WO2021226977A1 - 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台 - Google Patents

一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台 Download PDF

Info

Publication number
WO2021226977A1
WO2021226977A1 PCT/CN2020/090397 CN2020090397W WO2021226977A1 WO 2021226977 A1 WO2021226977 A1 WO 2021226977A1 CN 2020090397 W CN2020090397 W CN 2020090397W WO 2021226977 A1 WO2021226977 A1 WO 2021226977A1
Authority
WO
WIPO (PCT)
Prior art keywords
remote sensing
image
neural network
interpretation
mining
Prior art date
Application number
PCT/CN2020/090397
Other languages
English (en)
French (fr)
Inventor
张炜
Original Assignee
安徽中科智能感知产业技术研究院有限责任公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 安徽中科智能感知产业技术研究院有限责任公司 filed Critical 安徽中科智能感知产业技术研究院有限责任公司
Priority to PCT/CN2020/090397 priority Critical patent/WO2021226977A1/zh
Priority to CN202080000930.0A priority patent/CN111742329B/zh
Publication of WO2021226977A1 publication Critical patent/WO2021226977A1/zh
Priority to ZA2022/08476A priority patent/ZA202208476B/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/25Fusion techniques
    • G06F18/254Fusion techniques of classification results, e.g. of results related to same input data
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/25Fusion techniques
    • G06F18/253Fusion techniques of extracted features
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Definitions

