A method for analyzing pineapple leaf fibers based on hyperspectral image features

Abstract

The application provides a pineapple leaf fiber analysis method based on hyperspectral image features, and relates to the technical field of hyperspectral imaging analysis.The single-frame hyperspectral image of a detection area in a degumming process of pineapple leaf fiber is collected, and a pineapple leaf fiber area is extracted as an image to be analyzed;the pixel points in the image to be analyzed are subjected to spectral linear unmixing, the single-pixel spectral is decomposed into an overlapping form of gum spectral, fiber spectral and spectral residual error, and the spectral contribution coefficients corresponding to the gum and the fiber are solved;the initial boundary of the gum and the fiber is identified and corrected according to the contribution weight;the initial boundary of the gum and the fiber is corrected in combination with the spectral residual error of the non-intersection pixel points, so that the corrected boundary of the gum and the fiber is obtained;the degumming effect of the pineapple leaf fiber is evaluated in combination with the corrected boundary of the gum and the fiber corresponding to the current moment and the previous moment, the degumming boundary is extracted and corrected, and the continuous tracking of the boundary and the objective evaluation of the degumming effect are realized.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral imaging analysis technology, specifically to a method for analyzing pineapple leaf fibers based on hyperspectral image features. Background Technology

[0002] Pineapple leaf fiber has high utilization value in the fields of textiles, composite materials, and environmental protection materials. However, after being extracted from the leaves, pineapple leaf fiber usually needs to undergo degumming to remove the gum components on the fiber surface and between fiber bundles. The degree of degumming directly affects the dispersibility, softness, subsequent spinning adaptability, and finished product stability of pineapple leaf fiber. If degumming is insufficient, a large amount of gum residue will remain on the fiber surface or between fiber bundles, which can easily lead to difficulty in dispersing the fiber bundles and uneven subsequent processing. If the degumming process is not accurately judged, it may cause a lag in the control of the degumming process, affecting the processing quality of pineapple leaf fiber.

[0003] Current methods for assessing the degumming state of pineapple leaf fibers typically rely on manual observation, fixed degumming times, or sampling tests. Manual observation struggles to accurately distinguish between the surface gum state and the fiber's intrinsic state; fixed-time control fails to reflect the actual changes in different batches of pineapple leaf fibers during degumming; and sampling tests cannot continuously reflect the spatial movement of the degumming front. Therefore, existing methods are insufficient for timely and intuitive assessment of the actual progress in the transformation from gum to exposed fiber state during pineapple leaf fiber degumming, and also make it difficult to evaluate the degumming effect based on changes in the degumming boundary at different times.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for analyzing pineapple leaf fibers based on hyperspectral image features, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for analyzing pineapple leaf fibers based on hyperspectral image features, comprising the following steps: Step 1: Acquire a single-frame hyperspectral image of the detection area during the degumming process of pineapple leaf fibers. Use the maximum inter-class variance method to segment the image and extract the pineapple leaf fiber region as the image to be analyzed. Step 2: Screen pixel pairs with significant spectral differences within the image to be analyzed, and combine them with pre-established standard colloidal spectra and standard fiber spectra to determine the feature pixel pairs with the greatest spectral differences within the image to be analyzed; Step 3: Perform spectral linear unmixing on each pixel in the image to be analyzed, decompose the spectrum of a single pixel into the superposition of the gel spectrum, fiber spectrum and spectral residual, and solve for the spectral contribution coefficients of the gel and fiber. Degumming state of each pixel is determined according to the contribution weight, and the pixel at the interface between gel and fiber is identified. Connect all the pixel at the interface between gel and fiber to obtain the initial boundary between gel and fiber. The initial boundary between gel and fiber is corrected by combining the spectral residual of the non-interface pixel to obtain the corrected boundary between gel and fiber. Step 4: Obtain the modified boundaries of the gum and fiber at the current time and the previous time respectively, calculate the spatial movement amplitude of the modified boundaries of the gum and fiber, solve the boundary movement rate by combining the time interval between the two time points, and evaluate the degumming effect of pineapple leaf fiber based on the spatial movement amplitude and boundary movement rate.

[0007] Furthermore, the Otsu's method is used to segment the single-frame hyperspectral image, extracting the pineapple leaf fiber region as the image to be analyzed. The specific logic is as follows: extract the corresponding RGB image from the single-frame hyperspectral image, convert it to grayscale, count the number of pixels at each of the 256 grayscale levels to form a grayscale histogram, and calculate the inter-class variance of each threshold in the grayscale histogram. The threshold corresponding to the maximum inter-class variance is defined as the optimal segmentation threshold. The optimal segmentation threshold is used to segment the single-frame hyperspectral image into background and foreground parts. When segmenting into background and foreground parts, if the grayscale value of a pixel is less than or equal to the optimal segmentation threshold, the pixel is determined to belong to the background part; if it is greater than the optimal segmentation threshold, the pixel is determined to belong to the foreground part. The foreground part is then extracted as the image to be analyzed.

[0008] Furthermore, the specific logic of step 2 is as follows: For each pixel in the image to be analyzed, the spectral angular distance between the single pixel spectrum and the preset standard colloidal spectrum and standard fiber spectrum is calculated, and defined as colloidal angular distance and fiber angular distance respectively. Iterate through all pixel pairs formed by combinations of pixels in the image and filter the validity of each pixel pair: If the gelatinous angle distance of any pixel in a pixel pair is less than its fiber angle distance, the pixel is determined to be biased towards the gelatinous feature pixel. If the fiber angle distance of the other pixel is less than its gelatinous angle distance, the pixel is determined to be biased towards the fiber feature pixel. Then the pixel pair is considered to satisfy the gelatinous end-fiber end correspondence relationship and is a valid feature point pair. Within the entire range of valid pixel pairs, the spectral angular distance values ​​between pixels in each pair are compared, and the pixel pair with the largest spectral angular distance is identified as the feature pixel pair with the greatest spectral difference in the image to be analyzed.

