Appropriate modeling methods and feature selection algorithms must be selected to improve the accuracy of early and mid-term remote sensing detection of wheat stripe rust. In the current study, we explored the effectiveness of the random forest (RF) algorithm combined with the extreme gradient boosting (XGboost) method for early and mid-term wheat stripe rust detection based on the vegetation indices extracted from canopy level hyperspectral measurements. Initially, 21 vegetation indices that were related to the early and mid-term winter wheat stripe rust were calculated on the basis of canopy level hyperspectral reflectance. Subsequently, the optimal vegetation index combination for disease detection was determined using correlation analysis (CA) combined with RF algorithms. Then, the disease severity detection model of early and mid-term winter wheat stripe rust was constructed using XGBoost method based on the optimal vegetation index combination. For the evaluation and comparison of the initial results, three commonly used classification methods, namely, RF, backpropagation neural network (BPNN), and support vector machine (SVM), were utilized. The vegetation index combinations determined by the single CA algorithm were also used to construct detection models. Compared with the detection models based on the vegetation index combination obtained using the single CA algorithm, the overall accuracy of the four detection models based on the optimal vegetation index combination based on CA combined with RF algorithms increased by 16.1% (XGBoost), 9.7% (RF), 8.1% (SVM), and 8.1% (BPNN). Among the eight models, the XGBoost detection model based on the optimal vegetation index combination using CA combined with RF algorithms, CA-RF-XGBoost, achieved the highest overall accuracy of 87.1% and the highest kappa coefficient of 0.798. Our results indicate that the RF combined with XGBoost can improve the detection accuracy of early and mid-term winter wheat stripe rust effectively at canopy scale.

With the development of globalization and agriculture trade, as well as its own strong migratory capacity, fall armyworm (FAW) (Spodoptera frugiperda) (J.E. Smith) has invaded more than 70 countries, posing a serious threat to the production of major crops in these areas. FAW has now also been detected in Egypt in North Africa, putting Europe, which is separated from it only by the Mediterranean Sea, at high risk of invasion. Therefore, this study integrated multiple factors of insect source, host plant, and environment to provide a risk analysis of the potential trajectories and time periods of migration of FAW into Europe in 2016~2022. First, the CLIMEX model was used to predict the annual and seasonal suitable distribution of FAW. The HYSPLIT numerical trajectory model was then used to simulate the possibility of the FAW invasion of Europe through wind-driven dispersal. The results showed that the risk of FAW invasion between years was highly consistent (P<0.001). Coastal areas were most suitable for the expansion of the FAW, and Spain and Italy had the highest risk of invasion, with 39.08% and 32.20% of effective landing points respectively. Dynamic migration prediction based on spatio-temporal data can enable early warning of FAW, which is important for joint multinational pest management and crop protection.

Fusarium head blight, caused by a fungus, can cause quality deterioration and severe yield loss in wheat. It produces highly toxic deoxynivalenol, which is harmful to human and animal health. In order to quickly and accurately detect the severity of fusarium head blight, a method of detecting the disease using continuous wavelet analysis and particle swarm optimization support vector machines (PSO-SVM) is proposed in this paper. First, seven wavelet features for fusarium head blight detection were extracted using continuous wavelet analysis based on the hyperspectral reflectance of wheat ears. In addition, 16 traditional spectral features were selected using correlation analysis, including two continuous removal transformed spectral features, six differential spectral features, and eight vegetation indices. Finally, wavelet features and traditional spectral features were used as input features to construct fusarium head blight detection models in combination with the PSO-SVM algorithm, and the results were compared with those obtained using random forest (RF) and a back propagation neural network (BPNN). The results show that, under the same feature variables, the PSO-SVM detection method gave an overall higher accuracy than the BPNN detection method, while the overall accuracy of the RF detection model was the lowest. The overall accuracy of the RF, BPNN and PSO-SVM detection models with wavelet features was higher by 3.7%, 2.9% and 8.3% compared to the corresponding methodological models with traditional spectral features. The detection model with wavelet features combining the PSO-SVM algorithm gave the highest overall accuracies (93.5%) and kappa coefficients (0.903) in the six monitoring models. These results suggest that the PSO-SVM algorithm combined with continuous wavelet analysis can significantly improve the accuracy of fusarium head blight detection on the wheat ears scale.

