Image processing method and device based on artificial intelligence

Abstract

The invention relates to the technical field of image processing, and particularly discloses an image processing method and device based on artificial intelligence, and the method comprises the steps of super-resolution reconstruction, reconstruction effect judgment and the like. According to the method, super-resolution reconstruction is performed through a bilinear interpolation method, an elliptical area model based on a peak-to-noise ratio and a global similarity evaluation coefficient is constructed to dynamically evaluate a reconstruction effect, iterative optimization is performed through adaptive adjustment of scaling factor parameters, the balance between image resolution improvement and information fidelity is ensured, and after the reconstruction reaches the standard, the reconstruction efficiency is improved. A multi-dimensional feature fusion technology is adopted, a fit index is calculated in combination with form, texture and color feature indexes to realize accurate classification of crops, and the growth health state of the crops is evaluated through an elliptic cylinder volume model of a normalized vegetation index, an enhanced vegetation index and a photochemical reflection index. The accuracy and scientificity of remote sensing image recognition analysis are effectively improved, and the timeliness and reliability of agricultural remote sensing monitoring are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an image processing method and apparatus based on artificial intelligence. Background Technology

[0002] With the development of agricultural modernization, the concept of precision agriculture has gradually gained popularity. Its core lies in using advanced technologies to precisely monitor and manage the growth environment and conditions of crops, thereby improving agricultural production efficiency, reducing resource consumption, and increasing crop yields. In this process, processing and analyzing remote sensing images of the crop research area has become an important means of obtaining relevant crop information.

[0003] Traditional remote sensing images of crop research areas often suffer from low resolution due to factors such as sensor performance, shooting distance, and atmospheric conditions. This makes them insufficient for accurate identification and analysis of crop species and growth status. While some simple image magnification methods can increase image size to some extent, they lead to blurring and loss of detail, failing to truly improve image resolution and quality.

[0004] Traditional methods are often insufficiently comprehensive and accurate in evaluating the effectiveness of super-resolution reconstruction. A single evaluation metric cannot comprehensively consider the similarity of images in multiple aspects such as signal intensity, noise level, brightness, contrast, and structure, and therefore cannot accurately determine whether the reconstructed image meets the requirements for subsequent analysis.

[0005] In the field of crop object identification, previous methods have relied on single feature information, such as identification based solely on the color or shape of the crop. This single feature identification method has significant limitations and is easily affected by environmental factors, resulting in low accuracy and reliability of identification.

[0006] In crop growth assessment, traditional methods typically rely on a single spectral characteristic indicator to measure crop growth status, failing to comprehensively reflect the crop's growth condition from multiple dimensions. For example, using only the normalized difference in vegetation index (NDVI) cannot fully consider factors such as photosynthetic activity and physiological state, leading to inaccurate and unscientific assessment results. Moreover, when problems arise in crop growth, traditional methods often fail to detect and provide timely warnings, making it difficult for agricultural practitioners to take prompt action and potentially resulting in yield reductions and other losses. Summary of the Invention

[0007] The purpose of this invention is to provide an image processing method and apparatus based on artificial intelligence to solve the problems mentioned in the background.

[0008] The objective of this invention can be achieved through the following technical solutions: In a first aspect, this invention provides an image processing method based on artificial intelligence, comprising: S1. Acquisition of regional remote sensing images: acquiring remote sensing images of a crop research area, performing preprocessing operations on the remote sensing images of the crop research area to obtain a preprocessed remote sensing image of the crop research area, and marking it as the original remote sensing image of the crop research area.

[0009] S2. Super-resolution reconstruction: The original remote sensing image of the crop study area is super-resolution reconstructed using the bilinear interpolation method to obtain a high-resolution target image after reconstruction.

[0010] S3. Reconstruction effect judgment: The high-resolution target image is cropped and adjusted according to the size of the original remote sensing image to obtain a high-resolution target image after size processing. This image is used as the evaluation image. The peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the evaluation image are analyzed to judge the super-resolution reconstruction effect.

[0011] S4. Scaling Factor Adjustment: When the super-resolution reconstruction effect is determined to be poor, the preset horizontal and vertical scaling factor adjustment parameters are analyzed, and S2 and S3 are re-executed based on the horizontal and vertical scaling factor adjustment parameters. If the secondary super-resolution reconstruction effect is good, S5 is executed; if the secondary super-resolution reconstruction effect is poor, an early warning feedback is issued.

[0012] S5. Crop object recognition: When the super-resolution reconstruction effect is determined to be good, the crop types in the image to be evaluated are identified to obtain the crop classification image of the crop study area.

[0013] S6. Crop growth assessment: Spectral features are extracted from the regional images corresponding to each crop type to obtain the normalized vegetation index, enhanced vegetation index, and photochemical reflectance index of each pixel in the regional image corresponding to each crop type. The growth status of each crop type is then assessed and analyzed, and an early warning is issued if the growth status is poor.

[0014] In a second aspect, the present invention provides an image processing device based on artificial intelligence, comprising: an image acquisition unit, a super-resolution reconstruction module, a reconstruction effect judgment module, a scaling factor adjustment module, an object recognition module, a growth evaluation module, and a memory.

[0015] The image acquisition device is used to acquire remote sensing images of the crop study area, perform preprocessing operations on the remote sensing images of the crop study area to obtain preprocessed remote sensing images of the crop study area, and mark them as the original remote sensing images of the crop study area.

[0016] The super-resolution reconstruction module is used to perform super-resolution reconstruction of the original remote sensing images of the crop research area based on the bilinear interpolation method, so as to obtain a high-resolution target image after reconstruction.

[0017] The reconstruction effect judgment module is used to crop and adjust the high-resolution target image according to the size of the original remote sensing image to obtain a high-resolution target image after size processing, which is used as the image to be evaluated. The module analyzes the peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the image to be evaluated, and judges the super-resolution reconstruction effect accordingly.

