CN111812058A - Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology - Google Patents

Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology Download PDF

Info

Publication number
CN111812058A
CN111812058A CN202010480423.3A CN202010480423A CN111812058A CN 111812058 A CN111812058 A CN 111812058A CN 202010480423 A CN202010480423 A CN 202010480423A CN 111812058 A CN111812058 A CN 111812058A
Authority
CN
China
Prior art keywords
terahertz
image
leaf
cedrela sinensis
pesticide residues
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
CN202010480423.3A
Other languages
Chinese (zh)
Inventor
聂鹏程
瞿芳芳
陈卓怡
张慧
蔺磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Qiushi Agricultural Technology Development Co ltd
Shandong Industrial Technology Research Institute of ZJU
Original Assignee
Shandong Qiushi Agricultural Technology Development Co ltd
Shandong Industrial Technology Research Institute of ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Qiushi Agricultural Technology Development Co ltd, Shandong Industrial Technology Research Institute of ZJU filed Critical Shandong Qiushi Agricultural Technology Development Co ltd
Priority to CN202010480423.3A priority Critical patent/CN111812058A/en
Publication of CN111812058A publication Critical patent/CN111812058A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3563—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing solids; Preparation of samples therefor
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3581—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3581—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation
    • G01N21/3586—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation by Terahertz time domain spectroscopy [THz-TDS]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3563—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing solids; Preparation of samples therefor
    • G01N2021/3572—Preparation of samples, e.g. salt matrices

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Pathology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Immunology (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Toxicology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Multimedia (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The invention discloses a qualitative detection method of pesticide residues in toona sinensis based on a terahertz imaging technology, which comprises the following steps: (1) performing terahertz imaging on 8 types of cedrela sinensis leaf samples containing BZM pesticides; (2) dividing the obtained 9 types of terahertz spectrum data into a training data set and a test data set; constructing by using a training data set, and training the deep convolution neural network to generate a training model of the cedrela sinensis leaf image; inputting a test sample in the test data set into a training model of the image of the cedrela sinensis leaves, and judging the type of pesticide residues on the cedrela sinensis leaves to generate a test result; iteratively updating the training model of the cedrela sinensis leaf image according to the test result to obtain a recognition model of the cedrela sinensis leaf image; (4) and carrying out image visualization on the pesticide residues on the toona sinensis leaves to be detected by using the established recognition model, and displaying the types and the distribution conditions of the pesticide residues on the leaves.

Description

Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology
Technical Field
The invention relates to the field of plant detection, in particular to a qualitative detection method for pesticide residues in toona sinensis based on a terahertz imaging technology.
Background
The following background is provided to aid the reader in understanding the present invention and is not admitted to be prior art.
Toona sinensis is a high nutritive value vegetable rich in protein, calcium and vitamin C. BZM pesticides such as Benomyl (BNL), Carbendazim (BCM) and Thiabendazole (TBZ) are widely used in agricultural insecticides and fungicides to control parasitic infections and crop diseases, and can be used to control powdery mildew, leaf rust and blight of plants, etc. In agricultural production activities, the phenomena of overuse, abuse and mixed use of various pesticides are frequently prohibited, so that the pesticide has potential harm to human health and potential harm to the environment and ecological systems. In particular, the detection of various pesticide residues poses practical challenges.
Conventional laboratory analysis methods such as HPLC, GC-MS and LC-MS are widely used for detecting pesticide residues in agricultural product matrixes. These techniques simultaneously provide high throughput analysis of different types of pesticide residues and the detection results are very accurate. However, these methods are complicated, time-consuming and labor-consuming in the sample pretreatment process; traditional spectrometry, including visible-near infrared, ultraviolet, fluorescence, hyperspectral imaging, etc., has also been used to detect pesticide residues in agricultural substrates. Although this type of method has the advantage of being fast and non-destructive to the sample, the accuracy and sensitivity for detection of multicomponent pesticide residues is relatively low.
The terahertz spectrum and the imaging technology can be used as a part of a new detection method, and have the advantages of broadband property, penetrability, fingerprint absorption and the like. Terahertz absorption peaks of various pesticides can be used as unique features to distinguish each analyte in a mixture, thereby enabling the identification of multiple pesticides. The low energy level of terahertz radiation does not pose any known danger to the organism, making terahertz imaging a powerful and safe technique for biological applications. In addition, the acquired terahertz spectrum or imaging data is very rich, including not only time domain, frequency domain and amplitude information, but also phase information that cannot be obtained using other optical techniques. The terahertz fingerprint spectrum of the pesticide molecule provides characteristic information for the classification and identification of the target pesticide. The detection of pesticide residues in agricultural products based on the terahertz technology needs to synthesize terahertz fingerprint and machine learning data analysis results. Compared with the traditional method, the deep learning algorithm is a powerful tool and can better explore complex data characteristics on a high-abstraction level.
However, the terahertz imaging technology is not used for detecting pesticide residues in the toona sinensis, particularly multi-component mixed pesticide residues at present.
Disclosure of Invention
The invention aims to provide a method for quickly detecting pesticide residues in toona sinensis.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a qualitative detection method for pesticide residues in toona sinensis based on a terahertz imaging technology comprises the following steps:
(1) carrying out terahertz imaging on 8 types of toona sinensis leaf samples; the 8 types of Chinese toon leaf samples are respectively as follows: cedrela sinensis leaves as blank controls, cedrela sinensis leaves containing BNL residues, cedrela sinensis leaves containing BCM residues, cedrela sinensis leaves containing TBZ residues, cedrela sinensis leaves containing BNL and BCM residues, cedrela sinensis leaves containing BNL and TBZ residues, cedrela sinensis leaves containing BCM and TBZ, cedrela sinensis leaves containing BNL, BCM and TBZ;
(2) extraction of spectral data of blade in terahertz imaging