  • the invention relates to a method and platform for dynamic monitoring of typical ground features in mining based on multi-source remote sensing data fusion and deep neural network.
  • the purpose of the present invention is to provide a method and platform for dynamic monitoring of typical ground features in mining based on multi-source remote sensing data fusion and deep neural network, so as to solve the problem of poor recognition effect due to the low spatial resolution of remote sensing images in the prior art.
  • the mine environment is complex, and the existing machine recognition or manual recognition methods are used to separate and extract the problems of insufficient accuracy and real-time performance of typical ground objects.
  • the method for dynamic monitoring of typical ground features in mining based on multi-source remote sensing data fusion and deep neural network includes the following steps:
  • Step 1 Obtain time-series multi-source remote sensing data and basic mine data
  • Step 2 Remote sensing data processing and multi-source heterogeneous data fusion to obtain enhanced remote sensing image of high-resolution multi-spectral image
  • Step 3 Construct a deep neural network model, and use a combination of deep neural network and artificial visual interpretation to intelligently extract typical features
  • Step 4 Online dynamic monitoring of the extracted typical features.
  • the step 2 includes:
  • Step 2.1 Satellite remote sensing image processing
  • Step 2.2 Use the neural network to build a degradation model that degrades multi-spectral images into single-band images with different spatial resolutions and super-resolution that converts several low-spatial resolution single-band images into high-spatial-resolution multi-spectral images Rate synthesis model;
  • Step 2.3 Input the collected low-spatial resolution multispectral image into the super-resolution synthesis model, and obtain the enhanced remote sensing image with high spatial resolution after processing.
  • the step 2.2 includes:
  • Step 2.2.1 Obtain a single-band remote sensing image of the training area collected by the satellite, and collect basic mine data and single-band images with different spatial resolutions in the training area.
  • the single-band image is the same as the single-band remote sensing image.
  • a neural network takes the single-band image and basic mine data as input, and uses a single-band remote sensing image as output to train the neural network.
  • the basic mine data is geological and topographic information in the area, and the degradation model is obtained after training;
  • Step 2.2.2 Decompose the high-spatial-resolution multispectral image of the training area collected by the satellite into single-band images and input the degradation model to obtain single-band images with different spatial resolutions, and manually decompose the single-band images before decomposition
  • the low frequency part and high frequency part of the image are divided into blocks;
  • Step 2.2.3 Build a neural network, take a low-resolution single-band image as input, and use the segmented high-spatial-resolution single-band image as output, train the neural network, and the neural network also extracts low-resolution Image features, training the relationship between the extracted features and the high-frequency blocks in the high-spatial resolution single-band image.
  • the input single-band image can be matched with the best high-frequency block to obtain high-spatial resolution of the high-frequency part of the block
  • the step 3 includes:
  • Step 3.1 Manually mark various typical features contained in the remote sensing images obtained by training to form interpretation signs, and use the interpretation signs as samples for collection and management;
  • Step 3.2 Input the remote sensing image into the U-Net network for initial classification, and the obtained classification results are processed by the Mean-shift segmentation algorithm for edge optimization to improve the classification accuracy of the U-Net network for ground objects, and then pass the processed data through KNN
  • the algorithm is in the classification to reclassify the land creation category with insufficient feature distinguishability to obtain the classification result;
  • Step 3.3 According to the interpretation flags of step 3.1 and the classification results of step 3.2, manually interpret the classification results of step 3.2 to generate a large number of training tags, construct a neural network, and use remote sensing images and interpretation flags as input to train The label is used as the output, and the neural network is trained, and the machine interpretation model is obtained after training;
  • Step 3.4 The remote sensing image collected in step 2 is used to identify and interpret the logo, and the multi-time sequence remote sensing image and the interpretation logo collected later are used as input, and features are extracted through the machine interpretation model, and the typical features in the image are identified for dynamic use. monitor.
  • the step 3 further includes:
  • Step 3.5 Combine interpretation with machine interpretation and expert knowledge, and then perform manual visual interpretation on the results of machine interpretation, and verify with the ground conditions during manual visual interpretation. The interpretation accuracy of the results of machine interpretation is low. Of features.
  • the step 1 includes:
  • Step 1.1 Obtain multi-source satellite remote sensing images covering the mine at different times and resolutions, including multi-spectral images and single-band images;
  • Step 1.2 Use ground survey methods to obtain basic mine data.
  • the present invention also provides a dynamic monitoring platform for typical mining features based on multi-source remote sensing data fusion and deep neural network to realize the aforementioned dynamic monitoring of typical mining features based on multi-source remote sensing data fusion and deep neural network Monitoring method.
  • the dynamic monitoring platform includes a database, an image processing and interpretation module, a monitoring report module, a WebGIS system for mining dynamic monitoring, a visualization system for mining mining monitoring, and a mobile client for mining dynamic monitoring, among which:
  • Image processing and interpretation module used to fuse and enhance the acquired remote sensing data, and then interpret and identify typical features in the image for monitoring
  • Mine mining dynamic monitoring WebGIS system It is used to load interpretation results of different periods, and to identify the variation range of typical ground objects by superimposing the remote sensing interpretation results of multiple time series to realize the online display of the basic data and mining conditions of the mine;
  • Mine mining monitoring visualization system used to visualize the interpretation results of remote sensing on a large screen
  • Mine mining dynamic monitoring mobile client It is used to check the mining situation of the mine in time and upload the coordinate position and range of typical features when the ground investigator conducts field investigation and evidence collection;
  • Monitoring report module It is used to make monitoring situation maps and reports based on the interpretation results and the analysis results generated by dynamic monitoring, and give the monitoring results;
  • the database includes a supporting database and a monitoring result database.
  • the supporting database is used to store and manage remote sensing data and basic mine data
  • the monitoring result database is used to store and manage monitoring results.
  • the present invention has the following advantages: the present invention adopts a fusion processing method to enhance the spatial resolution of remote sensing images, so it can take advantage of the high spatial, high time, and high spectral resolution of satellite remote sensing images, through the fusion of multi-source remote sensing data Obtain more comprehensive image data characteristics.
  • a new method based on deep neural network is used to build a machine interpretation model, which speeds up the formation of training samples, can quickly obtain a large number of training samples, and the training samples have good correlation with the actual collected remote sensing images, and can combine basic data from different regions
  • model calculations not only greatly increases the speed of model construction, the efficiency of model calculation is also higher, and the interpretation time is greatly shortened.
  • a mining dynamic monitoring platform including a web system for mining dynamic monitoring, a visualization system for mining dynamic monitoring, a mobile client for mining dynamic monitoring, etc., to comprehensively view and analyze the monitoring results, and effectively improve the management level and monitoring effect of the mine.
  • Fig. 1 is a flow chart of a method for dynamic monitoring of typical ground features in mining based on multi-source remote sensing data fusion and deep neural network according to the present invention.
  • Fig. 2 is a flowchart of degradation and fusion enhancement of multi-source remote sensing images in the present invention.
  • Fig. 3 is a flow chart of intelligent recognition and extraction of typical features based on deep neural network in the present invention.
  • the present invention provides a method and platform for dynamic monitoring of typical mining features based on multi-source remote sensing data fusion and deep neural network, including the following steps:
  • the present invention provides a method for dynamic monitoring of typical ground features in mining based on multi-source remote sensing data fusion and deep neural network.
  • the method includes the following steps:
  • Step 1 Time-series multi-source remote sensing data acquisition and ground survey of basic mine data; specifically including the following steps:
  • Step 1.1 Obtain multi-source satellite remote sensing image maps covering the mine at different times and resolutions. Including domestic high score series (GF-1, GF-2), domestic resource satellite (ZY-3), foreign sentinel satellite (Sentinel-2A), Google satellite image, etc.
  • GF-1, GF-2 domestic high score series
  • ZY-3 domestic resource satellite
  • Sentinel-2A foreign sentinel satellite
  • Google satellite image etc.
