Necrotic small intestine characteristic wave band identification method based on least clipping square method

Abstract

The invention discloses a necrotic small intestine characteristic wave band identification method based on an LTS algorithm, and the method comprises the steps: converting normal small intestine hyperspectral data and necrotic small intestine hyperspectral data, and taking the converted data as an explanatory variable and a response variable of the LTS algorithm; the method comprises the following steps: determining an abnormal spectrum channel shared by an explanatory variable and a response variable by adopting an LTS algorithm to obtain a normal explanatory variable and a normal response variable after the abnormal spectrum channel is removed, constructing a regression model based on the normal explanatory variable and the normal response variable by adopting a least square method, and obtaining a weight through the regression model; the response variables are reconstructed on the basis of the weights, and finally the necrotic small intestine characteristic wave band is determined on the basis of the response variables and standardized residual errors of the reconstructed response variables on all the spectrum channels. The method has the advantages that the characteristic wave bands of the necrotic small intestine can be automatically identified, and the method is high in precision and low in labor cost.

Description

Technical Field

[0001] The invention belongs to the technical field of spectral analysis, and in particular relates to a method for identifying characteristic bands of necrotic small intestine based on a minimum clipping squares method. Background Art

[0002] Necrotizing enterocolitis (NEC) is a common gastrointestinal emergency in newborns. The mortality rate of newborns with necrotizing enterocolitis (NEC) can reach 30%, and the postoperative mortality rate of extremely low birth weight (ELBW) infants with the disease is still 50%. In the emergency surgical treatment of necrotizing enterocolitis (NEC), resection of gangrenous intestinal segments is still the main treatment method. Therefore, in clinical practice, appropriate technical means are needed to accurately identify necrotic intestinal segments.

[0003] Traditional methods for identifying necrotic intestinal segments mostly rely on the doctor's experience or microscopic observation after slicing the small intestinal tissue. However, these methods have the disadvantage of high labor costs. Therefore, there is a strong demand for convenient, fast, accurate and non-destructive differentiation of necrotic small intestine. In recent years, spectral technologies such as Fourier transform infrared spectroscopy and hyperspectroscopy have been widely used for the identification of necrotic small intestine due to their advantages of non-destructiveness, rapidity and high sensitivity, and have achieved high recognition accuracy. Spectral technology has become one of the powerful and effective auxiliary tools for clinical identification of necrotic small intestine. As an indispensable part of spectral analysis, the extraction and identification of spectral data features is very necessary. Feature extraction and identification can reduce the dimensionality of spectral data, highlight the useful information of the spectrum, remove redundant information, and help improve model generalization, prevent overfitting, and reduce computing time and computing resources.

[0004] Existing studies on the qualitative study of necrotic small intestine are very rich, but the research on the cause of small intestinal necrosis is still limited. The necrotic small intestine and the normal small intestine have different absorption of spectral radiation, which causes the difference between the necrotic small intestine and the normal small intestine to be characterized in a certain part of the spectral channel of the spectrum, that is, the characteristic band that distinguishes the necrotic small intestine. By identifying and distinguishing the characteristic bands of necrotic small intestine and analyzing the chemical bonds and functional groups corresponding to the characteristic bands, we can provide a theoretical basis and technical support for explaining the cause of small intestinal necrosis.

[0005] The existing identification methods of characteristic bands of necrotic small intestine rely on human experience or reference to existing literature for selection. Literature 1 "Zhang, et al. Discrimination Between normal and Necrotic Small Intestinal Tissue Using Hyperspectral Imaging and Unsupervised Classification. Journal of Biophotonics, 2023" selects the reflectance spectra of oxyhemoglobin and deoxyhemoglobin as characteristic bands by observing and comparing the average spectra of normal and necrotic small intestines. Literature 2 "Mehdorn, et al. Hyperspectral Imaging (HSI) in Acute Mesenteric Ischemia to Detect Intestinal Perfusion Deficits. Journal of Surgical Research, 2020" selects the reflectance spectrum of hemoglobin as the characteristic band by consulting relevant materials. Both methods for identifying characteristic bands are not reliable. The method in literature 1 is unreliable because the average spectra of normal and necrotic small intestines are very similar and the difference is small, so it is difficult to accurately identify the characteristic bands by observation. The reason why method 2 is unreliable is that the characteristic bands obtained by consulting the data may not be consistent with the characteristic bands in the actual experiment. This is because the differences in experimental samples and experimental environments will affect the positions of the actual characteristic bands.

[0006] Both methods have disadvantages such as low accuracy and high labor cost, and are unable to achieve automatic identification of characteristic bands of necrotic small intestine. Summary of the invention

[0007] The technical problem solved by the present invention is to provide a method for identifying characteristic bands of necrotic small intestine based on the minimum clipping squares method, which can realize automatic identification of characteristic bands of necrotic small intestine and has high accuracy and low labor cost.

[0008] The technical solution adopted by the present invention to solve the above technical problems is: a method for identifying characteristic bands of necrotic small intestine based on the LTS algorithm, by converting normal small intestine hyperspectral data and necrotic small intestine hyperspectral data as explanatory variables and response variables of the LTS algorithm, using the LTS algorithm to determine the abnormal spectral channels shared by the explanatory variables and the response variables to obtain normal explanatory variables and normal response variables after removing the abnormal spectral channels, then using the least squares method to build a regression model based on the normal explanatory variables and the normal response variables, obtaining weights through the regression model, and reconstructing the response variables based on the weights, and finally determining the characteristic bands of the necrotic small intestine based on the standardized residuals of the response variables and the reconstructed response variables on each spectral channel.

[0009] A method for identifying characteristic bands of necrotic small intestine based on LTS algorithm comprises the following steps:

[0010] Step 1: Acquire normal small intestine hyperspectral data and necrotic small intestine hyperspectral data with the same number of spectral channels through a hyperspectral data acquisition device. The normal small intestine hyperspectral data is a data matrix with n rows and p columns, and the necrotic small intestine hyperspectral data is a data matrix with m rows and p columns, where p represents the number of channels of the hyperspectral data acquisition device, n represents the number of normal small intestine samples in the normal small intestine hyperspectral data, m represents the number of necrotic small intestine samples in the necrotic small intestine hyperspectral data, n is an integer greater than or equal to 20, and m is an integer greater than 0; in the normal small intestine hyperspectral data, each row of data represents a normal small intestine sample, and each column of each normal small intestine sample represents a spectral channel of the normal small intestine sample, that is, each normal small intestine sample has p spectral channels. spectral channel; the sth row of data in the normal small intestine hyperspectral data is called the sth normal small intestine sample, s = 1, 2, ..., n, the tth column of the sth normal small intestine sample is called the tth spectral channel of the sth normal small intestine sample, t = 1, 2, ..., p; in the necrotic small intestine hyperspectral data, each row of data represents a necrotic small intestine sample, and each column in each necrotic small intestine sample represents a spectral channel of the necrotic small intestine sample, that is, each necrotic small intestine sample has p spectral channels; the hth row of data in the necrotic small intestine hyperspectral data is called the hth necrotic small intestine sample, and the tth column of the hth necrotic small intestine sample is called the tth spectral channel of the hth necrotic small intestine sample, h = 1, 2, ..., m; the normal small intestine hyperspectral data is recorded as X org , record the hyperspectral data of necrotic small intestine as Y org ;

