XVI International Botanical Congess

Neural networks are computing systems of highly-interconnected processing elements that can be trained to capture pattern recognition skills of experts. We are training specially-designed neural networks with venational features extracted from our high-resolution digital images of leaf specimens from the Wolfe collection at the Smithsonian Institution, numerous botanical gardens, herbaria and tropical specimens from the Amazon basin. Our system does not rely on dissection of flower or fruit material (often unavailable). Our goals are to 1) Replace plant keys (usually requiring a skilled botanist) with a completely automatic, portable system of digital camera and notebook computer that can be used by non-experts to identify plants in situ, eliminating the need to send specimens to be identified, 2) Create directly from plant material a 'virtual herbarium' image database that can be linked to other databases of medicinal and endangered plant species, agricultural weeds and toxic plants.