卷积神经网络在植被识别中的应用研究
Application of Convolutional Neural Network in Vegetation Recognition

DOI: 10.12677/CSA.2019.95094, PP. 841-848

Keywords: 深度学习，卷积神经网络，潘安湖湿地公园
Deep Learning, Convolutional Neural Network, Pan’an Lake Wetland Park

Full-Text   Cite this paper   Add to My Lib

Abstract:

大多数植物图片识别的方法，都是聚焦与植物图片的某一特征进行识别，例如叶径，叶长，花，果实，叶片。使用其植物某个器官进行识别，这样得出的结果并不可靠，因为实际自然界有非常多的植物有着极其相似的特征。本文通过选取整个植物图片作为训练样本，即提取植物的所有特征，基于卷积神经网络(Convolutional Neural Networks, CNN)中的AlexNet模型，利用GPU并行计算能力加快模型训练和图片识别速度。通过对潘安湖的5类植物数据集进行训练，训练得到正确精度为87.5%的模型，并且将此训练精度与最近邻(K-NearestNeighbor, KNN)和BP神经网络(Back Propagation, BP)两种分类算法训练得到的训练精度作比较，验证了模型的高可用性。以此模型为基础，应用Python开发了一款基于潘安湖湿地公园植物的植物APP识别软件。
Most plant image recognition methods focus on one feature of plant image, such as leaf diameter, leaf length, flower, fruit and leaf. Using one of its plant organs to identify, the result is not reliable, because there are many plants in nature with very similar characteristics. In this paper, the whole plant image is selected as training sample that is to extract all plant features. Based on AlexNet model of Convolutional Neural Networks (CNN), the parallel computing ability of GPU is used to speed up model training and image recognition. Through training the data sets of five kinds of plants in Pan’an Lake, the correct accuracy of the model is 87.5%. The training accuracy is com-pared with that of KNN nearest neighbor and BP neural network, which verifies the high availability of the model. Based on this model, a plant identification software named APP was developed by Python, which is based on the plant of Pan’an Lake Wetland Park.

References

[1]	杨瑞卿, 王千千, 徐德兰. 徐州潘安湖湿地公园植物多样性调查与分析[J]. 西北林学院学报, 2018, 33(3): 285-289.
[2]	尤肖虎, 潘志文, 高西奇, 等. 5G移动通信发展趋势与若干关键技术[J]. 中国科学: 信息科学, 2014, 44(5): 551-563
[3]	黄婕, 李浩锐, 黄皓琪, 胡小珍, 刘馨欣, 高凤连, 叶志鹏. 基于尺度不变特征变换算法的植物自动识别系统[J].计算机应用, 2016, 36(z2): 203-205.
[4]	秦丰, 刘东霞, 孙炳达, 阮柳, 马占鸿, 王海光. 基于深度学习和支持向量机的4种苜蓿叶部病害图像识别[J]. 中国农业大学学报, 2017, 22(7): 123-133.
[5]	Krizhevsky, A., Sutskever, I. and Hinton, G. (2012) ImageNet Classification with Deep Convolutional Neural Networks. Advances in Neural Information Processing Systems, 25.
[6]	Simonyan, K. and Zisserman, A. (2014) Very Deep Convolutional Networks for Large-Scale Image Recognition. Computer Science, arXiv: 1409.1556.
[7]	Szegedy, C., Liu, N.-W., Jia, N.-Y., et al. (2015) Going Deeper with Convolutions. 2015 IEEE Con-ference on Computer Vision and Pattern Recognition (CVPR). Boston, MA, 2015, 1-9. https://doi.org/10.1109/CVPR.2015.7298594
[8]	Targ, S., Almeida, D. and Lyman, K. (2016) Resnet in Resnet: Generalizing Residual Architectures. Computer Science, arXiv: 1603.08029.
[9]	Canziani, A., Paszke, A. and Culur-ciello, E. (2016) An Analysis of Deep Neural Network Models for Practical Applications. Computer Science, arXiv: 1605.07678.
[10]	黄杰贤, 刘燕, 杨冬涛. 基于BP神经网络的柚子分类研究[J]. 湖北农业科学, 2018, 57(24): 112-115.
[11]	刘应东, 牛惠民. 基于k-最近邻图的小样本KNN分类算法[J]. 计算机工程, 2011, 37(9): 198-200.

Full-Text