[0009] Furthermore, spectral linear unmixing is performed on each pixel in the image to be analyzed, decomposing the spectrum of a single pixel into a superposition of colloidal spectrum, fiber spectrum, and spectral residual. The specific steps are as follows: Among the identified feature pixel pairs with the greatest spectral difference, the measured spectrum of the pixel whose gelatinous angular distance is less than its fiber angular distance is taken as the gelatinous end spectrum, and the measured spectrum of the pixel whose fiber angular distance is less than its gelatinous angular distance is taken as the fiber end spectrum. The vector direction from the gelatinous end spectrum to the fiber end spectrum is determined as the degumming spectral axis. For any pixel in the image to be analyzed, the full-band spectrum of the pixel to be analyzed is projected onto the degummed spectral axis to obtain the projection position of the pixel to be analyzed on the degummed spectral axis. The fiber spectral contribution coefficient is calculated based on the relative proportion of the projection position between the colloid end spectrum and the fiber end spectrum, and the colloid spectral contribution coefficient is determined based on the coefficient normalization principle. Wherein, the sum of the spectral contribution coefficient of the gel and the spectral contribution coefficient of the fiber is always 1; When the projection position is located within the spectrum range of the gel end and the fiber end, the corresponding fiber spectrum contribution coefficient is calculated linearly based on the relative proportion of the projection position. When the projection position is located outside the spectrum of the gelatinous end, it is determined that the pixel point has no fiber component contribution, and the fiber spectrum contribution coefficient is determined to be 0. When the projection position is located outside the fiber end spectrum, the pixel is determined to be entirely fiber component, and the fiber spectrum contribution coefficient is set to 1. Based on the contribution coefficient of the gel spectral composition, the contribution coefficient of the fiber spectral composition, the gel end spectrum, and the fiber end spectrum, the fitted reconstructed spectrum of the pixel to be analyzed is reconstructed. The difference between the original measured spectrum of the pixel to be analyzed and the fitted reconstructed spectrum is calculated to obtain the spectral residual of the pixel, thus completing the linear unmixing decomposition of the single pixel spectrum.

[0010] Furthermore, the debonding state of each pixel is determined based on its contribution weight, the pixels at the interface between the adhesive and the fiber are identified, and the initial boundary between the adhesive and the fiber is obtained by connecting all the pixels at the interface between the adhesive and the fiber. The specific logic is as follows: The degumming state of each pixel is classified based on the contribution coefficients of the gel and fiber spectra obtained from solving each pixel. Pixels with a fiber contribution coefficient of 1 and a gel contribution coefficient of 0 are identified as pure fiber pixels. Pixels with a fiber contribution coefficient of 0 and a gel contribution coefficient of 1 are identified as pure gel pixels. Pixels with a mixed ratio where both the fiber and gel contribution coefficients are in the range of 0 to 1 are identified as gel-fiber transition pixels. Traverse all pixels in the image to be analyzed, extract all the transition pixels between the gel and the fiber. These transition pixels are the boundary pixels between the gel and the fiber. Connect all the identified boundary pixels between the gel and the fiber in sequence to form a continuous closed contour line, thereby obtaining the initial boundary between the gel and the fiber.

[0011] Furthermore, by combining the spectral residuals of non-intersecting pixels to correct the initial boundary between the adhesive and the fiber, the logic for obtaining the corrected boundary between the adhesive and the fiber is as follows: All pixels in the image to be analyzed are classified. Pixels that do not belong to the interface between the gel and the fiber are defined as non-interface pixels. For all non-gel and fiber interface pixels, call the obtained fitted reconstructed spectrum and spectral residual, calculate the spectral angular distance between the original measured spectrum and the fitted reconstructed spectrum of the pixel; if the spectral angular distance is greater than the preset spectral angular distance threshold, the pixel is determined to be a boundary recognition error pixel. Path cost is assigned to all non-adhesive and non-fiber interface pixels. If the non-interface pixel is a recognition error pixel, the path distance cost of the pixel is set to 1. If the non-boundary pixel is not a recognition error pixel, then the path distance cost of the pixel is set to positive infinity; The optimal path is solved based on the A* algorithm to obtain the shortest path between the two gels and the fiber initial boundary. If the sum of the path distance costs of all pixels on the shortest path between the two gels and the fiber initial boundary is not positive infinity, all pixels corresponding to the shortest path are updated to new gel and fiber boundary pixels. Traverse all initial boundary combinations of the gel and fiber, complete all optimal path iterations and updates, and obtain the optimized pixel points at the junction of the gel and fiber. By continuously connecting all optimized pixels at the interface between the adhesive and the fiber, the initial boundary is corrected and the contour is optimized, finally obtaining the corrected boundary between the adhesive and the fiber.

[0012] Furthermore, the modified boundaries of the gel and fiber at the current and previous times are obtained respectively. The spatial movement amplitude of the modified boundaries of the gel and fiber is calculated. The logic for solving the boundary movement rate by combining the time interval between the two times is as follows: Obtain the midpoint of the boundary pixel between the current and previous moments after the correction of the adhesive and fiber boundaries. Define the vector direction from the midpoint of the boundary pixel in the previous moment to the midpoint of the corresponding boundary pixel in the current moment as the boundary movement direction. For each boundary pixel of the adhesive and fiber in the previous moment, perform a translation judgment along the boundary movement direction. If the pixel intersects the current boundary of the adhesive and fiber after the correction within a range less than the preset boundary movement distance, the pixel is determined as a valid boundary pixel of the adhesive and fiber, and the translation distance corresponding to the pixel is determined as the single pixel boundary movement amplitude. Iterate through all boundary pixels of the adhesive and fiber in the previous moment, filter all valid boundary pixels, and calculate the arithmetic mean of their single pixel movement amplitudes. Define this average value as the spatial movement amplitude of the adhesive and fiber correction boundary between adjacent moments. The boundary space movement amplitude is calculated by comparing it with the acquisition time interval between the current moment and the previous moment, and finally the boundary movement rate between the gel and the fiber is obtained.

