一种基于卷积神经网络的草地植物识别方法

Grassland Plant Identification Method Based on Convolutional Neural Network

摘要: 现有对草地植物的图像识别主要集中于对叶片或大面积种群的识别，很少有从单株植物或小片群落的角度进行识别。本文针对上述问题，总结出三种适用于解决该问题的识别方法，改进和微调了现有基于卷积神经网络方法的预处理流程和网络模型来进行植物图像识别方法。本文采用近距离拍摄的高空间分辨率草地植物图片，设计实验对比分析了上述三种方法在识别标注样本数据集上的表现。实验结果表明，基于预训练模型的深度卷积神经网络方法同其他方法相比，在准确性上，具有显著的优越性。

Abstract: The existing image recognition of grassland plants is mainly focused on the identification of leaves or large area populations, and few of them are identified from the angle of single plant or small community. In this paper, three identification methods for solving the problem are summarized, and the process of preprocessing and network model based on the existing convolution neural network methods are improved and adjusted to carry out the method of plant image recognition. In this paper, the high spatial resolution grassland plant pictures are taken near the distance, and the performance of the above three methods on the identified annotation sample data sets is compared and analyzed. The experimental results show that the method of deep convolution neural network based on pre-training model is superior to other methods in identifying the accuracy of the data set of the labeled sample.