Quantitative evaluation method for early damage degree of apple based on hyperspectral imaging technology

Abstract

The application provides a kind of early injury degree quantitative evaluation method of apple based on hyperspectral imaging technology, after the original hyperspectral image of sample before and after injury is carried out black and white correction, the image of injury sensitive band is selected, pixel matching is carried out to the corresponding position of reference image and injury image, the correlation coefficient representing fruit injury distribution is calculated, the influence factor is calculated, and the early injury degree of fruit is determined.The beneficial effects of the application are fast, reliable and non-contact, simple and fast operation, the correlation coefficient and injury degree influence factor are proposed as evaluation standard, which has the advantages of convenience, speed and non-destructiveness, the spatial distribution of fruit injury can be directly obtained, early hidden injury can be effectively evaluated, and the injury distribution of fruit under impact load can be accurately reflected, which provides an important basis for evaluating the mechanical injury degree of fruit.

Description

Technical Field

[0001] This invention belongs to the field of fruit damage technology, and in particular relates to a method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology. Background Technology

[0002] Apples are rich in minerals and vitamins, possessing high nutritional and commercial value. With rising living standards, people are demanding higher quality apples. However, detecting early damage to apples has always been a challenge. Without skin damage, it's difficult to visually distinguish between healthy and damaged apples. Early damage increases the risk of fruit spoilage and, over time, can even contaminate surrounding produce. Currently, most economic losses are attributed to improper post-harvest fruit management and a lack of fruit sorting techniques. Due to limitations in measurement technology, quantitatively assessing the degree of early damage to apples remains a significant challenge. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method for quantitatively evaluating the degree of early damage in apples based on hyperspectral imaging technology, so as to solve the above or other problems existing in the prior art.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology. After black and white correction of the original hyperspectral images of the sample before and after damage, the image of the damage-sensitive band is selected, the corresponding positions of the reference image and the damage image are matched pixel by pixel, the correlation coefficient characterizing the distribution of fruit damage is calculated, the influencing factor is calculated, and the degree of early damage to the fruit is determined.

[0005] Furthermore, when performing pixel matching at corresponding locations in the reference image and the damaged image, the following steps are included:

[0006] The image is binarized to obtain the entire contour of the sample;

[0007] Construct the computation region;

[0008] The displacement search algorithm is used to find the target sub-region that best matches the reference sub-region in the damaged image, and the deformation of the calculation point is determined.

[0009] The position of each discrete calculation point in the reference image is determined by the image registration algorithm, and the displacement of the whole field is obtained so that the pixels before and after the damage correspond one by one.

[0010] Furthermore, the points furthest from the sample center in the four directions of the sample outline (up, down, left, and right) are used as boundaries to construct a rectangular frame as the calculation area.

[0011] Furthermore, in the step of determining the deformation amount of the calculation point by finding the target sub-region that best matches the reference sub-region in the damaged image through a displacement search algorithm, the deformation relationship representation function is:

[0012]

[0013] in, , ∆x and ∆y are the distances from point (x, y) to the center of the reference image. distance, This represents the coordinates of the center pixel of the reference image. This represents the coordinates of the center pixel of the damaged image. u represents the horizontal displacement of the image center point. represents the horizontal displacement gradient of an image sub-region, and v represents the vertical displacement of the image center point. This represents the vertical displacement gradient of an image sub-region.

[0014] Furthermore, the step of determining the deformation of the calculation point by finding the target sub-region that best matches the reference sub-region in the damaged image through a displacement search algorithm includes:

[0015] Obtaining sub-pixel displacement: Sub-pixel search is performed using an interpolation function, specifically a bicubic interpolation function, expressed as follows:

[0016]

[0017] Where x is the sub-pixel coordinate, x k and x k+1 These represent the left and right endpoints of the integer pixel interpolation interval, respectively. These are the interpolation coefficients, used in calculating the interpolation coefficients. The calculation is performed using three boundary conditions: , and exist Up continuous, In subinterval The above is a cubic algebraic polynomial. ;

[0018] Obtaining shape function parameters: The parameters are obtained iteratively using the inverse combination Gauss-Newton iteration algorithm.

