CN103674838B - A kind of fish fats content distribution detection method based on high light spectrum image-forming technology - Google Patents

Abstract

The invention discloses a method for detecting fish fat content distribution based on hyperspectral imaging technology, comprising the following steps: (1) collecting single-band spectral images of fish to be tested at twelve characteristic wavelengths; the twelve characteristic wavelengths 956nm, 975nm, 1033nm, 1081nm, 1129nm, 1153nm, 1205nm, 1234nm, 1282nm, 1354nm, 1522nm, 1627nm; The single-band spectral image in is converted into a reflectance image; (3) The fish fat content corresponding to each pixel in the fish reflectance image is calculated according to the formula. The fish fat content distribution detection method provided by the invention has high detection precision and short detection time, not only reduces environmental pollution, but also reduces detection cost to a certain extent.

Description

A detection method of fish fat content distribution based on hyperspectral imaging technology

technical field

The invention relates to the field of fat content detection, in particular to a method for detecting fish fat content distribution based on hyperspectral imaging technology.

Background technique

Hyperspectral imaging technology is a kind of integration of spectrum and image processing technology. It has the characteristics of spectrum and image at the same time, and can detect the scale accurately to the nanometer level. Due to these advantages, hyperspectral imaging technology has been used in the existing technology. It is widely used in many industries such as agriculture, food, petrochemical, pharmaceutical, feed, etc., especially has great application potential in the rapid detection and research of life information of agricultural products.

Fat is an important part of fish, which can provide essential fatty acids for the human body. It is a nutrient rich in heat energy and can be used as the main source of heat energy for the human body. Each gram of fat can provide 37.62kj (9kcal) heat energy in the body, which is more than carbohydrates And more than double the protein. By detecting and analyzing the fish fat content, the detection and continuous monitoring of the growth status of the fish can be realized, which is of great significance for improving the yield and quality of the fish.

The traditional detection of fish fat content uses the Soxhlet extraction method in the national food safety standard GB/T5009.6-2003 "Fat Determination of Foodstuffs". Although relatively reliable measurement results can be obtained according to this method, it is time-consuming (8-16h ) is laborious, not only consumes a large amount of solvent, but also seriously pollutes the environment, and requires a special instrument (Soxhlet extractor), which is inconvenient to use, and the final result can only obtain the average state of fish fat content, but cannot obtain fish fat content. Distribution of fat content.

The invention with authorized notification number CN101718692B discloses a method for rapidly detecting fat, protein, and dry matter content in bovine colostrum, including the following steps: (1) diluting the bovine colostrum sample with distilled water, wherein the dilution ratio for detecting the fat content is For bovine colostrum: water=3:5, the dilution ratio for detecting protein content is bovine colostrum: water=3:10, and the dilution ratio for detecting dry matter content is bovine colostrum: water=3:5; (2) use The Fourier transform infrared full-spectrum scanning dairy component analyzer detects the sample; 3) multiply the above-mentioned detection result by the corresponding dilution factor in step 1).

The method provided by the invention can only obtain the average content of fat, but not the distribution. Therefore, it is necessary to provide a detection method capable of rapidly detecting the distribution of fat content.

Contents of the invention

The invention provides a method for detecting fish fat content distribution based on hyperspectral imaging technology, which has high detection accuracy and short detection time, not only reduces environmental pollution, but also reduces detection cost to a certain extent.

A method for detecting fish fat content distribution based on hyperspectral imaging technology, comprising the following steps:

(1) Collect single-band spectral images of the fish to be tested at twelve characteristic wavelengths; the twelve characteristic wavelengths are 956nm, 975nm, 1033nm, 1081nm, 1129nm, 1153nm, 1205nm, 1234nm, 1282nm, 1354nm, 1522nm , 1627nm;

(2) Convert the single-band spectral image in step (1) into a reflectance image according to the linear relationship between the gray value of the single-band spectral image and the reflectance;

(3) Calculate the fish fat content corresponding to each pixel in the fish reflectance image according to the following formula;

Y=-273.066X ₁ +149.398X ₂ -239.908X ₃ +184.117X ₄ -118.958X ₅ +98.682X ₆ -105.307X ₇ +78.676X ₈ +82.497X ₉ -180.767X ₁₀ +42.217X _{11 +41.27} +26.707

In the formula: X _a represents the reflectance of a certain pixel in the reflectance image at the nm characteristic wavelength;

Y represents the fish fat content at the corresponding pixel.

