| 基于分布式卷积神经网络的遥感大数据分类方法 |
 点此下载全文 |
| 引用本文：王伟晨1 ，王瑾婷2.基于分布式卷积神经网络的遥感大数据分类方法[J].计算技术与自动化,2026,(2):141-148 |
| 摘要点击次数: 37 |
| 全文下载次数: 0 |
|
|
| 中文摘要:针对遥感大数据分类过程中分类准确率低、效率低、实时性差的问题，提出了一种基于分布式卷积神经网络的遥感大数据分类方法。设计了一种基于FPGA的分布式卷积神经网络架构，用于模型训练，提高计算效率；建立了并行多尺度特征融合分布式卷积神经网络模型，增加并行多尺度融合特征模块，以提高分类准确率，利用模型对遥感大数据进行分类；设计基于区块链的在线增量学习框架，动态更新数据，提高模型实时性。实验证明该方法的遥感大数据分类准确率为96.57%，相比于基于卷积神经网络的分类方法，分类准确率提高14.43%，验证其在遥感大数据分类方面具有有效性、优越性。 |
| 中文关键词:分布式卷积神经网络 遥感大数据分类 并行多尺度融合特征 基于区块链的在线增量学习 |
| |
| Remote Sensing Big Data Classification Method Based on Distributed Convolutional Neural Network |
|
|
| Abstract:To address the classification of remote sensing big data in the process of low classification accuracy, low speed and low reliability, this paper proposes a remote sensing big data classification method which is based on a distributed contextual neural network(DCNN). An FPGA-based distributed convolutional neural network architecture is designed for model training to improve the computational efficiency, and a parallel multiscale feature fusion distributed convolutional neural networks (PMFFDCNN) model is built, adding parallel multiscale fusion feature module to improve the classification accuracy, and utilizing the model to classify remote sensing big data. Designing blockchain-based online incremental learning framework to dynamically update the data and increase the model's real-time performance. Experiments prove that: the classification accuracy of remote sensing big data of this paper's method is 96.57%, which is 14.43% higher than that of the classification method based on consecutive neural nets, and this paper's method has the effectiveness and superiority in the classification of remote sensing big data. |
| keywords:distributed convolutional neural network remote sensing big data classification parallel multiscale fusion features online incremental learning based on blockchain |
| 查看全文 查看/发表评论 下载pdf阅读器 |
