shift parameters are recorded, including: Extract leaf brightness data from visible light images of tobacco leaves; determine pixel grayscale values ​​based on leaf brightness data; identify areas with pixel grayscale values ​​higher than a preset high-brightness threshold as high-brightness areas; In one embodiment, an industrial RGB camera is used to acquire visible light images of tobacco leaves, and the acquired RGB images are converted into grayscale images. Then, a sliding window approach is used to perform local brightness analysis on the tobacco leaf images, using a 9×9 pixel area as the local analysis window. The average brightness value within each window area is calculated, and leaf brightness data is generated based on the average brightness value. The corresponding pixel grayscale values ​​are calculated based on the leaf brightness data, and the grayscale distribution on the tobacco leaf surface is statistically analyzed. When the pixel grayscale value in a local area is higher than a preset high-brightness threshold, the corresponding area is identified as a high-brightness area.

[0023] In another embodiment, an industrial camera with a resolution of 1920×1080 was used to acquire images of 240 samples of Yunyan 87, Honghua Dajinyuan, and inferior tobacco leaves. The image acquisition distance was controlled at approximately 30cm, and the acquisition frequency was set to 25fps. During the image grayscale analysis, the RGB images were converted to 8-bit grayscale images, and the local average brightness value was calculated using an 11×11 pixel window. A highlight threshold of 210 was set. In a certain tobacco leaf image, a total of 38 highlight areas were detected, with the largest highlight area being 426 pixels and the smallest being 57 pixels. The average grayscale value of the highlight areas was 228.

[0024] Regions with pixel grayscale values ​​below a preset shadow threshold are identified as shadow regions; the brightness shift of the highlight and shadow regions is calculated; and the brightness shift is used as a brightness shift parameter.

[0025] In one embodiment, after identifying the bright areas, shadow area detection is performed on the grayscale image of the tobacco leaf. When the grayscale value of a local area pixel is lower than a preset shadow threshold, the corresponding area is identified as a shadow area to characterize the location area of ​​the tobacco leaf that is obstructed, backlit, or subject to local dark light interference. Subsequently, the average grayscale values ​​of the bright areas and shadow areas are calculated separately, and the brightness shift is calculated based on the difference between the average grayscale values ​​of the bright areas and the shadow areas. Combining the area proportions of the bright areas and the shadow areas, the brightness shift is normalized, and the normalized result is used as the brightness shift parameter.

[0026] In another embodiment, a shadow threshold of 65 was set. In 240 tobacco leaf samples, 52 shadow areas were detected, with an average grayscale value of 49 and a maximum shadow area of ​​518 pixels. In a certain tobacco leaf image, the average grayscale value of the highlight area was 228, and the average grayscale value of the shadow area was 49, with a grayscale difference of 179. The highlight area accounted for 13.2% of the total area, while the shadow area accounted for 17.6%. After normalization, the brightness shift was calculated to be 0.73, and this 0.73 was recorded as the brightness shift parameter for the corresponding tobacco leaf image.

[0027] Preferably, in step S2, applying the brightness compensation data to the visible light image of the tobacco leaf and detecting leaf aging characteristics includes: Brightness compensation data is applied to visible light images of tobacco leaves to generate corrected tobacco leaf images; leaf color parameters are determined based on the corrected tobacco leaf images; and yellowed areas are marked based on the leaf color parameters. In one embodiment, a raw visible light image of a tobacco leaf is acquired, and a brightness compensation matrix is ​​constructed based on a brightness offset parameter. Adaptive brightness correction is then performed on the bright and shadow areas in the image to generate a corrected tobacco leaf image. The corrected tobacco leaf image is then converted to the Lab color space, and the L channel brightness value and the color distribution data of the a and b channels are extracted. Leaf color parameters are determined based on the green-to-red color change in the a channel and the blue-to-yellow color change in the b channel. Next, a region growing algorithm is used to identify color anomaly regions, and yellowed regions are marked by combining the proportion of yellow hue, brightness uniformity, and the color change trend at the leaf edges. The area proportion of yellowed regions in the overall leaf is then calculated, and a leaf yellowing distribution map is constructed based on the distribution location of the yellowed regions.

[0028] In another embodiment, assuming the acquired tobacco leaf image resolution is 4096×3072 pixels and the brightness compensation coefficient ranges from 0.82 to 1.15; after image correction, the average leaf brightness increases from the original 168 to 192, and the brightness fluctuation in the shadow area is reduced from... Reduce to Extract the Lab color parameters, where the average value of the a channel is determined by... Change to The average value of the b channel increased from 18 to 31. After color thresholding, a total of 12 yellowing regions were identified, with the yellowing area at the leaf tip accounting for 18%, the yellowing area at the leaf margin accounting for 11%, and the yellowing area at the leaf base accounting for 4%. A leaf yellowing distribution map was generated based on the spatial distribution results of the yellowing regions.