Scanning a terahertz image of a leaf sample step by using a Region of Interest (ROI) as a shift operator, extracting a spectrum of each pixel in the ROI, taking an average value as a spectrum of the current ROI, and extracting a plurality of absorption spectra from the terahertz image of each leaf sample on the basis; in addition, a plurality of background spectra (BG) are extracted from the terahertz image of each type of cedrela sinensis leaf; thereby obtaining 9-type terahertz spectrum data;
(3) dividing the 9-class terahertz spectrum data obtained in the step (2) into a training data set and a test data set; constructing a DCNN deep convolution neural network by using the training data set, and training the DCNN deep convolution neural network to generate a training model of the cedrela sinensis leaf image; inputting the test sample in the test data set into a training model of the cedrela sinensis leaf image, and judging the type of pesticide residue on the cedrela sinensis leaf to generate a test result; iteratively updating the training model of the cedrela sinensis leaf image according to the test result to obtain a recognition model of the cedrela sinensis leaf image;
(4) and (4) carrying out image visualization on the pesticide residues on the toona sinensis leaves to be detected by using the identification model established in the step (3), so that the types and the distribution conditions of the pesticide residues on the leaves can be displayed.
Preferably, the specific operation of step (4) is: obtaining a terahertz image by the to-be-detected toona sinensis leaf according to the operation in the step (1), then extracting a terahertz absorption spectrum of each pixel point of the terahertz image, inputting the terahertz absorption spectrum into the identification model established in the step (3) to predict class labels, wherein the predicted class labels 1-9, namely BG, CK, BNL, BCM, TBZ, M1, M2, M3 and M4, are respectively represented by different colors, and the type and distribution condition of pesticide residues on the leaf can be visually displayed by restoring the predicted class labels at each pixel point into a two-dimensional image of complete pixels.
Preferably, when more than one type of category label is displayed in the blade region in the two-dimensional image obtained in step (4), the majority of label types are used as the final discrimination labels.
In the invention, each type of Chinese toon leaves can be provided with a plurality of same samples so as to enrich spectral data for modeling.
Preferably, in the step (1), 25 identical samples are set for each type of cedrela sinensis leaf, a THz-TDS system is used for imaging scanning of the cedrela sinensis leaf samples, a transmission imaging module is used for scanning point by point to obtain THz images of the leaf samples within the range of 0.1-3 THz, the scanning frequency is 15 Hz, and the image data size is 250 multiplied by 250 pixels;
in the step (2), the interested area of 3 x 3 pixels is adopted as a shift operator to scan on the blade step by step, the spectrum of each pixel in the interested area is extracted, the average value is taken as the spectrum of the current interested area, and on the basis, 100 absorption spectra are extracted from the terahertz image of each blade sample; 2500 pieces of background spectra (BG) were extracted from each category of terahertz image.
Preferably, in step (3), the DCNN deep convolutional neural network structure includes two convolutional modules and 5 fully-connected layers, each convolutional module includes a convolutional layer and a maximum pooling layer, the number of convolutional filters in the first and second convolutional layers is set to 32 and 64, respectively, the step size of a convolutional kernel is set to 1, the kernel size of the maximum pooling layer is set to 1 × 3, and the nonlinear property of the decision mapping function is normalized by using a corrected Linear Unit (ReLU); using a Batch Normalization (BN) function before each fully-connected layer and after each convolution module; processing the full connectivity data of the last layer by using a Softmax function to highlight the maximum value and limit the characteristic values of other nerve units below the maximum value; measuring the probability distribution distance between the DCNN output label and the real class label by using a classified cross entropy loss function; processing the loss function by adopting an adaptive moment estimation algorithm; the ratio of the training data set to the test data set is 70%: 30%, after all the connecting layers are subjected to batch normalization, adopting a jump-out method to further prevent overfitting; the training period, learning speed, beta1, and beta2 parameters are set to 500, 0.001, 0.9, and 0.99, respectively.
