With the improvement of people's living standards, consumers are paying more and more attention to the quality and safety of fruits and vegetables. As caused
On the surface of the fruit, there are black and white spots inside the rot, and the fruit is damaged or damaged due to transportation, etc., which seriously affects the health of the consumer. Therefore, the rapid and effective identification of black and white spots and bruises of fruits has important research value.
Hyperspectral image technology combines the advantages of spectral analysis and image processing. Many scholars at home and abroad have tested the internal and external quality characteristics of the research object. For example, Zhao Jiewen used hyperspectral image technology to detect slight damage of fruit, and the accuracy rate was 88.57%. Jasper G.
Tallada
The hyperspectral image technique was applied to test the surface damage of strawberry with different maturity, the surface defect of apple and the maturity of mango. Wang Yutian et al. used fluorescence spectroscopy to detect pesticide residues on the surface of fruits; Hu Shufen et al. used laser technology to test the pesticide residues on the surface of fruits; Xue Long et al. focused on the pesticide residues on the surface of fruits, and studied the high concentration of navel oranges. Using the hyperspectral image system with a spectral range of 425-725 nm, it was found that the detection of pesticide residues at higher concentrations was better. In this paper, hyperspectral image technology is used to detect black and white plaques and damaged areas of different fruits, so as to achieve the purpose of rapid identification of black and white spots and damaged areas of fruits.
two,
Test materials and methods
2.1
Rapid identification of orange spots and damage using imaging spectroscopy
Experimental Materials
In this study, oranges were used as research objects to analyze the black and white plaques and damaged areas of oranges. Among them, black and white spots and damage of oranges are intentionally formed by non-humans.
2.2 Experimental equipment
Hyperspectral imaging data acquisition adopts Sichuan Shuangli Hepu Technology Co., Ltd.
GaiaSorter
Hyperspectral sorter system. The system is mainly composed of a hyperspectral imager (V10E),
CCD
Camera, light source, black box, computer, structure and real map
1
. Experimental instrument parameter settings as table
1
.
table
1 GaiaSorter
Hyperspectral sorter system parameters
Figure
1
GaiaSorter
Hyperspectral sorter structure diagram and real map
2.3
Image processing analysis
use
SpecView
with
ENVI/IDL
Preprocessing and analysis of hyperspectral data, mirror transformation in preprocessing, black and white frame calibration
SpecView
In progress; analysis of other data in
ENVI/IDL
In progress.
three,
Rapid identification of orange spots and damage using imaging spectroscopy
Results and discussion
3
Spectral analysis of black spotted areas, normal areas and background of oranges
Taking the front and side of the orange as an example, taking 50 pixels around the orange dark spot area, the whiteboard area, the normal area, and the background, respectively, obtain the spectral reflectance of 50 pixels in the three different positions, and Find the reflectance mean of these 50 pixels, as shown in Figure 3. It can be seen from the figure that in the range of 580-700 nm, the spectral reflectance of the black spot, white spot and normal area of ​​the orange is more obvious, and the background in this spectral range, the spectral reflectance rises more slowly, so it can be This area quickly identifies oranges. Regardless of the positive or side spectrum of the orange, the spectral reflectance of the black spot region of the orange is lower than the white spot and normal area of ​​the orange in the range of 530-1000 nm. In the range of 400-1000 nm, the white spot area and the normal area are significantly different in the blue light band.
Figure 3 Spectral reflectance of black spotted areas, normal areas, and background of oranges
3.3
Minimum noise separation of oranges
The MNF transform is performed on the orange hyperspectral image that has been mirror-transformed and black-and-white frame calibrated (as shown in Figure 4, from left to right: apple, frontal orange, and side orange), respectively, with effective information-based bands and noise-based Bands, and are arranged in order of increasing signal to noise ratio. The main information of the original data is concentrated in the band with the large eigenvalue, and the band with the small eigenvalue is mainly dominated by noise. The majority of the eigenvalues ​​close to 0 is noise, and it is preferable to select a band having a high eigenvalue. As can be seen from Fig. 4, when the number of characteristic values ​​of the oranges is 7, the eigenvalues ​​tend to be zero and there is no significant change.
Figure 4 Flow chart of extraction of decayed areas and agricultural residual areas
3.4
Minimum noise separation transform
Due to the large number of bands of hyperspectral remote sensing data, there is a great correlation between the bands. In order to overcome the dimensionality disaster, the minimum noise separation transform is used for band selection to achieve the purpose of optimizing data, removing noise and data dimensionality reduction.
The Minimum Noise Separation Transform (MFF) is an improved method of principal component transformation (PCA). PCA is a linear transformation. After transformation, each principal component is independent of each other. As the principal component increases, the amount of information contained in the component decreases. The first principal component contains the largest amount of information, and the second principal component and The first principal component is irrelevant and contains the largest amount of information in the remaining components, and so on. However, PCA is sensitive to noise. In the transformed principal components, the signal-to-noise ratio of large information is not necessarily high. When the variance of the noise contained in the main component of a large amount of information is larger than the variance of the signal, the master The image quality formed by the component components is poor. In response to the inadequacy of the PCA transformation, Green and Berman proposed the Minimum Noise Separation Transform (MNF), which
Not only can the inner dimension (number of bands) of the image data be determined, the noise in the data can be separated, and the computational demand in subsequent processing can be reduced. The MNF transform is a linear transformation based on image quality, and the components of the transform result are arranged in accordance with the signal-to-noise ratio from large to small. Most of the noise is concentrated in components with small features after MNF transformation. Rather than the PCA transform, the variance is arranged from large to small, thus overcoming the effect of noise on image quality.