[0029] Leaf vein texture is identified based on the corrected tobacco leaf image; leaf edge curling is detected based on leaf vein texture; leaf aging level is assessed based on leaf edge curling and yellowing area; leaf aging characteristics are determined based on leaf aging level.

[0030] In one embodiment, multi-scale texture enhancement processing is performed on the corrected tobacco leaf image. The Frangi filtering algorithm is used to enhance the vein structure, and adaptive edge detection is used to identify the distribution areas of the main vein and lateral veins. Then, the central axis of the leaf is constructed based on the extension direction of the main vein, and the leaf edge contour is extracted based on the arrangement direction of the lateral veins. Next, the curvature change value and edge offset distance of the leaf edge contour at different positions are calculated to detect the leaf edge curling degree. Then, spatial overlap analysis is performed on the curled area and the yellowed area, and the leaf aging level is assessed based on the area ratio of the curled area, the degree of yellowing diffusion, and the continuity of the vein texture. When the leaf edge curling degree increases and the yellowed area expands towards the main vein, the leaf is determined to have entered a high aging stage, and the corresponding leaf aging characteristics are output according to the aging level.

[0031] In another embodiment, it is assumed that one main vein and 46 lateral veins are identified, with an average vein spacing of 3.8 mm. The edge curvature is calculated to range from 0.12 to 0.67, with the maximum curvature at the leaf tip reaching 0.71. Edge offset distances are calculated, with an average offset distance of 4.2 mm on the left side and 5.1 mm on the right. Subsequently, it is detected that the curled area accounts for 22% of the total leaf area, with a 16% overlap with the yellowed area. An aging scoring model is established based on the area of ​​the curled area, the proportion of yellowing diffusion, and the continuity of the vein texture. The leaf aging score is 0.78 (out of 1.0), indicating that the tobacco leaf is in a moderate to severe aging stage.

[0032] Preferably, identifying leaf vein texture based on a corrected tobacco leaf image includes: Identify leaf vein structure based on corrected tobacco leaf image; determine the main vein extension direction based on leaf vein structure; mark the lateral vein growth direction based on the main vein extension direction; calculate the vein branch angles in the main vein extension direction and the lateral vein growth direction. In one embodiment, grayscale enhancement and vein texture enhancement processing are performed on the corrected tobacco leaf image. A multi-scale Gabor filter is used to extract vein edge information, and a vein skeleton map is constructed using a morphological thinning algorithm. Subsequently, the main vein structure is identified based on the vein skeleton map, and the extension direction of the main vein is determined using Hough linear transform. Next, using the main vein as the reference axis, the lateral veins on both sides are segmented, and their growth directions are marked. Then, the angle between the growth direction of the lateral veins and the extension direction of the main vein is calculated, and this angle is used as the vein branching angle. Statistical analysis is performed on the vein branching angles in different regions to establish a tobacco leaf vein direction distribution model.

[0033] In another embodiment, assuming the acquired tobacco leaf image resolution is 3648×2736 pixels, after texture enhancement, a total of 1 main vein, 38 primary lateral veins, and 74 secondary lateral veins are identified. The angle of the main vein's extension direction is calculated using Hough linear transform. The average growth angle of the left-side primary lateral vein is... The average growth angle of the first-order lateral vein on the right side is Calculate the vein branching angle between the main vein and lateral veins, where the average branching angle in the leaf base region is... The average branching angle in the leaf region is The average branching angle in the leaf tip region is .

[0034] Calculate the leaf vein distribution interval parameters based on the leaf vein structure; determine the leaf vein arrangement density based on the leaf vein distribution interval parameters; determine the leaf vein texture based on the leaf vein branching angle and leaf vein arrangement density.

[0035] In one embodiment, based on the identified main vein and lateral vein structures, the spatial distance between adjacent lateral veins is extracted, and a vein distribution interval parameter is established according to the vein spacing in different regions. Subsequently, mean calculation and fluctuation analysis are performed on the vein spacing in different regions of the leaf to assess the uniformity of vein arrangement. Next, the vein arrangement density is calculated based on the number of veins per unit area, and a vein texture feature matrix is ​​established in conjunction with the vein branching angle. When the vein arrangement density is high and the vein branching angle is uniformly distributed, the vein texture is considered relatively dense; when the vein arrangement density decreases and the dispersion of the vein branching angle increases, the vein texture is considered relatively loose. Finally, the corresponding vein texture type is output based on the vein texture feature matrix.

