基于协同学的深度协同神经网络研究与应用

发布时间：2019-01-07 17:48

【摘要】：深度学习是当今科学技术领域最热门的话题之一,也是人工智能领域最为成功和有效的思想方法,基于深度学习理论的研究和应用也层出不穷。然而随着对深度学习领域研究的深入,深度学习领域所暴露出来的计算成本过高和训练成本过高等问题也亟待解决。本文主要研究基于协同学原理构建的深度协同神经网络是一种全新的深度学习网络模型,能够有效的减小传统深度学习中计算成本过高和训练成本过高等问题。本文的主要内容为:首先,介绍了协同学理论的基本思想、数学模型和相关重要概念。接着介绍了基于协同学理论的一类全新的神经网络模型:协同神经网络。阐述了协同神经网络的数学模型、结构模型和运行流程,并介绍了协同神经网络的几种基本算法:基于PFR模型的分类器算法、基于PFAP的分类器算法、SCAP算法和SCAPAP算法。详尽的说明了协同学和协同神经网络的各项特性。其次,介绍了目前传统的几种深度神经网络模型:卷积神经网络和深度信念网络,并分别介绍了他们的模型结构和运行过程。以协同学原理和传统深度神经网络的模型结构为基础构建深度协同神经网络,并详细描述了深度协同神经网络的模型结构、运行过程和算法步骤。为后续深度协同神经网络的研究与应用提供了充分的理论支持。最后,基于上述深度协同神经网络的模型,设计了在不同样本库下深度协同神经网络的各项性能实验,同时也将深度协同神经网络与传统的卷积神经网络和深度信念网络在相同样本库下进行实验测试,纵向对比它们的各项性能特点。综合实验结果表明深度协同神经网络不仅在识别效果上有不错的表现,而且在计算成本和运行效率上有着较好的表现。基于上述深度协同神经网络的性能特点,本文还自主设计、搭建、调试了无人机硬件平台,将深度协同神经网络应用于无人机的自主降落过程中对可降落标志物识别的飞行控制系统。本文基于仿真数据和无人机自主降落标志的识别,验证了所提出的深度协同神经网络模型的有效性,为后续深入研究无人机自主降落、空中避障打下了基础。
[Abstract]:Deep learning is one of the most popular topics in the field of science and technology, and it is also the most successful and effective ideological method in the field of artificial intelligence. However, with the in-depth research in the field of in-depth learning, the problems of high computational cost and high training cost exposed in the field of in-depth learning need to be solved. In this paper, the deep cooperative neural network based on the principle of synergetic learning is a new kind of deep learning network model, which can effectively reduce the problems of high computational cost and too high training cost in traditional deep learning. The main contents of this paper are as follows: firstly, the basic idea, mathematical model and related important concepts of Synergetics theory are introduced. Then, a new neural network model based on synergetic theory is introduced. The mathematical model, structure model and running flow of cooperative neural network are introduced, and several basic algorithms of cooperative neural network are introduced: classifier algorithm based on PFR model, classifier algorithm based on PFAP, SCAP algorithm and SCAPAP algorithm. The characteristics of Synergetics and Synergetic Neural Network are explained in detail. Secondly, this paper introduces several traditional depth neural network models: convolution neural network and depth belief network, and introduces their model structure and running process respectively. Based on the synergetic principle and the model structure of the traditional depth neural network, the deep cooperative neural network is constructed, and the model structure, running process and algorithm steps of the deep cooperative neural network are described in detail. It provides sufficient theoretical support for the research and application of the following deep cooperative neural networks. Finally, based on the model of the deep synergetic neural network mentioned above, the performance experiments of the deep cooperative neural network under different sample databases are designed. At the same time, the deep cooperative neural network, the traditional convolutional neural network and the depth belief network are tested under the same sample base, and their performance characteristics are compared longitudinally. The experimental results show that the deep cooperative neural network not only has a good performance in recognition effect, but also has a good performance in computing cost and running efficiency. Based on the performance of the deep cooperative neural network, the hardware platform of UAV is designed, built and debugged independently. The depth cooperative neural network is applied to the flight control system of UAV to recognize landable markers during autonomous landing. Based on the simulation data and the recognition of UAV autonomous landing marks, the effectiveness of the proposed deep cooperative neural network model is verified in this paper, which lays a foundation for further research on UAV autonomous landing and air obstacle avoidance.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183

【相似文献】