[0011] Step 2: Use window smoothing, baseline correction, and standard normal transformation to transform the normal small intestine hyperspectral data X orgPreprocessing is performed to obtain preprocessed normal small intestine hyperspectral data, which is recorded as X, where X includes n preprocessed normal small intestine samples, each row of data in X represents a preprocessed normal small intestine sample, each column in each preprocessed normal small intestine sample represents a spectral channel of the preprocessed normal small intestine sample, and each preprocessed normal small intestine has p spectral channels; the s-th row of data in X is called the s-th preprocessed normal small intestine sample, the t-th column of the s-th preprocessed normal small intestine sample is called its t-th spectral channel, the s-th preprocessed normal small intestine sample corresponds to the s-th normal small intestine sample, and the t-th spectral channel of the s-th preprocessed normal small intestine sample corresponds to the t-th spectral channel of the s-th normal small intestine sample;

[0012] Window smoothing, baseline correction, and standard normal transformation were used to analyze the hyperspectral data Y of necrotic small intestine. org Preprocessing is performed to obtain preprocessed necrotic small intestine hyperspectral data, which is recorded as Y, where Y includes m preprocessed necrotic small intestine samples, each row of data in Y represents a preprocessed necrotic small intestine sample, each column in each preprocessed necrotic small intestine sample represents a spectral channel of the preprocessed necrotic small intestine sample, and each preprocessed necrotic small intestine has p spectral channels; the hth row of data in Y is called the hth preprocessed necrotic small intestine sample, the tth column of the hth preprocessed necrotic small intestine sample is called its tth spectral channel, the hth preprocessed necrotic small intestine sample corresponds to the hth necrotic small intestine sample, and the tth spectral channel of the hth preprocessed necrotic small intestine sample corresponds to the tth spectral channel of the hth necrotic small intestine sample;

[0013] Step 3, transpose the preprocessed normal small intestine hyperspectral data X to obtain transposed normal small intestine hyperspectral data, which is recorded as X′, X′ includes n transposed normal small intestine samples, each column of data in X′ represents a transposed normal small intestine sample, each row in each transposed normal small intestine sample represents a spectral channel of the transposed normal small intestine sample, and each transposed normal small intestine has p spectral channels; the s-th column of data in X′ is called the s-th transposed normal small intestine sample, the t-th row of the s-th transposed normal small intestine sample is called its t-th spectral channel, the s-th transposed normal small intestine sample corresponds to the s-th normal small intestine sample, and the t-th spectral channel of the s-th transposed normal small intestine sample corresponds to the t-th spectral channel of the s-th normal small intestine sample;

[0014] The preprocessed necrotic small intestine hyperspectral data Y are transposed to obtain the transposed necrotic small intestine hyperspectral data, which is denoted as Y′. Y′ includes m transposed necrotic small intestine samples. Each column of data in Y′ represents a transposed necrotic small intestine sample. Each row in each transposed necrotic small intestine sample represents a spectral channel of the transposed necrotic small intestine sample. Each transposed necrotic small intestine has p spectral channels. The hth column of data in Y′ is called the hth transposed necrotic small intestine sample, denoted as y′.h , the t-th row of the h-th transposed necrotic small intestine sample is called its t-th spectral channel, the h-th transposed necrotic small intestine sample corresponds to the h-th necrotic small intestine sample, and the t-th spectral channel of the h-th transposed necrotic small intestine sample corresponds to the t-th spectral channel of the h-th necrotic small intestine sample;

[0015] Step 4: Use the Duplex algorithm to sample 20 transposed normal small intestine samples from the transposed normal small intestine hyperspectral data X′ to form a transposed normal small intestine sample matrix, denoted as D′. The transposed normal small intestine sample matrix D′ is a data matrix with p rows and 20 columns. Each column of data represents a transposed normal small intestine sample sampled by the Duplex algorithm. The data in the uth column of D′ is called the uth sampled transposed normal small intestine sample, and u=1, 2, …, 20.

[0016] Step 5, set the variable i and initialize i to 1;

[0017] Step 6: Determine the i-th abnormal spectral channel. The specific process is as follows:

[0018] S6.1. Take D′ as the explanatory variable of the LTS algorithm, y′ i As the response variable of the LTS algorithm, the LTS algorithm is first used to label the explanatory variable D′ and the response variable y′ i Common abnormal spectral channels, the number of channels marked as abnormal spectral channels is denoted as q i , remove its explanatory variable D′ and response variable y′ i The spectral channels marked as abnormal spectral channels in ;

[0019] The data matrix obtained by removing the abnormal spectral channel from D′ is called the normal explanatory variable y′ i The data matrix obtained after removing the abnormal spectral channels is called the normal response variable Normal explanatory variables pq i A data vector with 20 rows and columns, a normal response variable pq i A data vector with 1 row and 1 column;

[0020] S6.2. Least squares method based on normal explanatory variables With normal response variable Build regression models;

[0021] S6.3. Obtain normal explanatory variables through this regression model The weight w i , weight w iIt is a data vector with 20 rows and 1 column. The data in the u-th row is the weight value of the u-th sample transposed normal small intestine sample;

[0022] S6.4. Calculation of D′ and w i The product of is the predicted response variable, denoted as z′ i , that is, the i-th transposed necrotic small intestine sample y′ i The reconstruction data;

[0023] S6.5. Calculation of z′ i and y′ i The standardized residuals on p spectral channels are obtained by obtaining a standardized residual data r′ consisting of p standardized residuals. i , standardized residual data r′ i is a data vector with p rows and 1 column. Each row of data is a standardized residual. The jth row of data is the same as the i-th transposed necrotic small intestine sample y′. i Corresponding to the jth spectral channel, the jth row of data is the i-th transposed necrotic small intestine sample y′ i Normalized residual on the jth spectral channel, j = 1, 2, …, p;

[0024] S6.6. Calculation of r′ i The median and median absolute deviation (MAD) of the i-th transposed necrotic small intestine sample y′ are selected as the sum of the median and three times the median absolute deviation. i Threshold for classifying abnormal spectral channels;

[0025] S6.7. Construct a column vector L′ i , determine the i-th transposed necrotic small intestine sample y′ i Is the standardized residual on the jth spectral channel less than the i-th transposed necrotic small intestine sample y′ i The threshold for dividing the abnormal spectral channel is, if it is less than, then the column vector L′ i The jth row in is assigned a value of 0, and if it is not less than , it is assigned a value of 1; L′ i is a data vector with p rows and 1 column. The data in the jth row represents the i-th transposed necrotic small intestine sample y′ i Whether the jth spectral channel of is an abnormal spectral channel, if it is equal to 0, it is a normal spectral channel, if it is equal to 1, it is an abnormal spectral channel;

[0026] Step 7: Determine whether the current value of i is equal to m. If not, update the value of i by adding 1 to the current value of i, and then return to step 6. If equal, proceed to step 8.