Preferably, the spectrum in the frequency range of 0.2-2.2 THz with higher signal-to-noise ratio is intercepted and used for modeling in the step (3) and visualization analysis in the step (4) so as to eliminate the interference of spectral noise.
The invention has the advantages that: the terahertz imaging and the deep learning are simultaneously used for rapidly detecting various pesticide residues on the fresh leaves for the first time, and the terahertz imaging and the deep learning have great significance for the expansion of the terahertz application field.
Drawings
FIG. 1 is terahertz imaging of Cedrela sinensis leaves under different parameters;
FIG. 2 is a process of detecting various BZM pesticide residues in cedrela sinensis leaves by terahertz imaging and deep learning;
FIG. 3 is DCNN model performance: (a) the accuracy of the DCNN model is lost and is accurate in the DCNN iteration process, wherein in the step (a), in two histograms of the same type, the left histogram corresponds to a training set, and the right histogram corresponds to a testing set.
Fig. 4 is a DCNN visualization of different pesticide residues leaving toona sinensis leaves: (a) CK leaves without pesticide residues, (b) leaves with BNL residues, (c) leaves with BCM residues, (d) leaves with TBZ residues, (e) leaves with BNL and BCM residues, (f) leaves with BNL and TBZ residues, (g) leaves with BCM and TBZ residues, (h) leaves with BNL, BCM and TBZ residues.
Detailed Description
The present invention will be further described with reference to the structures or terms used herein. The description is given for the sake of example only, to illustrate how the invention may be implemented, and does not constitute any limitation on the invention.
Pesticide standard BNL (C)14H18N4O3，CAS：17804-35-2），BCM（C9H9N3O2CAS: 10605-21-7) and TBZ (C)10H6N2S2CAS: 148-79-8) were purchased from Sigma-Aldrich (st. louis, MO, USA).
Example 1
A terahertz imaging-based qualitative detection method for multi-component mixed pesticide residues in toona sinensis comprises the following steps:
step 1: preparation of cedrela sinensis leaf sample
Seven kinds of pesticide solutions were prepared with 100 ml of acetonitrile solvent, including three kinds of single component solutions (10 mg/L BNL, 10 mg/LBCM, 10mg/L TBZ), three kinds of two component solutions (10 mg/L BNL +10mg/L BCM (M1), 10mg/L BNL +10mg/L TBZ (M2), 10mg/L BCM +10mg/L TBZ (M3)), and one kind of three component solution (10 mg/L BNL +10 mg/LBCM +10mg/L TBZ (M4)). The blank Control (Control Check, CK) was 100 ml of pure acetonitrile without any pesticide. After the prepared pesticide solutions are uniformly mixed, 7 ml of each solution is respectively dripped on the whole surface of the cedrela sinensis leaves by a pipette, and the solution is dried at room temperature to fully volatilize acetonitrile. A total of 26 leaves were prepared in duplicate for each solution, resulting in a total of 208 toona sinensis leaf samples of eight categories (BNL, BCM, TBZ, M1, M2, M3, M4, CK).
Step 2: acquiring imaging data of a cedrela sinensis leaf sample: THz-TDS system CCT-1800 is used for carrying out terahertz spectrum detection of BZM pesticide and imaging scanning of cedrela sinensis leaves, and the imaging measurement specification parameters of the system are as follows: focal length 25.4 mm, spot size 1.25 mm, scanning range 50X 50 mm, scanning precision <0.1 mm.
And flatly fixing the blade on a sample frame by using a transparent adhesive tape, and scanning point by using a transmission imaging module by using the transparent adhesive tape as reference to obtain a terahertz image of the blade sample, wherein the scanning frequency is 15 Hz, and the image data size is 250 multiplied by 250 pixels. Fig. 1 shows terahertz time-domain spectral imaging of cedrela sinensis leaves at different delay times, and terahertz frequency-domain imaging and absorption spectral imaging of cedrela sinensis leaves at different frequency positions.
And step 3: extraction of spectral data of blade in terahertz imaging
