Waste landscaping detection system and method

Abstract

The invention relates to the technical field of detection, in particular to a waste landscaping detection system and method.The waste landscaping detection system comprises an image acquisition module, an image pre-fusion module, an image correction module, an image fusion module and a comprehensive feedback module, and the image correction module comprises a geometric correction sub-module and a radiation correction sub-module; according to the waste landscaping detection system and method provided by the invention, a set of complete detection process is constructed from image acquisition, preprocessing, correction, fusion and feedback, and the problem of inaccurate recognition caused by poor image quality in traditional detection is effectively solved through cooperative and efficient work of all modules, by improving the geometric accuracy of the image and optimizing the radiation characteristic, high-quality image data is provided for subsequent image fusion and recognition, the precision, reliability and efficiency of waste detection are improved, and powerful support is provided for waste landscaping treatment.

Description

Technical Field

[0001] The present invention relates to the technical field of detection, and specifically to a waste landscaping detection system and method. Background Art

[0002] With the acceleration of the urbanization process, the quantity of waste has increased sharply, imposing a huge pressure on the environment. By transforming waste into landscape elements with aesthetic value or specific functions, not only can landfill and incineration of garbage be reduced, but also the effective recycling of resources can be promoted. At the same time, it has strong social educational significance, which can enhance the public's environmental awareness and social responsibility. In addition, waste landscaping, as a new form of artistic expression, adds a unique cultural atmosphere to urban spaces, promotes the development of modern art, and reflects the win-win situation of ecological and social benefits .

[0003] For example, the invention patent with the publication number: CN110780055B is a modeling method for predicting the bioavailability of high-ring PAHs in soil based on spectral analysis, belonging to the technology of predicting the bioavailability of polycyclic aromatic hydrocarbons, and includes the following steps: (1) collection and pretreatment of soil samples, (2) oxidizing the soil samples by chemical oxidation method, (3) performing infrared spectral measurement on the original soil samples and their corresponding oxidized soil samples to obtain the proportion of infrared absorbance at each wavelength, (4) establishing the fitting curves of the concentrations of 16 PAHs and the infrared proportions, comparing the regression models of polycyclic aromatic hydrocarbons with different ring numbers and the proportion of absorbance at different wavelengths, (5) screening the regression models to obtain the regression model with the best correlation degree, and obtaining the functional relationship between the high-ring polycyclic aromatic hydrocarbons and the proportion of absorbance at the site.

[0004] For example, the invention patent with the publication number: CN105092819B is a spectral analysis method and device for the redox potential of soil above an oil and gas reservoir. The spectral analysis method includes: collecting soil samples at a set depth in the oil and gas reservoir and its peripheral shallow surface; measuring the redox potential of the soil samples; measuring the reflectance spectrum of the soil samples, eliminating the spectral transition of the soil samples at the set wavelength position, reprocessing the reflectance spectrum to generate three derivative spectral indexes: the first derivative of reflectance, the second derivative of reflectance, and the absorption depth of reflectance; calculating the correlation coefficients between the redox potential and the reflectance spectrum, the first derivative of reflectance, the second derivative of reflectance, and the absorption depth of reflectance; determining the best spectral indexes and modeling wavelengths for modeling according to the correlation coefficients; and establishing a regression model between the redox potential and the best spectral indexes by using the partial least squares method.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of the present application, it is found that the above technologies have at least the following technical problems: In the application practice of the current waste landscaping detection technology for the waste identification link, there are scale or angle deviations in the images captured by the sensors carried by the drones, resulting in low-quality images being collected, inaccurate waste identification, and thus affecting the accuracy and reliability of subsequent waste detection. Summary of the Invention

[0006] In view of the deficiencies of the prior art, the present invention provides a waste landscaping detection system and method, which can effectively solve the problems involved in the above background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: In the first aspect of the present invention, a waste landscaping detection system is provided, including: an image acquisition module, configured to mark the waste landscaping detection area as a target area and acquire multi-angle images of the target area.

[0008] An image pre-fusion module, configured to pre-fuse the multi-angle images of the target area to obtain a pre-fused image of the target area.

[0009] An image correction module, configured to determine whether the pre-fused image of the target area needs to be corrected and obtain a specified fused image of the target area, where the image correction module includes a geometric correction sub-module and a radiometric correction sub-module. The geometric correction sub-module is configured to determine whether the pre-fused image of the target area needs geometric correction, and the radiometric correction sub-module is configured to determine whether the pre-fused image of the target area needs radiometric correction.

[0010] An image fusion module, configured to fuse the specified fused image of the target area to obtain a multi-source data fused image of the target area.

[0011] A comprehensive feedback module, configured to perform image recognition on the multi-source data fused image of the target area to obtain the waste type and transmit it to a preset display window for reminder.

[0012] As a further method, the process of pre-fusing the multi-angle images of the target area to obtain a pre-fused image of the target area is specifically as follows: Obtain the basic information of the multi-angle images of the target area, including the noise level, contrast, brightness, and blur degree of each angle image, perform difference analysis on the noise level, brightness, contrast, and blur degree of each angle image with the noise level critical value, brightness critical value, contrast critical value, and blur degree critical value respectively to obtain the deviation degree, and introduce their corresponding weights respectively and perform coupling processing to obtain the image quality requirement processing index of each angle image.

[0013] Process the exponential screening based on the image quality requirements of each angle to obtain the processed images required for each angle. After preprocessing, perform pre-fusion processing to obtain the pre-fused image of the target area.

[0014] As a further method, for the process of processing the exponential screening based on the image quality requirements of each angle to obtain the processed images required for each angle, and then performing pre-fusion processing after preprocessing to obtain the pre-fused image of the target area: extract the threshold of the image quality requirement processing index from the system database, compare the exponential of the image quality requirement processing for each angle with the threshold of the image quality requirement processing index. If the exponential of the image quality requirement processing for a certain angle is greater than the threshold of the image quality requirement processing index, mark the image of this angle as the processed image required for the angle and perform preprocessing. If the exponential of the image quality requirement processing for a certain angle is less than or equal to the threshold of the image quality requirement processing index, there is no need to perform pre-fusion on the image of this angle.

[0015] After preprocessing, count all the images of each angle for pre-fusion to obtain the pre-fused image of the target area.

[0016] As a further method, for the process of determining whether the pre-fused image of the target area needs geometric correction: extract the geographical image of the target area, the feature point matching error threshold, and the pixel value difference threshold of the overlapping area from the system database. Mark the geographical image of the target area as the reference image of the target area. Obtain the feature point matching error and the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image. Compare the feature point matching error between the pre-fused image of the target area and the reference image with the feature point matching error threshold respectively, and compare the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image with the pixel value difference threshold of the overlapping area. When both the feature point matching error and the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image are lower than the threshold, there is no need to perform geometric correction on the pre-fused image of the target area. Otherwise, geometric correction needs to be performed on the pre-fused image of the target area. The specific process is as follows: use the feature point matching algorithm to perform geometric correction on the pre-fused image of the target area to obtain the geometrically corrected image of the target area, and evaluate the geometric correction effect, and perform corresponding geometric correction adjustment processing according to the evaluation results.

[0017] As a further method, the evaluation of the geometric correction effect is carried out in the following specific process: obtain the feature point matching error, pixel value difference in the overlapping area, structural similarity, and average error of the checkpoints not participating in the correction between the geometric correction image of the target area and the reference image. Compare the feature point matching error, pixel value difference in the overlapping area, structural similarity, and average error of the checkpoints not participating in the correction between the geometric correction image of the target area and the reference image with the extreme values of the feature point matching error, the extreme values of the pixel value difference in the overlapping area, the extreme values of the structural similarity, and the average error extreme values of the checkpoints not participating in the correction respectively, and set corresponding weight factors to reflect the relative importance of each index when evaluating the geometric correction effect. Then perform coupling processing to obtain the geometric correction effect evaluation index.

[0018] Extract the geometric correction effect index threshold from the system database. Compare the geometric correction effect evaluation index with the geometric correction effect index threshold. If the geometric correction effect evaluation index is higher than or equal to the geometric correction effect index threshold, input the geometric correction image of the target area into the image fusion module. If the geometric correction effect evaluation index is lower than the geometric correction effect index threshold, perform secondary correction on the geometric correction image of the target area.

