Research On Facility Disease Detection Method Based On Terahertz-hyperspectral And 3D Light Field Multi-dimensional Imaging

The identification and detection of crop diseases by means of information technology has always been the focus of research.A large number of studies have shown that the diseases of crops are usually located in the leaves of crops,and the disinfected crops will not only change their external morphology,but also destroy their internal components.At present,the detection of crop diseases still relies on the experience and judgment of professionals.The disadvantage is that it is time-consuming and labor-intensive and the accuracy is not high.In response to the existing problems,this article comprehensively uses knowledge in multiple fields such as three-dimensional light field imaging technology,hyperspectral detection,terahertz detection,machine learning,and multi-information fusion,and takes tomato and cucumber,which are common crops in greenhouses,as the research object.And the detection method of tomato mosaic disease was studied.The main research contents of this paper are as follows:(1)The classification and identification of cucumber and tomato diseases were studied by using hyperspectral imaging and terahertz time domain spectroscopy.Among them,the establishment method of hyperspectral effective spectrum,spectral pretreatment method,characteristic wavelength selection method and the establishment method of discriminant model are studied.Firstly,the effective spectral information of visible light(VS)and near infrared(NIR)is determined according to the original spectrum.By comparing and using the pretreatment methods of SG smoothing,wavelet transform and moving sliding window,the wavelet transform is determined as the optimal pretreatment methodBy using the pretreatment,segmentation and reconstruction technology of the point cloud data of the light field camera,the calculation method of the lesion area was studied.By comparing the leaf area measuring instrument,the average measurement error of cucumber leaves and tomato leaves was 6.23cm~2 and 1.93cm~2,which showed the high precision of the lesion quantization.Among them,the use of light field camera in the extraction of different canopy shape characteristic,studies the tomato plant morphological reconstruction method for the bottom of the canopy and leaf number of statistical method,through the reconstruction of point cloud and depth map segmentation,restore the tomato plants the lower canopy leaf shrivel morphological characteristics and the disease leaf quantity and the disease spot of statistical results.(2)The classification and identification of cucumber and tomato diseases were studied by using hyperspectral imaging and terahertz time domain spectroscopy.Among them,using hyperspectral imaging technology,the establishment method of effective spectrum,spectral pretreatment method,characteristic wavelength selection method and the establishment method of discriminant model are studied.First according to the original spectrum(VS)visible light and near infrared(NIR)under the spectrum information effectively,by SG smooth,wavelet transform,mobile sliding window used pretreatment methods such as contrast,determine the wavelet transform for the optimal pretreatment method,and then through the stepwise discriminant analysis method under visible light and near infrared band are selected features of wavelength and space,The linear discriminant model was established,and the results showed that the average recognition rate of cucumber powdery mildew was 93%and that of tomato Mosaic was 94.375%under the whole wave band.Then,through stepwise discriminant analysis method,the characteristic wavelengths in visible and near-infrared bands were selected and composed of the feature space,and the extraction and analysis methods of the feature images were studied to restore more details of the lesion samples.A linear discriminant model was established according to the characteristic bands.The results showed that the average recognition rate of cucumber powdery mildew was 93%and that of tomato Mosaic was 94.375%under the whole bands.The method of terahertz data preprocessing,the selection of terahertz effective spectrum,the selection of characteristic wavelength and the establishment of classification model are studied.Firstly,the original terahertz data was preprocessed by Python and the effective terahertz spectrum data set was established for subsequent processing.The effective terahertz spectrum was screened twice by IVSO-IRIV algorithm,and the terahertz feature space was formed.The extraction and analysis methods of terahertz feature images were studied to restore more details of the lesions.Then,the optimal regularization parameters and the regularization solution method were selected to establish the Ssparse Represent Classification(SRC)model.The results showed that the recognition accuracy of cucumber powdery mildey and tomato Mosaic was 87.78%and 89.58%at terahertz scale.(3)The establishment of multi-source information fusion recognition model.According to the feature information and recognition results extracted in the hyperspectral and terahertz scales,the DS evidence theory framework in the multi-source information fusion technology is used to fuse the information of the hyperspectral and terahertz scales,and the weight distribution and fusion rules are improved.To solve the conflict of evidence generated during the fusion process,a disease detection model based on multi-source information was established.The accuracy of the identification of cucumber powdery mildew and tomato mosaic disease was 94.35%and 94.89%,indicating that the fusion method is feasible.This paper uses the method of multi-source information detection to study the detection and identification methods of crop diseases in the facility environment,which provides a reference for the application of image processing,spectral analysis,and information fusion in agriculture.