CN-122024032-A - High-precision intelligent optical identification technology for target transgenic sugarcane

Abstract

The invention discloses an intelligent optical detection technology based on terahertz time-domain spectroscopy technology combined with an Isomap-KS-DCNN deep learning model, which successfully carries out classification and identification on five transgenic sugarcanes (target genes are anti-glufosinate genes and anti-glyphosate genes). The technology collects active excitation THz transmission spectra of five transgenic sugarcane in a range of 0.3-2.0 THz by using a terahertz time-domain spectroscopy system (THz-TDS), performs feature extraction on absorbance spectrum data by using an Isomap dimension reduction method to obtain a fingerprint feature matrix, and divides the obtained feature matrix into a training set and a testing set by using a Kennard-Stone (KS) algorithm and then inputs the training set and the testing set into a Deep Convolutional Neural Network (DCNN) classifier. Finally, a classification result diagram is obtained. The result shows that the classification accuracy of the Isomap-KS-DCNN model reaches 100%. According to the invention, THz absorbance data is utilized to carry out Isomap dimension reduction, then a feature matrix is utilized, a KS algorithm is applied to divide a training set and a testing set, and finally a DCNN classifier is utilized, so that the transgenic sugarcane can be subjected to intelligent detection rapidly, high-precision, environment-friendly and economical. The method uses an Isomap-KS-DCNN model, provides a brand-new rapid, high-precision, environment-friendly and economical measurement scheme for making and implementing corresponding detection and management standards of the transgenic sugarcane, can effectively improve the safety and effectiveness of the transgenic sugarcane, meets different requirements, and enhances the information transparency of the public to the transgenic sugarcane.

Inventors

• TU SHAN
• Pang Senhao
• ZHANG CHENG
• XIAO HANG
• LU MINGMING
• Tan Gunwu
• Su Jingkai
• Song zhongzhou
• ZHANG WENTAO
• LIU XIHUI
• WANG JUNGANG
• HU JUNHUI
• XIAO HUAPENG
• HE QILIN

Assignees

• 广西师范大学
• 苏靖凯

Dates

Publication Date

20260512

Application Date

20250221

Claims (7)

1. 1. A transgenic sugarcane high-precision intelligent optical detection technology based on an Isomap-KS-DCNN model is characterized by comprising the following steps: 1) Pretreatment of five transgenic sugarcane; 2) Acquiring original absorbance spectrum data of five transgenic sugarcane samples by using THz-TDS; 3) Performing dimension reduction processing on the collected original absorbance spectrum data by using Isomap to obtain input feature matrix data; 4) The training set and the test set are divided by using KS algorithm. 5) Performing DCNN classification analysis on the obtained input feature matrix data; 6) The classification results show a confusion matrix plot and a classification accuracy comparison plot for five transgenic sugarcane.
2. 2. The intelligent optical detection technology for transgenic sugarcane according to claim 1, wherein the pretreatment in the step (1) comprises grinding, filtering by using a 200-mesh sieve, weighing and tabletting, and finally drying in a vacuum drying oven.
3. 3. The intelligent optical detection technology of transgenic sugarcane according to claim 1, wherein the wave band in the step (2) is 0.3-2 thz.
4. 4. The intelligent optical detection technology for identifying transgenic sugarcane as claimed in claim 1, wherein the dimension of raw spectral data in step (3) is reduced from thousands of bits to single digits, and key features in raw absorbance spectral data are retained.
5. 5. The intelligent optical detection technology for identifying transgenic sugarcane according to claim 1, wherein the Isomap dimension reduction method in the step (3) calls an MDS algorithm on the geodesic distance to calculate a lowest dimension embedded matrix.
6. 6. The intelligent optical detection technology of transgenic sugarcane according to claim 1, wherein the DCNN classifier in the step (4) adopts a method of deep convolutional neural network.
7. 7. The intelligent optical detection technology of transgenic sugarcane according to claim 1, wherein the DCNN classification algorithm in the step (4) uses convolution calculation of a deep convolutional neural network as follows: A feature map representing the number of the s-th channel of the first layer; Representing a set of the ith filters of the ith channel of the first layer; as bias term, represent convolution operation; The full-connected layer output using the deep convolutional neural network approach can be expressed as: y=f(Wx+b) x is the input vector, n x 1;W is the weight matrix, m x n, b is the bias vector, m x 1;f is the activation function ReLU, y is the output vector, and m x 1.