[0019] As a further method, the secondary correction of the geometric correction image of the target area is carried out in the following specific process: mark the difference between the geometric correction effect evaluation index and the geometric correction effect index threshold as the geometric correction demand adjustment parameter. Extract the number of additional feature points corresponding to the geometric correction demand adjustment parameter from the system database. After adding the number of additional feature points to the initial number of feature points, obtain the number of secondary correction constraint feature points, and use it as the secondary correction constraint condition of the geometric correction model for correction and then analyze again to obtain the geometric correction effect evaluation index, which is recorded as the secondary geometric correction effect evaluation index.

[0020] Compare the secondary geometric correction effect evaluation index with the geometric correction effect index threshold. If the secondary geometric correction effect evaluation index is higher than or equal to the geometric correction effect index threshold, input the image of the target area after secondary geometric correction into the image fusion module. If the secondary geometric correction effect evaluation index is still lower than the geometric correction effect index threshold, it is necessary to switch the geometric correction model to perform re-geometric correction on the image of the target area after secondary geometric correction.

[0021] As a further method, the process of determining whether the pre-fusion image of the target area needs radiometric correction is as follows: Obtain the histogram similarity and average brightness difference between the pre-fusion image of the target area and the reference image. Compare the histogram similarity between the pre-fusion image of the target area and the reference image with the histogram similarity threshold respectively, and compare the average brightness difference between the pre-fusion image of the target area and the reference image with the average brightness difference threshold. When both the histogram similarity and the average brightness difference between the pre-fusion image of the target area and the reference image are lower than the thresholds, there is no need to perform radiometric correction on the pre-fusion image of the target area; otherwise, radiometric correction needs to be performed on the pre-fusion image of the target area. The specific process is as follows: Use radiometric calibration to perform radiometric correction on the pre-fusion image of the target area to obtain the radiometrically corrected image of the target area, and evaluate the radiometric correction effect, and perform corresponding radiometric correction adjustment processing according to the evaluation results.

[0022] As a further method, the process of evaluating the radiometric correction effect is as follows: Obtain the histogram similarity, average brightness difference, structural similarity, ground object reflectance difference, and signal-to-noise ratio of the radiometrically corrected image of the target area and the reference image. Compare the histogram similarity, average brightness difference, structural similarity, ground object reflectance difference, and signal-to-noise ratio of the radiometrically corrected image of the target area and the reference image with the histogram similarity extreme value, average brightness difference extreme value, structural similarity extreme value, ground object reflectance difference extreme value, and signal-to-noise ratio extreme value respectively, and set corresponding weight factors to reflect the relative importance of each index when evaluating the radiometric correction effect, and then perform coupling processing to obtain the radiometric correction effect evaluation index.

[0023] Extract the radiometric correction effect index threshold from the system database, and compare the radiometric correction effect evaluation index with the radiometric correction effect index threshold. If the radiometric correction effect evaluation index is higher than or equal to the radiometric correction effect index threshold, input the radiometrically corrected image of the target area into the image fusion module; if the radiometric correction effect evaluation index is lower than the radiometric correction effect index threshold, perform secondary correction on the radiometrically corrected image of the target area.

[0024] As a further method, the specific analysis process of performing secondary correction on the radiometrically corrected image of the target area is as follows: Mark the difference between the radiometric correction effect evaluation index and the radiometric correction effect index threshold as the radiometric correction demand adjustment parameter. Extract the demand adjustment calibration coefficient corresponding to the radiometric correction demand adjustment parameter from the system database. Replace the initial calibration coefficient with the demand adjustment calibration coefficient and use it as the secondary correction constraint condition of the radiometric correction model for correction, and then analyze again to obtain the radiometric correction effect evaluation index, which is recorded as the secondary radiometric correction effect evaluation index.

[0025] Compare the secondary radiation correction effect evaluation index with the radiation correction effect index threshold. If the secondary radiation correction effect evaluation index is higher than or equal to the radiation correction effect index threshold, input the image of the target area after secondary radiation correction into the image fusion module. If the secondary radiation correction effect evaluation index is still lower than the radiation correction effect index threshold, it is necessary to switch the radiation correction model to re-radiate correct the image of the target area after secondary radiation correction.

[0026] The second aspect of the present invention provides a waste landscaping detection method, including: S1. Mark the waste landscaping detection area as the target area, and collect multi-angle images of the target area.

[0027] S2. Pre-fuse the multi-angle images of the target area to obtain a pre-fused image of the target area.

[0028] S3. Determine whether the pre-fused image of the target area needs to be corrected, and obtain a specified fused image of the target area.

[0029] S4. Fuse the pre-fused image of the corrected target area to obtain a multi-source data fused image of the target area.

[0030] S5. Perform image recognition on the multi-source data fused image of the target area to obtain the waste type, and transmit it to a preset display window for reminder.

[0031] Compared with the prior art, the embodiments of the present invention at least have the following advantages or beneficial effects:

[0032] (1) By providing a waste landscaping detection system and method, the present invention constructs a complete detection process from image acquisition, preprocessing, correction, fusion to feedback. The modules work together efficiently, effectively solving the problem of inaccurate recognition caused by poor image quality in traditional detection, improving the accuracy, reliability and efficiency of detection, and providing strong support for waste landscaping treatment.

[0033] (2) By analyzing the basic information of the multi-angle images of the target area, the present invention screens out the images that need to be preprocessed and processes them specifically, which can improve the image quality, clearly present the details of the target area, provide a reliable basis for detection, and improve the accuracy and reliability. It can also optimize the fusion effect, reduce fusion errors and distortions, improve the detection accuracy and efficiency. It can also reduce the risk of misjudgment, accurately identify waste and landscape features, and provide a scientific basis for formulating treatment plans. It is also conducive to improving the intelligent level of the system and improving the timeliness and accuracy of detection.

[0034] (3) By improving the geometric accuracy of the image and optimizing the radiation characteristics, the present invention provides high-quality image data for subsequent image fusion and recognition. The improvement of image quality enables more accurate extraction of the characteristics of waste when identifying the type of waste, reduces the occurrence of misjudgment and missed judgment, thus significantly improving the accuracy and reliability of the waste landscape detection in the target area, and providing more reliable data support for waste treatment and landscape transformation. Brief Description of the Drawings

[0035] The present invention will be further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the following drawings.

[0036] Figure 1 It is a schematic diagram of the connection of the system modules of the present invention.

[0037] Figure 2 It is a schematic diagram of the method flow of the present invention. Detailed Embodiments

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0039] Referring to Figure 1 As shown, the first aspect of the present invention provides a waste landscape detection system, including: an image acquisition module for marking the waste landscape detection area as the target area and acquiring multi-angle images of the target area.

[0040] An image pre-fusion module for pre-fusing the multi-angle images of the target area to obtain a pre-fused image of the target area.

[0041] Specifically, the process of pre-fusing the multi-angle images of the target area to obtain a pre-fused image of the target area is as follows: obtaining the basic information of the multi-angle images of the target area, including the noise level, contrast, brightness, and blur degree of each angle image, performing difference analysis on the noise level, brightness, contrast, and blur degree of each angle image with the noise level critical value, brightness critical value, contrast critical value, and blur degree critical value respectively to obtain the deviation degree, and introducing their corresponding weights respectively and then performing coupling processing to obtain the processing index of the image quality requirements of each angle image.

[0042] Process the index screening based on the image quality requirements of each angle to obtain the processed images for each angle requirement, and after preprocessing, perform pre-fusion processing to obtain the pre-fused image of the target area.

[0043] Specifically, the system uses the weighted average algorithm to assign weights to the images of each angle according to the image quality requirement processing index of each angle, and performs weighted average calculation on the pixels at the same spatial position in the images of each angle to generate the pre-fused image of the target area.

[0044] It should be noted that in this embodiment, the noise level of the image can be measured by calculating the noise standard deviation of the image. That is, several small windows are selected in different regions of the image, and the standard deviation of the pixel values in each window is calculated. The image contrast refers to the range of the gray values of the image. The brightness of the image refers to the average gray value of the image. The blurriness of the image refers to the gradient amplitude of the image.

