Exploring deep neural networks for large wood detection in rivers: A novel approach

In recent years, there has been growing interest in using machine learning and deep learning algorithms to detect large wood (LW) in rivers. In this study, we employed the convolutional neural network (CNN) U-net for LW detection. We generated 774 RGB image labeled as LW and non-LW class for training the model. Model performance was evaluated using precision, recall, and F1 score metrics. Our results showed that the model successfully detected LW, achieving an F1 score of 0.75 and a general accuracy of 0.99 for both classes. The initial performance of the model can be attributed to its ability to learn patterns and features from high-dimensional data. Nevertheless, we only incorporated orthomosaic input variables for model training. The addition of more data can improve algorithm performance. Our study highlights the potential of CNNs to detect LW in rivers using remotely sensed data.