[0019] Furthermore, when calculating the correlation coefficient characterizing the distribution of fruit damage, the following are included:

[0020] Set point set M: After obtaining the calculation region, select a square sub-region with the center point x as the neighborhood as the point set M;

[0021] Define a reference sub-region: the reflectance distribution f(M) corresponding to the point set M is used as the reference sub-region, and the reflectance distribution of the sub-region corresponding to the damaged image is g(M);

[0022] Calculate the correlation coefficient: The correlation coefficient is used to evaluate the similarity between two images. The formula for calculating the correlation coefficient is as follows:

[0023]

[0024] in, , , , , This represents the correlation coefficient value. This represents the reflectance distribution of a sub-region of the reference image. This represents the reflectance distribution of a sub-region of a damaged image. This represents the average reflectance value of a sub-region of the reference image. This represents the average reflectance value of a damaged image sub-region. This represents the difference between the reflectance of a sub-region of the reference image and the average reflectance. This represents the difference between the reflectance of a damaged sub-region of the image and the average reflectance.

[0025] Furthermore, when calculating the impact factor, the formula for calculating the impact factor of the degree of damage is as follows:

[0026]

[0027] Where x represents the specified correlation coefficient threshold. This indicates the number of pixels in the sample tissue. This indicates the number of pixels in the sample image whose correlation coefficient value is lower than x.

[0028] Furthermore, when selecting images with damage-sensitive bands, the spectra of regions of interest are extracted from hyperspectral images of different damage areas and averaged. The peaks and troughs are selected as damage-sensitive bands, including 971nm, 1014nm, 1104nm, 1237nm and 1300nm.

[0029] Furthermore, when performing black-and-white correction on the original hyperspectral images before and after sample damage, the reflectance of the original hyperspectral images is calibrated using the following formula:

[0030]

[0031] Where I is the corrected hyperspectral image, I r For the original hyperspectral image, I d For a completely black calibration image, I w To calibrate an image on a whiteboard.

[0032] Furthermore, before performing black-and-white correction on the original hyperspectral images of the sample before and after damage, the following steps are also included:

[0033] Damage samples were obtained through impact testing. The impact test was conducted using a drop testing machine with the impact location at the equatorial region of the apple. The apple was placed on a drop platform and dropped freely from different heights onto a rigid plate.

[0034] Obtain raw hyperspectral images of the sample before and after damage, including raw hyperspectral data of the undamaged sample and hyperspectral data of the damaged sample.

[0035] By employing the above technical solution, original hyperspectral images of undamaged and damaged fruit are acquired. All original hyperspectral images undergo black-and-white correction. Sensitive band images are selected from the corrected hyperspectral images, and pixel matching is performed on corresponding positions in the reference and damaged images. The correlation coefficient of variables characterizing fruit damage distribution and the damage degree influencing factor characterizing damage magnitude are calculated. This method is rapid, reliable, non-contact, and simple to operate. The correlation coefficient and damage degree influencing factor are proposed as evaluation criteria, offering advantages of convenience, speed, and non-destructiveness. It directly provides the spatial distribution of fruit damage, effectively evaluates early-stage hidden damage, and accurately reflects the damage distribution of fruit under impact loads, providing an important basis for assessing the degree of mechanical damage to fruit. Attached Figure Description

[0036] Figure 1 This is a flowchart of an embodiment of the present invention;

[0037] Figure 2 This is an original spectral curve and a damage-sensitive band image of a hyperspectral image of an apple sample after reflectance correction, according to an embodiment of the present invention.

[0038] Figure 3 This is a damage distribution map obtained by correlation coefficient calculation according to an embodiment of the present invention;

[0039] Figure 4 UI under different drop heights and set thresholds according to an embodiment of the present invention x value. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0041] Figure 1The flowchart of an embodiment of the present invention is shown. This embodiment relates to a method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology. The method uses correlation coefficient and damage degree influencing factor as evaluation criteria, which has the advantages of being convenient, fast and non-destructive. It can directly obtain the spatial distribution of fruit damage and can effectively evaluate early hidden damage.

[0042] A quantitative evaluation method for early damage to apples based on hyperspectral imaging technology is proposed to realize the spatial distribution of fruit damage. In this method, after black and white correction of the original hyperspectral images of the sample before and after damage, images of the damage-sensitive band are selected, and pixel matching is performed on the corresponding positions of the reference image and the damage image. The correlation coefficient characterizing the distribution of fruit damage is calculated, a calculation formula for the degree of early damage to fruit is established, and the influencing factor of the degree of damage is calculated. The spatial distribution of fruit damage can be directly obtained, and early hidden damage can be effectively evaluated.

[0043] Specifically, let's take a sample of Red Fuji apples as an example to illustrate this, such as... Figure 1 As shown, it includes the following steps:

[0044] Obtaining Damaged Samples: Samples with mechanical damage are obtained through impact testing. During impact testing, a drop tester is used to obtain mechanical damage to the apple.