[0045] It should be noted that in this embodiment, the noise level critical value, the contrast critical value, the brightness critical value, and the blurriness critical value are all extracted from the system database.

[0046] In a specific embodiment, the system database is used to store the relevant data during the fusion process of the multi-angle images of the target area, specifically including: the noise level critical value, the contrast critical value, the brightness critical value, and the blurriness critical value, as well as the data extracted from the system database. The relevant data can be obtained by analyzing the historical image data of the target area, and the relevant auxiliary data can also be obtained through external data sources or databases.

[0047] In a specific embodiment, the numerical expression of the image quality requirement processing index of each angle is:

[0048]

[0049] In the formula, represents the image quality requirement processing index of the nth angle, Z n represents the noise level of the nth angle image, D n represents the contrast of the nth angle image, L n represents the brightness of the nth angle image, M n represents the blurriness of the nth angle image, represents the noise level critical value, represents the contrast critical value, represents the brightness critical value, represents the ambiguity critical value, ρ1 represents the angle image quality influence characteristic factor corresponding to the preset noise level, ρ2 represents the angle image quality influence characteristic factor corresponding to the preset contrast, ρ3 represents the angle image quality influence characteristic factor corresponding to the preset brightness, and ρ4 represents the angle image quality influence characteristic factor corresponding to the preset ambiguity.

[0050] It should be noted that a high noise level is usually accompanied by a low contrast because noise affects the dynamic range of the image and reduces the contrast. Low-brightness images usually also exhibit low contrast because when the overall brightness is low, the details in the image are compressed within a smaller brightness range. High noise levels and high ambiguities usually occur simultaneously, especially in photos taken under low-light conditions. At the same time, in practical applications, low-brightness images are more susceptible to blurring because the sensor may generate more noise and blurring under low-light conditions. When the noise level, contrast, brightness, and ambiguity of an image are higher, the corresponding image quality requirement processing index is larger, indicating that stronger image processing techniques are needed to improve the image quality.

[0051] It should be noted that in this embodiment, ρ1 represents the angle image quality influence characteristic factor corresponding to the preset noise level, ρ2 represents the angle image quality influence characteristic factor corresponding to the preset contrast, ρ3 represents the angle image quality influence characteristic factor corresponding to the preset brightness, and ρ4 represents the angle image quality influence characteristic factor corresponding to the preset ambiguity. They are respectively the numerical values representing the influence degree of the unit values of noise level, contrast, brightness, and ambiguity on the image quality, and can be directly obtained from the system database during use. Their corresponding relationships can be pre-set mapping relationships. For example, the noise level, contrast, brightness, and ambiguity respectively form a mapping set with the angle image quality influence characteristic factors corresponding to the preset noise level, contrast, brightness, and ambiguity in the system database. Inputting the real-time noise level, contrast, brightness, and ambiguity into the mapping set to obtain the corresponding angle image quality influence characteristic factors, and the mapping relationships therein can be one-to-one or many-to-one relationships. The value ranges of the above influence characteristic factors are all between 0 and 1.

[0052] Further, based on the screening of each angle image quality requirement processing index, each angle required processing image is obtained and preprocessed. The specific process is as follows: Extract the image quality requirement processing index threshold from the system database, compare each angle image quality requirement processing index with the image quality requirement processing index threshold. If the image quality requirement processing index of a certain angle is greater than the image quality requirement processing index threshold, then mark the image of this angle as the angle required processing image and perform preprocessing. If the image quality requirement processing index of a certain angle is less than or equal to the image quality requirement processing index threshold, then there is no need to perform pre-fusion on the image of this angle.

[0053] It should be understood that in this embodiment, image preprocessing specifically refers to processing and optimizing the basic quality problems of images, including: denoising: reducing the noise in the image to make the image cleaner; adjusting brightness and contrast: optimizing the overall illumination and gray-scale distribution of the image to enhance details; sharpening: strengthening the edges and details of the image to make the target information clearer.

[0054] After preprocessing, all angular images are statistically pre-fused to obtain a pre-fused image of the target area.

[0055] It should be explained that in this embodiment, the target area specified fused image refers to the image obtained by statistically pre-fusing all angular images after preprocessing all angular required processing images.

[0056] Specifically, once it is found that the image quality requirement processing index of a certain angular image is greater than the image quality requirement processing index threshold, the system immediately issues a warning and displays that the image needs to be preprocessed, that is, denoising, adjusting brightness, contrast, and sharpening the angular image, and then re-performing a loop judgment. If it is still not lower than or equal to the image quality requirement processing index threshold within the preset number of times, the angular image is marked as an abnormal image and image acquisition is re-performed for that angle.

[0057] It should be explained that in this embodiment, by analyzing the basic information such as the noise level, contrast, brightness, and blurriness of multi-angular images of the target area, and screening out the images that need to be preprocessed according to the quality requirement processing index, and specifically performing denoising, adjusting brightness and contrast, and sharpening and other processing on these images, the image quality can be improved, the details of the target area can be clearly presented, providing a reliable basis for detection, and improving accuracy and reliability. It can also optimize the fusion effect, reduce fusion errors and distortions, provide comprehensive and accurate image information, and improve detection accuracy and efficiency. It can also reduce the risk of misjudgment, accurately identify waste and landscape features, and provide a scientific basis for formulating treatment plans. It is also conducive to improving the intelligent level of the system and improving the timeliness and accuracy of detection.

[0058] An image correction module, used to determine whether the pre-fused image of the target area needs to be corrected, and obtain the specified fused image of the target area, where the image correction module includes a geometric correction sub-module and a radiometric correction sub-module. The geometric correction sub-module is used to determine whether the pre-fused image of the target area needs geometric correction, and the radiometric correction sub-module is used to determine whether the pre-fused image of the target area needs radiometric correction.

[0059] Specifically, to determine whether geometric correction is required for the pre-fused image of the target area, the specific process is as follows: Extract the geographical image of the target area, the feature point matching error threshold, and the pixel value difference threshold of the overlapping area from the system database. Mark the geographical image of the target area as the reference image of the target area. Obtain the feature point matching error and the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image. Compare the feature point matching error between the pre-fused image of the target area and the reference image with the feature point matching error threshold respectively, and compare the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image with the pixel value difference threshold of the overlapping area. When both the feature point matching error and the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image are lower than the threshold, there is no need to perform geometric correction on the pre-fused image of the target area; otherwise, geometric correction needs to be performed on the pre-fused image of the target area. The specific process is as follows: Use the feature point matching algorithm to perform geometric correction on the pre-fused image of the target area to obtain the geometrically corrected image of the target area, and evaluate the geometric correction effect, and perform corresponding geometric correction adjustment processing according to the evaluation result.

[0060] It should be explained that the feature point matching error between the pre-fused image of the target area and the reference image can be obtained by calculating the sum of the squares of the coordinate differences of all matching feature point pairs, then taking the average value, and finally taking the square root, that is, the root mean square error.

[0061] It should be explained that the pixel value difference of the overlapping area between the pre-fused image of the target area and the reference image can be obtained by calculating the sum of the squares of the differences between the corresponding pixel values of the pre-fused image and the reference image in the overlapping area, and then dividing by the total number of pixels in the overlapping area, that is, the mean square error.

[0062] Furthermore, to evaluate the geometric correction effect, the specific process is as follows: Obtain the feature point matching error, the pixel value difference of the overlapping area, the structural similarity, and the average error of the checkpoints not participating in the correction between the geometrically corrected image of the target area and the reference image. Compare the feature point matching error, the pixel value difference of the overlapping area, the structural similarity, and the average error of the checkpoints not participating in the correction between the geometrically corrected image of the target area and the reference image with the extreme value of the feature point matching error, the extreme value of the pixel value difference of the overlapping area, the extreme value of the structural similarity, and the extreme value of the average error of the checkpoints not participating in the correction respectively, and set corresponding weight factors to reflect the relative importance of each index when evaluating the geometric correction effect, and then perform coupling processing to obtain the geometric correction effect evaluation index.

[0063] It should be explained that in this embodiment, the structural similarity is obtained by the system calculating the luminance comparison function, the contrast comparison function, and the structure comparison function to obtain the final similarity value.