As a preference, the step of obtaining the linear relationship between the gray value of the single-band spectral image and the reflectance in the step (2) is as follows:

2-1. Collect the reference single-band spectral images of at least three diffuse reflectance standard plates at twelve characteristic wavelengths, and calculate the gray value of each reference single-band image. In the visible and near-infrared spectral range, the adopted Diffuse reflectance standards have reflectances that differ from each other;

2-2. For each characteristic wavelength, linearly fit the gray value and reflectance of the corresponding reference single-band image to obtain a linear relationship between the gray value and reflectance.

Preferably, there are three to twelve standard diffuse reflection plates.

At each characteristic wavelength, each diffuse reflectance standard plate corresponds to a single-band image, each single-band image corresponds to a gray value, the gray value of the diffuse reflectance standard plate is the independent variable, and the diffuse reflectance standard plate The albedo is the dependent variable, and the relationship between the gray value and the albedo is obtained by linear fitting.

The more the number of diffuse reflection standard plates, the more accurate the relationship between the gray value obtained by linear fitting and the reflectivity, and the corresponding time-consuming is also longer. Preferably, the diffuse reflection standard plates are three, respectively 99% diffuse Reflection Standards, 75% Diffuse Reflectance Standards, and 2% Diffuse Reflectance Standards.

The 99% diffuse reflectance standard plate means that the reflectance of the diffuse reflectance standard plate is 99% in the entire visible and near-infrared spectral range.

The 75% diffuse reflectance standard plate means that the reflectance of the diffuse reflectance standard plate is 75% in the entire visible and near-infrared spectral range.

The 2% diffuse reflectance standard plate means that the reflectance of the diffuse reflectance standard plate is 2% in the entire visible and near-infrared spectral range.

The 99% diffuse reflection standard plate, 75% diffuse reflection standard plate and 2% diffuse reflection standard plate are used to cover the range of reflectance to the greatest extent, making the linear relationship between gray value and reflectance more accurate.

In order to intuitively know the distribution information of fish fat content, preferably, in the step (3), after calculating the fish fat content corresponding to each pixel in the fish reflectance image, a fish fat content distribution map is drawn.

Compared with the prior art, the present invention has the following beneficial technical effects:

a) Collect hyperspectral images of fish at a few selected characteristic wavelengths, use multiple linear regression analysis to obtain the relationship between fish fat content and pixel reflectance in hyperspectral images, quickly and accurately detect fish fat content distribution, and save time.

b) No chemical material is used, no physical and chemical analysis is required, the detection cost is reduced, and the environment is not polluted.

c) It can analyze larger samples and multi-species samples, and can detect the distribution of fish fat content online in real time.

d) The spectral information features of the fish fat content distribution and the spectral imaging information features are fused on the feature layer to obtain an intuitive image of the fish fat content distribution, which is convenient for further analysis.

Description of drawings

Fig. 1 is the relationship figure of reflectivity and wavelength of three diffuse reflection standard plates;

Figure 2 is a distribution diagram of fish fat content.

detailed description

Example 1

(1) Establish the relationship between the gray value of the single-band spectral image and the reflectance

1-a. Collect the reference single-band spectral images of three diffuse reflectance standard plates at twelve characteristic wavelengths (each diffuse reflectance standard plate collects a reference single-band spectral image at each characteristic wavelength), and calculate each Gray value of a reference single-band image.

In the entire visible-near-infrared spectrum range, the reflectances of the diffuse reflectance standard plates used are 99%, 75%, and 2%, respectively. As shown in Figure 1, the three diffuse reflectance standard plates are Diffuse reflectance varies across the range and is the same at all wavelengths for each diffuse reflectance standard.

1-b. For each characteristic wavelength, linearly fit the gray value and reflectance of the corresponding reference single-band image to obtain a linear relationship between the gray value and reflectance.

For each characteristic wavelength, there are corresponding three sets of gray value and reflectance value (each set of gray value and reflectance is obtained from the reference single-band image collected by the same diffuse reflectance standard plate), the gray value is the independent variable, and the reflectance is the dependent variable. Linear fitting is performed on the three sets of gray values and reflectance values to obtain a linear relationship between the gray value and the reflectance value.

In the method of the invention, the acquisition of the linear relationship between the gray value and the reflectance value, the collection of the hyperspectral image and the distribution of the fish fat content are all automatically completed through the ENVI programming program.