[0045] When conducting impact tests on a drop tester, the impact location and the marking location are determined. The impact location is in the equatorial region of the apple, and the marking location is selected at the top of the apple.

[0046] The impact test begins. An apple is placed on a platform and dropped from a predetermined height onto a rigid plate. At the moment of impact, the apple is caught by hand as it bounces off to avoid secondary impact.

[0047] During the drop test, samples were dropped freely from different heights. The height and number of samples were selected based on actual needs and no specific requirements were specified here. For example, 30 regular-shaped Fuji apples without any bruises were selected. The selection criteria were: sample height approximately 7.5cm, equatorial diameter approximately 10cm, mass 260g ± 20g, and sample storage temperature 20 ± 2℃.

[0048] The same sample was marked with different numbers, each corresponding to a different drop height. For example, the same sample was marked with three directions, with numbers 1, 2 and 3 representing different directions, corresponding to drop heights of 0.3 m, 0.6 m and 0.9 m respectively.

[0049] Obtain the raw hyperspectral images of the samples, including the raw hyperspectral data of the undamaged sample and the raw hyperspectral data of the damaged sample. This process specifically includes the following steps:

[0050] Obtain raw hyperspectral data of undamaged samples: Use a spectroscopic camera to acquire hyperspectral images of undamaged samples and obtain raw hyperspectral data;

[0051] Obtain raw hyperspectral data of the damaged sample: Use a spectral camera to acquire hyperspectral images of the damaged sample and obtain raw hyperspectral data;

[0052] During the acquisition of the original hyperspectral image of the damaged sample, the sample to be acquired is placed on the moving platform of the hyperspectral imaging system, with the damaged area facing the spectral camera. The overall information is acquired by moving the platform. After adjusting the relevant parameters of the hyperspectral acquisition system, the hyperspectral image of the damaged sample is acquired.

[0053] The parameters are set as follows: the distance between the sample and the lens of the spectrophotometer is 30-50cm, the exposure time is 15-25ms, the forward speed of the moving platform is 0.75-1.20cm / s, and the acquisition distance is 8-16cm.

[0054] Due to the dark current in the CCD camera and the non-uniformity of illumination, the reflectance of all acquired raw hyperspectral images must first be corrected. This involves black-and-white correction of the raw hyperspectral images, including the following steps:

[0055] Under the same conditions and parameter settings as when acquiring the original hyperspectral image, a white calibration image was acquired by acquiring a white calibration plate, which is a standard polytetrafluoroethylene (PTFE) with a reflectivity close to 100%. A completely black calibration image with a reflectivity of 0% was acquired by covering the camera lens with a camera cover, which is completely opaque and has a reflectivity of 0%.

[0056] The reflectance of the original hyperspectral image is calculated using the following formula during calibration:

[0057]

[0058] Where I is the corrected hyperspectral image, I r For the original hyperspectral image, I d For a completely black calibration image, I w To calibrate an image on a whiteboard.

[0059] Image selection for damage-sensitive bands: In the corrected hyperspectral images, the spectra of regions of interest are extracted from different damage areas, and then averaged to serve as representatives of samples with different degrees of damage. For example... Figure 2 As shown, the average spectral curves of normal and damaged samples exhibit a consistent trend.

[0060] The reason for choosing damage-sensitive wavelengths is the enormous amount of data and redundant information from hundreds of wavelength images. By selecting characteristic wavelengths, the computational load can be reduced, and a small number of images representing the most significant damage to the sample can be identified. Peaks and troughs are set as damage-sensitive bands.

[0061] In this embodiment, as Figure 2 The images shown are for the corresponding bands, with damage-sensitive bands including 971nm, 1014nm, 1104nm, 1237nm, and 1300nm.

[0062] When performing pixel matching between corresponding locations in the reference image and the damaged image, the following steps are included:

[0063] Obtain the entire contour of the sample: Binarize the image to obtain the entire contour of the sample;

[0064] Constructing the computational region: Using the points furthest from the center of the sample in the four directions of the sample outline (up, down, left, and right) as boundaries, construct a rectangular frame as the computational region.

[0065] The displacement search algorithm is used to find the target sub-region that best matches the reference sub-region in the damaged image, and the deformation of the calculation point is determined.

[0066] In this step, the transformation relation is represented by a first-order function, which is:

[0067]

[0068] in, , ∆x and ∆y are the distances from point (x, y) to the center of the reference image. distance, This represents the coordinates of the center pixel of the reference image. This represents the coordinates of the center pixel of the image after damage, where u represents the horizontal displacement of the image center. represents the horizontal displacement gradient of an image sub-region, and v represents the vertical displacement of the image center point. This represents the vertical displacement gradient of an image sub-region.