[0013] Furthermore, the logic for evaluating the degumming effect of pineapple leaf fibers based on spatial movement amplitude and boundary movement rate is as follows: if the spatial movement amplitude is greater than the preset spatial movement amplitude threshold and the boundary movement rate is greater than the preset boundary movement rate threshold, then the degumming effect of pineapple leaf fibers is qualified; otherwise, the degumming effect of pineapple leaf fibers is unqualified.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention acquires hyperspectral images of the detection area during the degumming process of pineapple leaf fibers and extracts the pineapple leaf fiber portion as the image to be analyzed based on the Otsu's method. This distinguishes the background of the degumming liquid, non-fiber areas, and the pineapple leaf fiber analysis area, allowing subsequent analysis to focus on the pineapple leaf fiber body area. Furthermore, this invention analyzes the spectral differences between pixels in the image to be analyzed and combines standard gum spectra and standard fiber spectra to determine gum pixels and fiber pixels, providing a clear spectral reference basis for the degumming state analysis and reducing the errors caused by judging the degumming state based solely on a single grayscale value or manual observation.

[0015] This invention also characterizes the spectrum of each pixel in the image to be analyzed as a superposition of the spectra of the gum pixel, the spectra of the fiber pixel, and the spectral residuals. Based on the degumming state of adjacent pixels, it identifies the pixel at the interface between the gum and the fiber, thus forming the initial boundary between the gum and the fiber. Then, it corrects the initial boundary between the gum and the fiber by combining the spectral residuals of the non-gum-fiber interface pixels, obtaining the corrected boundary between the gum and the fiber. Thus, the transformation from the gum to the exposed fiber state during the degumming process of pineapple leaf fiber can be represented as a traceable boundary. By obtaining the combined movement of the corrected boundary between the gum and the fiber at the current moment and the previous moment, and evaluating the degumming effect accordingly, it can reflect the actual situation of the degumming front advancing over time, thereby improving the continuity, spatiality, and accuracy of the degumming effect judgment. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is the standard fiber spectrum of the present invention; Figure 3 This is the standard colloidal spectrum of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Example: Please see Figures 1-3 The present invention provides a technical solution: A method for analyzing pineapple leaf fibers based on hyperspectral image features, comprising the following steps: Step 1: Acquire a single-frame hyperspectral image of the detection area during the degumming process of pineapple leaf fibers. Use the maximum inter-class variance method to segment the image and extract the pineapple leaf fiber region as the image to be analyzed. For example, in a hyperspectral image, each pixel is not a single gray value, but a multi-band spectral vector, specifically represented as: in, For the first spectral vector of each pixel The index of the pixel in the image to be analyzed. , The total number of pixels in the image to be analyzed. It refers to the number of bands in the hyperspectral image. Indicates the first The pixel at the th point Reflectance at each wavelength band , This is an index for the hyperspectral image bands.

[0020] Furthermore, the Otsu's method is used to segment the single-frame hyperspectral image, and the pineapple leaf fiber region is extracted as the image to be analyzed. The specific logic is as follows: The corresponding RGB image is extracted from a single frame of hyperspectral image. Since a hyperspectral image contains images corresponding to multiple bands, the images corresponding to the R, G, and B bands can be directly extracted to obtain the RGB image. After grayscale conversion, the number of pixels at each of the 256 grayscale levels is counted to construct a grayscale histogram. Based on the grayscale histogram, different grayscale levels are selected sequentially as candidate thresholds, and the inter-class variance corresponding to each candidate threshold is calculated. For any candidate threshold, pixels with grayscale values ​​less than or equal to the candidate threshold are classified into the first type of region, and pixels with grayscale values ​​greater than the candidate threshold are classified into the second type of region. The pixel proportion and grayscale mean of the first and second type of regions are calculated respectively, and the inter-class variance is calculated based on the difference in grayscale mean between the two types of regions. After traversing all candidate thresholds, the grayscale level corresponding to the maximum value of the inter-class variance is determined as the optimal segmentation threshold. After determining the optimal segmentation threshold, the single-frame hyperspectral image acquired during the degumming process of pineapple leaf fibers is segmented using this threshold. Because the pineapple leaf fiber area is relatively brighter and the background area is relatively darker in the grayscale image during degumming, for each frame of the hyperspectral image, if the grayscale value of a pixel is less than or equal to the optimal segmentation threshold, that pixel is identified as part of the background; if the grayscale value of a pixel is greater than the optimal segmentation threshold, that pixel is identified as part of the foreground. The background includes the degumming liquid background, the supporting background, or other non-pineapple leaf fiber areas, and the foreground corresponds to the area where the pineapple leaf fibers are located.

[0021] Finally, based on the threshold segmentation results, the corresponding pixel regions of the foreground part in the original hyperspectral image are extracted, and the full-band spectral information of each pixel in the region is retained. The foreground part is used as the image to be analyzed for subsequent spectral feature analysis, gum and fiber state identification, and degumming boundary calculation. Thus, before performing subsequent spectral unmixing and degumming state judgment, the interference of the background region on the spectral analysis results can be eliminated, so that the subsequent analysis is focused on the pineapple leaf fiber body region, improving the accuracy and stability of the analysis of the pineapple leaf fiber degumming process.

[0022] Step 2: Screen pixel pairs with significant spectral differences within the image to be analyzed, and combine them with pre-established standard colloidal spectra and standard fiber spectra to determine the feature pixel pairs with the greatest spectral differences within the image to be analyzed; Furthermore, the specific logic of step 2 is as follows: For each pixel in the image to be analyzed, the spectral angular distance between the single pixel spectrum and the preset standard colloidal spectrum and standard fiber spectrum is calculated, and defined as colloidal angular distance and fiber angular distance respectively. For the standard gel spectrum, a manually confirmed gel region is selected as the gel calibration region; for the standard fiber spectrum, a manually confirmed pure fiber region is selected as the fiber calibration region. The gel calibration region and the fiber calibration region are obtained, and the average value of the multi-band spectral vector of each pixel in the gel calibration region is obtained as the standard gel spectrum; similarly, the average value of the multi-band spectral vector of each pixel in the fiber calibration region is obtained as the standard fiber spectrum.

