Search

• A-BiGRU enables more accurate modeling of crop growth to achiever high accuracy. • The Attention mechanism can focus on the critical growth period of the crop. • The bidirectional GRU structure is better than GRU. • Image reconstruction can effectively fill the data gap. Timely and accurate access to regional scale crop plant area and spatial distribution is essential for regional agricultural production and food security, especially in the context of global population growth and climate change. Deep learning has become prevalent in crop mapping under complex conditions due to its powerful feature extraction and nonlinear ability. This study proposes a time-series image classification framework using Attention-based Bidirectional Gated Recurrent Unit (A-BiGRU) to map rice, maize, and soybean in Fujin, China, from reconstructed Sentinel-2 time-series images. Firstly, the reconstructed Sentinel-2 time-series images with 10 spectral dimensions and 22 temporal dimensions were obtained by linear interpolation and Savitzky-Golay (SG) filter. Then, a neural network, the A-BiGRU was developed to identify different crops by taking advantage of their unique growth patterns. The attention mechanism enables temporal neural networks to focus on the critical growth periods of crops. Additionally, the structure of GRU is simpler than that of long short-term memory (LSTM) and simple recurrent neural network (SRNN), which reduces the number of parameters and alleviates overfitting. Compared to GRU, BiGRU can fully uses the time-series information of the entire crop growth cycle. To assess the effectiveness of the proposed method, we compared two deep learning methods (LSTM and SRNN) and three widely used non-deep learning classifiers (Spectral Angle Mapping (SAM), Support Vector Machine (SVM)) and eXtreme Gradient Boosting (XGBoost). The results demonstrate that A-BiGRU achieved the highest accuracy, with an overall accuracy of 0.9804, a macro F1 score of 0.9788 and a kappa score of 0.9714. We also selected four typical regions and compared the classification results with optical images, which showed that the proposed method has a good recognition effect. Therefore, the A-BiGRU method is capable of achieving high-precision crop mapping. [ABSTRACT FROM AUTHOR]

• The MP-Net model aggregates context information. • The MP-Net model reduces the feature losses. • The MP-Net model efficiently and precisely classifies crops. Accurate crop classification map is of great significance in various fields such as the survey of agricultural resource, the analysis of existing circumstance on land application, the yield estimation of crop and the disaster warning. The methods based on machine learning and deep learning are popularly used in crop classification and recognition of remote sensing images. However, the crop classification task based on neural networks still faces significant challenges due to the spatial and temporal distribution of crops and the inherent characteristics of remote sensing images. Therefore, this study proposes the multi-layer pyramid crop classification network (MP-Net) to solve the above problems. To reduce the feature loss during the crop extraction, the proposed model uses the pyramid pooling module to improve the ability of global information acquisition, and the information concatenation module to retain the upper features. Using the GF-6 and Sentinel-2 satellite data, the proposed model was tested in Erhai Lake Basin and Beian City. Compared with other five deep learning models, such as FCN, SegNet, U-Net, PSPNet and DeepLabv3+, the experimental results indicate that the proposed model achieves the highest accuracy in both study areas. Meanwhile, the proposed model has the advantages of short training time and high efficiency under the same running conditions. Overall, this study is beneficial to improve the efficiency and accuracy of crop classification task in the unbalanced temporal and spatial distribution. It also brings a feasible scheme for crop classification tasks in complex growing areas. The code has been publicly available at https://github.com/Xu-Chang-Hong/MP-Net. [ABSTRACT FROM AUTHOR]