In the aspect of spectrum extraction, terahertz images of 208 toona sinensis leaf samples are divided into two groups. Wherein, 200 terahertz images (25 images each including 1 CK group and 7 pesticide residue groups) of 8 pesticide residue types are subjected to leaf spectrum extraction and residue type identification. To extract the spectrum from the image, a progressive scan is performed on the leaf using the Region of Interest (ROI) (3 × 3 pixels) as a shift operator, extracting the spectrum of each pixel in the ROI, and taking the average as the spectrum at the current ROI position. On this basis, 100 spectra were extracted from the terahertz image of each leaf. 20000 absorption spectra (2500 spectra for each of 8 classes) were thus co-extracted from the terahertz images of 200 leaves. To avoid segmenting the leaf images, and to facilitate image visualization, 2500 background spectra (BG) were extracted from 25 images of each category. Therefore, 22500 spectra (9 categories: 1 BG group, 1 CK group and 7 pesticide residue group) were extracted in total for establishing a qualitative identification model of pesticide multi-residue types. The remaining 8 terahertz images (one image for each type, 1 CK group and 7 pesticide residue groups) were used to achieve visualization of the distribution of various pesticide residues on the leaves. In order to eliminate the interference of spectral noise, only the spectrum in the frequency range of 0.2-2.2 THz with higher signal-to-noise ratio is intercepted from the range of 0.1-3 THz for modeling and visual analysis.
And 4, step 4: DCNN
Dividing the 9-class terahertz spectrum data obtained in the step (2) into a training data set and a test data set; constructing a DCNN deep convolution neural network by using the training data set, and training the DCNN deep convolution neural network to generate a training model of the cedrela sinensis leaf image; inputting the test sample in the test data set into a training model of the cedrela sinensis leaf image, and judging the type of pesticide residue on the cedrela sinensis leaf to generate a test result; and iteratively updating the training model of the cedrela sinensis leaf image according to the test result to obtain the recognition model of the cedrela sinensis leaf image. Fig. 2 is a flow chart for detecting various BZM pesticide residues in cedrela sinensis leaves by utilizing terahertz imaging and DCNN.
Specifically, the present example identifies multiple pesticide residues on fresh toona sinensis leaves using a DCNN model consisting of a convolutional layer, a pooling layer, and a full junction layer. The DCNN framework is provided with two convolution modules and 5 full connection layers and is used for processing 9 types of input terahertz spectrum data (BG, CK and 7 pesticide residues), wherein 70% (15750) of spectrum data are randomly selected for training and establishing a model, and the rest 30% (6750) of spectrum data are used for testing and verifying the model. Each convolution module includes convolution layers and a max-pooling layer, and the number of convolution filters in the first and second convolution layers is set to 32 and 64, respectively. In order to quickly understand the local information of the spectral data and reduce the dimensionality of the spectral data, the step size of the convolution kernel is set to 1, the kernel size of the maximum pool layer is set to 1 × 3, and the nonlinear properties of the decision mapping function are normalized by using a corrected Linear Unit (ReLU). A Batch Normalization (BN) function is used before each fully connected layer and after each convolution module to increase learning speed and reduce initialization requirements. The full connectivity data of the last layer is processed using the Softmax function to highlight the maximum and limit the eigenvalues of other neural units below the maximum. And measuring the probability distribution distance between the DCNN output label and the real class label by using a classification cross entropy loss function. And processing the loss function by adopting an adaptive moment estimation algorithm. In order to prevent overfitting in the training process, 30% of training sets are selected for verification, and after batch normalization is carried out on all connecting layers, overfitting is further prevented by adopting a jump-out method. The training period, learning speed, beta1, and beta2 parameters are set to 500, 0.001, 0.9, and 0.99, respectively.