[0023] like Figure 2 , Figure 3 As shown, standard colloidal spectrum Recorded as: in, For the standard colloidal spectrum Reflectance in each wavelength band; The spectral angular distance between this pixel and the standard colloidal spectrum for: The spectral angular distance between the single-pixel spectrum and the standard fiber spectrum is similar; simply replace the standard gel spectrum with the standard fiber spectrum. Spectral angular distance does not compare the absolute brightness of two spectra, but rather the similarity of their spectral vectors; the smaller the spectral angular distance, the more similar the two spectra; the larger the spectral angular distance, the greater the difference between the two spectra.

[0024] If you directly select two pixels with the largest spectral difference in an image, three problems may occur: First, the two selected pixels may both belong to the gel-like region, but the large spectral differences are caused by noise, shadows, reflections, or local contamination. Second, the two selected pixels may both belong to the fiber region, which does not represent the direction of change from gel to fiber. Third, the selected pixels may be anomalous, such as edge noise points, residual background points, or points with abnormal lighting.

[0025] Therefore, we must first use the "colloidal angular distance" and "fiber angular distance" to determine whether each pixel is closer to the colloidal material or the fiber material. Only when one pixel is biased towards the colloidal material and the other pixel is biased towards the fiber material can we consider that the two pixels form a "colloidal end - fiber end" correspondence. Only the pixel pairs selected in this way have the physical meaning required for subsequent analysis.

[0026] Iterate through all pixel pairs formed by combinations of pixels in the image and filter the validity of each pixel pair: If the gelatinous angle distance of any pixel in a pixel pair is less than its fiber angle distance, the pixel is determined to be biased towards the gelatinous feature pixel. If the fiber angle distance of the other pixel is less than its gelatinous angle distance, the pixel is determined to be biased towards the fiber feature pixel. Then the pixel pair is considered to satisfy the gelatinous end-fiber end correspondence relationship and is a valid feature point pair. Within the entire range of valid pixel pairs, the spectral angular distance values ​​between pixels in each pair are compared, and the pixel pair with the largest spectral angular distance is identified as the feature pixel pair with the greatest spectral difference in the image to be analyzed.

[0027] By using the above-mentioned effective feature point pair screening method, the present invention first uses preset standard colloidal spectrum and preset standard fiber spectrum to preliminarily determine the spectral attributes of each pixel in the image to be analyzed, so that the pixel pair participating in the subsequent comparison must simultaneously contain pixels with colloidal features and pixels with fiber features, thereby avoiding mistaking the spectral differences between similar pixels caused by noise, illumination changes or local reflection differences as spectral differences between colloidal and fiber.

[0028] Furthermore, among all valid feature point pairs that satisfy the correspondence between the gel-fiber end, the pixel pair with the largest spectral angular distance between the pixels is selected as the feature pixel pair with the greatest spectral difference, so that the determined gel-end spectrum and fiber-end spectrum have the greatest spectral discrimination. This can enhance the ability to distinguish between the gel contribution coefficient and the fiber contribution coefficient in the subsequent spectral linear unmixing process, improve the discrimination accuracy of gel pixels, fiber pixels, and gel-fiber transition pixels, and provide a more stable endmember spectral benchmark for the subsequent extraction of the boundary between gel and fiber.

[0029] Step 3: Perform spectral linear unmixing on each pixel in the image to be analyzed, decompose the spectrum of a single pixel into the superposition of the gel spectrum, fiber spectrum and spectral residual, and solve for the spectral contribution coefficients of the gel and fiber. Degumming state of each pixel is determined according to the contribution weight, and the pixel at the interface between gel and fiber is identified. Connect all the pixel at the interface between gel and fiber to obtain the initial boundary between gel and fiber. The initial boundary between gel and fiber is corrected by combining the spectral residual of the non-interface pixel to obtain the corrected boundary between gel and fiber. Furthermore, spectral linear unmixing is performed on each pixel in the image to be analyzed, decomposing the spectrum of a single pixel into a superposition of colloidal spectrum, fiber spectrum, and spectral residual. The specific steps are as follows: Among the identified feature pixel pairs with the greatest spectral differences, the measured spectra of pixels with a gelatinous angular distance smaller than their fiber angular distance are taken as the gelatinous end spectra, and the measured spectra of pixels with a fiber angular distance smaller than their gelatinous angular distance are taken as the fiber end spectra. A spectral vector pointing from the gel-end spectrum to the fiber-end spectrum is constructed as the principal axis of the spectral state of degumming evolution. The vector expression is: in, The spectral vector pointing from the gel-end spectrum to the fiber-end spectrum; The spectral vector at the gel end; The fiber end spectral vector; For the spectral vector of each pixel Project it onto arrive On this spectral axis, the projection coefficient is calculated, which characterizes the relative contribution of the fiber components: in: For the first The projection coefficient of a pixel on the degummed spectral axis; the closer the projection coefficient is to 1, the higher the fiber ratio within the pixel; the closer it is to 0, the more adhesive residue there is.

[0030] in, This indicates the offset of the pixel relative to the pure adhesive end. It performs a dot product, the purpose of which is to calculate the projection of this offset onto the fiber end in the direction from the gelatinous end to the fiber end; that is, it does not look at... Overall distance It's not about how far it has traveled, but rather how far it has moved along the direction of delamination; the deviation unrelated to the direction of delamination; then divide by... This is for the purpose of normalization; because The square of the length of the spectral axis from the gel end to the fiber end, divided by it, yields... It then becomes a relative degree of contribution.