[0064] It should be understood that for obtaining the structural similarity, the system adopts a decomposition method similar to the structural similarity index. The purpose of the brightness comparison function is to measure the average brightness difference of the local regions of two images. By calculating the average brightness values of the corresponding local windows of the two images to be compared (denoted as μk and μy), and then obtaining the brightness similarity value through the formula L(k,y)=(2×μk×μy + C1) / (μk2 + μy2 + C1), where C1 is a constant to prevent the denominator from being zero.

[0065] The purpose of the contrast comparison function is to measure the similarity of the contrast within the local regions of two images. Calculate the standard deviations of the local windows of the two images respectively (denoted as σk and σy), and use the formula C(k,y)=(2×σk×σy + C2) / (σk2 + σy2 + C2) to obtain the contrast similarity value, where C2 is also a small constant to prevent division by zero.

[0066] The purpose of the structure comparison function is to reflect the similarity of the local structural information of two images. Calculate the covariance of the corresponding local windows of the two images (denoted as σky), combine their respective standard deviations, and obtain the structure similarity value through the formula S(k,y)=(σky + C3) / (σk×σy + C3). Usually, the constant C3 can be set to C2 / 2 to maintain numerical consistency with the contrast comparison function.

[0067] Finally, multiply the calculation results of the above three comparison functions, that is, SSIM(x,y)=L(x,y)×C(x,y)×S(x,y) to obtain the final index reflecting the comprehensive similarity of the two images in terms of brightness, contrast, and structure. It should be noted that the above calculation method draws on the basic idea of the SSIM index in principle, but specific parameters (such as the values of C1, C2, C3) and window sizes can be optimized and adjusted according to system requirements and actual application scenarios to achieve the best image quality evaluation effect.

[0068] It should be explained that in this embodiment, the extreme value of the feature point matching error, the extreme value of the pixel value difference in the overlapping region, the average error extreme value of the check point, and the extreme value of the structural similarity are all extracted from the system database.

[0069] Extract the geometric correction effect index threshold from the system database, compare the geometric correction effect evaluation index with the geometric correction effect index threshold. If the geometric correction effect evaluation index is higher than or equal to the geometric correction effect index threshold, input the geometric correction image of the target area into the image fusion module. If the geometric correction effect evaluation index is lower than the geometric correction effect index threshold, perform secondary correction on the geometric correction image of the target area.

[0070] In a specific embodiment, the numerical expression of the geometric correction effect evaluation index is:

[0071]

[0072] In the formula, JX represents the geometric correction effect evaluation index, e represents the natural constant, and α 1 represents the feature point matching error between the geometrically corrected image of the target area and the reference image, and α 2 represents the pixel value difference in the overlapping area between the geometrically corrected image of the target area and the reference image, and α 4 represents the structural similarity between the geometrically corrected image of the target area and the reference image, and α 3 represents the average error of the checkpoints not involved in the correction between the geometrically corrected image of the target area and the reference image, and Δα 1 represents the extreme value of the feature point matching error, and Δα 2 represents the extreme value of the pixel value difference in the overlapping area, and Δα 4 represents the extreme value of the structural similarity, and Δα 3 represents the extreme value of the average error of the checkpoints. σ1 represents the weight factor of the geometric correction effect corresponding to the preset feature point matching error, σ2 represents the weight factor of the geometric correction effect corresponding to the preset pixel value difference in the overlapping area, σ3 represents the weight factor of the geometric correction effect corresponding to the preset average error of the checkpoints, and σ4 represents the weight factor of the geometric correction effect corresponding to the preset structural similarity.

[0073] It should be noted that the average error of the checkpoints not involved in the correction between the geometrically corrected image of the target area and the reference image can be obtained by calculating the sum of the squares of the coordinate differences of all pairs of checkpoints not involved in the correction, then taking the average, and finally taking the square root, that is, the root mean square error of the coordinates of all pairs of checkpoints not involved in the correction.

[0074] It should be noted that a high feature point matching error will lead to an increase in the pixel value difference in the overlapping area. If the feature points are not accurately aligned, even subsequent correction processes cannot completely eliminate this inconsistency, resulting in an increase in the pixel value difference in the overlapping area. The feature point matching error directly affects the structural similarity. If the feature point matching is inaccurate, the corrected image will be distorted or deformed, reducing the overall structural similarity. The pixel value difference will directly affect the structural similarity. Even if the overall structure of the image remains the same, but if the pixel value difference is too large, the visual inconsistency will still affect the structural similarity. A high average error indicates that the correction model fails to fit the entire image well, which usually leads to poor feature point matching error, pixel value difference, and structural similarity. The corresponding geometric correction effect evaluation index is smaller, indicating that the geometric correction effect on the pre-fused image of the target area is worse.

[0075] It should be noted that in this embodiment, σ1 represents the weight factor of the geometric correction effect corresponding to the preset feature point matching error, σ2 represents the weight factor of the geometric correction effect corresponding to the preset pixel value difference in the overlapping area, σ3 represents the weight factor of the geometric correction effect corresponding to the preset average error of the check points, and σ4 represents the weight factor of the geometric correction effect corresponding to the preset structural similarity. They respectively represent the numerical values of the influence degrees of the unit numerical values of the feature point matching error, the pixel value difference in the overlapping area, the average error of the check points, and the structural similarity on the image geometric correction effect. When used, they can be directly obtained from the system database, and their corresponding relationships can be pre-set mapping relationships. For example, the feature point matching error, the pixel value difference in the overlapping area, the average error of the check points, and the structural similarity respectively form a mapping set with the weight factors of the geometric correction effects corresponding to the preset feature point matching error, the pixel value difference in the overlapping area, the average error of the check points, and the structural similarity in the system database. The real-time feature point matching error, the pixel value difference in the overlapping area, the average error of the check points, and the structural similarity are input into the mapping set to obtain the corresponding weight factors of the geometric correction effects. The mapping relationships therein can be one-to-one or many-to-one relationships. The value ranges of the above influence weight factors are all between 0 and 1.

[0076] Further, perform secondary correction on the geometrically corrected image of the target area. The specific process is as follows: Mark the difference between the geometric correction effect evaluation index and the geometric correction effect index threshold as the geometric correction demand adjustment parameter. Extract the number of additional feature points corresponding to the geometric correction demand adjustment parameter from the system database. After adding the number of additional feature points to the initial number of feature points, obtain the number of feature points for secondary correction constraint, and use it as the secondary correction constraint condition of the geometric correction model to perform correction and then analyze again to obtain the geometric correction effect evaluation index, which is recorded as the secondary geometric correction effect evaluation index.

[0077] Compare the secondary geometric correction effect evaluation index with the geometric correction effect index threshold. If the secondary geometric correction effect evaluation index is higher than or equal to the geometric correction effect index threshold, input the image of the target area after secondary geometric correction into the image fusion module. If the secondary geometric correction effect evaluation index is still lower than the geometric correction effect index threshold, it is necessary to switch the geometric correction model to perform re-geometric correction on the image of the target area after secondary geometric correction.

[0078] It should be noted that this system stores multiple geometric correction models. Based on the initial geometric correction model, sort the higher-order geometric correction models in ascending order, and sequentially switch to the higher-order geometric correction models in the sequence to perform re-geometric correction on the image of the target area after secondary geometric correction.

[0079] It should be noted that in this embodiment, the geometric correction model is a mathematical model used to correct geometric deformations in images, including: polynomial correction model, collinearity equation correction model, rational function model, and thin plate spline function correction model; among them, the polynomial correction model describes the relationship between image coordinates and actual geographic coordinates by establishing a polynomial function. This model is simple and flexible, applicable to various types of geometric deformations, but for complex terrains and deformations, higher-order polynomials may be required for accurate fitting. The collinearity equation correction model is based on the collinearity condition equation of photogrammetry, taking into account factors such as the position and attitude of the sensor and the earth's curvature, and can accurately describe the geometric relationship between the image and the ground. It is usually used for the correction of aerospace remote sensing images, with high accuracy, but requires the relevant parameters of the sensor to be known. The rational function model approximately describes the relationship between image coordinates and geographic coordinates using a rational polynomial function. The rational function model has high generality and flexibility, does not require an accurate sensor physical model, is applicable to various types of remote sensing data, and has been widely used in practical applications. The thin plate spline function correction model is an interpolation method based on spline functions, which can adaptively adjust the interpolation function according to the distribution of control points, and has a good correction effect for images with large local deformations. It is often used to process images with non-linear deformations, such as medical images or remote sensing images that have undergone complex transformations.

