Study On Features Of Plant Leaf Image Recognition And Realization Of Online Recognition System

The most common organisms on Earth are plants, which are crucial to the maintenance of atmospheric composition, nutrient cycling and other ecosystem processes. Plant classification is one of the foundations of plant research, it is very important in many areas of plant research.This paper begins with the discussion of the possibility of leaf image classification, and introduces some leaf image classification methods. Then we proposed a new approach for plant leaf classification, which treats histogram of oriented gradients (HOG) as a new representation of shape, and uses the Maximum Margin Criterion (MMC) for dimensionality reduction. We named this method as HOG-MMC.The proposed method was tested on two different datasets. The first is Swedish leaf dataset which contains isolated leaves from 15 different Swedish tree species, with 75 leaves per species. The second dataset is ICL leaf dataset, where all the plants were collected from Botanical Garden in Hefei, the capital city of Anhui Province of China by Intelligent Computing Laboratory (ICL), Chinese Academy of Sciences, China. We compared this algorithm with a classic shape classification method named Inner-Distance Shape Context (IDSC) and other traditional leaf classification methods on Swedish leaf dataset and ICL dataset. The proposed method achieves better performance compared with IDSC. Especially on ICL_B subset which has little variety between different leaf classes, HOG-MMC gets better recognition accuracy rate than the other methods by more than 10%.Finally, two online leaf image classification systems are realized based on PC and ARM.