[0031] For any pixel in the image to be analyzed, the full-band spectrum of the pixel to be analyzed is projected onto the degummed spectral axis to obtain the projection position of the pixel to be analyzed on the degummed spectral axis. The fiber spectral contribution coefficient is calculated based on the relative proportion of the projection position between the colloid end spectrum and the fiber end spectrum, and the colloid spectral contribution coefficient is determined based on the coefficient normalization principle. By using the above-mentioned linear spectral unmixing method, this invention converts the full-band high-dimensional spectrum of each pixel in the image to be analyzed into a one-dimensional projection position along the direction from the colloid end spectrum to the fiber end spectrum, and calculates the colloid spectral contribution coefficient and the fiber spectral contribution coefficient based on the projection position; thereby, the complex multi-band spectral differences are transformed into proportional parameters that can characterize the degree of degumming, so that the degumming state of each pixel can be quantitatively described. Furthermore, the present invention limits the fiber spectral contribution coefficient to between 0 and 1, and determines the gel spectral contribution coefficient based on the coefficient normalization principle, so that the sum of the gel spectral contribution coefficient and the fiber spectral contribution coefficient is always 1, thereby avoiding negative contribution values ​​or out-of-range contribution values ​​caused by noise, reflection or local abnormal spectra. When the projection position is located between the gelatinous end spectrum and the fiber end spectrum, it indicates that the pixel contains both gelatinous and fiber spectral features. Therefore, the fiber spectral contribution coefficient is calculated linearly based on their relative positions. When the projection position is located outside the gelatinous end, it indicates that the pixel is closer to the gelatinous state, and the fiber spectral contribution coefficient is set to 0. When the projection position is located outside the fiber end, it indicates that the pixel is closer to the fully fiber state, and the fiber spectral contribution coefficient is set to 1. Thus, the pixels in the image to be analyzed can be distinguished into pure gelatinous pixels, pure fiber pixels, and gelatinous-fiber transition pixels, providing a quantitative basis for subsequent identification of gelatinous-fiber interface pixels and extraction of degumming boundaries.

[0032] Based on the contribution coefficient of the gel spectral composition, the contribution coefficient of the fiber spectral composition, the gel end spectrum, and the fiber end spectrum, the fitted reconstructed spectrum of the pixel to be analyzed is reconstructed. The difference between the original measured spectrum of the pixel to be analyzed and the fitted reconstructed spectrum is calculated to obtain the spectral residual of the pixel, thus completing the linear unmixing decomposition of the single pixel spectrum.

[0033] Furthermore, the debonding state of each pixel is determined based on its contribution weight, the pixels at the interface between the adhesive and the fiber are identified, and the initial boundary between the adhesive and the fiber is obtained by connecting all the pixels at the interface between the adhesive and the fiber. The specific logic is as follows: The degumming state of each pixel is classified based on the contribution coefficients of the gel and fiber spectra obtained from solving each pixel. Pixels with a fiber contribution coefficient of 1 and a gel contribution coefficient of 0 are identified as pure fiber pixels. Pixels with a fiber contribution coefficient of 0 and a gel contribution coefficient of 1 are identified as pure gel pixels. Pixels with a mixed ratio where both the fiber and gel contribution coefficients are in the range of 0 to 1 are identified as gel-fiber transition pixels. When the fiber spectral contribution coefficient of a pixel is 1 and the gel spectral contribution coefficient is 0, it means that the spectrum of the pixel can be explained almost entirely by the fiber end spectrum, and no longer reflects the contribution of the gel end spectrum, so it can be identified as a pure fiber pixel. Conversely, when the fiber spectral contribution coefficient is 0 and the gel spectral contribution coefficient is 1, it means that the spectrum of the pixel can be explained almost entirely by the gel end spectrum, and no longer reflects the contribution of the fiber end spectrum, so it can be identified as a pure gel pixel. When both the fiber spectral contribution coefficient and the gel spectral contribution coefficient are between 0 and 1, it means that the spectrum of the pixel cannot be explained entirely by either the gel end spectrum or the fiber end spectrum, but has both gel and fiber spectral characteristics.

[0034] The degumming process is essentially a process of gradually peeling off the adhesive and gradually exposing the fibers. Pure adhesive pixels represent areas that have not been fully degummed, pure fiber pixels represent areas that have been relatively fully degummed, and adhesive-fiber transition pixels are exactly in between the two states.

[0035] Traverse all pixels in the image to be analyzed, extract all the transition pixels between the gel and the fiber. These transition pixels are the boundary pixels between the gel and the fiber. Connect all the identified boundary pixels between the gel and the fiber in sequence to form a continuous closed contour line, thereby obtaining the initial boundary between the gel and the fiber.

[0036] Furthermore, by combining the spectral residuals of non-intersecting pixels to correct the initial boundary between the adhesive and the fiber, the logic for obtaining the corrected boundary between the adhesive and the fiber is as follows: All pixels in the image to be analyzed are classified. Pixels that do not belong to the interface between the gel and the fiber are defined as non-interface pixels. For all non-gel and fiber interface pixels, call the obtained fitted reconstructed spectrum and spectral residual, calculate the spectral angular distance between the original measured spectrum and the fitted reconstructed spectrum of the pixel; if the spectral angular distance is greater than the preset spectral angular distance threshold, the pixel is determined to be a boundary recognition error pixel. During the degumming process of pineapple leaf fibers, the gum is gradually removed, exposing the fibers. Pixels located between the gum and fiber regions typically contain both gum and fiber spectral features. However, due to factors such as uneven lighting, local reflection, degumming liquid interference, and sensor noise during actual data acquisition, some pixels at the transition between gum and fiber may not be initially identified as boundary pixels. Therefore, this invention performs a spectral reconstruction consistency judgment on non-gum-fiber boundary pixels. If the spectral angular distance between the original measured spectrum and the fitted reconstructed spectrum of a non-boundary pixel is greater than a preset spectral angular distance threshold, it indicates that the actual spectral features of the pixel cannot be accurately explained by the linear combination of the current gum end spectrum and fiber end spectrum. This pixel may have been misclassified as a non-boundary pixel, and is therefore identified as a boundary recognition error pixel.

[0037] The logic for setting the preset spectral angular distance threshold is as follows: Select historical pineapple leaf fiber images with expert-calibrated boundaries for gum and fiber correction; use the aforementioned boundary extraction method to generate the corresponding initial boundaries for gum and fiber; in the expert-calibrated boundary pixel set, remove pixels that overlap with the algorithm's initial boundaries; the remaining pixels are the difference boundary pixels missed by the algorithm. Calculate the spectral angular distance of all difference pixels and take the upper quartile as the preset spectral angular distance threshold. This method preserves a high level of spectral difference in historical samples while avoiding numerical disturbances caused by extreme noise and abnormal pixel reflections. It effectively identifies transition pixels missed by the initial boundaries while ensuring the stability and accuracy of boundary correction.