[0027] Step 8: Use L′1 to L′m Construct the matrix L′, L′ h is the h-th column data of L′;

[0028] Step 9: Count the frequency of the number “1” in the j-th row of data in L′, that is, the number of necrotic small intestine samples marked as abnormal in the j-th spectral channel of the necrotic small intestine hyperspectral data, and record it as m j , compare the number of necrotic small intestine samples m marked as abnormal in the jth spectral channel of the necrotic small intestine hyperspectral data j The size of the number of necrotic small intestine samples m in the necrotic small intestine hyperspectral data, if m j If m is equal to m, then the jth spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is the characteristic band of necrotic small intestine. j If m is less than m and greater than 0, the jth spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is useless information caused by objective factors such as noise. j If it is equal to 0, the j-th spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is neither the characteristic band of the necrotic small intestine nor the useless information caused by objective factors such as noise, thus achieving high-precision recognition of the characteristic bands of the necrotic small intestine.

[0029] Compared with the prior art, the advantage of the present invention is that the normal small intestine hyperspectral data and the necrotic small intestine hyperspectral data are converted into explanatory variables and response variables of the LTS algorithm, the abnormal spectral channels shared by the explanatory variables and the response variables are determined by the LTS algorithm, and then the normal explanatory variables and the normal response variables after removing the abnormal spectral channels are obtained, and then the least squares method is used to build a regression model based on the normal explanatory variables and the normal response variables, and the weights are obtained through the regression model, and the response variables are reconstructed based on the weights, and finally, based on the standardized residuals of the response variables and the reconstructed response variables on each spectral channel, the abnormal spectral channels of the necrotic small intestine hyperspectral data are divided, and finally the high-precision identification of the characteristic bands of the necrotic small intestine is determined, thereby the present invention can realize the automatic identification of the characteristic bands of the necrotic small intestine, and has higher accuracy and lower labor cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the characteristic band identification method of necrotic small intestine based on LTS algorithm of the present invention;

[0031] Figure 2 When the characteristic band identification method of necrotic small intestine based on LTS algorithm of the present invention is used to identify the characteristic band of necrotic small intestine of a rabbit, the normal small intestine hyperspectral data and the necrotic small intestine hyperspectral data are derived from the same rabbit.

[0032] Figure 3When the characteristic band identification method of necrotic small intestine based on LTS algorithm of the present invention is used to identify the characteristic band of necrotic small intestine of rabbits, the result diagram of the characteristic band identification of necrotic small intestine of rabbits whose normal small intestine hyperspectral data and necrotic small intestine hyperspectral data are from different sources;

[0033] Figure 4 The result diagram of the necrotic small intestine characteristic band recognition of rabbit 1 at 10, 30, 60 and 90 minutes of partial small intestine necrosis is shown when the necrotic small intestine characteristic band recognition method based on the LTS algorithm of the present invention is used to recognize the necrotic small intestine characteristic band of rabbit 1;

[0034] Figure 5 The result diagram of the necrotic small intestine characteristic band recognition of rabbit 2 at 10, 30, 60 and 90 minutes of partial small intestine necrosis is shown when the necrotic small intestine characteristic band recognition method based on the LTS algorithm of the present invention is used to recognize the necrotic small intestine characteristic band of rabbit 2;

[0035] Figure 6 The characteristic band recognition result diagram of the necrotic small intestine of rabbit 3 at 10, 30, 60 and 90 minutes after partial small intestine necrosis is shown when the characteristic band recognition method of the necrotic small intestine based on the LTS algorithm of the present invention is used to recognize the characteristic band of the necrotic small intestine of rabbit 3. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below with reference to the accompanying drawings.

[0037] Embodiment 1: A method for identifying characteristic bands of necrotic small intestine based on the LTS algorithm, wherein the normal small intestine hyperspectral data and the necrotic small intestine hyperspectral data are converted and used as the explanatory variables and response variables of the LTS algorithm, and the abnormal spectral channels shared by the explanatory variables and the response variables are determined by the LTS algorithm to obtain normal explanatory variables and normal response variables after removing the abnormal spectral channels, and then a regression model is constructed based on the normal explanatory variables and the normal response variables using the least squares method, and weights are obtained through the regression model, and the response variables are reconstructed based on the weights, and finally the characteristic bands of the necrotic small intestine are determined based on the standardized residuals of the response variables and the reconstructed response variables on each spectral channel.

[0038] Embodiment 2: Figure 1 As shown, a method for identifying characteristic bands of necrotic small intestine based on LTS algorithm includes the following steps:

[0039] Step 1: Acquire normal small intestine hyperspectral data and necrotic small intestine hyperspectral data with the same number of spectral channels through a hyperspectral data acquisition device. The normal small intestine hyperspectral data is a data matrix with n rows and p columns, and the necrotic small intestine hyperspectral data is a data matrix with m rows and p columns, where p represents the number of channels of the hyperspectral data acquisition device, n represents the number of normal small intestine samples in the normal small intestine hyperspectral data, m represents the number of necrotic small intestine samples in the necrotic small intestine hyperspectral data, n is an integer greater than or equal to 20, and m is an integer greater than 0; in the normal small intestine hyperspectral data, each row of data represents a normal small intestine sample, and each column of each normal small intestine sample represents a spectral channel of the normal small intestine sample, that is, each normal small intestine sample has p spectral channels. spectral channel; the sth row of data in the normal small intestine hyperspectral data is called the sth normal small intestine sample, s = 1, 2, ..., n, the tth column of the sth normal small intestine sample is called the tth spectral channel of the sth normal small intestine sample, t = 1, 2, ..., p; in the necrotic small intestine hyperspectral data, each row of data represents a necrotic small intestine sample, and each column in each necrotic small intestine sample represents a spectral channel of the necrotic small intestine sample, that is, each necrotic small intestine sample has p spectral channels; the hth row of data in the necrotic small intestine hyperspectral data is called the hth necrotic small intestine sample, and the tth column of the hth necrotic small intestine sample is called the tth spectral channel of the hth necrotic small intestine sample, h = 1, 2, ..., m; the normal small intestine hyperspectral data is recorded as X org , record the hyperspectral data of necrotic small intestine as Y org ;

[0040] Step 2: Use window smoothing, baseline correction, and standard normal transformation to transform the normal small intestine hyperspectral data X org Preprocessing is performed to obtain preprocessed normal small intestine hyperspectral data, which is recorded as X, where X includes n preprocessed normal small intestine samples, each row of data in X represents a preprocessed normal small intestine sample, each column in each preprocessed normal small intestine sample represents a spectral channel of the preprocessed normal small intestine sample, and each preprocessed normal small intestine has p spectral channels; the s-th row of data in X is called the s-th preprocessed normal small intestine sample, the t-th column of the s-th preprocessed normal small intestine sample is called its t-th spectral channel, the s-th preprocessed normal small intestine sample corresponds to the s-th normal small intestine sample, and the t-th spectral channel of the s-th preprocessed normal small intestine sample corresponds to the t-th spectral channel of the s-th normal small intestine sample;

[0041] Window smoothing, baseline correction, and standard normal transformation were used to analyze the hyperspectral data Y of necrotic small intestine. orgPreprocessing is performed to obtain preprocessed necrotic small intestine hyperspectral data, which is recorded as Y, where Y includes m preprocessed necrotic small intestine samples, each row of data in Y represents a preprocessed necrotic small intestine sample, each column in each preprocessed necrotic small intestine sample represents a spectral channel of the preprocessed necrotic small intestine sample, and each preprocessed necrotic small intestine has p spectral channels; the hth row of data in Y is called the hth preprocessed necrotic small intestine sample, the tth column of the hth preprocessed necrotic small intestine sample is called its tth spectral channel, the hth preprocessed necrotic small intestine sample corresponds to the hth necrotic small intestine sample, and the tth spectral channel of the hth preprocessed necrotic small intestine sample corresponds to the tth spectral channel of the hth necrotic small intestine sample;

[0042] Step 3, transpose the preprocessed normal small intestine hyperspectral data X to obtain transposed normal small intestine hyperspectral data, which is recorded as X′, X′ includes n transposed normal small intestine samples, each column of data in X′ represents a transposed normal small intestine sample, each row in each transposed normal small intestine sample represents a spectral channel of the transposed normal small intestine sample, and each transposed normal small intestine has p spectral channels; the s-th column of data in X′ is called the s-th transposed normal small intestine sample, the t-th row of the s-th transposed normal small intestine sample is called its t-th spectral channel, the s-th transposed normal small intestine sample corresponds to the s-th normal small intestine sample, and the t-th spectral channel of the s-th transposed normal small intestine sample corresponds to the t-th spectral channel of the s-th normal small intestine sample;

[0043] The preprocessed necrotic small intestine hyperspectral data Y are transposed to obtain the transposed necrotic small intestine hyperspectral data, which is denoted as Y′. Y′ includes m transposed necrotic small intestine samples. Each column of data in Y′ represents a transposed necrotic small intestine sample. Each row in each transposed necrotic small intestine sample represents a spectral channel of the transposed necrotic small intestine sample. Each transposed necrotic small intestine has p spectral channels. The hth column of data in Y′ is called the hth transposed necrotic small intestine sample, denoted as y′. h , the t-th row of the h-th transposed necrotic small intestine sample is called its t-th spectral channel, the h-th transposed necrotic small intestine sample corresponds to the h-th necrotic small intestine sample, and the t-th spectral channel of the h-th transposed necrotic small intestine sample corresponds to the t-th spectral channel of the h-th necrotic small intestine sample;

[0044] Step 4: Use the Duplex algorithm to sample 20 transposed normal small intestine samples from the transposed normal small intestine hyperspectral data X′ to form a transposed normal small intestine sample matrix, denoted as D′. The transposed normal small intestine sample matrix D′ is a data matrix with p rows and 20 columns. Each column of data represents a transposed normal small intestine sample sampled by the Duplex algorithm. The data in the uth column of D′ is called the uth sampled transposed normal small intestine sample, and u=1, 2, …, 20.

[0045] Step 5, set the variable i and initialize i to 1;

[0046] Step 6: Determine the i-th abnormal spectral channel. The specific process is as follows:

[0047] S6.1. Take D′ as the explanatory variable of the LTS algorithm, y′ i As the response variable of the LTS algorithm, the LTS algorithm is first used to label the explanatory variable D′ and the response variable y′ i Common abnormal spectral channels, the number of channels marked as abnormal spectral channels is denoted as q i , remove its explanatory variable D′ and response variable y′ i The spectral channels marked as abnormal spectral channels in ;

[0048] The data matrix obtained by removing the abnormal spectral channel from D′ is called the normal explanatory variable y′ i The data matrix obtained after removing the abnormal spectral channels is called the normal response variable Normal explanatory variables pq i A data vector with 20 rows and columns, a normal response variable pq i A data vector with 1 row and 1 column;

[0049] S6.2. Least squares method based on normal explanatory variables With normal response variable Build regression models;

[0050] S6.3. Obtain normal explanatory variables through this regression model The weight w i , weight w i It is a data vector with 20 rows and 1 column. The data in the u-th row is the weight value of the u-th sample transposed normal small intestine sample;

[0051] S6.4. Calculation of D′ and w i The product of is the predicted response variable, denoted as z′ i , that is, the i-th transposed necrotic small intestine sample y′ i The reconstruction data;

[0052] S6.5. Calculation of z′ i and y′ i The standardized residuals on p spectral channels are obtained by obtaining a standardized residual data r′ consisting of p standardized residuals. i , standardized residual data r′ iis a data vector with p rows and 1 column. Each row of data is a standardized residual. The jth row of data is the same as the i-th transposed necrotic small intestine sample y′. i Corresponding to the jth spectral channel, the jth row of data is the i-th transposed necrotic small intestine sample y′ i Normalized residual on the jth spectral channel, j = 1, 2, …, p;

[0053] S6.6. Calculation of r′ i The median and median absolute deviation (MAD) of the i-th transposed necrotic small intestine sample y′ are selected as the sum of the median and three times the median absolute deviation. i Threshold for classifying abnormal spectral channels;

[0054] S6.7. Construct a column vector L′ i , determine the i-th transposed necrotic small intestine sample y′ i Is the standardized residual on the jth spectral channel less than the i-th transposed necrotic small intestine sample y′ i The threshold for dividing the abnormal spectral channel is, if it is less than, then the column vector L′ i The jth row in is assigned a value of 0, and if it is not less than , it is assigned a value of 1; L′ i is a data vector with p rows and 1 column. The data in the jth row represents the i-th transposed necrotic small intestine sample y′ i Whether the jth spectral channel of is an abnormal spectral channel, if it is equal to 0, it is a normal spectral channel, if it is equal to 1, it is an abnormal spectral channel;

[0055] Step 7: Determine whether the current value of i is equal to m. If not, update the value of i by adding 1 to the current value of i, and then return to step 6. If equal, proceed to step 8.

[0056] Step 8: Use L′1 to L′ m Construct the matrix L′, L′ h is the h-th column data of L′;

[0057] Step 9: Count the frequency of the number “1” in the j-th row of data in L′, that is, the number of necrotic small intestine samples marked as abnormal in the j-th spectral channel of the necrotic small intestine hyperspectral data, and record it as m j , compare the number m of abnormal spectral channels in the jth spectral channel of m necrotic small intestine samples j The size of the number of necrotic small intestine samples m in the necrotic small intestine hyperspectral data, if m j If m is equal to m, then the jth spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is the characteristic band of the necrotic small intestine. jIf m is less than m and greater than 0, the jth spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is useless information caused by objective factors such as noise. j If it is equal to 0, the j-th spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is neither the characteristic band of the necrotic small intestine nor the useless information caused by objective factors such as noise, thus achieving high-precision recognition of the characteristic bands of the necrotic small intestine.

[0058] Three rabbits were prepared, respectively called rabbit 1, rabbit 2 and rabbit 3. Professional doctors surgically blocked the blood supply to the small intestine of rabbit 1, rabbit 2 and rabbit 3, causing the blocked small intestine of rabbit 1, rabbit 2 and rabbit 3 to become necrotic, while the remaining small intestine remained normal. The blocked blood supply and necrotic small intestine of rabbit 1, rabbit 2 and rabbit 3 were necrotic small intestine, while the remaining small intestine that remained normal were normal small intestine. A hyperspectral data acquisition device with a channel number p of 116 was used to acquire the hyperspectral data of the normal small intestine of the remaining small intestine of rabbit 1 10 minutes after the partial small intestine of rabbit 1 became necrotic, and the hyperspectral data of the necrotic small intestine of rabbit 1 10 minutes after the partial small intestine of rabbit 1 became necrotic, respectively, and recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 30 minutes after the small intestine of one part of the rabbit was necrotic and the hyperspectral data of the necrotic small intestine of one part of the rabbit 30 minutes after the small intestine of one part of the rabbit was necrotic, and they were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit after 60 minutes of partial necrosis and the hyperspectral data of the necrotic small intestine of the rabbit after 60 minutes of partial necrosis, which were recorded as and The hyperspectral data acquisition device was used to obtain the normal small intestine hyperspectral data of the remaining small intestine of the rabbit 90 minutes after the small intestine of one part was necrotic and the hyperspectral data of the necrotic small intestine of one part of the rabbit 90 minutes after the small intestine of one part was necrotic, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 10 minutes after the necrosis of the two parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit 10 minutes after the necrosis of the two parts of the small intestine, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 30 minutes after the necrosis of the two parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit 30 minutes after the necrosis of the two parts of the small intestine, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit after 60 minutes of necrosis of the two parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit after 60 minutes of necrosis of the two parts of the small intestine, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 90 minutes after the necrosis of the two parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit 90 minutes after the necrosis of the two parts of the small intestine, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 10 minutes after the necrosis of the three parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit 10 minutes after the necrosis of the three parts of the small intestine, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 30 minutes after the small intestine of the rabbit 3 parts was necrotic and the hyperspectral data of the necrotic small intestine of the rabbit 30 minutes after the small intestine of the rabbit 3 parts was necrotic, and they were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit after 60 minutes of necrosis of the three parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit after 60 minutes of necrosis of the three parts of the small intestine, which were recorded as and The hyperspectral data acquisition device was used to obtain the hyperspectral data of the normal small intestine of the remaining small intestine of the rabbit 90 minutes after the necrosis of the three parts of the small intestine and the hyperspectral data of the necrotic small intestine of the rabbit 90 minutes after the necrosis of the three parts of the small intestine, which were recorded as and At this time, a total of and and and and and and and and and and and and and and and and There are 12 groups of hyperspectral data in total, and each group of hyperspectral data includes normal small intestine hyperspectral data and necrotic small intestine hyperspectral data. Necrotic small intestine hyperspectral data including 100 necrotic small intestine samples are randomly selected from the necrotic small intestine hyperspectral data of each group of hyperspectral data, that is, m is 100, and normal small intestine hyperspectral data including n normal small intestine samples are randomly selected from the normal small intestine hyperspectral data. The necrotic small intestine characteristic band recognition method based on the LTS algorithm of this embodiment is used to recognize the characteristic bands of the necrotic small intestine of rabbits.

[0059] When the characteristic band identification method of necrotic small intestine based on LTS algorithm of the present invention is used to identify the characteristic band of necrotic small intestine of rabbit, the characteristic band identification result of necrotic small intestine of rabbit whose normal small intestine hyperspectral data and necrotic small intestine hyperspectral data are from the same rabbit is shown in the figure below: Figure 2 As shown; when the characteristic band recognition method of necrotic small intestine based on LTS algorithm of the present invention is used to recognize the characteristic band of necrotic small intestine of rabbits, the characteristic band recognition results of necrotic small intestine of rabbits whose normal small intestine hyperspectral data and necrotic small intestine hyperspectral data are from different sources are shown in FIG. Figure 3 As shown; when the characteristic band recognition method of necrotic small intestine based on LTS algorithm of the present invention is used to recognize the characteristic band of necrotic small intestine of rabbit, the characteristic band recognition results of necrotic small intestine of rabbit 1 at 10, 30, 60 and 90 minutes of partial small intestine necrosis are shown in FIG. Figure 4 As shown; when the characteristic band recognition method of necrotic small intestine based on LTS algorithm of the present invention is used to recognize the characteristic band of necrotic small intestine of rabbit, the characteristic band recognition results of necrotic small intestine of rabbit 2 at 10, 30, 60 and 90 minutes of partial small intestine necrosis are shown in FIG. Figure 5 As shown; when the characteristic band recognition method of necrotic small intestine based on LTS algorithm of the present invention is used to recognize the characteristic band of necrotic small intestine of rabbit, the characteristic band recognition results of necrotic small intestine of rabbit 3 at 10, 30, 60 and 90 minutes of partial small intestine necrosis are shown in FIG. Figure 6 shown.

[0060] Figure 2 The normal small intestine hyperspectral data and necrotic small intestine hyperspectral data of rabbit 1 were selected as the research objects. The normal small intestine hyperspectral data in step 1 were obtained from the normal hyperspectral data of the remaining small intestine of rabbit 1 after part of the small intestine was necrotic for 30 minutes. 10917 normal small intestine samples were randomly selected from the experiment. The hyperspectral data of necrotic small intestine in step 1 were obtained from the hyperspectral data of necrotic small intestine of rabbit 1 after 30 minutes of small intestine necrosis. 100 necrotic small intestine samples were randomly selected from the data. The horizontal axis "Wavelength / nm" is the spectral channel of the necrotic small intestine samples, and the vertical axis "The number of flagged" is the number of necrotic small intestine samples that were judged to be abnormal in each spectral channel. Figure 2It can be observed that the number of necrotic small intestine samples marked as abnormal near 763nm is equal to 100. Therefore, it can be judged that the area near 763nm is the characteristic band of necrotic small intestine, that is, the normal small intestine hyperspectral data and the necrotic small intestine hyperspectral data come from the same rabbit and the characteristic band of necrotic small intestine after the same necrosis time is near 763nm. According to reference 3 "SJ Matcher and C.E. Cooper," Absolute quantification of deoxyhaemoglobin concentration in tissue near infrared spectroscopy," Phys. Med. Biol. 39, 1295-1312 (1994) " and reference 4 "JMMurkin and M. Arango," Near-infrared spectroscopy as an index of brain and tissue oxygenation," British Journal of Anaesthesia 103, i3-i13 (2009) ", it can be learned that the maximum absorption peak of deoxyhaemoglobin is near 763nm, which means that the necrosis of the small intestine of rabbits may be caused by hypoxia. This result is consistent with reference 5 "TP Fox and C. Godavitarne," What Really Causes Necrotising Enterocolitis? "ISRN Gastroenterology 2012,1–9(2012)." and the research results of document 6 "B.Gane,BVBhat,B.Adhisivam,R.Joy,P.Prasadkumar,P.Femitha,and B.Shruti,"Risk Factorsand Outcome in Neonatal Necrotising Enterocolitis,"Indian J Pediatr 81,425–428(2014)." are consistent, so it can be determined that the recognition of the characteristic bands of necrotic small intestine after the same necrosis time by the present invention for the normal small intestine hyperspectral data and the necrotic small intestine hyperspectral data from the same rabbit is correct.

[0061] Figure 3 The normal small intestine hyperspectral data of rabbit 1 and the necrotic small intestine hyperspectral data of rabbit 2 were selected as the research objects. The normal small intestine hyperspectral data in step 1 were obtained from the normal hyperspectral data of the remaining small intestine of rabbit 1 30 minutes after partial small intestine necrosis. 10917 normal small intestine samples were randomly selected from the experiment. The hyperspectral data of necrotic small intestine in step 1 were obtained from the hyperspectral data of necrotic small intestine of rabbit 2 after 30 minutes of small intestine necrosis. 100 necrotic small intestine samples were randomly selected from the data. The horizontal axis "Wavelength / nm" is the spectral channel of the necrotic small intestine samples, and the vertical axis "The number of flagged" is the number of necrotic small intestine samples that were judged to be abnormal in each spectral channel. Figure 3It can be observed that the number of necrotic small intestine samples marked as abnormal near 763nm is equal to 100. Therefore, it can be judged that the area near 763nm is the characteristic band of necrotic small intestine, that is, the characteristic band of necrotic small intestine after the same necrosis time of rabbits with different sources of normal small intestine hyperspectral data and necrotic small intestine hyperspectral data is near 763nm. According to reference 3 "SJ Matcher and C.E. Cooper," Absolute quantification of deoxyhaemoglobin concentration in tissue near infrared spectroscopy," Phys. Med. Biol. 39, 1295-1312 (1994) " and reference 4 "JMMurkin and M. Arango," Near-infrared spectroscopy as an index of brain and tissue oxygenation," British Journal of Anaesthesia 103, i3-i13 (2009) ", it can be learned that the maximum absorption peak of deoxyhaemoglobin is near 763nm, which means that the necrosis of the small intestine of rabbits may be caused by hypoxia. This result is consistent with reference 5 "TP Fox and C. Godavitarne," What Really Causes Necrotising Enterocolitis? "ISRN Gastroenterology 2012,1–9(2012)." and the research results of document 6 "B.Gane,BVBhat,B.Adhisivam,R.Joy,P.Prasadkumar,P.Femitha,and B.Shruti,"Risk Factorsand Outcome in Neonatal Necrotising Enterocolitis,"Indian J Pediatr 81,425–428(2014)." are consistent, so it can be determined that the present invention is correct in identifying the characteristic bands of necrotic small intestine of rabbits with different sources of normal small intestine hyperspectral data and necrotic small intestine hyperspectral data after the same necrosis time.

[0062] Figure 4Rabbit 1 was selected as the research subject, and a stacked bar chart of the number of necrotic small intestine samples obtained when the characteristic bands of necrotic small intestine of rabbit 1 at 10, 30, 60, and 90 minutes of partial small intestinal necrosis were drawn, wherein when the characteristic bands of necrotic small intestine of rabbit 1 at 10 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 10912, when the characteristic bands of necrotic small intestine of rabbit 1 at 30 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 10917, when the characteristic bands of necrotic small intestine of rabbit 1 at 60 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 10571, and when the characteristic bands of necrotic small intestine of rabbit 1 at 90 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 10456. Figure 4 In the figure, the horizontal axis "Wavelength / nm" is the spectral channel of the necrotic small intestine sample, and the vertical axis "The number of flagged" is the number of necrotic small intestine samples that are judged to be abnormal in each spectral channel. Figure 4It can be observed that the number of necrotic small intestine samples marked as abnormal near 763nm after partial small intestine necrosis 10, 30, 60, and 90 minutes is all equal to 100. Therefore, it can be judged that the vicinity of 763nm is the characteristic band of necrotic small intestine, that is, the characteristic band of the necrotic small intestine of Rabbit 1 after partial small intestine necrosis 10, 30, 60, and 90 minutes is near 763nm. According to reference 3 "SJ Matcher and CE Cooper," Absolute quantification of deoxyhaemoglobin concentration in tissue near infrared spectroscopy," Phys. Med. Biol. 39, 1295-1312 (1994) " and reference 4 "JMMurkin and M. Arango," Near-infrared spectroscopy as an index of brain and tissue oxygenation," British Journal of Anaesthesia 103, i3-i13 (2009) ", it can be learned that the maximum absorption peak of deoxyhaemoglobin is near 763nm, which means that the necrosis of the small intestine of rabbits may be caused by hypoxia. This result is consistent with reference 5 "TP Fox and C. Godavitarne," What Really Causes Necrotising Enterocolitis? "ISRN Gastroenterology 2012,1-9(2012)." and the research results of document 6 "B.Gane,BVBhat,B.Adhisivam,R.Joy,P.Prasadkumar,P.Femitha,and B.Shruti,"RiskFactors and Outcome in Neonatal Necrotising Enterocolitis,"Indian J Pediatr81,425-428(2014)." are consistent, so it can be determined that the present invention is correct in identifying the characteristic bands of necrotic small intestine of rabbits 10, 30, 60, and 90 minutes after partial small intestinal necrosis.

[0063] Figure 5Rabbit 2 was selected as the research subject, and a stacked bar chart of the number of necrotic small intestine samples obtained when the characteristic bands of necrotic small intestine of rabbit 2 at 10, 30, 60, and 90 minutes of partial small intestinal necrosis were drawn. When rabbit 2 identified the characteristic band of necrotic small intestine at 10 minutes of partial small intestinal necrosis, the value of m was 100 and the value of n was 10239. When rabbit 2 identified the characteristic band of necrotic small intestine at 30 minutes of partial small intestinal necrosis, the value of m was 100 and the value of n was 10854. When rabbit 2 identified the characteristic band of necrotic small intestine at 60 minutes of partial small intestinal necrosis, the value of m was 100 and the value of n was 10668. When rabbit 2 identified the characteristic band of necrotic small intestine at 90 minutes of partial small intestinal necrosis, the value of m was 100 and the value of n was 10520. Figure 5 In the figure, the horizontal axis "Wavelength / nm" is the spectral channel of the necrotic small intestine sample, and the vertical axis "The number of flagged" is the number of necrotic small intestine samples that are judged to be abnormal in each spectral channel. Figure 5It can be observed that the number of necrotic small intestine samples marked as abnormal near 763nm after partial small intestine necrosis 10, 30, 60, and 90 minutes is all equal to 100. Therefore, it can be judged that the vicinity of 763nm is the characteristic band of necrotic small intestine, that is, the characteristic band of the necrotic small intestine of Rabbit 2 after partial small intestine necrosis 10, 30, 60, and 90 minutes is near 763nm. According to reference 3 "SJ Matcher and CE Cooper," Absolute quantification of deoxyhaemoglobin concentration in tissue near infrared spectroscopy," Phys. Med. Biol. 39, 1295-1312 (1994) " and reference 4 "JMMurkin and M. Arango," Near-infrared spectroscopy as an index of brain and tissue oxygenation," British Journal of Anaesthesia 103, i3-i13 (2009) ", it can be learned that the maximum absorption peak of deoxyhaemoglobin is near 763nm, which means that the necrosis of the small intestine of rabbits may be caused by hypoxia. This result is consistent with reference 5 "TP Fox and C. Godavitarne," What Really Causes Necrotising Enterocolitis? "ISRN Gastroenterology 2012,1-9(2012)." and the research results of document 6 "B.Gane,BVBhat,B.Adhisivam,R.Joy,P.Prasadkumar,P.Femitha,and B.Shruti,"RiskFactors and Outcome in Neonatal Necrotising Enterocolitis,"Indian J Pediatr81,425-428(2014)." are consistent, so it can be determined that the present invention is correct in identifying the characteristic bands of necrotic small intestine after 10, 30, 60, and 90 minutes of necrosis of two parts of small intestine in rabbits.

