For centuries, herbarium specimens have been essential in documenting plant occurrences and studying (historical) biodiversity and environmental impacts. The digitization and online availability of herbaria have led to a growing need for automated tools to enrich these collections. Studying herbarium sheets involves significant manual labor, as they typically contain no morphological information about the specimens and their organs. Current herbarium processing methods are either not granular enough, inaccurate, or only work for a limited number of species. We aim to solve these issues by developing robust computer vision methods to process and analyze herbarium specimens. We will leverage the millions of publicly available herbarium images paired with taxonomic information to develop a novel vision transformer that generalizes to a broad range of species. We will use this model to fully segment herbarium sheets, producing specimen and plant organ masks (leaves, fruits, etc.). From these segments, we will extract morphological features that are crucial for understanding ecological contexts. Our existing expertise in this field forms the foundation for this research and we will use the novel digital specimen standards proposed in DiSSCo and TDWG. By applying our methods to millions of publicly available herbarium sheets and integrating the results into these repositories, we enable researchers to efficiently study historical biodiversity and environmental conditions.