[0069] When using the displacement search algorithm, sub-pixel displacements are required to ensure matching accuracy. To obtain these sub-pixel displacements, an interpolation function is used for sub-pixel search. Specifically, the interpolation function is a bicubic interpolation function, expressed as follows:

[0070]

[0071] Where x is the sub-pixel coordinate, x k and x k+1These represent the left and right endpoints of the integer pixel interpolation interval, respectively. The interpolation coefficients are used in the calculation of coefficients. The calculation is performed using three boundary conditions: , and exist Up continuous, In subinterval The above is a cubic algebraic polynomial. .

[0072] Obtaining shape function parameters: The inverse combination Gauss-Newton iteration algorithm is used to iteratively obtain the shape function parameters. This algorithm directly obtains displacement field information. By eliminating the influence of displacement, the images before and after the damage are matched one by one.

[0073] An image registration algorithm is used to determine the position of each discrete calculation point in the reference image in the target image, thereby obtaining the displacement of the entire field so that the pixels before and after the damage correspond one-to-one.

[0074] Calculating the correlation coefficient characterizing the distribution of fruit damage includes the following steps:

[0075] Define the point set M: After selecting the calculation region, select a square sub-region with the center point x as the neighborhood as the point set M;

[0076] Define a reference sub-region: the reflectance distribution f(M) corresponding to the point set M is used as the reference sub-region, and the reflectance distribution of the sub-region corresponding to the damaged image is g(M);

[0077] Calculate the correlation coefficient: The correlation coefficient is used to evaluate the similarity between two images. The formula for calculating the correlation coefficient is as follows:

[0078]

[0079] in, , , , , This represents the correlation coefficient value. This represents the reflectance distribution of a sub-region of the reference image. This represents the reflectance distribution of a sub-region of a damaged image. This represents the average reflectance value of a sub-region of the reference image. This represents the average reflectance value of a damaged image sub-region. This represents the difference between the reflectance of a sub-region of the reference image and the average reflectance. This represents the difference between the reflectance of a damaged sub-region of the image and the average reflectance.

[0080] Each point in the image represents a reflectance value, and the coordinates of the n points are... .

[0081] Calculate influencing factors, establish a formula for calculating the degree of early fruit damage, and define the damage degree influencing factor characterizing the magnitude of damage. The formula for calculating the damage severity influencing factor is:

[0082]

[0083] Where x represents the specified correlation coefficient threshold. This indicates the number of pixels in the sample tissue. This indicates the number of pixels in the sample image whose correlation coefficient value is lower than x.

[0084] The degree of damage to the apple sample can be reflected by statistically analyzing the ratio of the number of pixels with a small correlation coefficient to the total number of pixels in the apple sample.

[0085] like Figure 3 As shown, the correlation coefficients of apple samples at different drop heights in five damage-sensitive bands are calculated.

[0086] When the correlation coefficient C fg A value of 1 indicates that the spectral reflectance of the reference sample and the damaged sample are completely consistent.

[0087] When the drop height is 0.3 m, the correlation coefficient C fg The number of low-value items is relatively small;

[0088] When the drop height is 0.6 m, the correlation coefficient C fg The number of low values ​​is increasing;

[0089] When the drop height is 0.9 m, the correlation coefficient C fg The number of pixels with low values ​​increased significantly.

[0090] Depend on Figure 3 It can be clearly seen that as the drop height increases, the damaged area gradually increases, and the correlation between images decreases accordingly. The correlation coefficient field obtained from different drop heights can effectively reflect the early damage characteristics of apple samples.

[0091] After qualitatively visualizing the distribution, the damage needs to be quantitatively assessed using given indicators. It should be specified at what threshold the collected area is defined as the damaged zone. Figure 4The data shows how damage assessment metrics change with drop height when different thresholds are set. When the threshold is set to 0.6, it can be seen that over 20% of the points have a correlation coefficient below 0.6, meaning the similarity before and after the injury decreases. Therefore, selecting an appropriate threshold is crucial. As the drop height increases, the UI... x The value is increasing. (UI) x The higher the value, the more severe the damage. This aligns with the basic pattern of apple damage during a drop.