[0064] Figure 6Rabbit 3 was selected as the research subject, and a stacked bar chart of the number of necrotic small intestine samples obtained when the characteristic bands of necrotic small intestine of rabbit 3 at 10, 30, 60, and 90 minutes of partial small intestinal necrosis were drawn. When the characteristic bands of necrotic small intestine of rabbit 3 at 10 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 3428. When the characteristic bands of necrotic small intestine of rabbit 3 at 30 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 3568. When the characteristic bands of necrotic small intestine of rabbit 3 at 60 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 4389. When the characteristic bands of necrotic small intestine of rabbit 3 at 90 minutes of partial small intestinal necrosis were identified, the value of m was 100 and the value of n was 4600. Figure 6 In the figure, the horizontal axis "Wavelength / nm" is the spectral channel of the necrotic small intestine sample, and the vertical axis "The number of flagged" is the number of necrotic small intestine samples that are judged to be abnormal in each spectral channel. Figure 6It can be observed that the number of necrotic small intestine samples marked as abnormal near 763nm after partial small intestine necrosis 10, 30, 60, and 90 minutes is all equal to 100. Therefore, it can be judged that the vicinity of 763nm is the characteristic band of necrotic small intestine, that is, the characteristic band of the necrotic small intestine of rabbit 3 after partial small intestine necrosis 10, 30, 60, and 90 minutes is near 763nm. According to reference 3 "SJ Matcher and CE Cooper," Absolute quantification of deoxyhaemoglobin concentration in tissue near infrared spectroscopy," Phys. Med. Biol. 39, 1295-1312 (1994) " and reference 4 "JMMurkin and M. Arango," Near-infrared spectroscopy as an index of brain and tissue oxygenation," British Journal of Anaesthesia 103, i3-i13 (2009) ", it can be learned that the maximum absorption peak of deoxyhaemoglobin is near 763nm, which means that the necrosis of the small intestine of rabbits may be caused by hypoxia. This result is consistent with reference 5 "TP Fox and C. Godavitarne," What Really Causes Necrotising Enterocolitis? "ISRN Gastroenterology 2012,1-9(2012)." and the research results of document 6 "B.Gane,BVBhat,B.Adhisivam,R.Joy,P.Prasadkumar,P.Femitha,and B.Shruti,"RiskFactors and Outcome in Neonatal Necrotising Enterocolitis,"Indian J Pediatr81,425-428(2014)." are consistent, so it can be determined that the present invention is correct in identifying the characteristic bands of necrotic small intestine after 10, 30, 60, and 90 minutes of necrosis of the three parts of the small intestine of rabbits.

[0065] In summary, the method for identifying characteristic bands of necrotic small intestine based on the minimum clipping squares method of the present invention can realize automatic identification of characteristic bands of necrotic small intestine with high accuracy and low labor cost.

Claims

1. A method for identifying characteristic bands of necrotic small intestine based on LTS algorithm. Features The normal small intestine hyperspectral data and the necrotic small intestine hyperspectral data were converted as the explanatory variables and response variables of the LTS algorithm. The LTS algorithm was used to determine the abnormal spectral channels shared by the explanatory variables and the response variables, and then the normal explanatory variables and normal response variables were obtained after removing the abnormal spectral channels. Then, the least squares method was used to build a regression model based on the normal explanatory variables and the normal response variables. The weights were obtained through the regression model, and the response variables were reconstructed based on the weights. Finally, the characteristic bands of the necrotic small intestine were determined based on the standardized residuals of the response variables and the reconstructed response variables on each spectral channel.