[0036] In another embodiment, it is assumed that the spacing between adjacent lateral veins in the middle region of the leaf ranges from 2.6 mm to 5.8 mm, with an average spacing of 3.9 mm; the average spacing in the leaf tip region is 2.4 mm, and the average spacing in the leaf base region is 5.1 mm. The vein arrangement density is calculated using a 10 mm × 10 mm area as the statistical unit, with 18 veins in the middle region, 24 veins in the leaf tip region, and 13 veins in the leaf base region. Subsequently, combined with the vein branching angle analysis results, the vein angle dispersion in the middle region of the leaf is obtained as follows: The dispersion of the leaf tip region is The dispersion of the leaf base region is Based on the leaf vein arrangement density and leaf vein branching angle, a leaf vein texture feature matrix was constructed, and the leaf vein texture of the tobacco leaf was determined to be a "high-density regular texture". This leaf vein texture was then used as an important structural feature for identifying tobacco varieties.

[0037] Preferably, detecting leaf edge curling based on leaf vein texture includes: Extract main vein data and lateral vein distribution data based on leaf vein texture; establish the leaf center extension direction based on main vein data; identify leaf edge contour based on lateral vein distribution data; calculate curvature change value based on leaf edge contour. In one embodiment, the vein texture in a tobacco leaf image is enhanced by employing a directional gradient filtering algorithm to extract the structural information of the main vein and lateral veins, and then generating a vein structure map through skeletonization. Subsequently, the main vein data is extracted based on its continuous direction, and the leaf center extension direction is established using the main vein centerline. Next, based on the distribution trend of the lateral veins towards the leaf edge, contour tracking is performed on the leaf edge region to identify the leaf edge contour. Then, the rate of change of the edge tangent direction is calculated point-by-point along the leaf edge contour, and curvature change values ​​are established based on the directional changes between adjacent edge points. The curvature change values ​​of different regions are statistically analyzed to identify the local bending trend of the leaf edge.

[0038] In another embodiment, assuming the acquired tobacco leaf image resolution is 4000×3000 pixels, after vein texture enhancement, one main vein, 42 primary lateral veins, and 86 secondary lateral veins are identified. The leaf's central extension direction is established based on the main vein's centerline, with an extension length of 168.5 mm. The leaf edge contour is identified based on the lateral vein distribution, extracting a total of 2450 edge contour points. Subsequently, local curvature changes are calculated in groups of 10 edge points, with an average curvature of 0.18 in the central region, 0.64 in the tip region, and a maximum curvature of 0.79 on the left edge. A leaf edge bending trend map is constructed based on the curvature change results and used for subsequent leaf curling region identification and analysis.

[0039] Identify the curved area of ​​the blade edge based on the curvature change value; calculate the edge offset distance in the direction extending from the blade center based on the blade edge profile; determine the blade edge curl based on the curved area of ​​the blade edge and the edge offset distance.

[0040] In one embodiment, a curvature threshold range is established based on the curvature change value in the blade edge profile. Regions with curvature change values ​​higher than a preset curvature threshold are identified as blade edge curvature regions. Then, using the blade center extension direction as a reference axis, the vertical distance between the blade edge profile and the reference axis is calculated, and this distance is used as the edge offset distance. Next, a symmetry analysis is performed on the edge offset distances on the left and right sides of the blade to assess the degree of blade edge deformation. When the area of ​​the edge curvature region increases and the edge offset distance continues to rise, the degree of blade curling is determined to be enhanced. The distribution of the edge curvature region and the edge offset distance are weighted and fused to generate the blade edge curl degree.

[0041] In another embodiment, assuming a curvature bending threshold of 0.52, nine edge bending regions were identified out of 2450 edge contour points. The bending area at the leaf tip accounted for 14%, and the bending area on the left side accounted for 11%. Edge offset distances were calculated using the main vein centerline as the reference axis, yielding an average offset distance of 4.8 mm for the left edge, 5.3 mm for the right edge, and a maximum offset distance of 9.6 mm at the leaf tip. The edge offset distances were then normalized, and a curling degree calculation model was established based on the edge bending region area, resulting in a leaf edge curling degree of 0.73 (out of 1.0). Based on the curling degree level, the tobacco leaf was determined to have significant edge curling, and this curling characteristic was used as an important morphological basis for leaf aging analysis.

[0042] Preferably, in step S3, spectral reflectance characteristic analysis is performed based on near-infrared spectral data, and the reflectance parameters of the characteristic bands are recorded, including: Chlorophyll-sensitive and moisture absorption-sensitive bands were extracted from near-infrared spectral data; chlorophyll content was assessed in the chlorophyll-sensitive bands; and moisture content of tobacco leaves was assessed in the moisture absorption-sensitive bands. In one embodiment, a near-infrared spectral acquisition device is used to scan the tobacco leaves, acquiring near-infrared spectral data in the range of 900–1700 nm. The raw spectra are then smoothed and baseline corrected to reduce noise interference. Subsequently, chlorophyll-sensitive bands are extracted in the range of 680–750 nm, and chlorophyll content is assessed based on the position of the chlorophyll absorption edge and the trend of red-edge reflectance changes. Simultaneously, moisture absorption-sensitive bands are extracted near 970 nm, 1200 nm, and 1450 nm, and the moisture content of the tobacco leaves is assessed based on the depth of the moisture absorption valley and the degree of reflectance attenuation.