[0092] By employing the above technical solution, original hyperspectral images of undamaged and damaged fruit are acquired. All original hyperspectral images undergo black-and-white correction. Sensitive band images are selected from the corrected hyperspectral images, and pixel matching is performed on corresponding positions in the reference and damaged images. The correlation coefficient of variables characterizing fruit damage distribution and the damage degree influencing factor characterizing damage magnitude are calculated. This method is fast, reliable, non-contact, and simple to operate. The correlation coefficient and damage degree influencing factor are proposed as evaluation criteria, offering advantages of convenience, speed, and non-destructiveness. It can directly obtain the spatial distribution of fruit damage, effectively evaluate early latent damage, and accurately reflect the damage distribution of fruit under impact loads, providing an important basis for assessing the degree of mechanical damage to fruit.

[0093] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A method for quantitatively evaluating the degree of early damage in apples based on hyperspectral imaging technology, characterized in that: After black and white correction of the original hyperspectral images of the sample before and after damage, the image of the damage-sensitive band is selected, and pixel matching is performed on the corresponding positions of the reference image and the damaged image. The correlation coefficient characterizing the distribution of fruit damage is calculated, the influence factor is calculated, and the degree of early damage to the fruit is determined. The reference image is the image of the damage-sensitive band of the original hyperspectral image of the undamaged sample after black and white correction. The calculation of the correlation coefficient characterizing the distribution of fruit damage includes: Define point set M: After obtaining the calculation region, select a square sub-region as the point set M with the center point H as the neighborhood; Setting: The reflectance distribution of the reference sub-region corresponding to the point set M is f(M), and the reflectance distribution of the sub-region corresponding to the damaged image is g(M); Calculate the correlation coefficient: The correlation coefficient is used to evaluate the similarity between two images. The formula for calculating the correlation coefficient is as follows: ， in, , , , ; When calculating the impact factor, the formula for calculating the impact factor of the degree of damage is: Where s represents the specified correlation coefficient threshold. This indicates the number of pixels in the sample tissue. This indicates the number of pixels in the sample image whose correlation coefficient value is lower than s.

2. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to claim 1, characterized in that: The pixel matching process for corresponding positions in the reference image and the damaged image includes the following steps: The image is binarized to obtain the entire contour of the sample; Construct the computation region; The displacement search algorithm is used to find the target sub-region that best matches the reference sub-region in the damaged image, and the deformation of the calculation point is determined. The position of each discrete calculation point in the reference image is determined by the image registration algorithm, and the displacement of the whole field is obtained so that the pixels before and after the damage correspond one by one.

3. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to claim 2, characterized in that: The calculation area is constructed by taking the points farthest from the center of the sample in the four directions of top, bottom, left, and right as the boundaries.

4. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to claim 2 or 3, characterized in that: In the step of determining the deformation amount at the calculation point by finding the target sub-region that best matches the reference sub-region in the damaged image using a displacement search algorithm, the deformation relationship representation function is as follows: ， in, , .

5. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to claim 4, characterized in that: The step of determining the deformation of the calculation point by finding the target sub-region that best matches the reference sub-region in the damaged image using a displacement search algorithm includes: Obtaining sub-pixel displacement: Sub-pixel search is performed using an interpolation function, specifically a bicubic interpolation function, expressed as follows: ， Among them, in calculating the interpolation coefficients The calculation is performed using three boundary conditions: , and exist Up continuous, In subinterval The above is a cubic algebraic polynomial. ; Obtaining shape function parameters: The parameters are obtained iteratively using the inverse combination Gauss-Newton iteration algorithm.

6. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to any one of claims 1-3 and 5, characterized in that: When selecting images with damage-sensitive bands, the spectra of regions of interest are extracted from hyperspectral images of different damage areas and averaged. The peaks and troughs are selected as damage-sensitive bands, including 971nm, 1014nm, 1104nm, 1237nm and 1300nm.

7. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to claim 6, characterized in that: When performing black-and-white correction on the original hyperspectral images before and after sample damage, the reflectance of the original hyperspectral image is calibrated using the following formula: ， Where I is the corrected hyperspectral image, I r For the original hyperspectral image, I d For a completely black calibration image, I w To calibrate an image on a whiteboard.

8. The method for quantitatively evaluating the degree of early damage to apples based on hyperspectral imaging technology according to any one of claims 1-3, 5 and 7, characterized in that: Before performing black-and-white correction on the original hyperspectral images of the sample before and after damage, the following steps are also included: Damage samples were obtained through impact testing. The impact test was conducted using a drop testing machine with the impact location at the equatorial region of the apple. The apple was placed on a drop platform and dropped freely from different heights onto a rigid plate. Obtain raw hyperspectral images of the sample before and after damage, including raw hyperspectral data of the undamaged sample and hyperspectral data of the damaged sample.

Images