• Using hyperspectral data to retrieve wheat chlorophyll content. • Ridge regression algorithm and Gradient Boosting Regression Tree algorithm are used to establish the model. • The nonlinear inversion model obtained by the gradient boosting regression tree algorithm has higher accuracy than the linear model. Chlorophyll is the basic component of plant leaves. It is closely related to the use of solar radiation, the absorption of atmospheric carbon dioxide and the process of photosynthesis. The level of chlorophyll content has a very important indication of photosynthetic efficiency and development status of plants. Because hyperspectral data has the problem of large amount of data and high redundancy, this paper deals with hyperspectral data from two aspects to establish a prediction model of leaf chlorophyll content in different growth stages of winter wheat. On the one hand, this article does not perform dimensionality reduction processing on hyperspectral data. That is, for the full-band spectral data, the Ridge regression algorithm (abbreviated as Ridge) is used to establish a multiple linear prediction model. Use the Gradient Boosting Regression Tree algorithm (abbreviated as GBRT) and the feedforward neural network based on Back Propagation algorithm (abbreviated as BP) to establish two multiple nonlinear prediction models. On the other hand, dimensionality reduction is performed on hyperspectral data. Use Elastic Net algorithm (abbreviated as EN) to reduce the dimensionality of hyperspectral data to extract sensitive bands, and then use GBRT algorithm and feedforward neural network based on Back Propagation algorithm to establish multivariate nonlinear prediction models (abbreviated as EN-GBRT, EN-BP). After the model accuracy test, the conclusions are as follows: (1) The nonlinear prediction model obtained by the gradient boosting regression tree algorithm has higher accuracy and universality than the linear model in each growth period of winter wheat (R2 GBRT ＞R2 Ridge). (2) The EN-GBRT prediction model has greater advantages in predicting the chlorophyll content of winter wheat at each growth stage (R2 EN-GBRT ＞R2 EN-BP) within two nonlinear prediction models. (3) Comprehensive comparison of the prediction accuracy of GBRT and EN-GBRT models in each growth period. In the jointing period, heading period, flowering period, filling period, milking period and maturity period of winter wheat, the accuracy of prediction using the GBRT model was higher (R2 jointing = 0.81, R2 heading = 0.50, R2 flowering = 0.74, R2 filling = 0.41, R2 milking = 0.84, R2 maturity = 0.95). In the whole growth period of winter wheat, the prediction accuracy of EN-GBRT model was better (R2 whole = 0.85). [ABSTRACT FROM AUTHOR]

• Using hyperspectral data to invert the nitrogen content of different gradients in maize leaves. • Perform dimensionality reduction processing on information-rich hyperspectral data to reduce data redundancy and avoid over-fitting. • The elastic net algorithm is selected for dimensionality reduction, and the convex combination of L1 norm and L2 norm is used as the penalty term. • Nine inversion models were established, among which the best model was EN-PLSR, R2 = 0.96, RMSE = 0.19. Fast, accurate, and non-destructive detection of the nitrogen (N) content in corn leaves is of great significance for the precise dynamic management of nitrogen fertilizer application for corn. Hyperspectral data can provide an important means for detecting the nitrogen content in plants. Existing research has mainly focused on using various vegetation indices or 3–5 band combinations to estimate leaf nitrogen content, ignoring the different in spectral characteristics of hyperspectral data and failing to characterize most of the spectral information. Some scholars have used principal component analysis and wavelet analysis dimensionality reduction algorithms, but used different bands for these models. Therefore, more and different inversion models need to be introduced to improve the use of spectral data and increase the universality of the model. The present study selected three different methods to reduce data dimensionality, including the Successful Projections Algorithm (SPA) and the Least Absolute Shrinkage and Selection Operator (LASSO) and the Elastic Net (EN) algorithms. Then the processed spectral reflectance information and observational data for synchronous leaf nitrogen content were used to construct an inversion model used to predict leaf nitrogen content. Nine inversion models were constructed based on different dimensionality reduction and regression methods. Based on the coefficient of determination (R 2) and root mean square error (RMSE), the accuracy of each model was tested. The main results follow: (1) Dimensionality reduction processing of hyperspectral data can effectively prevent data from overfitting, limit the correlation between adjacent frequency bands, and reduce data redundancy. An EN dimensionality reduction algorithm (EN-Partial Least Squares Regression (PLSR)) model R 2 = 0.96, RMSE = 0.19) was better than a SPA (SPA-PLSR model R 2 = 0.90, RMSE = 0.26) and LASSO (LASSO-PLSR model R 2 = 0.89, RMSE = 0.37) dimensionality reduction algorithm. (2) For the same dimensionality reduction method, the accuracy of the regression model based on PLSR was higher than that of other models. Among the nine inversion models in this paper, the EN-PLSR inversion model has the best fitting effect (R 2 = 0.96, RMSE = 0.19). (3) Obvious changes in nitrogen content have little effect on the overall hyperspectral reflectance curve. This study provides a reference for high-efficiency and non-destructive testing of corn nitrogen content using hyperspectral technology. [ABSTRACT FROM AUTHOR]