[0043] In another embodiment, it is assumed that near-infrared spectroscopy was performed on 20 tobacco leaves, yielding a total of 400 sets of spectral data. A distinct chlorophyll absorption edge was detected near 705 nm. The red edge of mature tobacco leaves was concentrated between 712 and 718 nm, with an average reflectance of 0.42; the red edge of immature tobacco leaves was concentrated between 700 and 706 nm, with an average reflectance of 0.31. Simultaneously, in the 1450 nm moisture absorption band, the average absorption valley depth of mature tobacco leaves was 0.18, while that of immature tobacco leaves was 0.27. The chlorophyll content and moisture content of the tobacco leaves were obtained based on the changes in the chlorophyll absorption edge and the moisture absorption valley.

[0044] The maturity of tobacco leaves is determined based on chlorophyll content and moisture content; a standard spectral reflectance curve is matched based on the maturity of the tobacco leaves; and the reflectance parameters of characteristic bands are recorded based on the standard spectral reflectance curve.

[0045] In one embodiment, a tobacco leaf maturity evaluation model is established based on chlorophyll content and tobacco leaf moisture content. When chlorophyll content decreases and moisture content remains stable, the tobacco leaf is determined to be in the mature stage; when chlorophyll content is high and water absorption is strong, the tobacco leaf is determined to be in the immature stage. Subsequently, corresponding standard spectral reflectance curves are matched according to different maturity stages, and corresponding reflectance characteristics are recorded in the chlorophyll-sensitive band and the water absorption-sensitive band to generate characteristic band reflectance parameters.

[0046] In another embodiment, assuming that based on chlorophyll content and tobacco leaf moisture content, 20 tobacco leaves are divided into 8 mature leaves, 7 semi-mature leaves, and 5 immature leaves. The standard spectral reflectance curves for the mature stage form a stable reflectance plateau around 720 nm, with a reflectance range of 0.40–0.46; the semi-mature stage corresponds to a reflectance range of 0.33–0.39; and the immature stage corresponds to a reflectance range of 0.25–0.32. Simultaneously, in the 1450 nm moisture absorption band, the average reflectance is 0.18 for the mature stage, 0.23 for the semi-mature stage, and 0.29 for the immature stage. Characteristic band reflectance parameters are recorded based on the standard spectral reflectance curves corresponding to each maturity stage.

[0047] Preferably, assessing chlorophyll content in the chlorophyll-sensitive wavelength range includes: Evaluate chlorophyll absorption-side migration status in chlorophyll-sensitive bands; assess chlorophyll synthesis activity based on chlorophyll absorption-side migration status; identify chlorophyll-rich regions based on chlorophyll synthesis activity; In one embodiment, the near-infrared spectral data of tobacco leaves is preprocessed, including noise reduction, smoothing filtering, and reflectance normalization, to improve the identification accuracy of chlorophyll-sensitive bands. Subsequently, chlorophyll-sensitive bands are extracted in the 680–750 nm range, and the chlorophyll absorption edge position is determined based on the reflectance change curve. Then, the migration state of the chlorophyll absorption edge towards longer or shorter wavelengths is detected to assess changes in chlorophyll absorption capacity. When the chlorophyll absorption edge shifts towards longer wavelengths, it is determined that chlorophyll synthesis activity is enhanced; when the chlorophyll absorption edge shifts towards shorter wavelengths, it is determined that chlorophyll synthesis activity is weakened. Next, chlorophyll-rich regions are identified based on the spatial distribution of chlorophyll synthesis activity within the leaf area, and the reflectance change characteristics corresponding to the chlorophyll-rich regions are recorded.

[0048] In another embodiment, assuming near-infrared spectroscopy analysis is performed on 30 tobacco leaves, the changes in chlorophyll absorption edge are detected in the range of 705–725 nm. Among these, the chlorophyll absorption edge of 18 leaves shifted from 708 nm to 716 nm, with an average shift of 8 nm, corresponding to a reflectance increase range of 0.36–0.44, which was identified as a region of enhanced chlorophyll synthesis activity; the chlorophyll absorption edge of the remaining 12 leaves fell back from 712 nm to 704 nm, with an average shift of [missing information]. At 8 nm, the reflectance decreased to a range of 0.29–0.34, which was identified as a region of reduced chlorophyll synthesis activity. Based on the changes in reflectance across leaf regions, 25 chlorophyll-rich regions were identified, with the average reflectance of the regions on both sides of the main vein reaching 0.46 and the average reflectance of the edge regions reaching 0.38, thus completing the identification of chlorophyll-rich regions.