[0080] It should be noted that the polynomial correction model belongs to the basic function model. It can handle some simple geometric deformations, such as translation, rotation, scaling, and mild non-linear deformations. In cases where the accuracy requirement is not high, or the image deformation is relatively regular, the polynomial correction model can quickly and effectively complete the correction task. For example, for some images that have been simply scanned or mildly geometrically transformed, the polynomial correction model can meet the basic correction requirements.

[0081] The rational function model can be classified as an intermediate function model. The rational function model can perform relatively accurate corrections on various types of remote sensing data without the need for an accurate sensor physical model. It is applicable to most conventional remote sensing image correction tasks, can handle geometric deformations caused by relatively complex terrains and sensor attitude changes, and achieves a good balance between accuracy and generality.

[0082] The collinearity equation correction model and the thin plate spline function correction model belong to the advanced function models. The collinearity equation correction model is based on strict photogrammetry principles, taking into account various complex factors, and can achieve high-precision geometric correction, especially suitable for the correction of aerospace remote sensing images with extremely high accuracy requirements. The thin plate spline function correction model, on the other hand, has unique advantages in dealing with local non-linear deformations, can perform fine corrections for complex deformations in the image, and is often used for medical image processing or special remote sensing application scenarios with high requirements for image quality.

[0083] Specifically, to determine whether the pre-fusion image of the target area needs radiometric calibration, the specific process is as follows: Obtain the histogram similarity and the average brightness difference between the pre-fusion image of the target area and the reference image. Compare the histogram similarity between the pre-fusion image of the target area and the reference image with the histogram similarity threshold, and compare the average brightness difference between the pre-fusion image of the target area and the reference image with the average brightness difference threshold. When both the histogram similarity and the average brightness difference between the pre-fusion image of the target area and the reference image are lower than the thresholds, there is no need to perform radiometric calibration on the pre-fusion image of the target area; otherwise, radiometric calibration is required for the pre-fusion image of the target area. The specific process is as follows: Use radiometric calibration to perform radiometric calibration on the pre-fusion image of the target area to obtain the radiometrically calibrated image of the target area, and evaluate the radiometric calibration effect, and perform corresponding radiometric calibration adjustment processing according to the evaluation result.

[0084] It should be explained that in this embodiment, the average brightness difference is the difference in the average brightness values of all pixels in the pre-fusion image of the target area and the reference image, and the histogram similarity is used to measure the similarity between the histograms of the pre-fusion image of the target area and the reference image, which can be represented by calculating the Bhattacharyya distance between the histograms of the pre-fusion image of the target area and the reference image.

[0085] Furthermore, to evaluate the radiometric calibration effect, the specific process is as follows: Obtain the histogram similarity, average brightness difference, structural similarity, land surface reflectance difference, and signal-to-noise ratio of the radiometrically calibrated image of the target area between the radiometrically calibrated image of the target area and the reference image. Compare the histogram similarity, average brightness difference, structural similarity, land surface reflectance difference, and signal-to-noise ratio of the radiometrically calibrated image of the target area between the radiometrically calibrated image of the target area and the reference image with the histogram similarity extreme value, average brightness difference extreme value, structural similarity extreme value, land surface reflectance difference extreme value, and signal-to-noise ratio extreme value respectively, and set corresponding weight factors to reflect the relative importance of each index when evaluating the radiometric calibration effect, and then perform coupling processing to obtain the radiometric calibration effect evaluation index.

[0086] It should be explained that in this embodiment, the land surface reflectance difference refers to the difference in the land surface reflectance corresponding to the radiometrically calibrated image of the target area and the reference image. First, according to the radiance value of the image and relevant radiometric calibration parameters, convert the pixel values in the image into land surface reflectance. Then, for the areas of the same land surface type in the two images, calculate the difference in their reflectance. The signal-to-noise ratio is the ratio of the signal power to the noise power. The signal-to-noise ratio of the radiometrically calibrated image of the target area reflects the relative intensity of the useful signal and noise in the image.

[0087] It should be noted that in this embodiment, the extreme value of the average brightness difference, the extreme value of the ground object reflectivity difference, the extreme value of the signal-to-noise ratio, the extreme value of the histogram similarity, and the extreme value of the structural similarity are all extracted from the system database.

[0088] The threshold of the radiation correction effect index is extracted from the system database, and the radiation correction effect evaluation index is compared with the radiation correction effect index threshold. If the radiation correction effect evaluation index is higher than or equal to the radiation correction effect index threshold, the radiation correction image of the target area is input into the image fusion module. If the radiation correction effect evaluation index is lower than the radiation correction effect index threshold, the radiation correction image of the target area is corrected twice.

[0089] In a specific embodiment, the numerical expression of the radiation correction effect evaluation index is:

[0090]

[0091] In the formula, FX represents the radiation correction effect evaluation index, e represents the natural constant, β 4 represents the histogram similarity between the radiation correction image of the target area and the reference image, β 1 represents the average brightness difference between the radiation correction image of the target area and the reference image, β 5 represents the structural similarity between the radiation correction image of the target area and the reference image, β 2 represents the ground object reflectivity difference between the radiation correction image of the target area and the reference image, β 3 represents the signal-to-noise ratio of the radiation correction image of the target area, Δβ 4 represents the extreme value of the histogram similarity, Δβ 1 represents the extreme value of the average brightness difference, Δβ 5 represents the extreme value of the structural similarity, Δβ 2 represents the extreme value of the ground object reflectivity difference, Δβ 3 represents the extreme value of the signal-to-noise ratio, σ5 represents the radiation correction effect influence weight factor corresponding to the preset average brightness difference, σ6 represents the radiation correction effect influence weight factor corresponding to the preset ground object reflectivity difference, σ7 represents the radiation correction effect influence weight factor corresponding to the preset signal-to-noise ratio, σ8 represents the radiation correction effect influence weight factor corresponding to the preset histogram similarity, and σ9 represents the radiation correction effect influence weight factor corresponding to the preset structural similarity.

[0092] It should be explained that a low histogram similarity is accompanied by a large average brightness difference. This is because the histogram reflects the overall brightness distribution of an image. If the histograms of two images are significantly different, their average brightnesses also tend to be different. A low histogram similarity also leads to a decrease in structural similarity. Since the change in brightness distribution affects the local contrast and detail performance of the image, it reduces the overall structural consistency. The average brightness difference directly affects the structural similarity. Even if the overall structure of the image remains consistent, if the brightness difference is too large, the visual inconsistency will still affect the structural similarity. A large difference in ground object reflectance means that the ground object features in the image are not accurately aligned, resulting in a decrease in structural similarity. Especially in remote sensing applications, the accuracy of reflectance is crucial for ground object recognition. A high signal-to-noise ratio usually means better image quality. However, during the radiation correction process, if the histogram similarity, average brightness difference, structural similarity, and ground object reflectance difference perform poorly, a high signal-to-noise ratio may mean that too much noise has been introduced during the correction process or the noise has not been effectively removed. A high signal-to-noise ratio masks the problem of reflectance difference. However, if the reflectance difference is large, it indicates that the true reflectance characteristics of the ground objects have not been effectively maintained during the correction process, despite the low noise level. The lower the histogram similarity and structural similarity, and the higher the average brightness difference, ground object reflectance difference, and image signal-to-noise ratio, the smaller the corresponding radiation correction effect evaluation index, indicating that the radiation correction effect of the pre-fused image of the target area is worse. By comprehensively considering these parameters and their interrelationships, the radiation correction effect can be more comprehensively evaluated and optimized to ensure high-quality correction of the pre-fused image of the target area.