[0038] Through the above processing, the present invention can filter out abnormal pixels that may be missed or misclassified by the initial boundary from non-boundary pixels, providing candidate pixel basis for subsequent boundary path correction, thereby improving the integrity and accuracy of the identification of the adhesive and fiber boundary and reducing the problems of initial boundary breakage, offset or local missingness.

[0039] Path cost is assigned to all non-adhesive and non-fiber interface pixels. If the non-interface pixel is a recognition error pixel, the path distance cost of the pixel is set to 1. The optimal path is solved based on the A* algorithm to obtain the shortest path between the two gels and the fiber initial boundary. If the sum of the path distance costs of all pixels on the shortest path between the two gels and the fiber initial boundary is not positive infinity, all pixels corresponding to the shortest path are updated to new gel and fiber boundary pixels. Traverse all initial boundary combinations of the gel and fiber, complete all optimal path iterations and updates, and obtain the optimized pixel points at the junction of the gel and fiber. By continuously connecting all optimized pixels at the interface between the adhesive and the fiber, the initial boundary is corrected and the contour is optimized, finally obtaining the corrected boundary between the adhesive and the fiber.

[0040] The A* algorithm is a current technique for finding the shortest path between two gels and fibers at their initial boundary. When searching for the shortest path, the A* algorithm starts from the starting point and adds the distance cost of the already traveled paths from the current position to the starting point to the estimated cost (usually calculated using Manhattan distance or Euclidean distance) from the current position to the ending point to obtain an evaluation function. It prioritizes expanding the grid point with the smallest evaluation function value, sequentially checking its accessible neighboring grid points and updating their costs, skipping grids with positive infinity distance costs, until it expands to the ending point. Then, it backtracks along the recorded parent node to obtain the path with the smallest cumulative cost. If, after skipping grids with positive infinity distance costs, there is no path that allows the starting point to reach the ending point, then the shortest path from the starting point to the ending point does not exist. The sum of the path distance costs of all pixels on the combined shortest path from the starting point to the ending point is recorded as positive infinity.

[0041] It should be noted that, since this invention requires finding the shortest path between the initial boundaries of the two gels and the fiber, when running the A* algorithm, any boundary is selected as the starting boundary, and the adjacent boundary is selected as the ending boundary. A pixel on the starting boundary is arbitrarily selected as the starting point, and any pixel on the ending boundary is selected as the ending point. The A* algorithm is run, traversing all combinations of starting and ending points. The sum of the distance costs of all pixels on the optimal path is taken as the shortest path distance cost. The shortest path between the initial boundaries of the two gels and the fiber is found by finding the shortest path with the minimum distance cost among all combinations of starting and ending points. This is existing technology and will not be elaborated upon here.

[0042] Step 4: Obtain the modified boundaries of the gum and fiber at the current time and the previous time respectively, calculate the spatial movement amplitude of the modified boundaries of the gum and fiber, solve the boundary movement rate by combining the time interval between the two time points, and evaluate the degumming effect of pineapple leaf fiber based on the spatial movement amplitude and boundary movement rate.

[0043] Furthermore, the modified boundaries of the gel and fiber at the current and previous times are obtained respectively. The spatial movement amplitude of the modified boundaries of the gel and fiber is calculated. The logic for solving the boundary movement rate by combining the time interval between the two times is as follows: Obtain the midpoint of the boundary pixel between the current and previous moments after the correction of the adhesive and fiber boundaries. Define the vector direction from the midpoint of the boundary pixel in the previous moment to the midpoint of the corresponding boundary pixel in the current moment as the boundary movement direction. For each boundary pixel of the adhesive and fiber in the previous moment, perform a translation judgment along the boundary movement direction. If the pixel intersects the current boundary of the adhesive and fiber after the correction within a range less than the preset boundary movement distance, the pixel is determined as a valid boundary pixel of the adhesive and fiber, and the translation distance corresponding to the pixel is determined as the single pixel boundary movement amplitude. Iterate through all boundary pixels of the adhesive and fiber in the previous moment, filter all valid boundary pixels, and calculate the arithmetic mean of their single pixel movement amplitudes. Define this average value as the spatial movement amplitude of the adhesive and fiber correction boundary between adjacent moments. The boundary space movement amplitude is calculated by comparing it with the acquisition time interval between the current moment and the previous moment, and finally the boundary movement rate between the gel and the fiber is obtained.

[0044] The degumming process of pineapple leaf fibers is not determined by looking at the boundary between the gum and the fiber in a single frame of an image, but by determining whether the boundary actually advances between adjacent time points. By acquiring the midpoint of the boundary pixels at the current and previous moments after the correction of the adhesive and fiber boundaries, the main direction of boundary movement can be determined by the overall change in the position of the two boundaries, avoiding errors in direction judgment caused by noise, burrs, or uneven local degumming of individual pixels. Then, each boundary pixel at the previous moment is translated along this direction, and it is determined whether it intersects with the current moment boundary within a preset boundary movement distance. This allows for the selection of truly effective boundary pixels participating in boundary advancement, eliminating abnormal jump points, mismatched points, and locally erroneous boundaries. Finally, the average movement distance of all effective boundary pixels yields the overall spatial movement amplitude of the entire adhesive-fiber boundary between adjacent moments. Ratioing this spatial movement amplitude to the time interval between two image acquisition frames gives the boundary movement rate. Thus, the dynamic changes of gradual adhesive removal and fiber exposure during pineapple leaf fiber degumming can be transformed into quantifiable spatial movement and movement rate, improving the continuity, objectivity, and accuracy of the degumming effect evaluation.

[0045] Furthermore, the logic for evaluating the degumming effect of pineapple leaf fibers based on spatial movement amplitude and boundary movement rate is as follows: if the spatial movement amplitude is greater than the preset spatial movement amplitude threshold and the boundary movement rate is greater than the preset boundary movement rate threshold, then the degumming effect of pineapple leaf fibers is qualified; otherwise, the degumming effect of pineapple leaf fibers is unqualified.

