Advancements in remote sensing for invasive plant mapping along the Guadiana River: The role of CNN2D

Abstract

Remote sensing images are crucial for finding spatial and temporal patterns in vegetation. The Guadiana river (located in South West Spain) has been invaded since the early 2000s by floating aquatic plants with rapid reproductive capacity, causing negative effects to the ecosystem. In this study, we advance our previous work by employing a two-dimensional convolutional neural network (CNN2D) which integrates both spectral and spatial information of each pixel. This approach marks a significant improvement over our prior use of one-dimensional CNN (CNN1D) that relied solely on spectral data. The performance of the CNN2D in detecting invasive aquatic plants is demonstrated to be superior, particularly due to its ability to analyze spectral-spatial features. Our experimental results highlight the enhanced efficacy in invasive plant detection when spatial context is considered alongside spectral information.