森林工程2018，Vol.34Issue(1)：56-59,4.
基于卷积神经网络和树叶纹理的树种识别研究
Study on Tree Species Identification Based on Convolution Neural Network and Leaf Texture Image
摘要
Abstract
This experiment uses the computer image engineering technology and convolution neural network to extract and classify the leaf image information .The major research direction of this study is the vein texture characteristics of broad-leaved leaves in a sin-gle background .In this experiment , the convolution neural network is adopted .This algorithm has high accuracy in image information recognition , without manual definition of leaf texture features , which can reduce labor cost .The network is defined by Caffe platform and is initialized by a set of random parameters which conform to the standard normal distribution .The BP neural network back propa-gation algorithm and the random gradient are used to reduce the algorithm training for convolution neural networks .Through the train-ing of the network , the final recognition rate is above 95％, which exceeds the recognition rate of the human eye .It provides an effec-tive classification method for tree species identification research .The convolution neural network eliminates the work of manually defi-ning the features of leaves , and the final recognition effect is basically up to the standard of applicable .关键词
卷积神经网络/树种识别/树叶纹理/图像工程/阔叶Key words
Convolution neural network/tree species identification/leaf texture/image engineering/broad leaf分类
农业科技引用本文复制引用
赵鹏超,戚大伟..基于卷积神经网络和树叶纹理的树种识别研究[J].森林工程,2018,34(1):56-59,4.基金项目
国家自然科学基金项目(31570712) （31570712）
黑龙江省教育厅科学技术研究项目(12543019) （12543019）
黑龙江省自然科学基金项目(C201338) （C201338）
高校科研基金项目(2572014CB30) （2572014CB30）
中央高校基本科研业务费专项资金资助项目(2572016AB26) （2572016AB26）