[0018] The scaling factor adjustment module is used to analyze the preset horizontal and vertical scaling factor adjustment parameters when the super-resolution reconstruction effect is determined to be poor. Based on the horizontal and vertical scaling factor adjustment parameters, the super-resolution reconstruction module and the reconstruction effect judgment module are re-executed. If the secondary super-resolution reconstruction effect is good, the object recognition module is executed. If the secondary super-resolution reconstruction effect is poor, an early warning feedback is sent to the early warning feedback module for early warning feedback.

[0019] The object recognition module is used to identify the types of crops in the image to be evaluated when the super-resolution reconstruction effect is deemed good, and to obtain a crop classification image of the crop study area.

[0020] The growth assessment module is used to extract spectral features from the regional images corresponding to each crop type, and obtain the normalized vegetation index, enhanced vegetation index and photochemical reflectance index of each pixel in the regional image corresponding to each crop type. Then, it assesses and analyzes the growth status of each crop type. If the growth status is poor, it sends an early warning feedback to the early warning feedback module.

[0021] The early warning feedback module is used to provide early warning feedback for poor secondary super-resolution reconstruction results and poor growth status of various crop types.

[0022] The memory is used to store the pixel values ​​of the original remote sensing images, the image feature information corresponding to each crop type, and the normalized vegetation index range, enhanced vegetation index range, and photochemical reflectance index range of pixels in the image corresponding to healthy crops.

[0023] The beneficial effects of this invention are:

[0024] This invention employs a bilinear interpolation method for super-resolution reconstruction, generating high-resolution target images from original remote sensing images based on a preset scaling factor. This provides more detailed information for remote sensing images of crop research areas, facilitating more accurate identification of crop types and growth status, and improving the monitoring and analysis capabilities for crops. Furthermore, by comprehensively analyzing the peak noise ratio and global similarity evaluation coefficient between the original remote sensing image and the image to be evaluated, and combining this with the calculation of reconstruction effect coefficients using ellipse construction, the super-resolution reconstruction effect can be judged more comprehensively and accurately, ensuring that the reconstructed image meets the quality requirements for subsequent identification and analysis.

[0025] This invention addresses the issue of unsatisfactory super-resolution reconstruction by adjusting the scaling factor based on the difference between the reconstruction effect coefficient and a set threshold. This adaptive adjustment mechanism continuously optimizes the super-resolution reconstruction process, gradually reducing the error between the reconstructed image and the original image. This results in a reconstructed image that is closer to the ideal state in terms of resolution and quality, laying a better foundation for subsequent tasks such as crop object recognition. It also avoids reconstruction failures due to inappropriate initial parameters, saving manpower and time costs and improving reconstruction effectiveness. Furthermore, the invention provides early warning feedback for cases where the super-resolution reconstruction is still unsatisfactory after a second attempt, promptly alerting relevant personnel to potential problems in the current image reconstruction and the need for further data checks or method adjustments. This prevents inaccurate subsequent analysis results due to erroneous or low-quality image data.

[0026] This invention divides the image to be evaluated into sub-images and comprehensively considers multi-dimensional feature information such as morphology, texture, and color, as well as various indicators, to meticulously depict the characteristics of crops. Then, it compares the sub-images with the feature indicators of various crop types, calculates multiple fitting factors and fit indices, and determines the crop type of the sub-image by the maximum fit index. This multi-feature fusion and quantitative comparison method greatly improves the accuracy and reliability of crop identification and avoids the limitations of single feature identification.

[0027] This invention comprehensively assesses crop growth status from multiple dimensions, including vegetation cover, photosynthetic activity, and physiological state, by extracting spectral features closely related to crop growth, such as the normalized difference in vegetation index, enhanced vegetation index, and photochemical reflectance index. Compared to single-indicator assessments, this approach is more scientific and comprehensive, and can more accurately reflect the true growth status of crops. Simultaneously, by utilizing the spectral index range of moderately healthy crops, it calculates the ideal reference index and ideal deviation index, and performs ratio analysis with the spectral index of actual crop pixels, quantifying the growth status assessment process. This quantitative comparison method makes the assessment results more objective, reduces subjective judgment errors, and facilitates comparison of growth status among different crop types. Early warning feedback is provided when growth status is poor, helping agricultural practitioners to identify problems early and take timely measures such as irrigation, fertilization, and pest and disease control, reducing the risk of crop yield reduction and ensuring agricultural production efficiency. Attached Figure Description

[0028] The invention will now be further described with reference to the accompanying drawings.

[0029] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0030] Figure 2 This is a diagram showing the connection of the device modules of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 As shown, a first aspect of the present invention provides an image processing method based on artificial intelligence, comprising:

[0033] S1. Acquisition of Regional Remote Sensing Images: Acquire remote sensing images of the crop study area and perform preprocessing operations (including radiometric correction, geometric correction, and image enhancement) on the remote sensing images of the crop study area to obtain preprocessed remote sensing images of the crop study area, and mark them as the original remote sensing images of the crop study area.

[0034] It should be noted that radiometric correction aims to eliminate radiometric errors in remote sensing images caused by sensor characteristics, atmospheric transmission, and other factors, ensuring that the image's grayscale values ​​accurately reflect the reflection or emission characteristics of ground objects. Geometric correction aims to correct geometric distortions in remote sensing images, matching the location, shape, and size of ground objects in the image to their actual geographic locations for accurate geolocation and spatial analysis. Image enhancement aims to improve the visual effect of the image, highlight useful information, suppress noise and interference, and facilitate subsequent image analysis and interpretation. Specifically, remote sensing images of the target crop research area can be processed using tools provided in ENVI or Erdas software.

[0035] S2. Super-resolution reconstruction: The original remote sensing image of the crop study area is super-resolution reconstructed using the bilinear interpolation method to obtain a high-resolution target image after reconstruction.

