Extraction of leaf vein features based on artificial neural network - studies on the living plant identification Ⅰ

Abstract

葉片的識別是識別植物的重要組成部分,特別在野外識別植物活體尤其重要。葉脈的脈序是植物的內在特征,包含有重要的遺傳信息。但由于葉脈本身的多樣性,利用單一特征的圖像處理方法難以有效地提取葉脈。為了充分利用圖像的信息,本文提出了一種基于人工神經網絡的葉脈提取方法。該方法利用邊緣梯度、局部對比度和鄰域統計特征等10個參數來描述像素的鄰域特征,并將其作為神經網絡的輸入層。實驗結果表明,與傳統方法相比,經過訓練的神經網絡能夠更準確地提取葉脈圖像,為進一步的葉片識別打下了良好的基礎。 Leaf recognition is an important step for plant computerized identification, especially for fieldliving plants. Previous researches were mainly focused on leaf recognition by utilizing the peripheralcontour of the leaf while ignoring the leaf venation that actually contains important genetic information.Conventional thresholding-based methods cannot extract the information accurately due to high diversityof leaf veins. In this paper, an approach based on artificial neural network learning is proposed to extractleaf venation. Ten features including edge gradients, local contrast and statistical features are extractedfrom a window centered at the image pixel and used to train a neural network classifier. Compared withconventional thresholding-based methods, the trained neural network is capable of extracting more accu-rate modality of leaf venation for subsequent leaf recognition.