[0049] The chlorophyll degradation trend is assessed based on the chlorophyll absorption and migration status; chlorophyll attenuation regions are identified based on the chlorophyll degradation trend; chlorophyll enrichment content is calculated based on the chlorophyll enrichment regions; chlorophyll attenuation content is calculated based on the chlorophyll attenuation regions; and the chlorophyll enrichment content and chlorophyll attenuation content are weighted and calculated to record the chlorophyll content.

[0050] In one embodiment, the chlorophyll reflectance curve is analyzed based on the migration state of the chlorophyll absorption edge. When the chlorophyll absorption edge continuously moves towards shorter wavelengths and the reflectance decreases, it is determined that the chlorophyll has entered the degradation stage. Subsequently, chlorophyll degradation regions are identified based on the reflectance attenuation amplitude in different areas, and the chlorophyll attenuation content is calculated by combining the region area and the degree of reflectance change. At the same time, the chlorophyll enrichment content is calculated based on the reflectance enhancement amplitude and regional distribution range in the chlorophyll enrichment regions. Finally, the chlorophyll enrichment content and chlorophyll attenuation content are weighted according to preset weights to generate the overall chlorophyll content of the tobacco leaf.

[0051] In another embodiment, it is assumed that 14 out of 30 tobacco leaves showed a significant chlorophyll degradation trend, with the chlorophyll absorption edge decreasing from 710 nm to 702 nm and the average reflectance decreasing from 0.41 to 0.30. Thirty-two chlorophyll attenuation regions were identified, with an average attenuation rate of 22% at the leaf tip and 18% at the leaf margin. Simultaneously, in the chlorophyll-rich regions, the average reflectance near the main vein increased to 0.48, resulting in an average chlorophyll enrichment content of 0.63; in the chlorophyll attenuation regions, the average chlorophyll attenuation content was 0.27. Subsequently, the chlorophyll enrichment content and chlorophyll attenuation content were weighted at a ratio of 0.7 and 0.3, respectively, yielding an overall average chlorophyll content of 0.52.

[0052] Preferably, assessing chlorophyll degradation trends based on chlorophyll absorption edge migration status includes: The wavelength shift direction of chlorophyll absorption edge is extracted based on the migration state of chlorophyll absorption edge; the chlorophyll absorption value is evaluated based on the wavelength shift direction; and the chlorophyll decay stage is identified based on the chlorophyll absorption value. In one embodiment, spectral reflectance data of tobacco leaves in the 680nm–760nm wavelength range are acquired using a near-infrared hyperspectral imaging device, and dark current correction and Savitzky-Golay smoothing filtering are performed on the raw spectrum. Subsequently, the chlorophyll absorption edge position is determined by calculating the first derivative of the spectrum. The changing trend of the absorption edge center wavelength was analyzed based on continuous time-series data acquisition. A shift in the absorption edge from long-wavelength to short-wavelength direction was defined as a "blue shift," and vice versa as a "red shift." Chlorophyll absorbance was calculated based on the wavelength shift direction. Specifically, the ratio of the difference between the absorption peak near 705 nm and the reference reflection peak at 750 nm is statistically analyzed to characterize the chlorophyll's ability to absorb spectral energy. Then, the chlorophyll decay stage is identified based on the rate of change in chlorophyll absorption value: when the chlorophyll absorption value continuously decreases and the rate of decrease exceeds a preset threshold, the tobacco leaf is determined to have entered the chlorophyll decay stage; when the absorption value tends to stabilize, it is determined to be in the stable maturation stage.

[0053] In another embodiment, it is assumed that 20 tobacco leaves were subjected to continuous hyperspectral detection for 5 days, with a detection wavelength range of 680nm to 760nm and a spectral resolution of 1nm. Calculations showed that the chlorophyll absorption edge of 12 leaves gradually shifted from an initial 724nm to 715nm, forming a clear blue shift trend; the absorption edge of the remaining 8 leaves fluctuated around 720nm. The chlorophyll absorbance values ​​were calculated, ranging from 0.42 to 0.83, with 7 leaves having absorbance values ​​below 0.50. A chlorophyll attenuation threshold of 0.55 was set; when two consecutive detection results were below 0.55, the chlorophyll attenuation stage was determined. Five leaves were identified as being in the initial attenuation stage, four in the moderate attenuation stage, and three in the severe attenuation stage.

[0054] The chlorophyll degradation rate was calculated during the chlorophyll decay phase; the chlorophyll loss range was marked during the chlorophyll decay phase; and the chlorophyll degradation trend was assessed based on the chlorophyll loss range and the chlorophyll degradation rate.

