空间海量异构遥感数据快速筛选和智能技术

Rapid Screening of Massive Heterogeneous Remote SensingData and Intelligent Service Technology

针对海量异构遥感数据中目标物快速筛选问题,提出了一种基于深度学习框架的遥感数据快速筛选和智能技术。使用地面系统与图像金字塔联合处理多源异构数据集,目视选定对遥感数据的背景去噪和特征增强操作方式,便于输入三通道卷积神经网络进行训练、学习以及多源数据融合;筛选出处理效率及准确度最高的训练模型,提高数据融合效果与数据处理模型泛化能力;同时使用网络完成对载荷获取的海量异构数据进行特征提取,并利用错误识别结果更新特征库;最后针对连续多帧的点目标数据集进行实验。结果表明:该技术可实现目标物快速准确识别目的。

To solve the problem of fast screening of objects in massive heterogeneous remote sensing data,a deep learning framework based fast screening of remote sensing data and intelligent service technology were proposed.The ground system and the image pyramid were used to jointly process the multi-source heterogeneous data sets.The background denoising of the remote sensing data and the characteristics enhancement operation mode were selected visually to facilitate the training and learning of the three-channel convolution neural network,and the fusion of the multi-source data.The training model with the highest processing efficiency and accuracy was selected to improve the performance of data fusion and the generalization ability of the data processing model.At the same time,the network was used to complete the feature extraction of the massive heterogeneous data obtained from the payload,and the feature library was updated with the error recognition results.Finally,the experiment was carried out on the point target data set of continuous multiple frames,and the results showed that the technology could achieve the goal of fast and accurate target recognition.