Files
Abstract
Recent advances in computer vision, most notably deep convolutional neural networks (CNNs), are exploited to identify and localize various plant species in salt marsh images. Three distinct approaches are explored that provide estimations of abundance and spatial distribution at varying levels of granularity in terms of spatial resolution. Overall, a clear trade-off is observed between the CNN estimation quality and the spatial resolution of the underlying estimation thereby offering guidance for ecological applications of CNN-based approaches to automated plant identification and localization in salt marsh images. A novel way to train neural networks for semantic image segmentation, termed as Compositional Sparse Network (CSN), is also conceptualized and tested. By leveraging the properties of dynamic expansion, interconnection richness, and sparsity, a CSN is used as the backbone for the DeepLab-V3 architecture. Since CSN is analogous to Neural Architecture Search (NAS), it is also compared to a NAS-based semantic image segmentation approach.