Non-destructive Detection Of Pesticide Residues On Broccoli Based On Hyperspectral Imaging

Broccoli is a characteristic vegetable in Zhejiang Province,which is not only nutritious but also a certain degree of anti-cancer and disease prevention,loved by the people.The broccoli produced in our country not only meets the needs of the domestic market but also is exported to the international market.Therefore,the problem of food safety of broccoli not only affects the health of the Chinese consumers but also affects the reputation of China's international market.In this paper,hyperspectral image technology was used to identify the types of pesticide residues on broccoli surface and the quantitative non-destructive detection of abamectin pesticides.The main contents are as follows:(1)Different types of pesticide residues and different concentrations of abamectin pesticide solutions were respectively dispensed on the surface of broccoli and numbered.The average reflectance spectrum of the region of interest(ROI)is extracted based on its image information,and the original spectral data is preprocessed by piecewise multiplicative scatter correction(PMSC),aim to eliminate light scattering of the average spectral information.(2)A method for the identification of pesticide residues on broccoli was proposed based on hyperspectral image technology.First of all,four groups of broccoli(180 particles in total)were used as experimental samples,which respectively contained imidacloprid,abamectin and propineb as the first third groups,and the last group was sprayed with water.The samples were scanned by hyperspectral image system in the range of 383.70-1032.69 nm.Then,the principal component analysis(PCA)and the successive projections algorithm(SPA)were used to select the characteristic spectra.Select the first 10 principal components and 8 characteristic wavelengths(458.51,500.02,522.13,551.77,614.04,720.32,769.08,818.26nm).Finally,a pesticide residue detection model based on full-band and characteristic spectra was established using four classification algorithms: Mahalanobis distance(MD),Least Squares Support Vector Machine(LSSVM),Artificial Neural Network(ANN)and Extreme Learning Machine(ELM).The results show that SPA-ELM model has the best recognition effect,and the correct rates of training set and test set are 98.33% and 96.67%.(3)A method for the identification of different concentrations of abamectin pesticide solutions on broccoli was proposed based on hyperspectral image technology.A total of 100 hyperspectral images(900-1700 nm)of broccoli witch sprayed abamectin pesticide in different concentrations were collected and the spectra were pretreated.Using liquid chromatography-mass spectrometry method according to the GB23200.20-2016 standard for spraying 5 different concentrations of broccoli pesticides broccoli specific residue detection.Select the first nine principal component spectra and 14 characteristic wavelengths(978.04,1039.71,1059.22,1098.29,1114.59,1140.68,1153.74,1170.07,1225.68,1310.96,1363.59,1422.92,1548.364,1611.73nm),Mahalanobis distance(MD),least square support vector machine(LSSVM),artificial neural networks(ANN)and extreme learning machine(ELM)models were created to predict different concentrations of abamectin pesticide solutions from full spectra and characteristic band date.The results show that ELM model based on full spectra has the best recognition effect,which accuracy is 72%.(4)Aiming to solve the problem of low recognition rate of abamectin low concentration residue on broccoli surface,research on classification the hyperspectral data of low concentration residues of Abamectin in broccoli based on Convolutional Neural Network.A method of converting spectral information into gray-scale image is proposed,based on the study of convolution neural network.The convolution network is used to study the change of texture information between gray-scale images of different data to classify the data.Contrasting the convolution neural network model with different depths,the classification model with four layers of network is finally selected.The results show that the highest accuracy rate of 84.9% based on convolutional neural network is 10.1% higher than the accuracy of the model based on extreme learning machine model,which proves the feasibility of using convolutional neural network to classify the hyperspectral data of low concentration pesticide residues between 24.25?g/Kg and 170.03?g/Kg on broccoli surface,and provides a new direction for making more efficient models and algorithms.