[0093] It should be noted that in this embodiment, σ5 represents the weight factor of the radiation correction effect corresponding to the preset average brightness difference, σ6 represents the weight factor of the radiation correction effect corresponding to the preset difference in ground object reflectivity, σ7 represents the weight factor of the radiation correction effect corresponding to the preset signal-to-noise ratio, σ8 represents the weight factor of the radiation correction effect corresponding to the preset histogram similarity, and σ9 represents the weight factor of the radiation correction effect corresponding to the preset structural similarity. They respectively represent the numerical values of the influence degree of the unit values of the average brightness difference, the difference in ground object reflectivity, the signal-to-noise ratio, the histogram similarity, and the structural similarity on the image radiation correction effect, and can be directly obtained from the system database during use. Their corresponding relationships can be pre-set mapping relationships. For example, the average brightness difference, the difference in ground object reflectivity, the signal-to-noise ratio, the histogram similarity, and the structural similarity respectively form a mapping set with the weight factors of the radiation correction effect corresponding to the preset average brightness difference, the difference in ground object reflectivity, the signal-to-noise ratio, the histogram similarity, and the structural similarity in the system database. The real-time average brightness difference, the difference in ground object reflectivity, the signal-to-noise ratio, the histogram similarity, and the structural similarity are input into the mapping set to obtain the corresponding weight factors of the radiation correction effect, and the mapping relationships therein can be one-to-one or many-to-one relationships. The value ranges of the above-mentioned weight factors of the influence are all between 0 and 1.

[0094] Furthermore, perform secondary correction on the radiation correction image of the target area. The specific analysis process is as follows: Mark the difference between the radiation correction effect evaluation index and the radiation correction effect index threshold as the radiation correction requirement adjustment parameter. Extract the requirement adjustment calibration coefficient corresponding to the radiation correction requirement adjustment parameter from the system database. Replace the initial calibration coefficient with the requirement adjustment calibration coefficient and use it as the secondary correction constraint condition of the radiation correction model for correction, and then analyze again to obtain the radiation correction effect evaluation index, which is recorded as the secondary radiation correction effect evaluation index.

[0095] Compare the secondary radiation correction effect evaluation index with the radiation correction effect index threshold. If the secondary radiation correction effect evaluation index is higher than or equal to the radiation correction effect index threshold, input the image of the target area after secondary radiation correction into the image fusion module. If the secondary radiation correction effect evaluation index is still lower than the radiation correction effect index threshold, it is necessary to switch the radiation correction model to perform re-radiation correction on the image of the target area after secondary radiation correction.

[0096] It should be noted that this system stores multiple radiation correction models. Based on the initial radiation correction model, sort the higher-order radiation correction models in ascending order, and sequentially switch to the higher-order radiation correction models in the sequence to perform re-radiation correction on the image of the target area after secondary radiation correction.

[0097] It should be explained that the radiation correction model is used to correct the radiation errors in remote sensing images caused by factors such as sensor characteristics and atmospheric conditions, including: histogram matching method, regression analysis method, atmospheric correction model, solar angle correction model, flat field method, and internal average relative reflectance method; among them, the histogram matching method is to match the histogram of the image to be corrected with the histogram of the reference image to make their gray-scale distributions similar, so as to adjust the radiation value of the image. This method is simple and easy to implement, and can quickly improve the overall brightness and contrast of the image, but the correction effect on the complex radiation differences in the image is limited. The regression analysis method refers to using known radiation correction sources or ground measured data to establish a regression relationship between the image radiation value and the true radiation value, and then correcting the image. It can take into account the statistical relationship between the image radiation value and the actual physical quantity to a certain extent, but it depends on accurate sample data and has weak processing ability for complex radiation errors. Atmospheric correction models, such as the 6S model: Second Simulation of the Satellite Signal in the Solar Spectrum, that is, the second simulation model of satellite signal in the solar spectrum; the MODTRAN model: Moderate Resolution Transmittance Model, that is, the medium-resolution transmittance model; the ATCOR model: ATmospheric CORrection, that is, the atmospheric correction model, which is specially developed for the atmospheric correction of satellite and aerial remote sensing data; the FLAASH model: Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes, that is, the fast line-of-sight atmospheric analysis model of spectral hypercubes, which is mainly used for the atmospheric correction of hyperspectral data. These models are based on the theory of atmospheric radiation transfer, consider the absorption, scattering, etc. of electromagnetic radiation by the atmosphere, and simulate the influence of the atmosphere on radiation by inputting atmospheric parameters, sensor parameters, and surface parameters, etc., so as to realize the atmospheric correction of remote sensing images and restore the true reflectance or radiation brightness of ground objects. Such models have clear physical meanings and relatively high correction accuracies, but they require accurate atmospheric and surface parameters, and the calculation process is relatively complex. The solar angle correction model is used to correct the radiation differences of ground objects caused by the changes in the solar altitude angle and azimuth angle. By calculating the angle between the sun's rays and the ground, the radiation values in the image are normalized, so that the images obtained at different times and in different seasons are comparable. This model can effectively correct the radiation changes caused by the solar angle, but the correction effect on the radiation errors caused by other factors is not obvious. The flat field method refers to assuming that there is one or more uniform flat field areas in the image, whose reflectance or radiation characteristics are known and constant, and estimating and correcting the radiation errors in the image through the analysis of the flat field areas.This method is applicable to images with obvious flat-field regions. However, if the flat-field region is not properly selected or there are errors, it will affect the correction effect. The internal average relative reflectance method refers to eliminating the systematic radiation error in the image by calculating the relative reflectance relationship between different ground objects within the image. It does not require external reference data but uses the information of the image itself for correction. However, it has certain requirements for the scene content of the image, and the correction accuracy is relatively limited.

[0098] It should be explained that the histogram matching method and the internal average relative reflectance method can be classified as basic functional models. The histogram matching method mainly improves the visual effect by adjusting the gray distribution of the image, and the correction of radiation error is relatively rough. The internal average relative reflectance method uses the internal information of the image for relative correction, with limited correction accuracy and insufficient ability to handle complex radiation errors. They are usually applicable to situations where the requirement for correction accuracy is not high, or only preliminary radiation adjustment of the image is needed, such as quickly browsing images or general applications with relatively low requirements for image quality.

[0099] The regression analysis method and the flat-field method belong to intermediate functional models. The regression analysis method establishes the relationship between the radiation value and the true value with the help of sample data, and can consider the statistical law of radiation to a certain extent. The flat-field method uses the known flat-field region for radiation correction, with certain pertinence and accuracy. These two methods are applicable to some scenarios where there are certain requirements for correction accuracy but the parameters required for complex physical models are not available, such as relatively simple radiation correction of remote sensing images when detailed atmospheric parameters are lacking.

[0100] The atmospheric correction model and the solar angle correction model belong to advanced functional models. The atmospheric correction model is based on strict physical principles, can comprehensively consider various effects of the atmosphere on radiation, and accurately invert the true radiation characteristics of ground objects. The solar angle correction model accurately corrects for the important factor of the solar angle, and has a significant effect on eliminating radiation differences caused by changes in the solar position. They are applicable to application scenarios such as scientific research and quantitative analysis with high requirements for radiation accuracy, such as ecological environment monitoring, geological and mineral exploration, etc., where accurate acquisition of the true radiation information of ground objects is required.

[0101] It should be explained that in this embodiment, by improving the geometric accuracy of the image and optimizing the radiation characteristics, high-quality image data is provided for subsequent image fusion and recognition. The improvement of image quality enables more accurate extraction of the characteristics of waste when identifying the type of waste, reducing the occurrence of misjudgment and missed judgment, thus significantly improving the accuracy and reliability of waste landscape detection in the target area, and providing more reliable data support for waste treatment and landscape transformation.