  • Step 1.2 Ground survey of basic mine data, including topographic maps and DEM data of the mining area, mineral resource planning data, geological structure data, mineral resource distribution, mineral exploration data, mining rights, prospecting rights data, geology and hydrology, etc.
  • Step 2 Remote sensing data processing and fusion of multi-source heterogeneous data to obtain enhanced remote sensing images of high-resolution multi-spectral images. Specifically include the following steps:
  • Step 2.1 Satellite remote sensing image processing.
  • the remote sensing data processing flow includes operations such as radiation calibration, atmospheric correction, orthorectification, image registration, and image enhancement.
  • Step 2.2 Use the neural network to build a degradation model that degrades multi-spectral images into single-band images with different spatial resolutions and super-resolution that converts several low-spatial resolution single-band images into high-spatial-resolution multi-spectral images Rate synthesis model. Specifically include the following steps:
  • Step 2.2.1 Obtain a single-band remote sensing image of the training area collected by the satellite, and collect basic mine data and single-band images with different spatial resolutions in the training area.
  • the single-band image is the same as the single-band remote sensing image.
  • the neural network takes the single-band image and basic mine data as input, and uses the single-band remote sensing image as output to train the neural network.
  • the basic mine data is geological and topographic information in the region, and the degradation model is obtained after training.
  • Step 2.2.2 Decompose the high-spatial-resolution multispectral image of the training area collected by the satellite into single-band images and input the degradation model to obtain single-band images with different spatial resolutions, and manually decompose the single-band images before decomposition
  • the low-frequency part and the high-frequency part of the image are divided into blocks.
  • Step 2.2.3 Build a neural network, take a low-resolution single-band image as input, and use the segmented high-spatial-resolution single-band image as output, train the neural network, and the neural network also extracts low-resolution Image features, training the relationship between the extracted features and the high-frequency blocks in the high-spatial resolution single-band image.
  • the input single-band image can be matched with the best high-frequency block to obtain high-spatial resolution of the high-frequency part of the block
  • Step 2.3 Input the collected low-spatial resolution multispectral image into the super-resolution synthesis model, and use the prior knowledge obtained by the multi-layer neural network to match the optimal high-frequency block, and after processing, obtain the enhanced remote sensing image with high spatial resolution .
  • the single-band image with high spatial resolution is combined with the basic data to generate a single-band image with low spatial resolution through the degradation model, thereby quickly generating a large number of training samples related to the basic data. Therefore, when constructing a super-resolution synthesis model, the training efficiency High and reliable.
  • the super-resolution synthesis model can effectively improve the spatial resolution, and the remote sensing image of a high-resolution multi-spectral image is synthesized through multiple single-band images, so that the processed image has both high spatial resolution and multi-spectrum Features, maintaining the texture and tone information of the ground features are helpful for interpretation, and can well distinguish the ground feature information developed by the mine.
  • Step 3 Construct a deep neural network model, and use a combination of deep neural network and manual visual interpretation to intelligently extract typical features;
  • typical features include mining sites, transit sites, solid waste, mine buildings, and water bodies
  • Important features such as pollution, restoration and treatment; specifically include the following steps:
  • Step 3.1 Manually mark typical features such as mining sites, transit sites, solid waste, mining buildings, water pollution, restoration and treatment, to form interpretation signs, and use the interpretation signs as samples for collection management.
  • the interpretation signs are An important basis for interpreting different features and extracting target information.
  • Step 3.2 Segmentation of mine objects based on U-Net. Due to the irregularities and complex features of the mine, the high-resolution remote sensing image classification through the U-Net network, this process makes the classification results merge more shallow detailed information and deep robust information, and then through the Mean-shift segmentation algorithm (mean shift (Tracking algorithm) performs edge optimization processing on the obtained probability map results (classification results) to improve the classification accuracy of the U-Net network. Then, the KNN algorithm (K nearest neighbor classification algorithm, k-NearestNeighbor) is used to reclassify the mine feature categories that are prone to misclassification due to insufficient feature discrimination in the first step of the classification, so as to better obtain the high-resolution images. The essential characteristics of the ground features of the mine.
  • the Mean-shift segmentation algorithm mean shift (Tracking algorithm) performs edge optimization processing on the obtained probability map results (classification results) to improve the classification accuracy of the U-Net network.
  • KNN algorithm K nearest neighbor classification algorithm, k-NearestNeigh
  • Step 3.3 According to the interpretation flag of step 3.1 and the classification result of step 3.2, manually interpret the classification result of step 3.2 to generate a large number of training tags (classification results of the model), construct a neural network, and interpret the remote sensing image
  • the logo is used as input
  • the training label is used as output
  • the neural network is trained
  • the machine interpretation model is obtained after training.
  • the model uses the deep neural network framework to extract features after convolutional layer, pooling layer, activation function, and fully connected layer parameters are set. After training, it can achieve tasks such as target recognition and dynamic monitoring.
  • Step 3.4 The remote sensing image collected in step 2 is used to identify and interpret the logo, and the multi-time sequence remote sensing image and the interpretation logo collected later are used as input, and features are extracted through the machine interpretation model, and the typical features in the image are identified for dynamic use. monitor.
  • Step 3.5 Interpretation of classification results and ground verification.
  • the accuracy of the interpretation result is ensured.
  • Step 4 Perform online dynamic monitoring of the extracted typical features, and form a mining dynamic monitoring system based on the extracted dynamic remote sensing images corresponding to the typical features.
  • the present invention provides a dynamic monitoring platform for typical mining features based on multi-source remote sensing data fusion and deep neural network to realize the aforementioned dynamic monitoring of typical mining features based on multi-source remote sensing data fusion and deep neural network method.
  • the dynamic monitoring platform includes a database, an image processing and interpretation module, a monitoring report module, a mining dynamic monitoring WebGIS system, a mining mining monitoring visualization system, and a mining dynamic monitoring mobile phone client.
  • Image processing and interpretation module used to fuse and enhance the acquired remote sensing data, and then interpret and identify typical features in the image for monitoring. The method used is as described in step 2 and step 3.
  • Mine mining dynamic monitoring WebGIS system It is used to load the interpretation results of different periods, and through the superimposed analysis of the multi-time sequence remote sensing interpretation results to identify the variation range of typical ground features, to realize the online display of the basic data and mining conditions of the mine.
  • Mine mining monitoring visualization system used to visualize the interpretation results of remote sensing on a large screen to assist managers in commanding and decision-making;
  • Mine mining dynamic monitoring mobile phone client It is used to check the mining situation in time and accurately upload the coordinate position and range of typical features when ground investigators conduct field surveys and collect evidence.
  • Monitoring report module According to the interpretation results and dynamic monitoring of mining, at least one monitoring map (interpretation map, legend, scope, area, etc.) and report (including analysis and statistical information) are produced every quarter, but when In the event of an emergency, the monitoring results shall be given in time.
  • the database includes a support database and a monitoring result database.
  • the support database is used to store and manage remote sensing data and basic mine data.
  • the monitoring result database is used to store and manage monitoring results (Monthly, quarterly, and yearly monitoring results). .
  • the invention utilizes the characteristics of high space, high time, and high spectral resolution of satellite remote sensing images, and obtains more comprehensive image data characteristics through the fusion of multi-source remote sensing data.
  • Establish a framework for intelligent segmentation and deep neural network of mine objects establish an interpretation method combining artificial intelligence + expert knowledge, and automatically segment and intelligently extract important mining features.
  • Construction of a mining dynamic monitoring platform including a web system for mining dynamic monitoring, a visualization system for mining dynamic monitoring, and a mobile client for mining dynamic monitoring. It can more accurately, dynamically and intelligently monitor the mining scope and area of the mine, and improve the management level and monitoring effect of the mine through the monitoring platform and system.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台,方法包括以下步骤:步骤1:获取时序多源遥感数据以及矿山基础数据;步骤2:遥感数据处理并将多源异构数据融合得到高分辨率多光谱图像的增强遥感图像;步骤3:构造深度神经网络模型,采用深度神经网络与人工目视解译相结合的方法对典型地物进行智能提取;步骤4:对提取的典型地物进行在线动态监测和数据分析管理。该方法能提高遥感数据图像的空间分辨率增强图像后更容易识别典型地物,同时采用的机器解译方式能高效准确地识别出典型地物,从而能实时地对矿山的典型地物进行在线监测。

