An Android Application for Plant Identification

Authors:

Abstract
This paper presents an Android application to automatically identify plant species using a single leaf image as input. At the pre-processing phase, we proposed an improved segmentation method to eliminate the noise caused by capturing on non-uniform background so we can obtain the binary image which only contains the leaf shape. Then, several morphological features and Hu moment invariants descriptors were extracted as inputs of a joint classifier which combines the back propagation neural network(BPNN) with a weighted k-nearest-neighbor (KNN) to distinguish 220 species of plants. The outputs of the joint classifier are the top ten species that best match the query leaf image. At the end, we implemented these algorithms on Android OS and the application we developed has been downloaded about a million times.

This publication has 15 references indexed in Scilit: