Research On Heavy Metal Detection Method Of Lettuce Leaves Based On Depth Features Of Hyperspectral Image

In recent years,the pollution of heavy metals in vegetables has been more serious,and heavy metals can harm people's health through the food chain.At present,hyperspectral imaging technology has been successfully applied to the research of vegetable quality and safety detection.As a nonlinear method,deep learning can introduce the inherent non-linearity of neural networks to fully mine the nonlinear information hidden behind massive data,thereby reducing the redundancy of data information.In this paper,visible-near infrared hyperspectral imaging technology was used to detect heavy metals(cadmium,lead,cadmium and lead)in lettuce leaves.The heavy metal detection mechanism of hyperspectral imaging technology and the optimization and improvement of the neural network structure of the stacked automatic encoder were studied,and the optimized deep network structure was applied to the detection of heavy metals in lettuce leaves.The details were as follows:(1)The parameters of the external biomass(plant height,diameter,leaf color and texture)and internal physiological and biochemical(leaf SPAD value,nitrogen content and moisture content)of lettuce samples were measured and analyzed under different heavy metal(cadmium,lead,cadmium and lead)levels.The analysis results showed that with the increase of the stress concentration of different heavy metals(cadmium,lead,cadmium and lead),the internal and external parameters of the lettuce samples increased first and then decreased.In addition,the hyperspectral image information in the visible and near infrared band is mainly related to the external phenotype,internal macro molecules(protein,water)and cell arrangement of plants.Therefore,the analysis of the changes of the internal and external parameters of lettuce leaves under different heavy metals can provide a theoretical support for the application of visible near infrared hyperspectral image technology in the detection of different heavy metals in lettuce leaves.(2)SAE-LSSVR algorithm(stacked auto encoder coupled to partial least squares support vector machine regression)was proposed and applied to the detection of Cd in lettuce leaves.The whole region of lettuce leaf sample spectral data was collected and preprocessed with different spectral pre-treatment methods.Successive projections algorithm(SPA),partial least square regression(PLSR)and stacked auto encoder(SAE)were used to acquire the optimum wavelength,respectively.Besides,the characteristic wavelengths were used to build partial least squares support vector machine regression(LSSVR)models.Furthermore,the best prediction performance for detecting Cd content in lettuce leaves was obtained by Savitzky-Golay combined with first derivative(SG-1st)pre-processing method,with R_p~2of 0.9487,RMSEP of0.01049 mg/kg and RPD of 3.330 using SAE-LSSVR method.(3)WT-SAE algorithm(wavelet transform combined with stacked auto encoder)was proposed and applied to the detection of Pb in lettuce leaves.The whole region of lettuce leaf sample spectral data was collected and preprocessed.In addition,the WT-SAE algorithm was used to complete the extraction of spectral depth features,and support vector machine regression(SVR)was used for regression modeling.Furthermore,the best prediction performances for detecting lead(Pb)concentration in lettuce leaves was obtained from the raw data set,with the coefficient of determination for prediction(R_p~2)of 0.9590,root mean square error for prediction(RMSEP)of 0.05587 mg/kg and residual predictive deviation(RPD)of 3.251 using db5 as wavelet basis function with wavelet fifth layer decomposition.(4)WT-SCAE algorithm(wavelet transform combined with stack convolution auto encoder)was proposed and applied to the detection of compound heavy metals(Cd and Pb)in lettuce leaves.WT was used to decompose the visible-near infrared(400.68-1001.61nm)hyperspectral image of lettuce sample in the multi-scale transform to acquire the optimal wavelet decomposition layers of Cd and Pb content prediction,and then using SCAE to perform deep feature learning on spectral data under optimal wavelet decomposition layer.SVR models established by the deep features obtained by WT-SCAE achieved reasonable performance with R_p~2of 0.9319,RMSEP of 0.04988 mg/kg and RPD of 3.187 for Cd content,and with R_p~2of 0.9418,RMSEP of 0.04123 mg/kg and RPD of 3.214 for Pb content.The heavy metal detection method of lettuce leaves based on the depth features of hyperspectral images was proposed in this paper,it can achieve the effective extraction of the depth spectral features and the rapid nondestructive detection of heavy metals.The research results of this paper can also provide theoretical and methodological basis for the monitoring of high-quality planting of other green economic crops and the development of inorganic detection equipment.