[0036] Specifically, the implementation process of super-resolution reconstruction based on the bilinear interpolation method is as follows:

[0037] 201: Determine the size of the original remote sensing image and the target image: Extract the image size corresponding to the original remote sensing image of the crop study area, and determine the expression for the image size through super-resolution reconstruction according to the preset horizontal and vertical scaling factors. The image size corresponding to the high-resolution target image after super-resolution reconstruction is identified from the formula, where s 水平 s 竖直 W represents the scaling factors in the horizontal and vertical directions, respectively. original H original W represents the width and height of the original remote sensing image, respectively. target H target These represent the width and height of the target image, respectively, within the corresponding image dimensions.

[0038] It should be noted that image size refers to the number of pixels contained in an image in the horizontal and vertical directions, usually expressed in the form of "width × height". For example, an image with a size of 1920×1080 means that it has 1920 pixels in the horizontal direction and 1080 pixels in the vertical direction.

[0039] 202: Establish coordinate mapping relationship: For each pixel point (x) in the high-resolution target image target ,y target (0≤x) target ≤W target ,0≤y target ≤H target ), which is mapped to the corresponding location (x) in the original remote sensing image through a scaling factor. original ,yoriginal The specific mapping calculation formula is as follows:

[0040] It should be noted that since the mapped coordinates may not be integers, interpolation is needed to determine the corresponding pixel values ​​in the original remote sensing image mapped by the scaling factor.

[0041] 203: Determine adjacent pixels: In the original remote sensing image, obtain four pixels adjacent to the corresponding positions mapped to the original remote sensing image, and denote the coordinates of these four pixels as (x1, y1), (x1, y2), (x2, y1), and (x2, y2), respectively. y2 = y1 + 1, This indicates a round-down operation.

[0042] 204: Bilinear Interpolation Calculation: Extract the pixel values ​​of these four pixels from the original remote sensing image stored in memory, denoted as f(x1,y1), f(x1,y2), f(x2,y1), and f(x2,y2) respectively. Then, calculate the values ​​using the bilinear interpolation formula f(x1,y1), f(x1,y2), and f(x2,y2). target ,y target )=(1-ε)(1-γ)f(x1,y1)+(1-ε)γf(x1,y2)+ε(1-γ)f(x2,y1)+εγf(x2,y2), to obtain the pixel value f(x) of the pixel in the high-resolution target image. target ,y target ), where ε=x original -x1,γ=y original -y1.

[0043] 205: Traverse all pixels of the target image: Perform calculations on each pixel in the high-resolution target image in the same manner as steps 202-204, thereby completing the interpolation calculation of the entire image and generating a high-resolution remote sensing image.

[0044] S3. Reconstruction effect judgment: The high-resolution target image is cropped and adjusted according to the size of the original remote sensing image to obtain a high-resolution target image after size processing. This image is used as the evaluation image. The peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the evaluation image are analyzed to judge the super-resolution reconstruction effect.

[0045] It should be noted that the high-resolution target image is cropped and adjusted to the size of the original remote sensing image. Through precise size calibration, the high-resolution target image is converted to a state consistent with the size of the original remote sensing image, generating a size-processed high-resolution target image. This image will be used as the evaluation image in subsequent evaluations. This process ensures that the reconstruction effect evaluation is carried out under the same scale benchmark, improving the accuracy and comparability of the evaluation.

[0046] Specifically, the process of analyzing the peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the image to be evaluated is as follows:

[0047] The original remote sensing image and the image to be evaluated are both normalized for pixel values. Pixels at the same coordinates in the original remote sensing image and the same coordinates in the image to be evaluated are extracted and their differences are calculated to obtain the pixel difference. At the same time, the pixel differences of all pixels are squared and these squared values ​​are accumulated to obtain the sum of squared pixel differences. The sum of squared pixel differences is divided by the total number of pixels in the image to obtain the mean square error.

[0048] It should be noted that pixel value normalization is performed simultaneously on the original remote sensing image and the image to be evaluated, so that the image pixel values ​​are unified to a specific standard range, eliminating the differences in pixel values ​​caused by different image acquisition devices or processing procedures.

[0049] It should be noted that mean squared error reflects the average degree of difference between the corresponding pixel values ​​of the original remote sensing image and the image to be evaluated. The smaller the mean squared error value, the closer the corresponding pixel values ​​of the original remote sensing image and the image to be evaluated are.

[0050] From the formula Calculate the peak noise ratio η between the original remote sensing image and the image to be evaluated. It is represented as mean square error, and max represents the maximum value of the dynamic range of image pixel values.

[0051] In a specific embodiment, since the original remote sensing image and the image to be evaluated have been normalized for pixel values, the dynamic range of the image pixel values ​​is [0,1], and the value of max is 1.

[0052] It should be noted that a higher peak signal-to-noise ratio (PSNR) means a smaller difference between the image to be evaluated and the original remote sensing image, a higher degree of matching between the image to be evaluated and the original remote sensing image in terms of signal strength and noise level, and a better super-resolution reconstruction effect.

[0053] The program slides across the original remote sensing image and the image to be evaluated row by row and column by column according to a preset local window size. Within each window position, the mean, variance, and covariance of the pixels in the original remote sensing image window and the image to be evaluated window are obtained. At the same time, the two windows at corresponding positions form a window group.

[0054] The variance of pixel values ​​between two windows in each window group is compared with a set variance threshold. If the variance of pixel values ​​between two windows in a window group is less than the set variance threshold, the window group is determined to be an invalid window group; otherwise, the window group is determined to be a valid window group. The number of valid window groups is obtained, and each valid window group is numbered in a preset order and labeled as 1, 2, ... i... n.

[0055] According to the formula Calculate the brightness similarity index between the original remote sensing image and the image to be evaluated corresponding to each effective window group. Contrast Similarity Index and structural similarity index R represents the original remote sensing image, and T represents the image to be evaluated. These represent the mean pixel value of the original remote sensing image and the mean pixel value of the image to be evaluated for each effective window group, respectively. These are represented as the pixel variance of the original remote sensing image and the pixel variance of the image to be evaluated corresponding to each effective window group, respectively. It is represented as the pixel covariance between the original remote sensing image and the image to be evaluated for each effective window group.

[0056] According to the formula Calculate the global similarity evaluation coefficient ψ between the original remote sensing image and the image to be evaluated. 全局 (R,T).

[0057] It should be noted that when calculating the global similarity evaluation coefficient, the variance of the pixel values ​​of the two windows within each window group is compared with a set variance threshold to filter out the effective window groups and remove the invalid window groups. This reduces the interference of invalid data on the evaluation results, making the evaluation results more accurately reflect the similarity between images and improving the accuracy and reliability of the evaluation.

[0058] It should be noted that the global similarity evaluation coefficient comprehensively reflects the degree of similarity of the entire image in terms of brightness, contrast, and structure. The closer the value is to 1, the higher the overall similarity between the reconstructed image and the original image, and the more ideal the super-resolution reconstruction effect.

[0059] Specifically, the process for evaluating the super-resolution reconstruction effect is as follows:

[0060] The peak noise ratio (PNR) and global similarity evaluation coefficient between the original remote sensing image and the image to be evaluated are converted into lengths according to a preset ratio, resulting in the lengths of the PNR and global similarity evaluation coefficients. Ellipses are then constructed with the length of the PNR as the major axis and the length of the global similarity evaluation coefficient as the minor axis. The area of ​​the ellipse is then extracted, and the numerical value of the ellipse area is marked as the reconstruction effect coefficient.

[0061] The reconstruction effect coefficient is compared with the set reconstruction effect coefficient threshold. If the reconstruction effect coefficient is greater than or equal to the set reconstruction effect coefficient threshold, the super-resolution reconstruction effect is judged to be good; otherwise, the super-resolution reconstruction effect is judged to be poor.

[0062] In one specific embodiment, the present invention employs a bilinear interpolation method for super-resolution reconstruction, which can generate a high-resolution target image from the original remote sensing image based on a preset scaling factor. For remote sensing images of crop research areas, this can obtain more detailed information, helping to more accurately identify crop types, growth status, etc., and improving the monitoring and analysis capabilities of crops. Furthermore, by comprehensively analyzing the peak noise ratio and global similarity evaluation coefficient between the original remote sensing image and the image to be evaluated, and combining this with the calculation of reconstruction effect coefficients using ellipse construction, the super-resolution reconstruction effect can be judged more comprehensively and accurately, ensuring that the reconstructed image meets the quality requirements for subsequent identification and analysis.

[0063] S4. Scaling Factor Adjustment: When the super-resolution reconstruction effect is determined to be poor, the preset horizontal and vertical scaling factor adjustment parameters are analyzed, and S2 and S3 are re-executed based on the horizontal and vertical scaling factor adjustment parameters. If the secondary super-resolution reconstruction effect is good, S5 is executed; if the secondary super-resolution reconstruction effect is poor, an early warning feedback is issued.

[0064] Specifically, the process of analyzing the preset horizontal and vertical scaling factor adjustment parameters is as follows:

[0065] When the super-resolution reconstruction effect is determined to be poor, the reconstruction effect coefficient is extracted and the difference between it and the set reconstruction effect coefficient threshold is calculated to obtain the reconstruction effect coefficient difference.

[0066] The reconstruction effect coefficient difference is matched with the reconstruction effect coefficient difference corresponding to each set of scaling factor adjustment parameters to obtain the scaling factor adjustment parameter set, which includes scaling factor adjustment parameters in the horizontal and vertical directions.

[0067] In one specific embodiment, the present invention adjusts the scaling factor in a targeted manner based on the difference between the reconstruction effect coefficient and a set threshold when the super-resolution reconstruction effect is poor. This adaptive adjustment mechanism can continuously optimize the super-resolution reconstruction process, gradually reduce the error between the reconstructed image and the original image, and make the reconstructed image closer to the ideal state in terms of resolution and quality. This lays a better foundation for subsequent work such as crop object recognition, avoids reconstruction failure due to inappropriate initial parameters, saves manpower and time costs, and improves reconstruction effect. Furthermore, it provides early warning feedback for cases where the effect is still poor after a second super-resolution reconstruction, promptly reminding relevant personnel that there are problems with the current image reconstruction and that further data checks or method adjustments may be necessary, avoiding inaccurate subsequent analysis results due to erroneous or low-quality image data.

[0068] S5. Crop object recognition: When the super-resolution reconstruction effect is determined to be good, the crop types in the image to be evaluated are identified to obtain the crop classification image of the crop study area.

[0069] Specifically, methods for identifying different crop types in the image to be evaluated include:

[0070] The image to be evaluated is divided into sub-images according to a set proportional size. Image feature information of each sub-image is obtained, including morphological, texture, and color feature information. Furthermore, the numerical values ​​of various indicators of the morphological, texture, and color feature information of each sub-image are obtained and denoted as follows: j represents the number of the j-th sub-image in the image to be evaluated, j = 1, 2, ..., m; a1 represents the number of the a1-th indicator of morphological feature information, a1 = 1, 2, ..., b1, where b1 is the total number of a1 indicator numbers; a2 represents the number of the a2-th indicator of texture feature information, a2 = 1, 2, ..., b2, where b2 is the total number of a2 indicator numbers; a3 represents the number of the a3-th indicator of color feature information, a3 = 1, 2, ..., b3, where b3 is the total number of a3 indicator numbers.

[0071] It should be noted that dividing the image to be evaluated into sub-images according to a set ratio and size reduces the amount of data processed per word, lowers the complexity and difficulty of calculation, which not only helps to improve the accuracy of crop object recognition and processing, but also effectively improves the processing speed and efficiency.

[0072] It should be noted that the morphological feature information is obtained by feature extraction based on the edge detection algorithm, the texture feature information is obtained by feature extraction based on the gray-level co-occurrence matrix, and the color feature information is obtained by feature extraction based on converting the image color space to HSV.

[0073] It should be noted that the various indicators of morphological characteristics include, but are not limited to, leaf length, plant height, canopy structure, and leaf area.