[0046] The logic for setting the spatial movement amplitude threshold and boundary movement rate threshold is as follows: Obtain a sample group of pineapple leaf fibers with historically assessed quality by experts, and acquire the corresponding spatial movement amplitude and boundary movement rate. Using the normalized spatial movement amplitude and boundary movement rate as features, cluster the sample group into two classes based on the K-means algorithm. Specifically: construct boundary feature vectors using spatial movement amplitude and boundary movement rate as features; randomly initialize two boundary feature vectors as cluster centers; calculate the Euclidean distance between each boundary feature vector and each cluster center, and assign it to the cluster center with the closest Euclidean distance, forming two initial clusters; then, recalculate the mean of all Euclidean distances within each cluster, using it as the new cluster center; iterate the above assignment and update steps repeatedly until the maximum number of iterations is reached. Clusters with the highest boundary movement rates are defined as qualified clusters, and those with the lowest boundary movement rates are defined as unqualified clusters. The maximum value of the spatial movement amplitude of unqualified clusters and the minimum value of the spatial movement amplitude of qualified clusters are obtained, and the average of these maximum and minimum values ​​is used as the spatial movement amplitude threshold. The maximum value of the boundary movement rate of unqualified clusters and the minimum value of the boundary movement rate of qualified clusters are obtained, and the average of these maximum and minimum values ​​is used as the boundary movement rate threshold. The maximum value in the unqualified cluster represents the highest boundary that historical unqualified samples could reach on this indicator, while the minimum value in the qualified cluster represents the lowest boundary that historical qualified samples could reach on this indicator. These two values ​​together constitute the "critical transition interval between qualified and unqualified samples." Taking the average of these two values ​​as the threshold is equivalent to setting the judgment boundary in the middle between the upper limit of unqualified samples and the lower limit of qualified samples. This avoids the threshold being too low, which would cause some unqualified samples to be misjudged as qualified, and also avoids the threshold being too high, which would cause some qualified samples to be misjudged as unqualified. As a result, the spatial movement amplitude threshold and the boundary movement rate threshold are closer to the actual boundary positions of the two types of degumming states in historical samples.

[0047] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0048] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0049] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0050] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that cannot be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for analyzing pineapple leaf fibers based on hyperspectral image features, characterized in that, The specific steps include: Step 1: Acquire a single-frame hyperspectral image of the detection area during the degumming process of pineapple leaf fibers. Use the maximum inter-class variance method to segment the image and extract the pineapple leaf fiber region as the image to be analyzed. Step 2: Screen pixel pairs with significant spectral differences within the image to be analyzed, and combine them with pre-established standard colloidal spectra and standard fiber spectra to determine the feature pixel pairs with the greatest spectral differences within the image to be analyzed; The specific logic of step 2 is as follows: For each pixel in the image to be analyzed, the spectral angular distance between the single pixel spectrum and the preset standard colloidal spectrum and standard fiber spectrum is calculated, and defined as colloidal angular distance and fiber angular distance respectively. Iterate through all pixel pairs formed by combinations of pixels in the image and filter the validity of each pixel pair: If the gelatinous angle distance of any pixel in a pixel pair is less than its fiber angle distance, the pixel is determined to be biased towards the gelatinous feature pixel. If the fiber angle distance of the other pixel is less than its gelatinous angle distance, the pixel is determined to be biased towards the fiber feature pixel. Then the pixel pair is considered to satisfy the gelatinous end-fiber end correspondence relationship and is a valid feature point pair. Within the entire range of valid pixel pairs, the spectral angular distance values ​​between pixels in each pair are compared, and the pixel pair with the largest spectral angular distance is identified as the feature pixel pair with the greatest spectral difference in the image to be analyzed. Step 3: Perform spectral linear unmixing on each pixel in the image to be analyzed, decompose the spectrum of a single pixel into the superposition of the gel spectrum, fiber spectrum and spectral residual, and solve for the spectral contribution coefficients of the gel and fiber. Degumming state of each pixel is determined according to the contribution weight, and the pixel at the interface between gel and fiber is identified. Connect all the pixel at the interface between gel and fiber to obtain the initial boundary between gel and fiber. The initial boundary between gel and fiber is corrected by combining the spectral residual of the non-interface pixel to obtain the corrected boundary between gel and fiber. The debonding state of each pixel is determined based on its contribution weight. Pixels at the interface between the adhesive and fiber are identified, and the initial boundary between the adhesive and fiber is obtained by connecting all these pixels. The specific logic is as follows: The degumming state of each pixel is classified based on the contribution coefficients of the gel and fiber spectra obtained from solving each pixel. Pixels with a fiber contribution coefficient of 1 and a gel contribution coefficient of 0 are identified as pure fiber pixels. Pixels with a fiber contribution coefficient of 0 and a gel contribution coefficient of 1 are identified as pure gel pixels. Pixels with a mixed ratio where both the fiber and gel contribution coefficients are in the range of 0 to 1 are identified as gel-fiber transition pixels. The algorithm iterates through all pixels in the image to be analyzed, extracting all transition pixels between the gel and fiber layers. These transition pixels are the boundaries between the gel and fiber layers. All identified boundary pixels are then connected sequentially to form a continuous closed contour line, thus obtaining the initial boundary between the gel and fiber layers. The initial boundary is then corrected using the spectral residuals of non-boundary pixels. The logic for obtaining the corrected boundary is as follows: All pixels in the image to be analyzed are classified. Pixels that do not belong to the interface between the gel and the fiber are defined as non-interface pixels. For all non-gel and fiber interface pixels, call the obtained fitted reconstructed spectrum and spectral residual, calculate the spectral angular distance between the original measured spectrum and the fitted reconstructed spectrum of the pixel; if the spectral angular distance is greater than the preset spectral angular distance threshold, the pixel is determined to be a boundary recognition error pixel. Path cost is assigned to all non-adhesive and non-fiber interface pixels. If the non-interface pixel is a recognition error pixel, the path distance cost of the pixel is set to 1. If the non-boundary pixel is not a recognition error pixel, then the path distance cost of the pixel is set to positive infinity; The optimal path is solved based on the A* algorithm to obtain the shortest path between the two gels and the fiber initial boundary. If the sum of the path distance costs of all pixels on the shortest path between the two gels and the fiber initial boundary is not positive infinity, all pixels corresponding to the shortest path are updated to new gel and fiber boundary pixels. Traverse all initial boundary combinations of the gel and fiber, complete all optimal path iterations and updates, and obtain the optimized pixel points at the junction of the gel and fiber. By continuously connecting all the optimized pixels at the junction of the adhesive and the fiber, the initial boundary is corrected and the contour is optimized, and finally the corrected boundary of the adhesive and the fiber is obtained. Step 4: Obtain the modified boundaries of the gum and fiber at the current time and the previous time respectively, calculate the spatial movement amplitude of the modified boundaries of the gum and fiber, solve the boundary movement rate by combining the time interval between the two time points, and evaluate the degumming effect of pineapple leaf fiber based on the spatial movement amplitude and boundary movement rate.