3.4.
1
Black and white patch area recognition of oranges based on MNF
Figure
5
The front side of the orange, the original side view (hyperspectral RGB color synthesis), and the gray level map of the seven characteristic values ​​before the MNF transformation are listed. From the gradation map of the eigenvalues ​​of the MNF transformation of the frontal orange, the first eigenvalue gray map can better distinguish the background and the orange black spots. However, the background and the orange shading cannot be distinguished from each other; the second and third features The luminance portion of the grayscale image is dark spots, but the non-spotted oranges are also incorrectly recognized as black spots; the fourth eigenvalue grayscale image can better recognize the black spots and white spots of the oranges, that is, the brighter portions are The black spots and white spots of oranges have a good recognition effect; the gray scales of the eigenvalues ​​of the 5th, 6th, 7th and later are unable to correctly identify the dark spots and white spots.
Figure
5
RGB original image of the front of the orange and the grayscale map of the first 7 MNF eigenvalues
As shown
6
From the gradation map of the eigenvalues ​​of the MNF transform of the side orange, the first eigenvalue gray map can better distinguish the background and the orange; the second and third eigenvalue gray map recognition effects are not satisfactory, black and white spots, background The fourth eigenvalue grayscale image can identify the orange black spot, but also mistakenly identify some spotless oranges as black spots; the fifth eigenvalue grayscale image can better recognize the oranges. Black and white spots, damaged areas, but some backgrounds are mistakenly recognized as black and white spots. The grayscale maps of the eigenvalues ​​at the sixth, seventh, and subsequent levels do not correctly identify dark spots, white spots, and damaged areas.
Figure
6
RGB original image of orange measuring surface and grayscale map of the first 7 MNF characteristic values
3.6 Rapid identification of orange spots and damaged areas based on vegetation index and threshold segmentation
According to the variation of spectral reflectance of black and white spotted area, damaged area, normal area and background of Fig. 3, the vegetation index NDVI (706, 590) was constructed to remove the background and mask MNF5. Finally, the gray density was used to segment and the orange spot was represented by red. , damaged area, yellow represents slight damage or tiny orange spots, as shown
7
Shown. It can be seen from the figure that both the vegetation index and the threshold segmentation method can quickly and accurately identify the spots and damaged areas, regardless of the front or side of the orange.
Figure
7
Rapid identification of orange spots and damage areas based on vegetation index and threshold segmentation
four
discuss
The application of hyperspectral imaging technology to the rapid identification of fruit spots and damaged areas has demonstrated the superiority of its "combination of maps". Although the color of fruit damage and the pigmentation of the fruit epidermis can be recognized by the naked eye, in industrial production, it is time-consuming and labor-intensive to select non-injured and spotless fruits by manpower alone. Using imaging hyperspectral technology, the spectral reflectance of different fruits is obtained, and the characteristic bands of damage and spots are found out. The vegetation index is constructed by using characteristic bands to realize the rapid and effective identification of fruit damage and spotted areas, and to automatically select high-quality fruits. purpose. The results of this study show that using hyperspectral imaging technology, using minimum noise separation, vegetation index and other methods can effectively identify fruit damage and spot areas, but the minimum noise separation method is more complex, the operation speed is slower, not suitable for industrial use. Application in production, and the vegetation index algorithm is simple, and only four bands can be used for four operations to achieve rapid identification of fruit damage and spots.
Privacy Hydrogel Screen Protector
The Privacy Screen Protector can display a black screen directly in front of the screen at an angle greater than 30° to effectively block the sight of people next to it, while achieving a perfect balance between black screen privacy and daytime clarity.
The Screen Protector can protect the edges and gaps of the display so that it can extend to the entire screen surface, thereby achieving maximum coverage without any exposed space.
The Self-Healing Screen Protector can provide the best protection for your phone from drops, bumps, scratches and normal wear and tear. Using an oleophobic waterproof coating can prevent sweat and grease from remaining on your fingerprints, keeping you simple all day long.
The 0.14mm thick Ultra-Thin Protective Film has a "real touch" feel and ensures fast response performance.
If you want to know more about Privacy Screen Protector products, please click the product details to view the parameters, models, pictures, prices and other information about Privacy Screen Protector.
Whether you are a group or an individual, we will try our best to provide you with accurate and comprehensive information about the Privacy Screen Protector!
Anti-peep Screen Protector, Privacy Screen Protector, Anti-spy Screen Protector, Privacy Protective Film, Privacy Hydrogel Film,Anti-peeping Screen Protector
Shenzhen Jianjiantong Technology Co., Ltd. , https://www.jjtscreenprotector.com