Description

High-precision intelligent optical identification technology for target transgenic sugarcane Technical Field The invention relates to an intelligent optical detection technology of target transgenic sugarcane, in particular to a method for classifying and identifying transgenic sugarcane with different target genes of the same receptor by using an Isomap-KS-DCNN deep learning model based on THz time domain spectroscopy technology. Background Sugar cane is the largest commercial crop in tropical and subtropical regions, producing 75% of the world's sugar, but biotic stresses including diseases, harmful weeds and pests and abiotic stress factors including seasonal aberrations in the monsoon, low temperature, salinity, alkalinity and heavy metal elements contained in the soil have many negative effects on the production yield and quality of sugar cane. Among them, application of herbicides is considered as an effective method for improving yield and quality of sugarcane, but herbicides of glufosinate and glyphosate having low cost, low toxicity and high efficiency are currently non-selective broad-spectrum herbicides, which inevitably cause toxic effects on target crops during application, so that they can be used only on crops resistant to glufosinate genes and glyphosate genes. Although the traditional breeding method successfully produces a large amount of commercial products, the problems of long variety development time, more complex improvement of polyploid crop sugarcane than diploid and the like exist. There is a need for more efficient, immediate new technologies to address these challenges. Modern genetic engineering techniques may enable scientists to transfer genes for desired traits (e.g., resistance to disease, insects, and pests, or adverse environmental stresses) in a more accurate, predictable, and controllable manner as compared to selective breeding. Humans have made significant breakthroughs in the agricultural field through plant transgenesis, and the creation of plant biotechnology can help agriculture achieve higher yields in a more sustainable manner. However, studies have shown that the use of transgenic techniques can result in unexpected changes in plant physiology, which in turn can lead to intentional or unintentional expression of metabolites by transgenic plants (e.g., lignin, cuticle, coat, etc.), which can be detrimental to non-target transgenic plants in fully exposed conditions, creating negative ecological interactions. Therefore, it is important to perform rapid, nondestructive and accurate classification detection on the planted transgenic plants of different target genes in agriculture. The THz spectrum is between Microwave (MW) and Infrared (IR) and has a frequency range of 0.1 to 10.0THz. Intermolecular and intramolecular rotation and vibration at THz frequencies provide unique fingerprints for many materials. Different biological macromolecules have specificity in structure and arrangement, so chemical components of crops can be analyzed through THz spectral fingerprint patterns. THz is also suitable for nondestructive, accurate and rapid detection of crops because of the characteristics of low energy and high penetration of the wave band. Successful implementation of THz detection systems, however, requires the development of reliable and efficient identification algorithms. The massive amounts of complex data generated by THz spectral acquisition often contain a large amount of redundant information, which often leads to the so-called "dimension disaster" problem. Feature extraction and data dimension reduction techniques play a vital role in processing these high-dimensional data. By effectively extracting key information and reducing the dimensionality of data, the computational complexity can be reduced, and the efficiency and accuracy of subsequent analysis can be improved, so that the challenges in high-dimensional data processing are better solved. The Isomap algorithm builds on the classical MDS algorithm. The nonlinear manifold learning dimension reduction method has the core idea of keeping the geodesic distance between data points, effectively keeping the global geometry of the data and having advantages when processing different kinds of data with complex boundaries. Deep Convolutional Neural Network (DCNN) classification is a high-level classification method based on convolutional neural networks. The core idea is to automatically extract the characteristics in the data through the weight sharing and local connection mechanism, so as to accurately predict. At present, classification and identification of transgenic sugarcane with different target genes by using an Isomap-KS-DCNN classification model of THz-TDS are in a blank stage. Disclosure of Invention The invention aims to provide an Isomap-KS-DCNN deep learning intelligent optical detection technology for classifying and identifying transgenic sugarcane with different target genes of the same recepto