(2) Calculation of fish fat content distribution

2-a. Collect 150 fish, first use the hyperspectral image imaging system (ImSpectorV10E, Spectral Imaging Ltd., Oulu, Finland) to scan the single-band spectral images of each fish at twelve characteristic wavelengths; the twelve characteristic wavelengths are respectively It is 956nm, 975nm, 1033nm, 1081nm, 1129nm, 1153nm, 1205nm, 1234nm, 1282nm, 1354nm, 1522nm, 1627nm; each wavelength corresponds to a single-band spectral image, and then adopts GB/T5009.6-2003 "Fat in Food The method in "Determination" measures the fat content of 20 different regions of these 150 fish, that is, each fish is cut into 20 parts, and the 20 regions include all parts of the whole fish.

Among the 150 fish, 100 fish are randomly selected as the modeling set samples, and the remaining 50 fish are used as the prediction set.

2-b. According to the linear relationship between the gray value of the single-band spectral image and the reflectance, the single-band spectral image of each fish at twelve characteristic wavelengths is converted into a reflectance image.

Based on the relationship between the gray value at the twelve characteristic wavelengths and the reflectance in step (1), the single-band spectral image of the fish to be tested (each pixel in the single-band spectral image corresponds to a Each pixel has a different gray value) into a reflectance image, and each pixel in the reflectance image corresponds to a different reflectance.

For the 100 fish in the modeling set, the corresponding reflectance image can be obtained for the hyperspectral image of each region, and the average reflectance of each region can be obtained after averaging, and the fish fat content in each region (obtained by the national standard ) and the average reflectance fitting to obtain the relationship between the fish fat content and the average reflectance is shown in the following formula (I),

Y'=-273.066X' ₁ +149.398X' ₂ -239.908X' ₃ +184.117X' ₄ -118.958X' ₅ +98.682X' ₆ -105.307X' ₇ +78.676X' ₈ +82.497X' ₉ -180.767 X' ₁₀ +42.217X' ₁₁ +42.736X' ₁₂ +26.707 (I)

In the formula: _X'a represents the average reflectance of the reflectance image at the characteristic wavelength of anm;

Y' represents the fish fat content at the corresponding pixel.

The formula (I) obtained by fitting the average reflectance and fish fat content expresses the relationship between the average reflectance and fish fat content, and formula (I) also reflects the relationship between the reflectance at each pixel and the fat content. (I) Obtain formula (II) as follows:

Y=-273.066X ₁ +149.398X ₂ -239.908X ₃ +184.117X ₄ -118.958X ₅ +98.682X ₆ -105.307X ₇ +78.676X ₈ +82.497X ₉ -180.767X ₁₀ +42.217X _{11 +41.27} +26.707

In the formula: X _a represents the reflectance of a certain pixel in the reflectance image at the nm characteristic wavelength;

Y represents the fish fat content at the corresponding pixel.

Substitute the reflectance corresponding to each pixel in the reflectance image into formula (II) to calculate the fat content, and obtain the fat content at each pixel in the fish image to be tested, and then draw the fat of the fish accordingly Content distribution map, to obtain the fat content distribution information at each point of the fish, the fat content distribution of a region is shown in Figure 2.

Using the method of the present invention to detect the predicted fat content (obtained by substituting the average reflectance into formula (I)) of 50 fish in the prediction set (each fish corresponds to 20 areas) and the result of the real fat content detected by the national standard See Table 1 for comparison.

Table 1

data set Sample size correlation coefficient root mean square error modeling set 100 0.9045 1.4678 prediction set 50 0.8867 1.5236

As can be seen from Table 1, the prediction result of the detection method proposed by the present invention is highly correlated with the measured value of the national standard method.

Comparative example 1

Select twelve characteristic wavelengths, namely 952nm, 970nm, 1030nm, 1085nm, 1122nm, 1150nm, 1209nm, 1231nm, 1288nm, 1350nm, 1528nm, 1625nm, and establish fat content and reflectance in the same way based on these twelve characteristic wavelengths The relationship of is shown in formula (III):

Y=-270.235X ₁ +142.245X ₂ -235.235X ₃ +180.245X ₄ -115.245X ₅ +95.147X ₆ -109.365X ₇ +74.286X ₈ +86.347X ₉ -185.225X ₁₀ +48.365X _{11 +41.2} +25.247

(III)

In formula (III): X _a represents the reflectance of a certain pixel in the reflectance image at the characteristic wavelength of anm;

Y represents the fat content at the corresponding pixel.

