Development of an estimation system based on the relationship between plant genetic resource images and environmental information for the collection point

Abstract

Geographical information about collection points of plant genetic resources and photographs of plants and their habitats have been accumulated into a genetic resources database. Recently, climate data such as temperature and precipitation data have become available on the Internet. The characteristic shape and color of some plants and seeds are closely related to their habitat environment. Evaluation of the relationship between a plant’s image and its habitat environment as a feature quantity should make it possible to estimate the habitat of a plant from its morphological characteristics. The ability to estimate the habitat should increase exploration and plant collection efficiency. Deep learning was applied to the analysis of plant images. Then, a deep convolutional activation feature derived from the seed images by deep learning was used to develop an estimation system for classifying pairs of plant species. An estimation system for classifying the habitat climate from seed images was also developed.