[0055] In one embodiment, after identifying the chlorophyll decay stage, the chlorophyll decomposition rate is calculated based on chlorophyll absorbance values ​​from a continuous time series. Specifically, the chlorophyll absorbance values ​​at adjacent time points are... , Differential calculations are performed to obtain the absorption attenuation rate per unit time. The chlorophyll decomposition rate was then used as the chlorophyll absorption rate. Subsequently, pixel-level spectral mapping was performed on tobacco leaf images, marking areas with chlorophyll absorbance values ​​below a preset attenuation threshold as chlorophyll loss regions. The chlorophyll loss range was then determined using region connectivity analysis. A degradation trend assessment model was constructed based on the proportion of chlorophyll loss area to the total leaf area and the chlorophyll decomposition rate. When the chlorophyll decomposition rate was high and the loss range continued to expand, the leaf was determined to be in a rapid degradation trend; when the decomposition rate was low and the loss range changed slowly, it was determined to be in a stable degradation trend.

[0056] In another embodiment, assuming analysis of spectral data of tobacco leaves over seven consecutive days, the chlorophyll absorbance of one leaf decreased from 0.78 to 0.41, with an average decomposition rate of 0.053 / day; another leaf decreased from 0.74 to 0.66, with an average decomposition rate of 0.011 / day. The leaf areas were marked for chlorophyll loss, and the calculated chlorophyll loss area of ​​the first leaf increased from 12% to 46%, and that of the second leaf increased from 8% to 15%. A chlorophyll decomposition rate threshold of 0.03 / day and a chlorophyll loss range threshold of 30% were set. The first leaf, with a decomposition rate greater than 0.03 / day and a loss range exceeding 30%, was determined to have a rapid degradation trend; the second leaf was determined to have a slow degradation trend. Furthermore, statistical results showed that there were 6 rapidly degrading leaves, 9 slowly degrading leaves, and 5 moderately degrading leaves.

[0057] Preferably, assessing the moisture content of tobacco leaves in the moisture absorption-sensitive wavelength range includes: Calculate the peak water absorption of leaves in the water absorption-sensitive band; assess the water retention status of leaves based on the peak water absorption of leaves; assess the degree of water loss from leaf tissues based on the water retention status of leaves. In one embodiment, near-infrared hyperspectral imaging equipment is used to acquire reflectance spectral data of tobacco leaves in the range of 900 nm to 1700 nm, and whiteboard correction and noise filtering are performed on the raw spectra. Subsequently, the leaf moisture absorption peaks are extracted in the moisture absorption sensitive bands (such as around 970 nm, 1200 nm, and 1450 nm), and the leaf moisture absorption parameters are obtained by calculating the difference between the absorption peak depth and the baseline reflectance. The leaf's water retention status is assessed based on the changing trends of absorption peaks across multiple water-sensitive wavelengths. A high and relatively stable water absorption peak indicates good water retention, while a continuously decreasing absorption peak indicates weakened water retention. Subsequently, the degree of water loss in the leaf tissue is analyzed based on the water retention status. Specifically, the degree of water loss—mild, moderate, or severe—is determined by statistically analyzing the decrease in water absorption parameters per unit time and the changes in reflectance in local leaf regions.

[0058] In another embodiment, it is assumed that near-infrared spectroscopy was performed on 25 tobacco leaves for 6 consecutive days, with a detection wavelength range of 900nm to 1700nm and a spectral resolution of 2nm. Calculations showed that the average water absorption peak value at 1450nm decreased from 0.82 to 0.54 for 15 leaves, while the absorption peak value remained above 0.70 for the other 10 leaves. Setting a water retention threshold of 0.60, a decrease in water retention capacity was determined when the leaf water absorption peak value was below 0.60. The absorption peak decline rate per unit time was statistically analyzed. Eight leaves showed a decline rate greater than 0.05 / day and were identified as moderately dehydrated; five leaves showed a decline rate greater than 0.08 / day and were identified as severely dehydrated; the remaining 12 leaves were identified as slightly dehydrated.

[0059] Dehydrated areas of leaves are marked based on the degree of water loss in leaf tissue; the moisture content of tobacco leaves is calculated in the dehydrated areas.

[0060] In one embodiment, after assessing the degree of water loss in leaf tissue, pixel-level water loss regions are identified in tobacco leaves by combining spatial distribution information from hyperspectral images. Specifically, pixel regions with water absorption parameters below a preset water loss threshold are marked as dehydrated regions, and isolated noise points are removed using region connectivity analysis to obtain continuous dehydrated leaf regions. Subsequently, near-infrared reflectance data of corresponding pixels within the dehydrated regions are extracted, and the moisture content of the tobacco leaves is calculated using a pre-established moisture content regression model. The regression model can be established using partial least squares regression (PLSR) or support vector regression (SVR) to achieve a mapping relationship between spectral reflectance and the actual moisture content of the leaves. The moisture content of each dehydrated region is statistically averaged to obtain the overall moisture state of the tobacco leaves, providing a quantitative basis for analyzing the degree of tobacco leaf dryness.