[0074] It should be noted that the various indicators of texture feature information include, but are not limited to, contrast, correlation, energy, and entropy.

[0075] It should be noted that the various indicators of color feature information include, but are not limited to, hue, saturation, and brightness.

[0076] Extract the image feature information corresponding to each crop type stored in the memory, and obtain the values ​​of various indicators of morphological feature information, texture feature information, and color feature information corresponding to each crop type. Record the values ​​of each indicator of morphological feature information, texture feature information, and color feature information corresponding to each crop type as follows: f represents the number of the f-th type of crop, f = 1, 2, ..., g.

[0077] Specifically, methods for identifying different crop types in an image to be evaluated also include:

[0078] The values ​​of various indicators of morphological, texture, and color features corresponding to each crop type are compared with the values ​​of various indicators of morphological, texture, and color features of each sub-image. The deviations of each crop type and each sub-image in terms of morphological, texture, and color features are obtained and denoted as follows:

[0079] According to the formula Calculate the morphological feature information adaptation factor between each crop type and each sub-image. g represents the total number of crop types.

[0080] Similarly, the adaptation factors of texture feature information and color feature information between each crop type and each sub-image are obtained, and they are denoted as follows:

[0081] According to the formula Calculate the matching index between each crop type and each sub-image. e represents the natural constant, and τ represents the correction coefficient of the preset fit index, with a value of 0.927.

[0082] It should be noted that the correction coefficient of the fit index serves to amplify, reduce, or round the fit index to facilitate calculation and analysis.

[0083] By comparing the matching indices of each crop type with those of each sub-image, the crop type corresponding to the largest matching index is taken as the crop type of each sub-image, thus obtaining the region corresponding to each crop type in the image to be evaluated, and thus obtaining the crop classification image of the crop study area.

[0084] In one specific embodiment, the present invention divides the image to be evaluated into sub-images and comprehensively considers multi-dimensional feature information such as morphology, texture and color and their various indicators, which can meticulously depict the characteristics of crops. Then, it compares the feature indicators of each crop type and calculates multiple fitting factors and fit indices. The crop type of the sub-image is determined by the maximum fit index. This multi-feature fusion and quantitative comparison method greatly improves the accuracy and reliability of crop identification and avoids the limitations of single feature identification.

[0085] S6. Crop growth assessment: Spectral features are extracted from the regional images corresponding to each crop type to obtain the normalized vegetation index, enhanced vegetation index, and photochemical reflectance index of each pixel in the regional image corresponding to each crop type. The growth status of each crop type is then assessed and analyzed, and an early warning is issued if the growth status is poor.

[0086] It should be noted that the normalized vegetation index, enhanced vegetation index, and photochemical reflectance index of each pixel in the regional image corresponding to each crop type are calculated. The specific statistical process is as follows:

[0087] Spectral features were extracted from regional images corresponding to each crop type using different wavelength bands. This yielded spectral feature information for each pixel within the regional image corresponding to each crop type, and the near-infrared, red, blue, 531nm, and 570nm reflectances of each pixel were extracted. These reflectances were then calculated using the normalized vegetation index formula. The normalized vegetation index of each pixel in the regional image corresponding to each crop type was calculated.

[0088] It should be noted that vegetation has high reflectivity in the near-infrared band and high absorption in the red band. The near-infrared reflectivity and red reflectivity of healthy vegetation differ significantly. By normalizing the difference and sum of the two values, vegetation information can be highlighted and interference from other ground features can be suppressed. The larger the normalized vegetation index value, the more vigorous the crop growth, the higher the coverage, and the richer the chlorophyll content. The lower the ratio, the more likely the crop is to have poor growth conditions such as pests and diseases.

[0089] The enhanced vegetation index is calculated using the formula.

[0090]

[0091] The enhanced vegetation index of each pixel in the regional image corresponding to each crop type was calculated.

[0092] It should be noted that the enhanced vegetation index is based on the normalized vegetation index analysis and incorporates the blue light band, taking into account the influence of factors such as soil background and atmospheric aerosols on the vegetation index, which can more accurately reflect the true growth status of crops.

[0093] pass The photochemical reflectance index of each pixel in the regional image corresponding to each crop type was calculated.

[0094] It should be noted that the photochemical reflectance index is mainly based on the change in the reflectance of plant leaves to the green light band during photosynthesis. When plants are affected by pests and diseases, photosynthesis will be affected, causing the reflectance of leaves at 531nm and 570nm to change, and the photochemical reflectance index value will also change accordingly. It is used to assess the photosynthetic physiological status of crops and the degree of stress.

[0095] Specifically, the process of assessing and analyzing the growth status of each crop type is as follows:

[0096] Extract the normalized vegetation index (NVI), enhanced vegetation index (VEI), and photochemical reflectance index (PRI) intervals of pixels in the images corresponding to healthy crops stored in the memory. Obtain the upper and lower limits of the NVI intervals and calculate the mean to obtain the ideal reference NVI. Subtract the NVI from the lower limit of the NVI interval to obtain the ideal deviation NVI. Similarly, obtain the ideal reference enhanced vegetation index, ideal deviation enhanced vegetation index, ideal reference PRI, and ideal deviation PRI based on the enhanced vegetation index intervals and PRI intervals.

[0097] The normalized vegetation index of each pixel in the regional image corresponding to each crop type is calculated by subtracting the normalized vegetation index of the ideal reference and taking the absolute value. Then, the ratio analysis is performed with the ideal deviation normalized vegetation index to obtain the ratio analysis result one.

[0098] The difference between the enhanced vegetation index of each pixel in the regional image corresponding to each crop type and the ideal reference enhanced vegetation index is calculated, and the absolute value is then compared with the ideal deviation enhanced vegetation index to obtain the ratio analysis result two.

[0099] The absolute value of the difference between the photochemical reflectance index of each pixel in the regional image corresponding to each crop type and the ideal reference photochemical reflectance index is calculated and then compared with the ideal deviation photochemical reflectance index to obtain the ratio analysis result three.

