基于卷积神经网络的遥感图像分类

基于卷积神经网络的遥感图像分类(论文16000字)
摘要：随着计算机技术的发展，图像处理的技术也不断得到提高，从最初的传统图像处理技术，到现在的基于深度学习的图像处理技术。其中，图像分类是图像处理的基础，在各种领域有着广泛的应用。从2012年开始，深度学习成为机器学习中的热门方向，受到了广泛的关注，基于深度学习的图像处理也比传统方法更加精确，也更加方便，将会成为未来图像处理的主流技术。深度学习中的卷积神经网络自提出之后，也得到了很好的发展，本文结合遥感图像场景，对基于卷积神经网络的遥感图像分类进行研究，主要介绍了卷积神经网络的组成、经典模型和经常使用的平台框架；在本文图像分类研究上，基于VGG模型， 将VGG模型进行改动，用于遥感图像分类，最终的训练结果总体精度达到97%，实验过程中，出现了拟合不佳的情况，通过添加Dropout层，优化拟合情况。也基于宽度学习对图像进行了处理，并与深度学习做了简单的比较。
深度学习可以建立复杂的模型结构，得到更加适合图像分类的模型，由于时间和设备的限制，本文未能实现更优的模型结构。总体来说，深度学习还能取得更好的发展，在各个领域得到更好的应用。
关键词：卷积神经网络；遥感图像分类；深度学习；宽度学习；Keras。 [资料来源：http://www.THINK58.com]

Remote Sensing Image Classification Based on Convolutional Neural Network
ABSTRACT： With the development of computer technology, image processing technology has been improved constantly, from the original traditional image processing technology to the image processing technology based on deep learning. Among them, image classification is the basis of image processing, which has been widely used in various field. Since 2012, deep learning has become a popular direction in machine learning, and attracted wide attention. Image processing based on deep learning is more accurate and convenient than traditional methods. It will become the mainstream technology of image processing in the future. The convolution neural network in deep learning has been developed wellsince it was put forward, and this paper studies the classification of remote sensing images based on convolution neural network in combination with remote sensing image scene. The composition, classical model and commonly used platform framework of convolution neural network are mainly introduced. In this paper, based on VGG model, the VGG model is modified. For remote sensing image classification, the overall accuracy of the final training results reached 97%. In the experiment process, there was a poor fitting situation. Dropout layer was added to optimize the fitting situation.The image is also processed based on broad learning, and a simple comparison with deep learning is made.

Deep learning can build complex model structure and get more suitable model for image classification. Due to the limitation of time and equipment, this paper can not achieve better model structure.Generally speaking, deep learning can also achieve better development and better application in various fields.
Key Word：CNN； remote sensingclassification；deep learning；broadlearning；Keras.
 

[资料来源：http://THINK58.com]

目录
第一章 绪论    1
1.1选题的意义和目的    1
1.2课题研究现状    1
1.3 遥感图像分类的方法    2
1.4论文组织结构    3
第二章 卷积神经网络的理论    4
2.1 深度学习概述    4
2.2卷积神经网络    4
2.2.1 卷积神经网络的概述    4
2.2.2卷积神经网络的结构    5
2.2.3经典模型    11
2.2.4卷积神经网络的平台和工具    12
2.3本章小结    12
第三章 基于卷积神经网络的遥感图像分类    13
3.1概述    13
3.2实验环境与数据集    13
3.3实验过程及实验结果分析    14

3.3.1模型的搭建    14
3.3.2实验过程及结果分析    16
3.4本章小结    20
第四章 基于broad learning的图像分类    22
4.1 概述    22
4.2实验内容    22
4.2.1实验环境与数据集    22
4.2.2 实验过程    22
4.2.3结果分析    26
4.3 本章小结    26
第五章 总结与展望    27
5.1本文总结    27
5.2后续展望    27
参考文献    28
致谢    31 [资料来源：http://think58.com]