Description

一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台 技术领域
本发明涉及一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台。
背景技术
传统矿山监测,由于缺乏多源时空遥感数据的组织与信息挖掘,仅是对分幅影像的矿体位置、面积指标的定性或半定量描述,并未真正实现时空动态变化监测。如果引入多源时空遥感数据,高时间分辨率使得遥感数据更新频率更快,可以对矿山进行高时频动态监测;高光谱分辨率使得矿山及地物的分辨识别能力更加准确,但是空间分辨率往往达不到地面采集影像的清晰度,加上矿山地区环境复杂,现有技术难以在空间分辨率不高的图像上解译出典型地物,因此目前还没有可靠准确的机器解译识别方法。而且由于矿山数量众多、分布面广,矿山的环境问题复杂多样,导致传统的以人工解译为主的矿山典型地物解译周期长、成本高,影响矿山开采的动态监测。
因此,如何融合多源卫星遥感数据并对矿山标志物进行智能解译来实现矿山开采典型地物的动态监测已经成为一个急需解决的技术问题。
发明内容
本发明的目的在于提供一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台,以解决现有技术中因为遥感图像空间分辨率较低,识别效果较差,并且矿山环境复杂,采用现有机器识别或人工识别的方法,分隔提取典型地物的准确性和实时性都不足的问题。
所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,包括以下步骤:
步骤1:获取时序多源遥感数据以及矿山基础数据;
步骤2:遥感数据处理并将多源异构数据融合得到高分辨率多光谱图像的增强遥感图像;
步骤3:构造深度神经网络模型,采用深度神经网络与人工目视解译相结合的方法对典型地物进行智能提取;
步骤4:对提取的典型地物进行在线动态监测。
优选的,所述步骤2包括:
步骤2.1:卫星遥感影像处理;
步骤2.2:利用神经网络构建将多光谱图像降质为不同空间分辨率的单波段图像的降质模型和将若干低空间分辨率的单波段图像转化为高空间分辨率的多光谱图像的超分辨率合成模型;
步骤2.3:将采集的低空间分辨率的多光谱图像输入超分辨率合成模型,处理后得到高空间分辨率的增强遥感图像。
优选的,所述步骤2.2包括:
步骤2.2.1:获取卫星采集的训练区域的单波段遥感图像,并在训练区域通过采集矿山基础数据和不同空间分辨率的单波段图像,所述单波段图像与单波段遥感图像波段相同,构建神经网络,将所述单波段图像和矿山基础数据作为输入,将单波段遥感图像作为输出,对神经网络进行训练,矿山基础数据为区域内的地质地形信息,训练后得到所述降质模型;
步骤2.2.2:将卫星采集的训练区域高空间分辨率的多光谱图像分解为单波段图像后输入所述降质模型得到不同空间分辨率的单波段图像,通过人工方式将分解前的单波段图像中的低频部分和高频部分进行分块;
步骤2.2.3:构建神经网络,将低空间分辨率的单波段图像作为输入,将分块后的高空间分辨率的单波段图像作为输出,对神经网络进行训练,神经网络还提取低分辨率图像的特征,训练所提取特征与高空间分辨率的单波段图像中高频块的关联,训练后能让输入的单波段图像匹配最优的高频块得到对高频部分分块的高空间分辨率的单波段图像,并最后将各个高空间分辨率的单波段图像合成为高空间分辨率的多光谱增强遥感图像,从而得到所述超分辨率合成模型。
优选的,所述步骤3包括:
步骤3.1:对训练得到的遥感图像中包含的各种典型地物进行手动标识形成解译标志,并将解译标志作为样本进行采集管理;
步骤3.2:将遥感图像输入U-Net网络进行初次分类,得到的分类结果通过 Mean-shift分割算法进行边缘优化处理,提高U-Net网络对地物的分类精度,再将处理后的数据通过KNN算法进行在分类,以将特征区分性不足造地物类别进行再分类得到分类结果;
步骤3.3:根据步骤3.1的解译标志和步骤3.2的分类结果通过人工方式对步骤3.2的分类结果进行分类解译生成大量训练标签,构建神经网络,将遥感图像和解译标志作为输入,将训练标签作为输出,对神经网络进行训练,训练后得到机器解译模型;
步骤3.4:在步骤2采集的遥感图像标识解译标志,将之后采集的多时序遥感图像和解译标志作为输入,通过机器解译模型提取特征,识别出图像中的典型地物以用于动态监测。
优选的,所述步骤3还包括:
步骤3.5:机器解译与专家知识结合解译,对机器解译的结果再进行人工目视解译,人工目视解译时与地面情况进行验证,用于机器解译结果解译精度较低的地物。
优选的,所述步骤1包括:
步骤1.1:获取覆盖矿山的不同时间、不同分辨率的多源卫星遥感图像,包括多光谱图像和单波段图像;
步骤1.2:采地面调查方式获取矿山基础数据。
本发明还提供了一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测平台,以实现上述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,所述动态监测平台包括数据库、图像处理解译模块、监测报告模块、矿山开采动态监测WebGIS系统、矿山开采监测可视化系统和矿山开采动态监测手机客户端,其中:
图像处理解译模块:用于将获取的遥感数据进行融合增强,再解译识别出图像中的典型地物进行监测
矿山开采动态监测WebGIS系统:用于加载不同时期的解译结果,通过对多时序的遥感解译结果进行叠加分析识别出典型地物的变化范围,实现在线展示矿山的基础数据及开采情况;
矿山开采监测可视化系统:用于将遥感解译结果通过大屏可视化展现;
矿山开采动态监测手机客户端:用于及时查看矿山的开采情况和在地面调查人员进行野外调查取证时上传典型地物的坐标位置及范围;
监测报告模块:用于根据解译结果和动态监测产生的分析结果制作监测情况图件和报告,给出监测结果;
数据库包括支撑数据库和监测结果数据库,支撑数据库用于对遥感数据及矿山的基础数据进行储存和管理,监测结果数据库用于对监测结果进行储存和管理。
本发明具有如下优点:本发明采用融合处理方法对遥感图像的空间分辨率进行了增强,因此可以利用卫星遥感影像高空间、高时间、高光谱分辨率的特点,通过对多源遥感数据的融合获得更全面的影像数据特征。同时采用基于深度神经网络的新方法构建机器解译模型,加快了训练样本的形成,能快速得到大量训练样本,且训练样本与实际采集的遥感图像相关性好,并能结合不同地区的基本数据进行模型计算,不仅大大提高了模型构建速度,模型计算的效率也更高,大大缩短了解译时间,只需对部分机器解译可靠性较低的典型地物进行人工解译验证就能保证准确性,因此该方法对典型地物识别的效率和准确性都达到实时监测的要求,真正实现了对矿山典型地物的动态监测,监测矿山的开采范围及面积。
建设矿山开采动态监测平台,包括矿山开采动态监测Web系统、矿山开采动态监测可视化系统、矿山开采动态监测手机客户端等,全面查看和分析监测结果,有效提高矿山的管理水平和监测效果。
附图说明
图1为本发明一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法的流程图。
图2为本发明中多源遥感影像进行降质和融合增强的流程图。
图3为本发明中基于深度神经网络的对典型地物智能识别提取的流程图。
具体实施方式
下面对照附图,通过对实施例的描述,对本发明具体实施方式作进一步详细的说明,以帮助本领域的技术人员对本发明的发明构思、技术方案有更完整、准 确和深入的理解。