[0100] Using the first and second ratio analysis results as the major and minor axes, an ellipse is constructed. The center of the ellipse is selected, and a straight line with a length equal to the value corresponding to the third ratio analysis result is drawn with the center as the starting point. An elliptical cylinder is constructed with this straight line as the height of the ellipse and the base of the ellipse. The volume value of the elliptical cylinder is extracted, and the reciprocal of the volume value of the elliptical cylinder is used as the growth health coefficient to obtain the growth health coefficient of each crop type.

[0101] The growth health coefficient of each crop type is compared with the set growth health coefficient threshold. When the growth health coefficient of a certain crop type is less than the set growth health coefficient threshold, the crop type is judged to be growing well; otherwise, the crop type is judged to be growing poorly.

[0102] In one specific embodiment, this invention comprehensively assesses crop growth status from multiple dimensions, including vegetation cover, photosynthetic activity, and physiological state, by extracting spectral features closely related to crop growth, such as the normalized vegetation index, enhanced vegetation index, and photochemical reflectance index. This approach is more scientific and comprehensive than single-indicator assessments and can more accurately reflect the true growth status of crops. Simultaneously, by utilizing the spectral index range of moderately healthy crops, an ideal reference index and an ideal deviation index are calculated and compared with the spectral index of actual crop pixels, quantifying the growth status assessment process. This quantitative comparison method makes the assessment results more objective, reduces subjective judgment errors, and facilitates comparisons of growth status among different crop types. Early warning feedback is provided when growth status is poor, helping agricultural practitioners to identify problems early and take timely measures such as irrigation, fertilization, and pest and disease control to reduce the risk of crop yield reduction and ensure agricultural production efficiency.

[0103] Please see Figure 2 As shown, a second aspect of the present invention provides an image processing apparatus based on artificial intelligence, comprising: an image acquisition unit, a super-resolution reconstruction module, a reconstruction effect judgment module, a scaling factor adjustment module, an object recognition module, a growth evaluation module, an early warning feedback module, and a memory.

[0104] The image acquisition unit is connected to the super-resolution reconstruction module, the super-resolution reconstruction module is connected to the reconstruction effect judgment module, the reconstruction effect judgment module is connected to the scaling factor adjustment module and the object recognition module respectively, the scaling factor adjustment module is connected to the super-resolution reconstruction module and the early warning feedback module respectively, the object recognition module is connected to the growth evaluation module, the growth evaluation module is connected to the early warning feedback module, and the memory is connected to the super-resolution reconstruction module, the object recognition module and the growth evaluation module respectively.

[0105] The image acquisition device acquires remote sensing images of the crop study area and performs preprocessing operations on the remote sensing images of the crop study area to obtain preprocessed remote sensing images of the crop study area, which are then marked as the original remote sensing images of the crop study area.

[0106] The super-resolution reconstruction module performs super-resolution reconstruction on the original remote sensing images of the crop research area using the bilinear interpolation method, resulting in a high-resolution target image after reconstruction.

[0107] The reconstruction effect judgment module crops and adjusts the high-resolution target image according to the size of the original remote sensing image to obtain a high-resolution target image after size processing, which is used as the image to be evaluated. The module analyzes the peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the image to be evaluated, and judges the super-resolution reconstruction effect accordingly.

[0108] When the scaling factor adjustment module determines that the super-resolution reconstruction effect is not good, it analyzes the preset horizontal and vertical scaling factor adjustment parameters, and re-executes the super-resolution reconstruction module and the reconstruction effect judgment module based on the horizontal and vertical scaling factor adjustment parameters. If the secondary super-resolution reconstruction effect is good, the object recognition module is executed; if the secondary super-resolution reconstruction effect is not good, it sends an early warning feedback to the early warning feedback module.

[0109] When the object recognition module determines that the super-resolution reconstruction effect is good, it identifies the types of crops in the image to be evaluated and obtains the crop classification image of the crop study area.

[0110] The growth assessment module extracts spectral features from the regional images corresponding to each crop type to obtain the normalized vegetation index, enhanced vegetation index, and photochemical reflectance index of each pixel in the regional image corresponding to each crop type. Then, it assesses and analyzes the growth status of each crop type. If the growth status is poor, it sends an early warning feedback to the early warning feedback module.

[0111] The early warning feedback module provides early warning feedback for poor secondary super-resolution reconstruction results and poor growth status of various crop types.

[0112] The memory stores the pixel values ​​of the original remote sensing images, the image feature information corresponding to each crop type, and the normalized vegetation index range, enhanced vegetation index range, and photochemical reflectance index range of pixels in the image corresponding to healthy crops.

