Detection of Pesticide Residue Level in Grape Using Hyperspectral Imaging with Machine Learning

Abstract

Rapid and accurate detection of pesticide residue levels can help to prevent the harm of pesticide residue. This study used visible/near-infrared (Vis-NIR) (376–1044 nm) and near-infrared (NIR) (915–1699 nm) hyperspectral imaging systems (HISs) to detect the level of pesticide residues. Three different varieties of grapes were sprayed with four levels of pesticides. Logistic regression (LR), support vector machine (SVM), random forest (RF), convolutional neural network (CNN), and residual neural network (ResNet) models were used to build classification models for pesticide residue levels. The saliency maps of CNN and ResNet were conducted to visualize the contribution of wavelengths. Overall, the results of NIR spectra performed better than those of Vis-NIR spectra. For Vis-NIR spectra, the best model was ResNet, with the accuracy of over 93%. For NIR spectra, LR was the best, with the accuracy of over 97%, but SVM, CNN, and ResNet also showed closed and fine results. The saliency map of CNN and ResNet presented similar and closed ranges of crucial wavelengths. Overall results indicated deep learning performed better than conventional machine learning. The study showed that the use of hyperspectral imaging technology combined with machine learning can effectively detect the level of pesticide residues in grapes.

Discrimination of Pesticide Residue Levels on the Hami Melon Surface Using Multiscale Convolution

“…This was attributed to the moisture change in the Hami melon. The pesticide residues did not change the position of the spectral feature absorbance peaks, and it agreed with previous studies such as Ye et al [ 21 ], Yu et al [ 23 ], and Sun et al [ 36 ]. The spectral curves of different pesticide residue contents in different mature periods overlapped partially, and the difference was not obvious.…”

“…This result was obviously better than other models. However, it was lower in comparison to Ren et al [ 41 ], who used the chisquare test combined with linear discriminant analysis, and Ye et al [ 21 ], who used ResNet or logistic regression. However, they did not measure the actual residue contents, and divided residue levels according to the ratio of pesticides to water.…”

“…The padding strategy was selected to keep the size of the feature map. According to previous studies [ 21 , 22 , 23 ], the activation, objective, and loss functions were commonly used as rectified linear unit (ReLU), softmax, and multiclassification cross-entropy, respectively. The stochastic gradient descent (SGD) optimizer, with a momentum of 0.6 and decay of 1 × 10 −5 , was used to improve the training process.…”

“…The deep brief network (DBN) was used to select and extract spectral features, achieving the identification of fenvalerate and triazoline residues in lettuce leaves [ 20 ]. The residual neural network (ResNet) was shown to have a good effect on three-level residue discrimination in grapes [ 21 ]. Moreover, the one-dimensional convolutional neural network (1D-CNN) achieved the identification of pesticide residues on garlic chive leaves (λ-cyhalothrin, trichlorfon, phoxim, mixtures of trichlorfon and phoxim) [ 22 ], and also worked well on Hami melon (chlorothalonil, imidacloprid, and pyraclostrobin) [ 23 ].…”

Convolutional Neural Network for Object Detection in Garlic Root Cutting Equipment

“…In the study of identifying cocoa beans, a comparison of a deep computer vision system with a conventional computer vision system revealed that the former was more accurate [ 70 ]. Studies to detect pesticide residue levels have shown that deep learning outperforms traditional machine learning [ 71 ]. In a study of barley flour classification, machine learning methods were used to show superior predictive power compared to computer vision systems [ 72 ].…”