如图1-3所示,本发明提供了一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台,包括以下步骤:
本发明提供了一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,该方法包括下列步骤:
步骤1:时序多源遥感数据获取以及矿山基础数据地面调查;具体包括如下步骤:
步骤1.1:获取覆盖矿山的不同时间、不同分辨率的多源卫星遥感影像图。包括国产高分系列(GF-1、GF-2)、国产资源卫星(ZY-3)、国外哨兵卫星(Sentinel-2A)、Google卫星图等。
步骤1.2:矿山基础数据地面调查,包括矿区地形图和DEM数据、矿产资源规划数据、地质构造资料、矿产资源分布、矿产勘查数据、采矿权、探矿权数据、地质水文等。
步骤2:遥感数据处理并将多源异构数据融合得到高分辨率多光谱图像的增强遥感图像。具体包括如下步骤:
步骤2.1:卫星遥感影像处理。遥感数据处理流程包括辐射定标、大气校正、正射校正、图像配准、图像增强等操作。
步骤2.2:利用神经网络构建将多光谱图像降质为不同空间分辨率的单波段图像的降质模型和将若干低空间分辨率的单波段图像转化为高空间分辨率的多光谱图像的超分辨率合成模型。具体包括如下步骤:
步骤2.2.1:获取卫星采集的训练区域的单波段遥感图像,并在训练区域通过采集矿山基础数据和不同空间分辨率的单波段图像,所述单波段图像与单波段遥感图像波段相同,构建神经网络,将所述单波段图像和矿山基础数据作为输入,将单波段遥感图像作为输出,对神经网络进行训练,矿山基础数据为区域内的地质地形信息,训练后得到所述降质模型。
步骤2.2.2:将卫星采集的训练区域高空间分辨率的多光谱图像分解为单波段图像后输入所述降质模型得到不同空间分辨率的单波段图像,通过人工方式将分解前的单波段图像中的低频部分和高频部分进行分块。
步骤2.2.3:构建神经网络,将低空间分辨率的单波段图像作为输入,将分 块后的高空间分辨率的单波段图像作为输出,对神经网络进行训练,神经网络还提取低分辨率图像的特征,训练所提取特征与高空间分辨率的单波段图像中高频块的关联,训练后能让输入的单波段图像匹配最优的高频块得到对高频部分分块的高空间分辨率的单波段图像,并最后将各个高空间分辨率的单波段图像合成为高空间分辨率的多光谱增强遥感图像,从而得到所述超分辨率合成模型。
步骤2.3:将采集的低空间分辨率的多光谱图像输入超分辨率合成模型,利用多层神经网络获得的先验知识,匹配最优高频块,处理后得到高空间分辨率的增强遥感图像。
高空间分辨率的单波段图像结合基础数据通过降质模型生成低空间分辨率的单波段图像,由此快速生成了与基础数据相关的大量训练样本,因此构建超分辨率合成模型时,训练效率高,可靠性好。通过超分辨率合成模型能有效提高空间分辨率,通过多个单波段图像合成一幅高分辨率多光谱图像的遥感图像,使得处理后的图像既有较高的空间分辨率,又具有多光谱特征,保持其纹理和地物的色调信息有助于解译,能很好的区分矿山开发的地物信息。
步骤3:构造深度神经网络模型,采用深度神经网络与人工目视解译相结合的方法对典型地物进行智能提取;典型地物包括矿山开采场地、中转场地、固体废弃物、矿山建筑、水体污染、恢复治理等重要地物;具体包括如下步骤:
步骤3.1:对矿山开采场地、中转场地、固体废弃物、矿山建筑、水体污染、恢复治理等典型地物进行手动标识形成解译标志,并将解译标志作为样本进行采集管理,解译标志是判读不同地物和提取目标信息的重要依据。
步骤3.2:基于U-Net的矿山对象分割。由于矿山不规则、地物复杂,通过U-Net网络的高分遥感影像分类,此过程令分类结果融合更多浅层细节信息与深层的鲁棒信息,之后通过Mean-shift分割算法(均值移动跟踪算法)对获得概率图结果(分类结果)进行边缘优化处理,提高U-Net网络对地物的分类精度。然后,采用KNN算法(K最近邻分类算法,k-NearestNeighbor)对第一步分类由于特征区分性不足易于造成错分的矿山地物类别进行再分类,从而更好的获取高分辨率影像中的矿山地物本质特征。
步骤3.3:根据步骤3.1的解译标志和步骤3.2的分类结果通过人工方式对步骤3.2的分类结果进行分类解译生成大量训练标签(模型的分类结果),构建 神经网络,将遥感图像和解译标志作为输入,将训练标签作为输出,对神经网络进行训练,训练后得到机器解译模型。该模型利用深度神经网络框架经过卷积层、池化层、激活函数、全连接层参数设置后进行特征提取,经训练后能实现目标识别、动态监测等任务。
步骤3.4:在步骤2采集的遥感图像标识解译标志,将之后采集的多时序遥感图像和解译标志作为输入,通过机器解译模型提取特征,识别出图像中的典型地物以用于动态监测。
步骤3.5:分类结果解译与地面验证。通过人工智能与专家知识结合的解译方式,保证解译结果的精确化。先使用机器解译,对于解译精度较高的结果,人工可以较少干预;对于解译精度较差的结果,需要人工目视解译和地面验证后进行确认,从而实现了对典型地物的智能提取。
步骤4:对提取的典型地物进行在线动态监测,依据提取的对应典型地物的动态遥感图像形成矿山开采动态监测体系。
本发明提供了一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测平台,以实现前述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法。
所述动态监测平台包括数据库、图像处理解译模块、监测报告模块、矿山开采动态监测WebGIS系统、矿山开采监测可视化系统和矿山开采动态监测手机客户端。
图像处理解译模块:用于将获取的遥感数据进行融合增强,再解译识别出图像中的典型地物进行监测。所用方法如步骤2和步骤3所述。
矿山开采动态监测WebGIS系统:用于加载不同时期的解译结果,通过对多时序的遥感解译结果进行叠加分析识别出典型地物的变化范围,实现在线展示矿山的基础数据及开采情况。
矿山开采监测可视化系统:用于将遥感解译结果通过大屏可视化展现,辅助管理人员进行指挥决策;
矿山开采动态监测手机客户端:用于及时查看矿山的开采情况和在地面调查人员进行野外调查取证时精确上传典型地物的坐标位置及范围。
监测报告模块:根据矿山开采的解译结果及动态监测,至少每一季度制作一 份监测情况图件(解译图斑、图例、范围、面积等)及报告(包括分析统计信息),但当发生应急情况时,要及时给出监测结果。
数据库包括支撑数据库和监测结果数据库,支撑数据库用于对遥感数据及矿山的基础数据进行储存和管理,监测结果数据库用于对监测结果(对月、季、年每期监测结果)进行储存和管理。
本发明利用卫星遥感影像高空间、高时间、高光谱分辨率的特点,通过对多源遥感数据的融合获得更全面的影像数据特征。建立对遥感影像及矿山基础数据管理的监测支撑数据库和对月、季、年每期监测结果管理的动态监测结果数据库。搭建矿山对象智能分割和深度神经网络框架,建立人工智能+专家知识结合的解译方式,对矿山重要地物标识进行自动分割,智能提取。建设矿山开采动态监测平台,包括矿山开采动态监测Web系统、矿山开采动态监测可视化系统、矿山开采动态监测手机客户端等。可以更准确、动态、智能地监测矿山的开采范围及面积,并通过监测平台及系统提高矿山的管理水平和监测效果。
上面结合附图对本发明进行了示例性描述,显然本发明具体实现并不受上述方式的限制,只要采用了本发明的发明构思和技术方案进行的各种非实质性的改进,或未经改进将本发明构思和技术方案直接应用于其它场合的,均在本发明保护范围之内。

