基于Inception网络的花卉分类(论文11000字)
摘要：现如今,随着社会的进步和科学技术的蓬勃发展,机器学习[1]结合图像的分类与识别已经广泛地应用到了日常生活中的很多方面。目前花卉识别的研究工作相对较少,在过去的实验结果中,识别准确率都不高,需要对算法进一步改进。由于花卉图像的类间的差异较大，所以花卉的分类是困难的。传统的花卉分类方法和常见的卷积神经网络是难以提取花卉内在信息的，因此分类实验的准确率并不理想。优化的卷积神经网络[2]能够更好的提取图片的内在特征和底层信息，并且它是一个极好的图像分类方法，深层次的卷积神经网络可以在提取图像特征时减少错误。它可以直接使用图像作为输入对象，自动提取花卉特征。并且它具有在各种外部因素条件下良好的鲁棒性。因此，深层次的卷积神经网络在图像分类识别有明显的优势。但是普通的神经网络又有其局限性,需要对其进行改进,以提高识别准确率,本文对卷积神经网络的inception 网络进行改进,能够得到比传统神经网络更高的分类识别准确率。
关键字：花卉分类；卷积神经网络；Inception 网络；深度学习
Flower Classification based on inception Network
Abstract:Nowadays, with the progress of society and the vigorous development of science and technology, machine learning combined with image classification and recognition has been widely used in many aspects of daily life. At present, the research work of flower recognition is relatively few. In the past experimental results, the recognition accuracy is not high, so the algorithm needs to be further improved. Because there are great differences between the classes of flower images, the classification of flowers is difficult. The traditional flower classification method and the common convolution neural network are difficult to extract the inherent information of flowers, so the accuracy of classification experiment is not ideal. Convolution neural network can better extract the inherent features and bottom of the picture. Layer information, and it is an excellent image classification method, convolution neural network can reduce errors when extracting image features. It can directly use the image as the input object to automatically extract flower features. And it has good robustness in various external factors. Therefore, convolution neural network has obvious advantages in image classification and recognition. However, the ordinary neural network has its limitations, so it needs to be improved in order to improve the recognition accuracy. In this paper, the inception network of convolution neural network can obtain higher classification and recognition accuracy than the traditional neural network.
Key words:Flower classification; convolution neural network; Inception network; deep learning
目录
1绪论 1
1.1 研究背景和意义 1
1.2国内外研究状况 1
1.3 花卉分类识别技术的发展 2
1.4 本论文研究内容 2
1.5 本论文的结构安排 2
2 花卉分类算法概述 2
2.1 传统花卉分类 2
2.2花卉分类传统方法 3
2.3 传统花卉分类的缺点 3
2.4 基于机器学习的花卉分类 3
2.5 本章小结 4
3 Inception网络花卉分类方案 4
3.1 网络优化 4
3.2 模型训练 4
3.3 花卉分类流程 5
4 网络优化的设计与实现 6
4.1 网络概述 6
4.2 Inception网络的基本组成 6
4.3 Inception 网络模型 9
4.4 激活函数的优化 10
4.5 损失函数的优选 13
4.6 本章小结 14
5实验结果与分析 14
5.1 花卉数据集的来源 14
5.2 开发工具和涉及 15
5.3实验结果与分析 15
6总结与展望 20
参考文献 20
致 谢 22 |