[0102] It should be noted that in this embodiment, the geometric correction sub-module can determine whether the pre-fusion image of the target area needs geometric correction. It compares the feature point matching error and the pixel value difference in the overlapping area between the pre-fusion image and the reference image. When these values exceed the threshold, the feature point matching algorithm is used to correct the image. This can effectively correct the geometric distortion of the image caused by factors such as shooting angle and terrain, making the shape, position, and size of the waste in the image more accurate, avoiding misjudgment of the waste caused by geometric distortion, and laying a foundation for accurately identifying the waste subsequently. In addition, the effect of geometric correction will be evaluated. If the effect does not meet the standard, secondary correction will be performed to ensure that the image meets the detection requirements. The radiometric correction sub-module will determine whether the pre-fusion image of the target area needs radiometric correction. It compares the histogram similarity and the average brightness difference between the pre-fusion image and the reference image. When the difference between the two is large, radiometric calibration is used for correction. This can solve the problem of image radiometric difference caused by factors such as lighting conditions and sensor response differences, making the image brightness and color more uniform, enhancing the contrast between the waste and the background, making the features of the waste clearer, and facilitating identification. Similarly, the effect of radiometric correction is evaluated. If the effect is not good, secondary correction is performed to ensure the quality of the image after radiometric correction.

[0103] The image fusion module is used to fuse the specified fusion images of the target area to obtain the multi-source data fusion image of the target area.

[0104] Specifically, the wavelet transform algorithm is used to process each input image, converting them from the spatial domain to the frequency domain. In this process, the image is decomposed into multiple sub-bands, including a low-frequency sub-band and multiple high-frequency sub-bands. The low-frequency sub-band captures the main structure and contour information of the image, while the high-frequency sub-bands contain the detailed information of the image, such as edges and textures.

[0105] It should be noted that for the high-frequency sub-bands that reflect high-frequency information such as the details of the waste, by comparing the absolute values of the coefficients in the corresponding high-frequency sub-bands of each image, the coefficient with the larger absolute value is selected as the fused coefficient. This is because the coefficient with the larger absolute value usually corresponds to the important detailed features in the image, and in this way, the detailed information in the image can be retained and enhanced.

[0106] For the low-frequency sub-bands that reflect the overall contour of the image, fusion is performed by means of weighted average or the like. Weighted average can assign different weights to the low-frequency sub-band coefficients of each image according to the characteristics or importance of each image, and then calculate the weighted average value as the fused low-frequency sub-band coefficient. This method can integrate the overall information of each image and avoid losing important contour information during the fusion process.

[0107] After completing the coefficient fusion in the frequency domain, the fused frequency-domain data is converted back to the spatial domain through the inverse wavelet transform. The inverse transform process reconstructs the fused image based on the fused sub-band coefficients, that is, the multi-source data fusion image of the target area. The image obtained in this way combines the advantages of the original multiple images, retains the overall contour information, highlights the detail features, and is more conducive to subsequent analysis and processing of the target area, such as applications like waste identification and monitoring.

[0108] The comprehensive feedback module is used to perform image recognition on the multi-source data fusion image of the target area to obtain the waste type and transmit it to a preset display window for reminder.

[0109] Specifically, the comprehensive feedback module uses the Faster R-CNN deep learning model to quickly and accurately identify different types of waste based on the multi-source data fusion image of the target area, and transmit it to a preset display window for reminder. At the same time, data analysis is performed on the identified waste types, information such as the quantity and distribution area of different types of waste is statistically analyzed, and a detailed detection report is generated.

[0110] It should be explained that the waste types distinguished according to the waste landscaping detection method in this embodiment include: 1. Physical property waste, including wood, metal, stone, plastic, etc., which are mainly evaluated by physical properties such as size, shape, density, and hardness. 2. Chemical composition waste, including waste containing heavy metals (such as lead, mercury) or organic pollutants (such as polychlorinated biphenyls, dioxins), and the content of harmful substances needs to be determined through chemical analysis. 3. Biological safety waste, mainly organic waste such as plant residues and kitchen waste, and it is necessary to detect whether there are pathogenic microorganisms or pests. 4. Durability and stability waste, such as building material waste like bricks, concrete blocks, and tiles, and the compressive strength, weather resistance, and corrosion resistance need to be tested. 5. Environmentally impactful waste, including composite material waste such as waste electronic products and textiles, and its degradability and impact on soil and water bodies need to be evaluated. 6. Aesthetic and functional waste, such as highly malleable waste like glass bottles and ceramic fragments, and aesthetic features such as color, transparency, and surface texture and their functional adaptability need to be considered.

[0111] It should be noted that this application constructs a complete detection process from image acquisition, preprocessing, correction, fusion to feedback, effectively solving the problem of inaccurate recognition caused by poor image quality in traditional detection, improving the accuracy, reliability and efficiency of detection, and providing strong support for the landscape treatment of waste. Among them, the image acquisition module obtains multi-angle images, and the image pre-fusion module screens and preprocesses the images according to information such as noise and contrast, removes noise, adjusts brightness and contrast, and improves clarity. The geometric and radiometric correction sub-modules of the image correction module correct the geometric deformation and radiometric differences of the images respectively, enhancing the distinction between waste and the background. By correcting the image deviation, the system reduces the image error caused by shooting factors. Each module works together, and the comprehensive feedback module identifies the waste type based on the high-quality fused image, reducing misjudgment and missed judgment, making the detection result more accurate. The pre-fusion module screens images to reduce the processing of invalid data. Each module operates efficiently, and the correction module accurately evaluates the correction effect to avoid overprocessing, improving the overall detection efficiency of the system.

[0112] Referring to Figure 2 As shown, the second aspect of the present invention provides a waste landscape detection method. S1: Mark the waste landscape detection area as the target area, and collect multi-angle images of the target area.

[0113] S2: Perform pre-fusion on the multi-angle images of the target area to obtain a pre-fused image of the target area.

[0114] S3: Determine whether the pre-fused image of the target area needs to be corrected, and obtain a specified fused image of the target area.

[0115] S4: Perform fusion on the pre-fused image of the target area after correction to obtain a multi-source data fused image of the target area.

[0116] S5: Perform image recognition on the multi-source data fused image of the target area to obtain the waste type, and transmit it to a preset display window for reminder.

[0117] The above content is only an example and explanation of the structure of the present invention. Those skilled in the art of this technology can make various modifications or supplements to the described specific embodiments 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 by this claim book, they should all belong to the protection scope of the present invention.

Claims

1. A waste landscape detection system, characterized in that Including: An image acquisition module, which is used to mark the waste landscape detection area as the target area and acquire multi-angle images of the target area; An image pre-fusion module, which is used to pre-fuse the multi-angle images of the target area to obtain a pre-fused image of the target area. The system uses a weighted average algorithm to assign weights to each angle image according to the image quality requirement processing index for each angle. The pixels at the same spatial position in each angle image are calculated by weighted average to generate a pre-fused image of the target area; An image correction module, which is used to determine whether the pre-fused image of the target area needs to be corrected and obtain a specified fused image of the target area. The image correction module includes a geometric correction sub-module and a radiation correction sub-module. The geometric correction sub-module is used to determine whether the pre-fused image of the target area needs geometric correction, and the radiation correction sub-module is used to determine whether the pre-fused image of the target area needs radiation correction; An image fusion module, which is used to fuse the specified fused image of the target area to obtain a multi-source data fused image of the target area; A comprehensive feedback module, which is used to perform image recognition on the multi-source data fused image of the target area to obtain the waste type and transmit it to a preset display window for reminder; When obtaining the pre-fused image of the target area, it further includes: when it is found that the image quality requirement processing index of a certain angle image is greater than the image quality requirement processing index threshold, the system immediately issues a warning and displays that the image needs to be pre-processed, that is, denoising, adjusting brightness, contrast and sharpening the image of this angle, and then re-performing a loop judgment. If it is not lower than or equal to the image quality requirement processing index threshold within the preset number of times, the image of this angle is marked as an abnormal image and the image of this angle is re-acquired.

2. The waste landscape detection system according to claim 1, wherein: The specific process of pre-fusing the multi-angle images of the target area to obtain a pre-fused image of the target area is as follows: Obtain the basic information of the multi-angle images of the target area, including the noise level, contrast, brightness and blur degree of each angle image. Analyze the differences between the noise level, brightness, contrast and blur degree of each angle image and the noise level critical value, brightness critical value, contrast critical value and blur degree critical value respectively to obtain the deviation degree, and introduce their corresponding weights respectively and perform coupling processing to obtain the image quality requirement processing index of each angle image; Based on the image quality requirement processing index of each angle image, screen out the required processing images of each angle, perform pre-processing, and then perform pre-fusion processing to obtain a pre-fused image of the target area.