[0113] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. An image processing method based on artificial intelligence, characterized by, Comprise: S1. Remote sensing image acquisition of the region: collect the remote sensing image of the crop research area, and pretreat the remote sensing image of the crop research area to obtain the original remote sensing image; S2. Super-resolution reconstruction: the original remote sensing image is super-resolution reconstructed based on the bilinear interpolation method to obtain a high-resolution target image; S3. Reconstructing effect judgment: the high-resolution target image is cropped and adjusted according to the size of the original remote sensing image to obtain a high-resolution target image, which is used as an image to be evaluated, and the peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the image to be evaluated are analyzed to judge the super-resolution reconstruction effect; S4. Scaling factor adjustment: when the super-resolution reconstruction effect is not good, the horizontal and vertical scaling factor adjustment parameters are analyzed, and steps S2 and S3 are re-executed based on the horizontal and vertical scaling factor adjustment parameters, if the secondary super-resolution reconstruction effect is good, step S5 is executed, if the secondary super-resolution reconstruction effect is not good, a warning feedback is given; S5. Crop object recognition: when the super-resolution reconstruction effect is good, each crop type in the image to be evaluated is identified to obtain a crop classification image of the crop research area; S6. Crop growth evaluation: the spectral feature of the region image corresponding to each crop type is extracted to obtain the normalized vegetation index, enhanced vegetation index and photochemical reflection index of each pixel in the region image corresponding to each crop type, and then the growth status of each crop type is evaluated and analyzed, if the growth status is not good, a warning feedback is given. 2.The AI-based image processing method of claim 1, wherein, The implementation process of the super-resolution reconstruction based on the bilinear interpolation method is as follows: 201: determining the size of the original remote sensing image and the target image: extracting the image size corresponding to the original remote sensing image of the crop research area, determining the image size through the super-resolution reconstruction image size determination expression according to the preset horizontal direction and vertical direction scaling factor From which the image size corresponding to the high-resolution target image after the super-resolution reconstruction processing is identified, wherein s 水平 , s 竖直 Respectively represent the horizontal direction and the vertical direction scaling factor, W original , H original Respectively represent the width and height in the image size corresponding to the original remote sensing image, W target , H target Respectively represent the width and height in the image size corresponding to the target image; 202: Establish coordinate mapping relationship: for each pixel point (x target ,y target )(0≤x target ≤W target ,0≤y target ≤H target ) in the high-resolution target image, map to the corresponding position (x original ,y original ) in the original remote sensing image through the scaling factor, and the specific mapping calculation formula is 203: determining adjacent pixel points: obtaining four pixel points adjacent to the corresponding position mapped into the original remote sensing image in the original remote sensing image, and recording the coordinates of the four pixel points as (x1, y1), (x1, y2), (x2, y1), (x2, y2) respectively, wherein x2 = x1 + 1, y2 = y1 + 1, represents a down rounding operation; 204: Bilinear Interpolation Calculation: Extract the pixel values ​​of these four pixels from the original remote sensing image stored in memory, denoted as f(x1,y1), f(x1,y2), f(x2,y1), and f(x2,y2) respectively. Then, calculate the values ​​using the bilinear interpolation formula f(x1,y1), f(x1,y2), and f(x2,y2). target ,y target )=(1-ε)(1-γ)f(x1,y1)+(1-ε)γf(x1,y2)+ε(1-γ)f(x2,y1)+εγf(x2,y2), to obtain the pixel value f(x) of the pixel in the high-resolution target image. target ,y target ), where ε=x original -x1,γ=y original -y1; 205: traverse all pixels of the target image: calculate each pixel point in the high-resolution target image in the same way as steps 202-204, and then complete the interpolation calculation of the whole image to generate a high-resolution remote sensing image. 3.The AI-based image processing method of claim 1, wherein, The process of analyzing the peak noise ratio and global similarity evaluation coefficient of the original remote sensing image and the image to be evaluated is as follows: The pixel values of the original remote sensing image and the image to be evaluated are normalized, the pixels at the coordinates in the original remote sensing image and the pixels at the same coordinates in the image to be evaluated are extracted, and the difference between the two is calculated to obtain the pixel difference, the pixel difference of all pixel points is squared, and the square values are accumulated to obtain the pixel difference square sum, the pixel difference square sum is divided by the total number of pixels of the image to obtain the mean square error; The peak signal-to-noise ratio η between the original remote sensing image and the image to be evaluated is calculated by the formula is expressed as a mean square error, and max is expressed as the maximum value of the dynamic range of the image pixel value.​ According to the preset local window size, slide row by row and column by column on the original remote sensing image and the image to be evaluated, in each window position, obtain the mean, variance and covariance of the pixels in the original remote sensing image window and the pixels in the image to be evaluated window, and the two windows at the corresponding position form a window group; The variances of the pixel values of two windows in each window group are compared with a set variance threshold value, if the variances of the pixel values of two windows in a window group are both less than the set variance threshold value, the window group is determined as an invalid window group, otherwise, the window group is determined as a valid window group, the number of valid window groups is obtained, and each valid window group is numbered according to a preset order and sequentially marked as 1, 2,..., i,..., n; According to the formula The brightness similarity index between the original remote sensing image and the image to be evaluated corresponding to each effective window group is calculated The contrast similarity index And the structural similarity index R represents the original remote sensing image, and T represents the image to be evaluated, respectively represent the pixel mean of the original remote sensing image and the pixel mean of the image to be evaluated corresponding to each effective window group, respectively represent the pixel variance of the original remote sensing image and the pixel variance of the image to be evaluated corresponding to each effective window group, represent the pixel covariance of the original remote sensing image and the image to be evaluated corresponding to each effective window group; The global similarity evaluation coefficient ψglobal(R, T) between the original remote sensing image and the image to be evaluated is calculated according to the formula The global similarity evaluation coefficient ψglobal(R, T) between the original remote sensing image and the image to be evaluated is calculated according to the formula 4.The AI-based image processing method of claim 1, wherein, The process of evaluating the super-resolution reconstruction effect is as follows: The peak signal-to-noise ratio and the global similarity evaluation coefficient between the original remote sensing image and the image to be evaluated are converted into lengths according to a preset ratio, the length of the peak signal-to-noise ratio and the length of the global similarity evaluation coefficient are obtained, an ellipse is constructed with the length of the peak signal-to-noise ratio as the long axis and the length of the global similarity evaluation coefficient as the short axis, and then the area of the ellipse is extracted, and the numerical value of the area of the ellipse is marked as a reconstruction effect coefficient; The reconstruction effect coefficient is compared with a set reconstruction effect coefficient threshold value, if the reconstruction effect coefficient is greater than or equal to the set reconstruction effect coefficient threshold value, it is determined that the super-resolution reconstruction effect is good, otherwise, it is determined that the super-resolution reconstruction effect is poor. 5.The artificial intelligence-based image processing method of claim 1, wherein, The process of analyzing the preset horizontal direction and vertical direction scaling factor adjustment parameters is as follows: When it is determined that the super-resolution reconstruction effect is poor, the reconstruction effect coefficient is extracted, and a difference calculation is performed between the reconstruction effect coefficient and a set reconstruction effect coefficient threshold value, to obtain a reconstruction effect coefficient difference; The reconstruction effect coefficient difference is matched with the set reconstruction effect coefficient difference corresponding to each scaling factor adjustment parameter set, to obtain a scaling factor adjustment parameter set, and the scaling factor adjustment parameter set includes the horizontal direction and vertical direction scaling factor adjustment parameters. 6.The AI-based image processing method of claim 1, wherein, The method for identifying each crop type in the image to be evaluated includes: The image to be evaluated is divided into sub-images according to a set size of the scale, image feature information of each sub-image is obtained, morphological feature information, texture feature information and color feature information of each sub-image are obtained, and numerical values of each index of the morphological feature information, the texture feature information and the color feature information of each sub-image are obtained and numbered; Image feature information corresponding to each crop type stored in the memory is extracted, and numerical values of each index of the morphological feature information, the texture feature information and the color feature information corresponding to each crop type are obtained.