Claims (7)

  1. 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,其特征在于:包括以下步骤:
    步骤1:获取时序多源遥感数据以及矿山基础数据;
    步骤2:遥感数据处理并将多源异构数据融合得到高分辨率多光谱图像的增强遥感图像;
    步骤3:构造深度神经网络模型,采用深度神经网络与人工目视解译相结合的方法对典型地物进行智能提取;
    步骤4:对提取的典型地物进行在线动态监测。
  2. 根据权利要求1所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,其特征在于:所述步骤2包括:
    步骤2.1:卫星遥感影像处理;
    步骤2.2:利用神经网络构建将多光谱图像降质为不同空间分辨率的单波段图像的降质模型和将若干低空间分辨率的单波段图像转化为高空间分辨率的多光谱图像的超分辨率合成模型;
    步骤2.3:将采集的低空间分辨率的多光谱图像输入超分辨率合成模型处理后得到高空间分辨率的增强遥感图像。
  3. 根据权利要求1所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,其特征在于:所述步骤2.2包括:
    步骤2.2.1:获取卫星采集的训练区域的单波段遥感图像,并在训练区域通过采集矿山基础数据和不同空间分辨率的单波段图像,所述单波段图像与单波段遥感图像波段相同,构建神经网络,将所述单波段图像和矿山基础数据作为输入,将单波段遥感图像作为输出,对神经网络进行训练,矿山基础数据为区域内的地质地形信息,训练后得到所述降质模型;
    步骤2.2.2:将卫星采集的训练区域高空间分辨率的多光谱图像分解为单波段图像后输入所述降质模型得到不同空间分辨率的单波段图像,通过人工方式将分解前的单波段图像中的低频部分和高频部分进行分块;
    步骤2.2.3:构建神经网络,将低空间分辨率的单波段图像作为输入,将分块后的高空间分辨率的单波段图像作为输出,对神经网络进行训练,神经网络还提取低分辨率图像的特征,训练所提取特征与高空间分辨率的单波段图像中高频 块的关联,训练后能让输入的单波段图像匹配最优的高频块得到对高频部分分块的高空间分辨率的单波段图像,并最后将各个高空间分辨率的单波段图像合成为高空间分辨率的多光谱增强遥感图像,从而得到所述超分辨率合成模型。
  4. 根据权利要求1-3中任一所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,其特征在于:所述步骤3包括:
    步骤3.1:对训练得到的遥感图像中包含的各种典型地物进行手动标识形成解译标志,并将解译标志作为样本进行采集管理;
    步骤3.2:将遥感图像输入U-Net网络进行初次分类,得到的分类结果通过Mean-shift分割算法进行边缘优化处理,提高U-Net网络对地物的分类精度,再将处理后的数据通过KNN算法进行在分类,以将特征区分性不足造地物类别进行再分类得到分类结果;
    步骤3.3:根据步骤3.1的解译标志和步骤3.2的分类结果通过人工方式对步骤3.2的分类结果进行分类解译生成大量训练标签,构建神经网络,将遥感图像和解译标志作为输入,将训练标签作为输出,对神经网络进行训练,训练后得到机器解译模型;
    步骤3.4:在步骤2采集的遥感图像标识解译标志,将之后采集的多时序遥感图像和解译标志作为输入,通过机器解译模型提取特征,识别出图像中的典型地物以用于动态监测。
  5. 根据权利要求4所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,其特征在于:所述步骤3还包括:
    步骤3.5:机器解译与专家知识结合解译,对机器解译的结果再进行人工目视解译,人工目视解译时与地面情况进行验证,用于机器解译结果解译精度较低的典型地物。
  6. 根据权利要求5所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,其特征在于:所述步骤1包括:
    步骤1.1:获取覆盖矿山的不同时间、不同分辨率的多源卫星遥感图像,包括多光谱图像和单波段图像;
    步骤1.2:采地面调查方式获取矿山基础数据。
  7. 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监 测平台,其特征在于:以实现权利要求1-6所述的一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法,所述动态监测平台包括数据库、图像处理解译模块、监测报告模块、矿山开采动态监测WebGIS系统、矿山开采监测可视化系统和矿山开采动态监测手机客户端,其中:
    图像处理解译模块:用于将获取的遥感数据进行融合增强,再解译识别出图像中的典型地物进行监测
    矿山开采动态监测WebGIS系统:用于加载不同时期的解译结果,通过对多时序的遥感解译结果进行叠加分析识别出典型地物的变化范围,实现在线展示矿山的基础数据及开采情况;
    矿山开采监测可视化系统:用于将遥感解译结果通过大屏可视化展现;
    矿山开采动态监测手机客户端:用于及时查看矿山的开采情况和在地面调查人员进行野外调查取证时上传典型地物的坐标位置及范围;
    监测报告模块:用于根据解译结果和动态监测产生的分析结果制作监测情况图件和报告,给出监测结果;
    数据库包括支撑数据库和监测结果数据库,支撑数据库用于对遥感数据及矿山的基础数据进行储存和管理,监测结果数据库用于对监测结果进行储存和管理。
PCT/CN2020/090397 2020-05-15 2020-05-15 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台 WO2021226977A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
PCT/CN2020/090397 WO2021226977A1 (zh) 2020-05-15 2020-05-15 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台
CN202080000930.0A CN111742329B (zh) 2020-05-15 2020-05-15 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台
ZA2022/08476A ZA202208476B (en) 2020-05-15 2022-07-28 Method and platform for dynamically monitoring typical ground features in mining on the basis of multi-source remote sensing data fusion and deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/090397 WO2021226977A1 (zh) 2020-05-15 2020-05-15 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台