[0061] In another embodiment, it is assumed that dehydration areas were detected using hyperspectral images of 20 tobacco leaves, with a dehydration threshold set at a water absorption parameter below 0.48. The detection results showed that 11 leaves had obvious dehydration areas, with the dehydration area accounting for 18%–52% of the total leaf area; the remaining 9 leaves had dehydration areas below 10%. Based on the PLSR model, the water content of the dehydration areas was calculated, yielding the average water content for different leaf types: severely dehydrated areas 18.5%–24.3%, moderately dehydrated areas 25.6%–31.2%, and slightly dehydrated areas 32.5%–38.7%. Three leaves, with a dehydration area exceeding 45% and an average water content below 22%, were identified as severely dehydrated leaves.

[0062] Preferably, this specification also provides a tobacco variety identification system based on the fusion of visible light images and near-infrared spectra, used to perform the tobacco variety identification method based on the fusion of visible light images and near-infrared spectra as described above. This tobacco variety identification system based on the fusion of visible light images and near-infrared spectra includes: The data acquisition module 101 is used to acquire visible light images and near-infrared spectral data of tobacco leaves; perform illumination interference analysis based on the visible light images of tobacco leaves, and record brightness shift parameters. The leaf aging detection module 102 is used to generate brightness compensation data based on brightness offset parameters; apply the brightness compensation data to the visible light image of tobacco leaves; and detect leaf aging characteristics. Spectral analysis module 103 is used to perform spectral reflectance characteristic analysis based on near-infrared spectral data, record reflectance parameters of characteristic bands, and determine the spectral characteristics of tobacco based on the reflectance parameters of characteristic bands. The variety identification module 104 is used to perform feature fusion of leaf aging characteristics and tobacco spectral characteristics to generate fusion identification results; and to determine the tobacco variety based on the fusion identification results.

[0063] It should be noted that you should refer to [link / reference]. Figure 3 The figure illustrates the process of tobacco spectral feature extraction based on near-infrared spectroscopy. Near-infrared spectral curves in the range of 900-1700nm are obtained, and chlorophyll absorption, water absorption, and structural information reflectance features are extracted in characteristic bands such as 950nm, 1200nm, 1450nm, and 1650nm. Then, parameters such as reflectance, absorption depth, area, and slope of each characteristic band are analyzed and summarized to form a multi-band fused feature vector, generating tobacco spectral features for tobacco variety identification and analysis.

[0064] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0065] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for identifying tobacco varieties based on the fusion of visible light images and near-infrared spectra, characterized in that, Includes the following steps: Step S1: Acquire visible light images and near-infrared spectral data of tobacco leaves; Perform illumination interference analysis based on visible light images of tobacco leaves and record brightness shift parameters; Step S2: Generate brightness compensation data based on brightness offset parameters; apply the brightness compensation data to the visible light image of tobacco leaves and detect leaf aging characteristics; Step S3: Perform spectral reflectance characteristic analysis based on near-infrared spectral data and record the reflectance parameters of the characteristic bands; Determining the spectral characteristics of tobacco based on characteristic band reflectance parameters; Step S4: Perform feature fusion of leaf aging characteristics and tobacco spectral characteristics to generate fusion recognition results; determine the tobacco variety based on the fusion recognition results.

2. The tobacco variety identification method based on visible light image and near-infrared spectroscopy fusion according to claim 1, characterized in that, In step S1, illumination interference analysis is performed based on the visible light image of the tobacco leaf, and the brightness shift parameters are recorded, including: Extract leaf brightness data from visible light images of tobacco leaves; determine pixel grayscale values ​​based on leaf brightness data; identify areas with pixel grayscale values ​​higher than a preset high-brightness threshold as high-brightness areas; Regions with pixel grayscale values ​​below a preset shadow threshold are identified as shadow regions; the brightness shift of the highlight and shadow regions is calculated; and the brightness shift is used as a brightness shift parameter.

3. The tobacco variety identification method based on the fusion of visible light images and near-infrared spectra according to claim 1, characterized in that, In step S2, the brightness compensation data is applied to the visible light image of the tobacco leaves, and leaf aging characteristics are detected, including: Brightness compensation data is applied to visible light images of tobacco leaves to generate corrected tobacco leaf images; leaf color parameters are determined based on the corrected tobacco leaf images; and yellowed areas are marked based on the leaf color parameters. Leaf vein texture is identified based on calibrated tobacco leaf images; leaf edge curling is detected based on leaf vein texture; leaf aging level is assessed based on leaf edge curling and yellowing areas; and leaf aging characteristics are determined based on leaf aging level.