2. According to claim 1, a method for identifying characteristic bands of necrotic small intestine based on LTS algorithm, Features The following steps are involved: Step 1. Acquire normal small intestine hyperspectral data and necrotic small intestine hyperspectral data with the same number of spectral channels through a hyperspectral data acquisition device. The normal small intestine hyperspectral data is a data matrix with n rows and p columns, and the necrotic small intestine hyperspectral data is a data matrix with m rows and p columns, where p represents the number of channels of the hyperspectral data acquisition device, n represents the number of normal small intestine samples in the normal small intestine hyperspectral data, and m represents the number of necrotic small intestine samples in the necrotic small intestine hyperspectral data. n is an integer greater than or equal to 20, and m is an integer greater than 0. In the normal small intestine hyperspectral data, each row of data represents a normal small intestine sample, and each column of each normal small intestine sample represents a spectral channel of the normal small intestine sample, that is, each normal small intestine sample has p spectral channels. ; The s-th row of data in the normal small intestine hyperspectral data is called the s-th normal small intestine sample, s = 1, 2, ..., n, the t-th column of the s-th normal small intestine sample is called the t-th spectral channel of the s-th normal small intestine sample, t = 1, 2, ..., p; In the necrotic small intestine hyperspectral data, each row of data represents a necrotic small intestine sample, and each column in each necrotic small intestine sample represents a spectral channel of the necrotic small intestine sample, that is, each necrotic small intestine sample has p spectral channels; The h-th row of data in the necrotic small intestine hyperspectral data is called the h-th necrotic small intestine sample, and the t-th column of the h-th necrotic small intestine sample is called the t-th spectral channel of the h-th necrotic small intestine sample, h = 1, 2, ..., m; The normal small intestine hyperspectral data is denoted as X org , record the hyperspectral data of necrotic small intestine as Y org ; Step 2: Use window smoothing, baseline correction, and standard normal transformation to transform the normal small intestine hyperspectral data Xo rg Preprocessing is performed to obtain preprocessed normal small intestine hyperspectral data, which is recorded as X, where X includes n preprocessed normal small intestine samples, each row of data in X represents a preprocessed normal small intestine sample, each column in each preprocessed normal small intestine sample represents a spectral channel of the preprocessed normal small intestine sample, and each preprocessed normal small intestine has p spectral channels; the s-th row of data in X is called the s-th preprocessed normal small intestine sample, the t-th column of the s-th preprocessed normal small intestine sample is called its t-th spectral channel, the s-th preprocessed normal small intestine sample corresponds to the s-th normal small intestine sample, and the t-th spectral channel of the s-th preprocessed normal small intestine sample corresponds to the t-th spectral channel of the s-th normal small intestine sample; Window smoothing, baseline correction, and standard normal transformation were used to analyze the hyperspectral data Y of necrotic small intestine. org Preprocessing is performed to obtain preprocessed necrotic small intestine hyperspectral data, which is recorded as Y, where Y includes m preprocessed necrotic small intestine samples, each row of data in Y represents a preprocessed necrotic small intestine sample, each column in each preprocessed necrotic small intestine sample represents a spectral channel of the preprocessed necrotic small intestine sample, and each preprocessed necrotic small intestine has p spectral channels; the hth row of data in Y is called the hth preprocessed necrotic small intestine sample, the tth column of the hth preprocessed necrotic small intestine sample is called its tth spectral channel, the hth preprocessed necrotic small intestine sample corresponds to the hth necrotic small intestine sample, and the tth spectral channel of the hth preprocessed necrotic small intestine sample corresponds to the tth spectral channel of the hth necrotic small intestine sample; Step 3, transpose the preprocessed normal small intestine hyperspectral data X to obtain transposed normal small intestine hyperspectral data, which is recorded as X′, X′ includes n transposed normal small intestine samples, each column of data in X′ represents a transposed normal small intestine sample, each row in each transposed normal small intestine sample represents a spectral channel of the transposed normal small intestine sample, and each transposed normal small intestine has p spectral channels; the s-th column of data in X′ is called the s-th transposed normal small intestine sample, the t-th row of the s-th transposed normal small intestine sample is called its t-th spectral channel, the s-th transposed normal small intestine sample corresponds to the s-th normal small intestine sample, and the t-th spectral channel of the s-th transposed normal small intestine sample corresponds to the t-th spectral channel of the s-th normal small intestine sample; The preprocessed necrotic small intestine hyperspectral data Y are transposed to obtain the transposed necrotic small intestine hyperspectral data, which is denoted as Y′. Y′ includes m transposed necrotic small intestine samples. Each column of data in Y′ represents a transposed necrotic small intestine sample. Each row in each transposed necrotic small intestine sample represents a spectral channel of the transposed necrotic small intestine sample. Each transposed necrotic small intestine has p spectral channels. The hth column of data in Y′ is called the hth transposed necrotic small intestine sample, denoted as y′. h , the t-th row of the h-th transposed necrotic small intestine sample is called its t-th spectral channel, the h-th transposed necrotic small intestine sample corresponds to the h-th necrotic small intestine sample, and the t-th spectral channel of the h-th transposed necrotic small intestine sample corresponds to the t-th spectral channel of the h-th necrotic small intestine sample; Step 4: Use the Duplex algorithm to sample 20 transposed normal small intestine samples from the transposed normal small intestine hyperspectral data X′ to form a transposed normal small intestine sample matrix, denoted as D′. The transposed normal small intestine sample matrix D′ is a data matrix with p rows and 20 columns. Each column of data represents a transposed normal small intestine sample sampled by the Duplex algorithm. The data in the uth column of D′ is called the uth sampled transposed normal small intestine sample, where u=1, 2, ..., 20. Step 5, set the variable i and initialize i to 1; Step 6: Determine the i-th abnormal spectral channel. The specific process is as follows: S6.

1. Take D′ as the explanatory variable of the LTS algorithm, y′ i As the response variable of the LTS algorithm, the LTS algorithm is first used to label the explanatory variable D′ and the response variable y′ i Common abnormal spectral channels, the number of channels marked as abnormal spectral channels is denoted as q i , remove its explanatory variable D′ and response variable y′ i The spectral channels marked as abnormal spectral channels in ; The data matrix obtained by removing the abnormal spectral channel from D′ is called the normal explanatory variable y′ i The data matrix obtained after removing the abnormal spectral channels is called the normal response variable Normal explanatory variables pq i A data vector with 20 rows and columns, a normal response variable pq i A data vector with 1 row and 1 column; S6.

2. Least squares method based on normal explanatory variables With normal response variable Build regression models; S6.

3. Obtain normal explanatory variables through this regression model The weight w i , weight w i It is a data vector with 20 rows and 1 column. The data in the u-th row is the weight value of the u-th sample transposed normal small intestine sample; S6.

4. Calculation of D′ and w i The product of is the predicted response variable, denoted as z′ i , that is, the i-th transposed necrotic small intestine sample y′ i The reconstruction data; S6.

5. Calculation of z′ i and y′ i The standardized residuals on p spectral channels are obtained by obtaining a standardized residual data r′ consisting of p standardized residuals. i , standardized residual data r′ i is a data vector with p rows and 1 column. Each row of data is a standardized residual. The jth row of data is the same as the i-th transposed necrotic small intestine sample y′. i Corresponding to the jth spectral channel, the jth row of data is the i-th transposed necrotic small intestine sample y′ i Normalized residual on the jth spectral channel, j = 1, 2, ..., p; S6.

6. Calculation of r′ i The median and median absolute deviation (MAD) of the i-th transposed necrotic small intestine sample y′ are selected as the sum of the median and three times the median absolute deviation. i Threshold for classifying abnormal spectral channels; S6.

7. Construct a column vector L′ i , determine the i-th transposed necrotic small intestine sample y′ i Is the standardized residual on the jth spectral channel less than the i-th transposed necrotic small intestine sample y′ i The threshold for dividing the abnormal spectral channel is, if it is less than, then the column vector L′ i The jth row in is assigned a value of 0, and if it is not less than , it is assigned a value of 1; L′ i is a data vector with p rows and 1 column. The data in the jth row represents the i-th transposed necrotic small intestine sample y′ i Whether the jth spectral channel of is an abnormal spectral channel, if it is equal to 0, it is a normal spectral channel, if it is equal to 1, it is an abnormal spectral channel; Step 7: Determine whether the current value of i is equal to m. If not, update the value of i by adding 1 to the current value of i, and then return to step 6. If equal, proceed to step 8. Step 8: Use L′ 1 To L' m Construct the matrix L′, L′ h is the h-th column data of L′; Step 9: Count the frequency of the number "1" in the j-th row of data in L', that is, the number of necrotic small intestine samples marked as abnormal in the j-th spectral channel of the necrotic small intestine hyperspectral data, and record it as m j , compare the number of necrotic small intestine samples m marked as abnormal in the jth spectral channel of the necrotic small intestine hyperspectral data j The size of the number of necrotic small intestine samples m in the necrotic small intestine hyperspectral data, if m j If m is equal to m, then the jth spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is the characteristic band of necrotic small intestine. j If m is less than m and greater than 0, the jth spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is useless information caused by objective factors such as noise. j If it is equal to 0, the j-th spectral channel of each necrotic small intestine sample in the necrotic small intestine hyperspectral data is neither the characteristic band of the necrotic small intestine nor the useless information caused by objective factors such as noise, thus achieving high-precision recognition of the characteristic bands of the necrotic small intestine.

Images