Wheat stripe rust has a severe impact on wheat yield and quality. An effective prediction method is necessary for food security. In this study, we extract the optimal vegetation indices (VIs) sensitive to stripe rust at different time-periods, and develop a wheat stripe rust prediction model with satellite images to realize the multi-temporal prediction. First, VIs related to stripe rust stress are extracted as candidate features for disease prediction from time series Sentinel-2 images. Then, the optimal VI combinations are selected using sequential forward selection (SFS). Finally, the occurrence of wheat stripe rust in different time-periods is predicted using the support vector machine (SVM) method. The results of the features selected demonstrate that, before the jointing period, the optimal VIs are related to the biomass, pigment, and moisture of wheat. After the jointing period, the red-edge VIs related to the crop health status play important roles. The overall accuracy and Kappa coefficient of the prediction model, which is based on SVM, is generally higher than those of the k-nearest neighbor (KNN) and back-propagation neural network (BPNN) methods. The SVM method is more suitable for time series predictions of wheat stripe rust. The model obtained accuracy based on the optimal VI combinations and the SVM increased over time; the highest accuracy was 86.2%. These results indicate that the prediction model can provide guidance and suggestions for early disease prevention of the study site, and the method combines time series Sentinel-2 images and the SVM, which can be used to predict wheat stripe rust.

Wheat stripe rust poses a serious threat to wheat production. An effective prediction method is important for food security. In this study, we developed a prediction model for wheat stripe rust based on vegetation indices and meteorological features. First, based on time-series Sentinel-2 remote sensing images and meteorological data, wheat phenology (jointing date) was estimated using the harmonic analysis of time-series combined with average cumulative temperature. Then, vegetation indices were extracted based on phenological information. Meteorological features were screened using correlation analysis combined with independent t-test analysis. Finally, a random forest (RF) was used to construct a prediction model for wheat stripe rust. The results showed that the RF model using the input combination (phenological information-based vegetation indices and meteorological features) produced a higher prediction accuracy and a kappa coefficient of 88.7% and 0.772, respectively. The prediction model using phenological information-based vegetation indices outperformed the prediction model using single-date image-based vegetation indices, and the overall accuracy improved from 62.9% to 78.4%. These results indicated that the method combining phenological information-based vegetation indices and meteorological features can be used for wheat stripe rust prediction. The results of the prediction model can provide guidance and suggestions for disease prevention in the study area.

Hyperspectral images (HSIs) have been widely used in many fields of application, but it is still extremely challenging to obtain higher classification accuracy, especially when facing a smaller number of training samples in practical applications. It is very time-consuming and laborious to acquire enough labeled samples. Consequently, an efficient hybrid dense network was proposed based on a dual-attention mechanism, due to limited training samples and unsatisfactory classification accuracy. The stacked autoencoder was first used to reduce the dimensions of HSIs. A hybrid dense network framework with two feature-extraction branches was then established in order to extract abundant spectral–spatial features from HSIs, based on the 3D and 2D convolutional neural network models. In addition, spatial attention and channel attention were jointly introduced in order to achieve selective learning of features derived from HSIs. The feature maps were further refined, and more important features could be retained. To improve computational efficiency and prevent the overfitting, the batch normalization layer and the dropout layer were adopted. The Indian Pines, Pavia University, and Salinas datasets were selected to evaluate the classification performance; 5%, 1%, and 1% of classes were randomly selected as training samples, respectively. In comparison with the REF-SVM, 3D-CNN, HybridSN, SSRN, and R-HybridSN, the overall accuracy of our proposed method could still reach 96.80%, 98.28%, and 98.85%, respectively. Our results show that this method can achieve a satisfactory classification performance even in the case of fewer training samples.