The single-band spectral images of fish were obtained at twelve characteristic wavelengths, and the fat content of fish was calculated based on formula (III). The comparison with the real fat content detected by the national standard method is shown in Table 2.

Table 2

data set Sample size correlation coefficient root mean square error modeling set 100 0.7954 1.9638

prediction set 50 0.7682 2.065

Comparative example 2

Select twelve characteristic wavelengths, namely 955nm, 974nm, 1037nm, 1081nm, 1125nm, 1152nm, 1204nm, 1235nm, 1287nm, 1357nm, 1523nm, 1629nm, and establish fat content and reflectance in the same way based on these twelve characteristic wavelengths The relationship of is shown in formula (IV):

Y=-271.983X ₁ +144.324X ₂ -231.876X ₃ +181.765X ₄ -114.745X ₅ +92.356X ₆ -105.485X ₇ +73.348X ₈ +85.395X ₉ -183.145X ₁₀ +45.874X _{11 +41.68} +24.985 (IV)

In formula (IV): X _a represents the reflectance of a certain pixel in the reflectance image at the characteristic wavelength of anm;

Y represents the fat content at the corresponding pixel.

The single-band spectral images of fish were obtained at twelve characteristic wavelengths, and the fat content of fish was calculated based on formula (IV). The comparison with the real fat content detected by the national standard method is shown in Table 3.

table 3

data set Sample size correlation coefficient root mean square error modeling set 100 0.7256 2.2638 prediction set 50 0.7195 2.3461

From the results of Example 1 and Comparative Examples 1 and 2, the selection of characteristic wavelengths has an important influence on whether the detection of fish fat content is accurate. The present invention obtains a detection result with a high coefficient of determination by selecting a suitable characteristic wavelength, which is used for Quickly perform spatial distribution results of fish fat content.

Claims (4)

1., based on a fish fats content distribution detection method for high light spectrum image-forming technology, it is characterized in that, comprise the following steps:

(1) the single band spectrum picture of fish to be measured 12 characteristic wave strong points is gathered; Described 12 characteristic wavelengths are respectively 956nm, 975nm, 1033nm, 1081nm, 1129nm, 1153nm, 1205nm, 1234nm, 1282nm, 1354nm, 1522nm, 1627nm;

(2) according to the gray-scale value of single band spectrum picture and the linear relationship of reflectivity, the single band spectrum picture in step (1) is converted into albedo image;

The obtaining step of the gray-scale value of single band spectrum picture and the linear relationship of reflectivity is as follows:

2-1, collection at least three pieces of diffuse reflection on-gauge plates are at the benchmark single band spectrum picture of 12 characteristic wave strong points, ask for the gray-scale value of every width benchmark single band image, within the scope of visible and near infrared spectrum, the diffuse reflection on-gauge plate adopted has mutually different reflectivity;

2-2, for each characteristic wavelength, the gray-scale value of corresponding benchmark single band image and reflectivity are carried out linear fit, obtains the linear relationship of gray-scale value and reflectivity;

(3) the fish fats content in the albedo image of fish corresponding to each pixel is calculated according to following formula;

Y＝-273.066X ₁+149.398X ₂-239.908X ₃+184.117X ₄-118.958X ₅+98.682X ₆-105.307X ₇+78.676X ₈+82.497X ₉-180.767X ₁₀+42.217X ₁₁+42.736X ₁₂+26.707

In formula: X _arepresent in the albedo image of anm characteristic wave strong point, the reflectivity of a certain pixel;

Y represents the fish fats content at respective pixel point place.

2., as claimed in claim 1 based on the detection method of the fish fats content of high light spectrum image-forming technology, it is characterized in that, described diffuse reflection on-gauge plate is three ~ 12 pieces.

3., as claimed in claim 2 based on the detection method of the fish fats content of high light spectrum image-forming technology, it is characterized in that, described diffuse reflection on-gauge plate is three pieces, is respectively 99% diffuse reflection on-gauge plate, 75% diffuse reflection on-gauge plate and 2% diffuse reflection on-gauge plate.

4. as claimed in claim 1 based on the detection method of the fish fats content of high light spectrum image-forming technology, it is characterized in that, in described step (3), after calculating the fish fats content in the albedo image of fish corresponding to each pixel, draw fish fats content distribution figure.