Automatic Gully Detection: Neural Networks and Computer Vision

The Evaluation on the Credit Risk of Enterprises with the CNN-LSTM-ATT Model

Credit evaluation is a difficult problem in the process of financing and loan for small and medium-sized enterprises. Due to the high dimension and nonlinearity of enterprise behavior data, traditional logistic regression (LR), random forest (RF), and other methods, when the feature space is very large, it is easy to show low accuracy and lack of robustness. However, recurrent neural network (RNN) will have a serious gradient disappearance problem under long sequence training. This paper proposes a compound neural network model based on the attention mechanism to meet the needs of enterprise credit evaluation. The convolutional neural network (CNN) and the long short-term memory (LSTM) network were used to establish the model, using soft attention, the gradient propagates back to other parts of the model through the attention mechanism module. In the multimodel comparison experiment and three different enterprise data experiments, the CNN-LSTM-ATT model proposed in this paper is superior to the traditional models LR, RF, CNN, LSTM, and CNN-LSTM in most cases. The experimental results under multimodel comparison reflect the higher accuracy of the model, and the group test reflects the higher robustness of the model.

Mapping of Rill Erosion of the Middle Volga (Russia) Region Using Deep Neural Network

Soil erosion worldwide is an intense, poorly controlled process. In many respects, this is a consequence of the lack of up-to-date high-resolution erosion maps. All over the world, the problem of insufficient information is solved in different ways, mainly on a point-by-point basis, within local areas. Extrapolation of the results obtained locally to a more extensive territory produces inevitable uncertainties and errors. For the anthropogenic-developed part of Russia, this problem is especially urgent because the assessment of the intensity of erosion processes, even with the use of erosion models, does not reach the necessary scale due to the lack of all the required global large-scale remote sensing data and the complexity of considering regional features of erosion processes over such vast areas. This study aims to propose a new methodology for large-scale automated mapping of rill erosion networks based on Sentinel-2 data. A LinkNet deep neural network with a DenseNet encoder was used to solve the problem of automated rill erosion mapping. The recognition results for the study area of more than 345,000 sq. km were summarized to a grid of 3037 basins and analyzed to assess the relationship with the main natural-anthropogenic factors. Generalized additive models (GAM) were used to model the dependency of rill erosion density to explore complex relationships. A complex nonlinear relationship between erosion processes and topographic, meteorological, geomorphological, and anthropogenic factors was shown.

Classification Efficacy Using K-Fold Cross-Validation and Bootstrapping Resampling Techniques on the Example of Mapping Complex Gully Systems

The availability of aerial and satellite imageries has greatly reduced the costs and time associated with gully mapping, especially in remote locations. Regardless, accurate identification of gullies from satellite images remains an open issue despite the amount of literature addressing this problem. The main objective of this work was to investigate the performance of support vector machines (SVM) and random forest (RF) algorithms in extracting gullies based on two resampling methods: bootstrapping and k-fold cross-validation (CV). In order to achieve this objective, we used PlanetScope data, acquired during the wet and dry seasons. Using the Normalized Difference Vegetation Index (NDVI) and multispectral bands, we also explored the potential of the PlanetScope image in discriminating gullies from the surrounding land cover. Results revealed that gullies had significantly different (p < 0.001) spectral profiles from any other land cover class regarding all bands of the PlanetScope image, both in the wet and dry seasons. However, NDVI was not efficient in gully discrimination. Based on the overall accuracies, RF’s performance was better with CV, particularly in the dry season, where its performance was up to 4% better than the SVM’s. Nevertheless, class level metrics (omission error: 11.8%; commission error: 19%) showed that SVM combined with CV was more successful in gully extraction in the wet season. On the contrary, RF combined with bootstrapping had relatively low omission (16.4%) and commission errors (10.4%), making it the most efficient algorithm in the dry season. The estimated gully area was 88 ± 14.4 ha in the dry season and 57.2 ± 18.8 ha in the wet season. Based on the standard error (8.2 ha), the wet season was more appropriate in gully identification than the dry season, which had a slightly higher standard error (8.6 ha). For the first time, this study sheds light on the influence of these resampling techniques on the accuracy of satellite-based gully mapping. More importantly, this study provides the basis for further investigations into the accuracy of such resampling techniques, especially when using different satellite images other than the PlanetScope data.

A Methodological Comparison of Three Models for Gully Erosion Susceptibility Mapping in the Rural Municipality of El Faid (Morocco)

Erosion is the main threat to sustainable water and soil management in Morocco. Located in the Souss-Massa watershed, the rural municipality of El Faid remains an area where gully erosion is a major factor involved in soil degradation and flooding. The aim of this study is to predict the spatial distribution of gully erosion at the scale of this municipality and to evaluate the predictive capacity of three prediction methods (frequency ratio (FR), logistic regression (LR), and random forest (RF)) for the characterization of gullying vulnerability. Twelve predisposing factors underlying gully formation were considered and mapped (elevation, slope, aspect, plane curvature, slope length (SL), stream power index (SPI), composite topographic index (CTI), land use, topographic wetness index (TWI), normalized difference vegetation index (NDVI), lithology, and vegetation cover (C factor). Furthermore, 894 gullies were digitized using high-resolution imagery. Seventy-five percent of the gullies were randomly selected and used as a training dataset, whereas the remaining 25% were used for validation purposes. The prediction accuracy was evaluated using area under the curve (AUC). Results showed that the factor that most contributed to the prevalence of gullying was topographic (slope, CTI, LS). Furthermore, the fitted models revealed that the RF model had a better prediction quality, with the best AUC (91.49%). The produced maps represent a valuable tool for sustainable management, land conservation, and protecting human lives against natural hazards (floods).