4. The tobacco variety identification method based on the fusion of visible light images and near-infrared spectra according to claim 3, characterized in that, Based on the corrected tobacco leaf image, leaf vein texture identification includes: Identify leaf vein structure based on corrected tobacco leaf images; determine the main vein extension direction based on the leaf vein structure; mark the lateral vein growth direction based on the main vein extension direction; calculate the vein branch angles in the main vein extension direction and the lateral vein growth direction. Calculate the leaf vein distribution interval parameters based on the leaf vein structure; determine the leaf vein arrangement density based on the leaf vein distribution interval parameters; determine the leaf vein texture based on the leaf vein branching angle and leaf vein arrangement density.

5. The tobacco variety identification method based on visible light image and near-infrared spectroscopy fusion according to claim 3, characterized in that, Detecting leaf edge curling based on leaf vein texture includes: Extract main vein data and lateral vein distribution data based on leaf vein texture; establish the leaf center extension direction based on main vein data; identify leaf edge contour based on lateral vein distribution data; calculate curvature change value based on leaf edge contour. Identify the curved area of ​​the blade edge based on the curvature change value; calculate the edge offset distance in the direction extending from the blade center based on the blade edge profile; determine the blade edge curl based on the curved area of ​​the blade edge and the edge offset distance.

6. The tobacco variety identification method based on visible light image and near-infrared spectroscopy fusion according to claim 1, characterized in that, In step S3, spectral reflectance characteristic analysis is performed based on near-infrared spectral data, and the reflectance parameters of the characteristic bands are recorded, including: Chlorophyll-sensitive and moisture absorption-sensitive bands were extracted from near-infrared spectral data; chlorophyll content was assessed in the chlorophyll-sensitive bands; and moisture content of tobacco leaves was assessed in the moisture absorption-sensitive bands. The maturity of tobacco leaves is determined based on chlorophyll content and moisture content; a standard spectral reflectance curve is matched based on the maturity of the tobacco leaves; and the reflectance parameters of characteristic bands are recorded based on the standard spectral reflectance curve.

7. The tobacco variety identification method based on the fusion of visible light images and near-infrared spectra according to claim 6, characterized in that, Assessing chlorophyll content in chlorophyll-sensitive wavelengths includes: Evaluate chlorophyll absorption-side migration status in chlorophyll-sensitive bands; assess chlorophyll synthesis activity based on chlorophyll absorption-side migration status; identify chlorophyll-rich regions based on chlorophyll synthesis activity; The chlorophyll degradation trend is assessed based on the chlorophyll absorption and migration status; chlorophyll attenuation regions are identified based on the chlorophyll degradation trend; chlorophyll enrichment content is calculated based on the chlorophyll enrichment regions; chlorophyll attenuation content is calculated based on the chlorophyll attenuation regions; and the chlorophyll enrichment content and chlorophyll attenuation content are weighted and calculated to record the chlorophyll content.

8. The tobacco variety identification method based on the fusion of visible light images and near-infrared spectra according to claim 7, characterized in that, Assessing chlorophyll degradation trends based on chlorophyll uptake and migration status includes: The wavelength shift direction of chlorophyll absorption edge is extracted based on the migration state of chlorophyll absorption edge; the chlorophyll absorption value is evaluated based on the wavelength shift direction; and the chlorophyll decay stage is identified based on the chlorophyll absorption value. The chlorophyll degradation rate was calculated during the chlorophyll decay phase; the chlorophyll loss range was marked during the chlorophyll decay phase; and the chlorophyll degradation trend was assessed based on the chlorophyll loss range and the chlorophyll degradation rate.

9. The tobacco variety identification method based on the fusion of visible light images and near-infrared spectra according to claim 6, characterized in that, Assessing the moisture content of tobacco leaves using moisture absorption-sensitive wavelengths includes: Calculate the peak water absorption of leaves in the water absorption-sensitive band; assess the water retention status of leaves based on the peak water absorption of leaves; assess the degree of water loss from leaf tissues based on the water retention status of leaves. Dehydrated areas of leaves are marked based on the degree of water loss in leaf tissue; the moisture content of tobacco leaves is calculated in the dehydrated areas.

10. A tobacco variety identification system based on the fusion of visible light images and near-infrared spectra, characterized in that, For performing the tobacco variety identification method based on visible light image and near-infrared spectral fusion as described in claim 1, the tobacco variety identification system based on visible light image and near-infrared spectral fusion includes: The data acquisition module is used to acquire visible light images and near-infrared spectral data of tobacco leaves; it performs light interference analysis based on the visible light images of tobacco leaves and records brightness shift parameters. The leaf aging detection module is used to generate brightness compensation data based on brightness offset parameters; apply the brightness compensation data to visible light images of tobacco leaves, and detect leaf aging characteristics; The spectral analysis module is used to perform spectral reflectance characteristic analysis based on near-infrared spectral data, record reflectance parameters of characteristic bands, and determine the spectral characteristics of tobacco based on the reflectance parameters of characteristic bands. The variety identification module is used to fuse leaf aging characteristics and tobacco spectral characteristics to generate a fused identification result; the tobacco variety is determined based on the fused identification result.

Images