Wild plant data collection system based on distributed location

Abstract

Building accurate knowledge of the identity, the geographic distribution and the evolution of living species is very essential for sustainable development of the biodiversity as well as the whole society. However, basic information is often partially accessible for scientists to do research, plant data collection is one of the problems. To make plants collection become more available especially in the field environment with weak network signal, this paper proposed a novel method that combines plant recognition with distributed location information and designed a corresponding method based on Bayesian estimation to analyze plants with sick disease and unknown parts. A mobile application system has been designed and implemented which uploads plant photos to the cloud server for recognition based on the established plant library and speed up the process with the distributed location information. Meanwhile, to solve weak signal problem in the wild field, the library buffer of neighboring area where recent searching items exist are proposed in the mobile client be therefore that the application can finish matching locally and reduce the network bandwidth requirement. To illustrate the availability and practicability, different set of plants with leaves or flowers have been collected and the results show that the average recognizing accuracy via traditional method is larger than 84% and recognition time is less than 1.5s, and the accuracy will soar to more than 90% if neural network is used.