2. The method for analyzing pineapple leaf fibers based on hyperspectral image features according to claim 1, characterized in that: The Otsu's method is used to segment a single-frame hyperspectral image, and the pineapple leaf fiber region is extracted as the image to be analyzed. The specific logic is as follows: The corresponding RGB image is extracted from a single-frame hyperspectral image and converted to grayscale. The number of pixels at each of the 256 grayscale levels is counted to construct a grayscale histogram. The inter-class variance of each threshold in the grayscale histogram is calculated, and the threshold with the largest inter-class variance is defined as the optimal segmentation threshold. The single-frame hyperspectral image is segmented into background and foreground parts using the optimal segmentation threshold. When segmenting into background and foreground parts, if the grayscale value of a pixel is less than or equal to the optimal segmentation threshold, the pixel is determined to belong to the background part; if it is greater than the optimal segmentation threshold, the pixel is determined to belong to the foreground part. The foreground part is then extracted as the image to be analyzed.

3. The method for analyzing pineapple leaf fibers based on hyperspectral image features according to claim 1, characterized in that: For each pixel in the image to be analyzed, spectral linear unmixing is performed to decompose the spectrum of a single pixel into a superposition of colloidal spectrum, fiber spectrum, and spectral residual. The specific steps are as follows: Among the identified feature pixel pairs with the greatest spectral difference, the measured spectrum of the pixel whose gelatinous angular distance is less than its fiber angular distance is taken as the gelatinous end spectrum, and the measured spectrum of the pixel whose fiber angular distance is less than its gelatinous angular distance is taken as the fiber end spectrum. The vector direction from the gelatinous end spectrum to the fiber end spectrum is determined as the degumming spectral axis. For any pixel in the image to be analyzed, the full-band spectrum of the pixel to be analyzed is projected onto the degummed spectral axis to obtain the projection position of the pixel to be analyzed on the degummed spectral axis. The fiber spectral contribution coefficient is calculated based on the relative proportion of the projection position between the colloid end spectrum and the fiber end spectrum, and the colloid spectral contribution coefficient is determined based on the coefficient normalization principle. Wherein, the sum of the spectral contribution coefficient of the gel and the spectral contribution coefficient of the fiber is always 1; When the projection position is located within the spectrum range of the gel end and the fiber end, the corresponding fiber spectrum contribution coefficient is calculated linearly based on the relative proportion of the projection position. When the projection position is located outside the spectrum of the gelatinous end, it is determined that the pixel point has no fiber component contribution, and the fiber spectrum contribution coefficient is determined to be 0. When the projection position is located outside the fiber end spectrum, the pixel is determined to be entirely fiber component, and the fiber spectrum contribution coefficient is set to 1. Based on the contribution coefficient of the gel spectral composition, the contribution coefficient of the fiber spectral composition, the gel end spectrum, and the fiber end spectrum, the fitted reconstructed spectrum of the pixel to be analyzed is reconstructed. The difference between the original measured spectrum of the pixel to be analyzed and the fitted reconstructed spectrum is calculated to obtain the spectral residual of the pixel, thus completing the linear unmixing decomposition of the single pixel spectrum.

4. The method for analyzing pineapple leaf fibers based on hyperspectral image features according to claim 3, characterized in that: The logic for obtaining the modified boundaries of the gel and fiber at the current and previous times respectively, calculating the spatial displacement of the modified boundaries, and combining the time interval between the two times to solve for the boundary displacement rate is as follows: Obtain the midpoint of the boundary pixel between the current and previous moments after the correction of the adhesive and fiber boundaries. Define the vector direction from the midpoint of the boundary pixel in the previous moment to the midpoint of the corresponding boundary pixel in the current moment as the boundary movement direction. For each boundary pixel of the adhesive and fiber in the previous moment, perform a translation judgment along the boundary movement direction. If the pixel intersects the current boundary of the adhesive and fiber after the correction within a range less than the preset boundary movement distance, the pixel is determined as a valid boundary pixel of the adhesive and fiber, and the translation distance corresponding to the pixel is determined as the single pixel boundary movement amplitude. Iterate through all boundary pixels of the adhesive and fiber in the previous moment, filter all valid boundary pixels, and calculate the arithmetic mean of their single pixel movement amplitudes. Define this average value as the spatial movement amplitude of the adhesive and fiber correction boundary between adjacent moments. The boundary space movement amplitude is calculated by comparing it with the acquisition time interval between the current moment and the previous moment, and finally the boundary movement rate between the gel and the fiber is obtained.

5. The method for analyzing pineapple leaf fibers based on hyperspectral image features according to claim 4, characterized in that: The logic for evaluating the degumming effect of pineapple leaf fibers based on spatial movement amplitude and boundary movement rate is as follows: If the spatial movement amplitude is greater than the preset spatial movement amplitude threshold and the boundary movement rate is greater than the preset boundary movement rate threshold, then the degumming effect of the pineapple leaf fiber is considered qualified; otherwise, the degumming effect of the pineapple leaf fiber is considered unqualified.

Images