3. The waste landscape detection system according to claim 2, characterized in that: The specific process of screening out the required processing images of each angle based on the image quality requirement processing index of each angle image, performing pre-processing, and then performing pre-fusion processing to obtain a pre-fused image of the target area is as follows: Extract the threshold of the image quality requirement processing index from the system database. Compare the image quality requirement processing index of each angle with the threshold of the image quality requirement processing index. If the image quality requirement processing index of a certain angle is greater than the threshold of the image quality requirement processing index, mark the image of that angle as the image to be processed for angle requirements and perform preprocessing. If the image quality requirement processing index of a certain angle is less than or equal to the threshold of the image quality requirement processing index, there is no need to perform pre-fusion on the image of that angle; After preprocessing, count all the angle images for pre-fusion to obtain the pre-fused image of the target area.

4. The waste landscape detection system according to claim 1, characterized in that: Determine whether the pre-fused image of the target area needs geometric correction. The specific process is as follows: Extract the geographical image of the target area, the threshold of feature point matching error, and the threshold of pixel value difference in the overlapping area from the system database. Mark the geographical image of the target area as the reference image of the target area. Obtain the feature point matching error and the pixel value difference in the overlapping area between the pre-fused image of the target area and the reference image. Compare the feature point matching error between the pre-fused image of the target area and the reference image with the threshold of feature point matching error, and compare the pixel value difference in the overlapping area between the pre-fused image of the target area and the reference image with the threshold of pixel value difference in the overlapping area. When both the feature point matching error and the pixel value difference in the overlapping area between the pre-fused image of the target area and the reference image are lower than the corresponding thresholds, there is no need to perform geometric correction on the pre-fused image of the target area. Otherwise, geometric correction needs to be performed on the pre-fused image of the target area. The specific process is as follows: Use the feature point matching algorithm to perform geometric correction on the pre-fused image of the target area to obtain the geometrically corrected image of the target area, and evaluate the geometric correction effect, and perform corresponding geometric correction adjustment processing according to the evaluation result.

5. The waste landscape detection system according to claim 4, characterized in that: Evaluate the geometric correction effect. The specific process is as follows: Obtain the feature point matching error, the pixel value difference in the overlapping area, the structural similarity, and the average error of the checkpoints not participating in the correction between the geometrically corrected image of the target area and the reference image. Compare the feature point matching error, the pixel value difference in the overlapping area, the structural similarity, and the average error of the checkpoints not participating in the correction between the geometrically corrected image of the target area and the reference image with the extreme value of the feature point matching error, the extreme value of the pixel value difference in the overlapping area, the extreme value of the structural similarity, and the extreme value of the average error of the checkpoints not participating in the correction respectively, and set corresponding weight factors to reflect the relative importance of each index when evaluating the geometric correction effect, and then perform coupling processing to obtain the geometric correction effect evaluation index; Extract the threshold of the geometric correction effect index from the system database. Compare the geometric correction effect evaluation index with the threshold of the geometric correction effect index. If the geometric correction effect evaluation index is higher than or equal to the threshold of the geometric correction effect index, input the geometrically corrected image of the target area into the image fusion module. If the geometric correction effect evaluation index is lower than the threshold of the geometric correction effect index, perform secondary correction on the geometrically corrected image of the target area.

6. The waste landscape detection system according to claim 5, characterized in that: Perform secondary correction on the geometrically corrected image of the target area. The specific process is as follows: The difference between the geometric correction effect evaluation index and the geometric correction effect index threshold is marked as the geometric correction demand adjustment parameter. The number of demand-increasing feature points corresponding to the geometric correction demand adjustment parameter is extracted from the system database. After adding the number of demand-increasing feature points to the initial number of feature points, the number of secondary correction constraint feature points is obtained, which is used as the secondary correction constraint condition of the geometric correction model. After correction and re-analysis, the geometric correction effect evaluation index is recorded as the secondary geometric correction effect evaluation index; Compare the secondary geometric correction effect evaluation index with the geometric correction effect index threshold. If the secondary geometric correction effect evaluation index is higher than or equal to the geometric correction effect index threshold, the image of the target area after secondary geometric correction is input into the image fusion module. If the secondary geometric correction effect evaluation index is still lower than the geometric correction effect index threshold, the geometric correction model needs to be switched to re-geometric correct the image of the target area after secondary geometric correction.

7. The waste landscape detection system according to claim 1, characterized in that: The process of determining whether the pre-fusion image of the target area needs radiometric correction is as follows: Obtain the histogram similarity and average brightness difference between the pre-fusion image of the target area and the reference image. Compare the histogram similarity between the pre-fusion image of the target area and the reference image with the histogram similarity threshold, and compare the average brightness difference between the pre-fusion image of the target area and the reference image with the average brightness difference threshold. When both the histogram similarity and the average brightness difference between the pre-fusion image of the target area and the reference image are lower than the corresponding thresholds, there is no need to perform radiometric correction on the pre-fusion image of the target area. Otherwise, radiometric correction needs to be performed on the pre-fusion image of the target area. The specific process is as follows: Use radiometric calibration to perform radiometric correction on the pre-fusion image of the target area to obtain the radiometric correction image of the target area, and evaluate the radiometric correction effect, and perform corresponding radiometric correction adjustment processing according to the evaluation result.

8. The waste landscape detection system according to claim 7, wherein: The process of evaluating the radiometric correction effect is as follows: Obtain the histogram similarity, average brightness difference, structural similarity, land surface reflectance difference, and signal-to-noise ratio of the radiometric correction image of the target area and the reference image. Compare the histogram similarity, average brightness difference, structural similarity, land surface reflectance difference, and signal-to-noise ratio of the radiometric correction image of the target area and the reference image with the histogram similarity extreme value, average brightness difference extreme value, structural similarity extreme value, land surface reflectance difference extreme value, and signal-to-noise ratio extreme value respectively, and set corresponding weight factors to reflect the relative importance of each index when evaluating the radiometric correction effect, and then perform coupling processing to obtain the radiometric correction effect evaluation index; Extract the radiometric correction effect index threshold from the system database. Compare the radiometric correction effect evaluation index with the radiometric correction effect index threshold. If the radiometric correction effect evaluation index is higher than or equal to the radiometric correction effect index threshold, the radiometric correction image of the target area is input into the image fusion module. If the radiometric correction effect evaluation index is lower than the radiometric correction effect index threshold, perform secondary correction on the radiometric correction image of the target area.

9. The waste landscape detection system according to claim 8, wherein: The secondary correction of the radiometrically corrected image of the target area is specifically analyzed as follows: The difference between the radiometric correction effect evaluation index and the radiometric correction effect index threshold is marked as the radiometric correction requirement adjustment parameter. The requirement adjustment calibration coefficient corresponding to the radiometric correction requirement adjustment parameter is extracted from the system database. The requirement adjustment calibration coefficient replaces the initial calibration coefficient and is used as the secondary correction constraint condition of the radiometric correction model. After correction, the radiometric correction effect evaluation index is obtained again and denoted as the secondary radiometric correction effect evaluation index; The secondary radiometric correction effect evaluation index is compared with the radiometric correction effect index threshold. If the secondary radiometric correction effect evaluation index is higher than or equal to the radiometric correction effect index threshold, the image of the target area after secondary radiometric correction is input into the image fusion module. If the secondary radiometric correction effect evaluation index is still lower than the radiometric correction effect index threshold, it is necessary to switch the radiometric correction model to re-radiometrically correct the image of the target area after secondary radiometric correction.

10. A waste landscape detection method, applied to a waste landscape detection system according to any one of claims 1-9, characterized in that: Including: S1. Mark the waste landscape detection area as the target area, and collect multi-angle images of the target area; S2. Pre-fuse the multi-angle images of the target area to obtain the pre-fused image of the target area; S3. Determine whether the pre-fused image of the target area needs to be corrected, and obtain the specified fused image of the target area; S4. Fuse the pre-fused image of the target area after correction to obtain the multi-source data fused image of the target area; S5. Perform image recognition on the multi-source data fused image of the target area to obtain the waste type, and transmit it to the preset display window for reminder.

Images