7. The artificial intelligence-based image processing method of claim 6, wherein, The method for identifying each crop type in the image to be evaluated further includes: Numerical values of each index of the morphological feature information, the texture feature information and the color feature information corresponding to each crop type are compared with numerical values of each index of the morphological feature information, the texture feature information and the color feature information of each sub-image, to obtain deviations of each index of the morphological feature information, the texture feature information and the color feature information between each crop type and each sub-image; A morphological feature information adaptation factor of each crop type and each sub-image is calculated; A texture feature information adaptation factor and a color feature information adaptation factor of each crop type and each sub-image are analyzed; A fitting index of each crop type and each sub-image is calculated; The crop types are compared with each sub-image fitting index, and the crop type corresponding to the maximum fitting index is taken as the crop type of each sub-image, so as to obtain the region corresponding to each crop type in the image to be evaluated, and further obtain the crop classification image of the crop research region. 8.The AI-based image processing method of claim 1, wherein, The evaluation analyzes the growth status of each crop type as follows: The normalized vegetation index interval, the enhanced vegetation index interval and the photochemical reflectance index interval of the pixels in the image corresponding to the healthy crop stored in the storage are extracted, the upper limit value and the lower limit value of the normalized vegetation index interval are obtained and the mean value is calculated to obtain the ideal reference normalized vegetation index, and the ideal reference normalized vegetation index is subtracted from the lower limit value of the normalized vegetation index interval to obtain the ideal deviation normalized vegetation index. Similarly, according to the enhanced vegetation index interval and the photochemical reflectance index interval, the ideal reference enhanced vegetation index, the ideal deviation enhanced vegetation index, the ideal reference photochemical reflectance index and the ideal deviation photochemical reflectance index are obtained; The normalized vegetation index of each pixel in the region image corresponding to each crop type is subtracted from the ideal reference normalized vegetation index, and the absolute value is taken to perform ratio analysis with the ideal deviation normalized vegetation index to obtain ratio analysis result one; Similarly, the enhanced vegetation index and the photochemical reflectance index of each pixel in the region image corresponding to each crop type are analyzed respectively to obtain ratio analysis result two and ratio analysis result three; An ellipse is constructed with ratio analysis result one and ratio analysis result two as the major axis and the minor axis, the center of the ellipse is selected, a straight line with a length equal to the value corresponding to ratio analysis result three is drawn from the center as the starting point, and an elliptical cylinder is constructed with the straight line as the height and the ellipse as the base surface. The volume value of the elliptical cylinder is extracted, and the reciprocal of the volume value of the elliptical cylinder is taken as the growth health coefficient to obtain the growth health coefficient of each crop type. The growth health coefficient of each crop type is compared with the set growth health coefficient threshold value, and when the growth health coefficient of a certain crop type is less than the set growth health coefficient threshold value, it is determined that the growth status of the crop type is good, otherwise, it is determined that the growth status of the crop type is poor.

9. An apparatus for artificial intelligence-based image processing for performing the artificial intelligence-based image processing method according to any one of claims 1 to 8, characterized by, It includes: An image collector is used to collect remote sensing images of a crop research region, and to perform preprocessing operations on the remote sensing images of the crop research region to obtain preprocessed remote sensing images of the crop research region, and mark them as original remote sensing images of the crop research region; A super-resolution reconstruction module is used to perform super-resolution reconstruction on the original remote sensing images of the crop research region based on a bilinear interpolation method to obtain a high-resolution target image after reconstruction processing; A reconstruction effect judgment module is used to crop and adjust the high-resolution target image according to the size of the original remote sensing image to obtain a high-resolution target image after size processing, which is taken as an image to be evaluated. The peak noise ratio and the global similarity evaluation coefficient of the original remote sensing image and the image to be evaluated are analyzed, and the super-resolution reconstruction effect is judged. The scaling factor adjustment module is configured to analyze preset horizontal and vertical scaling factor adjustment parameters when the super-resolution reconstruction effect is determined to be poor, and re-perform the super-resolution reconstruction module and the reconstruction effect determination module based on the horizontal and vertical scaling factor adjustment parameters. If the secondary super-resolution reconstruction effect is good, the object recognition module is executed. If the secondary super-resolution reconstruction effect is poor, the object recognition module is sent to the early warning feedback module for early warning feedback. The object recognition module is configured to recognize each crop type in the image to be evaluated when the super-resolution reconstruction effect is determined to be good, and obtain a crop classification image of the crop research area. The growth evaluation module is configured to perform spectral feature extraction on the region image corresponding to each crop type, obtain the normalized vegetation index, enhanced vegetation index and photochemical reflectance index of each pixel in the region image corresponding to each crop type, and further evaluate and analyze the growth status of each crop type. If the growth status is poor, the growth evaluation module is sent to the early warning feedback module for early warning feedback. The early warning feedback module is configured to provide early warning feedback for the secondary super-resolution reconstruction effect being poor and the growth status of each crop type being poor. The memory is configured to store the pixel values of the original remote sensing image, store the image feature information corresponding to each crop type, and store the normalized vegetation index interval, enhanced vegetation index interval and photochemical reflectance index interval of the pixels in the image corresponding to the healthy crop.

Images