Publications (1)

Publication Number Publication Date
WO2021226977A1 true WO2021226977A1 (zh) 2021-11-18

Family

ID=72658096

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/090397 WO2021226977A1 (zh) 2020-05-15 2020-05-15 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台

Country Status (3)

Country Link
CN (1) CN111742329B (zh)
WO (1) WO2021226977A1 (zh)
ZA (1) ZA202208476B (zh)

Cited By (28)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111160128A (zh) * 2019-12-11 2020-05-15 中国资源卫星应用中心 一种基于对抗神经网络模型的遥感图像处理方法及系统
CN114241326A (zh) * 2022-02-24 2022-03-25 自然资源部第三地理信息制图院 一种渐进式遥感影像地物要素智能生产方法及系统
CN114384031A (zh) * 2022-01-12 2022-04-22 广西壮族自治区地理信息测绘院 星-空-地高光谱遥感水体重金属污染立体监测方法
CN114429588A (zh) * 2022-04-06 2022-05-03 四川省冶勘设计集团有限公司 应用于矿山环境恢复治理的数据处理方法及系统
CN114492210A (zh) * 2022-04-13 2022-05-13 潍坊绘圆地理信息有限公司 一种高光谱卫星星载数据智能解译系统及其实现方法
CN114780608A (zh) * 2022-04-06 2022-07-22 中国安全生产科学研究院 一种尾矿库数据在线调查系统
CN114943303A (zh) * 2022-06-16 2022-08-26 福州大学 一种基于多传感器遥感的时序aod重构方法
CN115588140A (zh) * 2022-10-24 2023-01-10 北京市遥感信息研究所 一种多光谱遥感图像多方向目标检测方法
CN115641442A (zh) * 2022-12-02 2023-01-24 耕宇牧星(北京)空间科技有限公司 一种基于融合网络的山体图像山脊线分割方法
CN115688053A (zh) * 2022-12-30 2023-02-03 速度时空信息科技股份有限公司 一种基于数据融合矿山环境动态监测管理方法及系统
CN115830446A (zh) * 2022-11-25 2023-03-21 中国水利水电科学研究院 一种动态水体产品融合方法、装置、设备及可读存储介质
CN115860975A (zh) * 2023-02-15 2023-03-28 南京航天宏图信息技术有限公司 一种基于卫星遥感的盐湖锂矿项目产能监测方法与装置
CN115984655A (zh) * 2022-12-19 2023-04-18 中国科学院空天信息创新研究院 一种基于跨域共享的多模态遥感数据融合方法
CN116091953A (zh) * 2023-04-11 2023-05-09 耕宇牧星(北京)空间科技有限公司 一种基于分组小波校准网络的建筑垃圾识别方法
CN116341986A (zh) * 2023-04-07 2023-06-27 北京一点五度科技有限公司 一种多源多时相传感器数据融合的碳汇项目审计与评级方法
CN116363502A (zh) * 2023-01-31 2023-06-30 中国科学院地理科学与资源研究所 融合多源地理大数据的采矿地多维信息获取方法及装置
CN116580309A (zh) * 2023-07-13 2023-08-11 湖北省地质环境总站 结合深度学习与面向对象分析的露天矿山采场提取方法
CN116958814A (zh) * 2023-07-24 2023-10-27 河北工程大学 一种湿地典型植物生长情况实时监测方法及系统
CN116977868A (zh) * 2023-06-07 2023-10-31 珠江水利委员会珠江水利科学研究院 一种基于特征匹配的影像乘积融合方法、系统及存储介质
CN117218558A (zh) * 2023-11-07 2023-12-12 北京市生态环境监测中心 一种多模态数据融合的工地扬尘精细化核算监管方法
CN117216641A (zh) * 2023-09-20 2023-12-12 四川省公路规划勘察设计研究院有限公司 一种高位崩滑地质灾害智能识别方法
CN117237256A (zh) * 2023-09-27 2023-12-15 中国自然资源航空物探遥感中心 一种浅海珊瑚礁监测数据采集方法、装置及设备
CN117315433A (zh) * 2023-11-30 2023-12-29 中国科学院空天信息创新研究院 基于分布一致性约束的遥感多模态多空间泛函映射方法
CN117422991A (zh) * 2023-10-07 2024-01-19 中色地科蓝天矿产(北京)有限公司 基于大数据的智能矿山检测系统、方法和可读存储介质
CN117422936A (zh) * 2023-12-15 2024-01-19 广州蓝图地理信息技术有限公司 一种遥感图像分类方法及系统
CN117576581A (zh) * 2024-01-17 2024-02-20 山东元鸿勘测规划设计有限公司 基于图像处理的地质勘探遥感监测方法
CN117592004A (zh) * 2024-01-19 2024-02-23 中国科学院空天信息创新研究院 Pm2.5浓度卫星监测方法、装置、设备及介质
CN117422991B (zh) * 2023-10-07 2024-06-04 中色地科蓝天矿产(北京)有限公司 基于大数据的智能矿山检测系统、方法和可读存储介质