Fig. 3 shows the recognition result of the DCNN model. The training and testing results of the DCNN for each class reached a very satisfactory level of accuracy ((a) of fig. 3). Fig. 3 (b) shows the loss and accuracy of the DCNN model during the iteration. After 250 iterations, the accuracy of the DCNN model gradually stabilized. Accordingly, after 250 iterations, the loss of the test set fluctuates to some extent but gradually decreases, and the loss of the training process decreases even after 250 iterations. Compared with a test set, the training set has higher precision and lower loss, and the robustness and the non-overfitting property of the DCNN model are proved. Particularly, the modeling accuracy rate of the CK sample without pesticide residues in training and testing reaches 100%. In contrast, DCNN's identification accuracy was lowest for leaf samples containing BNL pesticide residues, probably due to BNL instability and degradability. For BG samples (no leaf or pesticide), the accuracy of the training and test sets may not reach 100%, possibly due to interference from instrument noise. The modeling accuracy of the rest four multi-pesticide residue samples (M1, M2, M3 and M4) is high. The nine classes of samples were analyzed in combination, and the average recognition accuracy of the training set and the test set was 97.27% and 96.74%, respectively. These results indicate that the DCNN model can be an effective tool for identifying different pesticide residues in leaves.
And 5: image visualization of multi-component mixed pesticide distribution in leaves
The remaining 8 terahertz images (one image for each type, 1 CK group and 7 pesticide residue groups) in step 3 are used for realizing visualization of distribution of various pesticide residues on the leaves, and terahertz absorption spectra of the 8 terahertz images (each 225 × 225= 62500) are extracted (namely one terahertz absorption spectrum is extracted for each pixel).
And (3) selecting spectral data of a frequency band (0.2-2.2 THz) with a higher signal-to-noise ratio from the absorption spectra (0.1-3.0 THz) corresponding to each pixel point of the blade terahertz image as the input of the DCNN model obtained in the step (4), wherein the output value of the DCNN model is the predicted category label of the corresponding spectrum at each pixel point. Since the spectrum at each pixel point of the terahertz image is input into the DCNN model for calculation, the class label predicted at each pixel point can be restored to a two-dimensional image of 250 × 250 pixels. Fig. 4 shows a DCNN visualization of the eight cedrela sinensis leaves, clearly revealing the morphology of the leaves and the spatial distribution of pesticide residues on the leaves. The predicted class labels 1-9, BG, CK, BNL, BCM, TBZ, M1, M2, M3, and M4, are represented in different colors. The image visualization method shows the classification and distribution of various pesticide residues in the leaves, which cannot be observed by naked eyes in the original leaf sample. Although the BG region presents some class labels that are mispredicted, which may be due to interference from instrumental noise, the region of the leaf is well shown in the restored image. Fig. 4 (a) shows that the images of the k group of toona sinensis leaves, the result of class labeling at each pixel is substantially accurate, but there are some recognition errors on the veins and edges. As shown in fig. 4 (b), (c) and (d), the leaf images containing BNL, BCM and TBZ pesticide residues highly reflect their corresponding class labels. Fig. 4 (e) and (f) show the distribution of two mixed pesticide residues of M1 (BNL + BCM) and M2 (BNL + TBZ), respectively. Due to the instability of BNL, not only most of the mixed pesticide residue information of M1 and M2, but also information of a few single-component pesticides BCM and TBZ are shown in (e) and (f) of fig. 4. As shown in (g) of fig. 4, the imaging result of M3 (BCM + TBZ) was superior to that of M1 and M2. As shown in fig. 4 (h), the predicted label is more complex for M4 (BNL + BCM + TBZ) containing a three-component pesticide residue. These results indicate that as the number of pesticide components increases, the accuracy of the corresponding DCNN prediction at each pixel point decreases, which poses a challenge to the detection of multiple pesticide residues. However, by using most of the label types displayed in the blade area in the image as the final identification label, more reliable identification results of a plurality of mixed pesticide residue types can be obtained.

