Study On Identification Method Of Chilo Suppressalis Based On Terahertz Detection Technology

Chilo suppressalis is one of the most serious and frequent pests in rice growth.If not controlled in time and effectively,it will lead to a large decrease in rice production.Because the Chilo suppressalis is in the stem,there is a great technical bottleneck in traditional detection methods.Terahertz spectrum can provide qualitative and quantitative information of samples,and THz has the characteristics of low photon energy and non-ionizing radiation.Some non-polar products are transparent in THz band,but opaque in visible and infrared wavelengths.The main components of rice stem are cellulose and other non-polar materials,which have high penetration in terahertz band and low radiation energy of THz,so the high precision detection can be achieved.Therefore,the terahertz detection technology has great potential for the detection of rice suppressant.In this study,a large number of experiments and stoichiometric methods were used to establish a model and analyze the data to detect rice Chilo suppressalis.The specific research content is as follows:(1)Study on the identification and detection of eggs based on terahertz transmission imaging technology.The rice leaf area,eggs of the first day,eggs of the second day,eggs of the third day and eggs after hatching are taken as experimental samples.Five groups of absorbance parameters in the frequency domain of samples are extracted,and a linear discriminant analysis model(LDA)is established using different pretreatment methods.It is found that the LDA model established by standard normal variable transformation(SNV)preprocessing has the best effect,the accuracy of modeling set is 95.11%,and the accuracy of prediction set is 94%.Feature wavelength was extracted from the preprocessed data.Three feature wavelength screening methods,namely competitive adaptive reweighting method(CARS),invariant information elimination method(UVE)and continuous projection algorithm(SPA),were used to establish linear discrimination models.The results showed that:SNV-CARS-LDA model had the best effect,with the accuracy of 99.33% of the whole modeling set and 98% of the prediction set,which could effectively identify whether the rice leaves contained insect eggs and whether the insect eggs hatched.And in the image,the egg region and the leaf region can also be clearly separated by threshold segmentation.(2)Based on terahertz time-domain spectroscopy to detect the damage of chilo suppressant to rice,using the principle that the optical properties of plants can change under disease conditions,so as to realize the damage detection of Chilo suppressalis to rice.The hatching larvae were implanted into the growing rice,and the rice damage status was divided into normal,mild,moderate and severe groups at 0,2,4 and 6 days according to the different implantation time.Data were collected and analyzed in time domain and frequency domain spectrum.The absorbance parameters of four groups of samples were extracted.Qualitative discriminant models were built using three different pretreatment methods,namely asymmetric least squares method(As LS),iterative adaptive Weighted Penalty least Squares method(Air PLS)and Baseline estimation and Sparse Denoising method(BEADS).It was found that the BEADS had the best pretreatment effect.CARS,UVE and SPA were used to extract the characteristic wavelength of the preprocessed data.Finally,linear discriminant model(LDA),support vector machine model(SVM)and K-nearest neighbor algorithm model(KNN)were established.The results showed that: The modeling method of BEADS-CARSKNN was the best,with the accuracy of 95.45%,100%,100% and 100% respectively,and the accuracy of the whole prediction set reached 98.88%,which could effectively identify the damage degree of the early stage of the pest on rice.(3)The detection of chilo suppressant in rice based on terahertz spectrum and image.Preparation of rice stem and borers mixed pressing plate,analysis of time domain spectrum.Partial least squares(PLS)and least squares support vector machine(LS-SVM)quantitative detection models were established using CARS,UVE and SPA three different feature wavelength extraction methods.The results showed that:The prediction effect of CARS algorithm combined with LS-SVM model was better,the correlation coefficient Rp was0.9699,RMSEP was 0.0541.In view of the practical need of physical sample detection,terahertz imaging was used to detect rice Chilo suppressalis.Terahertz transmission images of rice straw and Chilo suppressalis samples were collected,and terahertz spectral data were extracted from the regions of interest for analysis and discrimination model was established.The results showed that the identification rate of the established SVM qualitative discrimination model could reach 100%.In order to explore the visual expression of the rapid detection of the rice suppressant,the images of the physical samples in the band of 0.5-1.5THz were extracted and the threshold segmentation was performed to clearly identify the larvae of the moth suppressant hidden in the rice stalk,thus achieving the identification and detection effect.In addition,in order to reduce the amount of imaging data,the SPA feature extraction method was further used,and only 10 wavelengths of the composite image was used to realize the visual detection of chilo suppressant larvae in rice straw.