Claims (6)

1. A qualitative detection method for pesticide residues in toona sinensis based on a terahertz imaging technology comprises the following steps:
(1) carrying out terahertz imaging on 8 types of toona sinensis leaf samples; the 8 types of Chinese toon leaf samples are respectively as follows: cedrela sinensis leaves as blank controls, cedrela sinensis leaves containing BNL residues, cedrela sinensis leaves containing BCM residues, cedrela sinensis leaves containing TBZ residues, cedrela sinensis leaves containing BNL and BCM residues, cedrela sinensis leaves containing BNL and TBZ residues, cedrela sinensis leaves containing BCM and TBZ, cedrela sinensis leaves containing BNL, BCM and TBZ;
(2) extraction of spectral data of blade in terahertz imaging
Scanning the terahertz image of each leaf sample step by using the region of interest as a shift operator, extracting the spectrum of each pixel in the region of interest, taking the average value as the spectrum of the current region of interest, and extracting a plurality of absorption spectra from the terahertz image of each leaf sample on the basis; in addition, a plurality of background spectra (BG) are extracted from the terahertz image of each type of cedrela sinensis leaf; thereby obtaining 9-type terahertz spectrum data;
(3) dividing the 9-class terahertz spectrum data obtained in the step (2) into a training data set and a test data set; constructing a DCNN deep convolution neural network by using the training data set, and training the DCNN deep convolution neural network to generate a training model of the cedrela sinensis leaf image; inputting the test sample in the test data set into a training model of the cedrela sinensis leaf image, and judging the type of pesticide residue on the cedrela sinensis leaf to generate a test result; iteratively updating the training model of the cedrela sinensis leaf image according to the test result to obtain a recognition model of the cedrela sinensis leaf image;
(4) and (4) carrying out image visualization on the pesticide residues on the toona sinensis leaves to be detected by using the identification model established in the step (3), so that the types and the distribution conditions of the pesticide residues on the leaves can be displayed.
2. The qualitative detection method for pesticide residues in toona sinensis based on the terahertz imaging technology as claimed in claim 1, characterized in that: the specific operation of the step (4) is as follows: obtaining a terahertz image by the to-be-detected toona sinensis leaf according to the operation in the step (1), then extracting a terahertz absorption spectrum of each pixel point of the terahertz image, inputting the terahertz absorption spectrum into the identification model established in the step (3) to predict class labels, wherein the predicted class labels 1-9, namely BG, CK, BNL, BCM, TBZ, M1, M2, M3 and M4, are respectively represented by different colors, and the type and distribution condition of pesticide residues on the leaf can be visually displayed by restoring the predicted class labels at each pixel point into a two-dimensional image of complete pixels.
3. The qualitative detection method for pesticide residues in toona sinensis based on the terahertz imaging technology as claimed in claim 2, characterized in that: and (5) when more than one type of label is displayed in the blade area in the two-dimensional image obtained in the step (4), taking the most label types as final distinguishing labels.
4. The qualitative detection method of pesticide residues in toona sinensis based on terahertz imaging technology as claimed in one of claims 1 to 3, characterized in that: in the step (1), 25 identical samples are set for each type of cedrela sinensis leaf, a THz-TDS system is used for imaging scanning of the cedrela sinensis leaf samples, a transmission imaging module is used for scanning point by point to obtain terahertz images of the leaf samples within the range of 0.1-3 THz, the scanning frequency is 15 Hz, and the size of image data is 250 multiplied by 250 pixels;
in the step (2), the interested area of 3 x 3 pixels is adopted as a shift operator to scan on the blade step by step, the spectrum of each pixel in the interested area is extracted, the average value is taken as the spectrum of the current interested area, and on the basis, 100 absorption spectra are extracted from the terahertz image of each blade sample; 2500 pieces of background spectra were extracted from each category of terahertz image.
5. The qualitative detection method of pesticide residues in toona sinensis based on terahertz imaging technology as claimed in one of claims 1 to 3, characterized in that: in the step (3), the DCNN deep convolutional neural network structure includes two convolutional modules and 5 fully-connected layers, each convolutional module includes a convolutional layer and a maximum pooling layer, the number of convolutional filters in the first convolutional layer and the second convolutional layer is set to be 32 and 64, respectively, the step length of a convolutional kernel is set to be 1, the kernel size of the maximum pooling layer is set to be 1 × 3, and a correction linear unit is used to normalize the nonlinear property of the decision mapping function; using a batch normalization function before each fully connected layer and after each convolution module; processing the full connectivity data of the last layer by using a Softmax function to highlight the maximum value and limit the characteristic values of other nerve units below the maximum value; measuring the probability distribution distance between the DCNN output label and the real class label by using a classified cross entropy loss function; processing the loss function by adopting an adaptive moment estimation algorithm; the ratio of the training data set to the test data set is 70%: 30%, after all the connecting layers are subjected to batch normalization, adopting a jump-out method to further prevent overfitting; the training period, learning speed, beta1, and beta2 parameters are set to 500, 0.001, 0.9, and 0.99, respectively.
6. The qualitative detection method of pesticide residues in toona sinensis based on terahertz imaging technology as claimed in one of claims 1 to 3, characterized in that: and (3) intercepting the spectrum in the frequency range of 0.2-2.2 THz for modeling in the step (3) and visual analysis in the step (4).
CN202010480423.3A 2020-05-30 2020-05-30 Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology Withdrawn CN111812058A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010480423.3A CN111812058A (en) 2020-05-30 2020-05-30 Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010480423.3A CN111812058A (en) 2020-05-30 2020-05-30 Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology

Publications (1)

Publication Number Publication Date
CN111812058A true CN111812058A (en) 2020-10-23

Family

ID=72847862

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010480423.3A Withdrawn CN111812058A (en) 2020-05-30 2020-05-30 Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology

Country Status (1)

Country Link
CN (1) CN111812058A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113344051A (en) * 2021-05-28 2021-09-03 青岛青源峰达太赫兹科技有限公司 Neural network classification method based on terahertz data
CN113740276A (en) * 2021-09-02 2021-12-03 福州大学 Fruit and vegetable pesticide residue visual real-time detection method and system based on multispectral detection system
CN114112979A (en) * 2021-12-02 2022-03-01 南京林业大学 Terahertz spectrum-based agricultural product pesticide residue quantitative detection method
WO2023208619A1 (en) * 2022-04-25 2023-11-02 Bayer Aktiengesellschaft Prediction of deposition structures of pesticides and/or nutrients on parts of plants

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113344051A (en) * 2021-05-28 2021-09-03 青岛青源峰达太赫兹科技有限公司 Neural network classification method based on terahertz data
CN113740276A (en) * 2021-09-02 2021-12-03 福州大学 Fruit and vegetable pesticide residue visual real-time detection method and system based on multispectral detection system
CN114112979A (en) * 2021-12-02 2022-03-01 南京林业大学 Terahertz spectrum-based agricultural product pesticide residue quantitative detection method
WO2023208619A1 (en) * 2022-04-25 2023-11-02 Bayer Aktiengesellschaft Prediction of deposition structures of pesticides and/or nutrients on parts of plants

Similar Documents

Publication Publication Date Title
Nie et al. Classification of hybrid seeds using near-infrared hyperspectral imaging technology combined with deep learning
Bauriegel et al. Early detection of Fusarium infection in wheat using hyper-spectral imaging
CN111812058A (en) Qualitative detection method for pesticide residues in toona sinensis based on terahertz imaging technology
Chandrasekaran et al. Potential of near-infrared (NIR) spectroscopy and hyperspectral imaging for quality and safety assessment of fruits: An overview
ElMasry et al. Recent applications of multispectral imaging in seed phenotyping and quality monitoring—An overview
Ravikanth et al. Extraction of spectral information from hyperspectral data and application of hyperspectral imaging for food and agricultural products
Kimuli et al. Utilisation of visible/near-infrared hyperspectral images to classify aflatoxin B1 contaminated maize kernels
ElMasry et al. Image analysis operations applied to hyperspectral images for non-invasive sensing of food quality–a comprehensive review
Yang et al. Spectral and image integrated analysis of hyperspectral data for waxy corn seed variety classification
Wang et al. Detection of external insect infestations in jujube fruit using hyperspectral reflectance imaging
Li et al. Hyperspectral imaging and their applications in the nondestructive quality assessment of fruits and vegetables
Kimuli et al. Application of SWIR hyperspectral imaging and chemometrics for identification of aflatoxin B1 contaminated maize kernels
Cozzolino Use of infrared spectroscopy for in-field measurement and phenotyping of plant properties: instrumentation, data analysis, and examples
Xiaobo et al. Independent component analysis in information extraction from visible/near-infrared hyperspectral imaging data of cucumber leaves
Al-Saddik et al. Assessment of the optimal spectral bands for designing a sensor for vineyard disease detection: the case of ‘Flavescence dorée’
Shao et al. Rapid classification of Chinese quince (Chaenomeles speciosa Nakai) fruit provenance by near-infrared spectroscopy and multivariate calibration
Gao et al. Improvement of chlorophyll content estimation on maize leaf by vein removal in hyperspectral image
Zhao et al. Detection of fungus infection on petals of rapeseed (Brassica napus L.) using NIR hyperspectral imaging
Jiang et al. Rapid and non-destructive detection of natural mildew degree of postharvest Camellia oleifera fruit based on hyperspectral imaging
Cao et al. Identification of species and geographical strains of Sitophilus oryzae and Sitophilus zeamais using the visible/near‐infrared hyperspectral imaging technique
Barbosa et al. A novel use of infra-red spectroscopy (NIRS and ATR-FTIR) coupled with variable selection algorithms for the identification of insect species (Diptera: Sarcophagidae) of medico-legal relevance
Soares et al. Classification of individual cotton seeds with respect to variety using near-infrared hyperspectral imaging
He et al. Classification of aflatoxin B1 naturally contaminated peanut using visible and near-infrared hyperspectral imaging by integrating spectral and texture features
Barré et al. Automated phenotyping of epicuticular waxes of grapevine berries using light separation and convolutional neural networks
Tee et al. Discovering the Ganoderma boninense detection methods using machine learning: a review of manual, laboratory, and remote approaches

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WW01 Invention patent application withdrawn after publication

Application publication date: 20201023

WW01 Invention patent application withdrawn after publication