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215135B (zh) * 2020-10-10 2022-07-08 东华理工大学 矿区开采与治理成效监测方法及装置
CN112732963A (zh) * 2021-01-15 2021-04-30 浙江大学 基于遥感大数据的跨界服务应用方法
CN113837972A (zh) * 2021-10-14 2021-12-24 中铁十九局集团矿业投资有限公司 一种基于多光谱遥感技术开采方法
CN114493963B (zh) * 2022-02-24 2023-04-07 中国科学院空天信息创新研究院 一种矿山生态环境监测与修复系统
CN114565046A (zh) * 2022-03-01 2022-05-31 中科海慧(北京)科技有限公司 一种基于遥感高光谱的矿产资源探测系统及探测方法
CN114612779A (zh) * 2022-03-14 2022-06-10 中科海慧(北京)科技有限公司 一种基于时空大数据分析的地质矿产勘察方法
CN115249331B (zh) * 2022-09-22 2022-12-20 山东世纪矿山机电有限公司 基于卷积神经网络模型的矿山生态安全识别方法
CN116030354B (zh) * 2023-03-29 2023-06-16 东华理工大学南昌校区 一种基于遥感数据融合的地质灾害分析方法及系统
CN116665066A (zh) * 2023-07-31 2023-08-29 平安科技(深圳)有限公司 遥感数据处理方法、平台、计算机设备和可读存储介质
CN117635219B (zh) * 2024-01-26 2024-04-26 长春黄金设计院有限公司 金属矿山生产大数据智能分析系统及方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001075937A (ja) * 1999-09-07 2001-03-23 Kdd Corp ニューラルネットワーク入力判別装置
CN101950361A (zh) * 2010-09-06 2011-01-19 中国科学院遥感应用研究所 一种基于光谱匹配度的遥感影像专题信息自适应提取方法
CN106295714A (zh) * 2016-08-22 2017-01-04 中国科学院电子学研究所 一种基于深度学习的多源遥感图像融合方法
CN107895136A (zh) * 2017-08-15 2018-04-10 东北大学 一种煤矿区域识别方法及系统
CN109685066A (zh) * 2018-12-24 2019-04-26 中国矿业大学(北京) 一种基于深度卷积神经网络的矿井目标检测与识别方法
CN109992641A (zh) * 2019-04-11 2019-07-09 中国矿业大学 一种面向煤矿无人工作面的高精度动态4dgis模型构建方法

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100538399C (zh) * 2007-09-27 2009-09-09 武汉大学 高分辨率遥感影像的分割和分类及变化检测一体化方法
CN101916436B (zh) * 2010-08-30 2011-11-16 武汉大学 一种多尺度空间投影遥感影像融合方法
US10262205B2 (en) * 2015-07-28 2019-04-16 Chiman KWAN Method and system for collaborative multi-satellite remote sensing
CN106651609B (zh) * 2016-09-27 2021-06-15 云南大学 一种矿产资源开发环境遥感监测辅助信息系统及方法
CN107273938B (zh) * 2017-07-13 2020-05-29 西安电子科技大学 基于双通道卷积阶梯网的多源遥感影像地物分类方法
CN107704807B (zh) * 2017-09-05 2020-12-29 北京航空航天大学 一种基于多源遥感序列影像的动态监测方法
CN107766471A (zh) * 2017-09-27 2018-03-06 中国农业大学 一种多源数据的组织管理方法与装置
CN107944477B (zh) * 2017-11-15 2019-07-02 北京师范大学 一种针对城市群的融合多源数据的遗传分类方法
CN109285133A (zh) * 2018-09-28 2019-01-29 哈尔滨工业大学 一种细节增强的遥感图像数据时空谱一体化融合方法
CN109509160A (zh) * 2018-11-28 2019-03-22 长沙理工大学 一种利用逐层迭代超分辨率的分层次遥感图像融合方法
CN110427995B (zh) * 2019-07-24 2023-11-17 中国科学院空天信息创新研究院 一种基于多源遥感数据的贝叶斯土壤水分估算方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001075937A (ja) * 1999-09-07 2001-03-23 Kdd Corp ニューラルネットワーク入力判別装置
CN101950361A (zh) * 2010-09-06 2011-01-19 中国科学院遥感应用研究所 一种基于光谱匹配度的遥感影像专题信息自适应提取方法
CN106295714A (zh) * 2016-08-22 2017-01-04 中国科学院电子学研究所 一种基于深度学习的多源遥感图像融合方法
CN107895136A (zh) * 2017-08-15 2018-04-10 东北大学 一种煤矿区域识别方法及系统
CN109685066A (zh) * 2018-12-24 2019-04-26 中国矿业大学(北京) 一种基于深度卷积神经网络的矿井目标检测与识别方法
CN109992641A (zh) * 2019-04-11 2019-07-09 中国矿业大学 一种面向煤矿无人工作面的高精度动态4dgis模型构建方法

Cited By (47)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111160128A (zh) * 2019-12-11 2020-05-15 中国资源卫星应用中心 一种基于对抗神经网络模型的遥感图像处理方法及系统
CN111160128B (zh) * 2019-12-11 2023-07-18 中国四维测绘技术有限公司 一种基于对抗神经网络模型的遥感图像处理方法及系统
CN114384031A (zh) * 2022-01-12 2022-04-22 广西壮族自治区地理信息测绘院 星-空-地高光谱遥感水体重金属污染立体监测方法
CN114241326B (zh) * 2022-02-24 2022-05-27 自然资源部第三地理信息制图院 一种渐进式遥感影像地物要素智能生产方法及系统
CN114241326A (zh) * 2022-02-24 2022-03-25 自然资源部第三地理信息制图院 一种渐进式遥感影像地物要素智能生产方法及系统
CN114780608A (zh) * 2022-04-06 2022-07-22 中国安全生产科学研究院 一种尾矿库数据在线调查系统
CN114429588B (zh) * 2022-04-06 2022-06-03 四川省冶勘设计集团有限公司 应用于矿山环境恢复治理的数据处理方法及系统
CN114780608B (zh) * 2022-04-06 2022-12-20 中国安全生产科学研究院 一种尾矿库数据在线调查系统
CN114429588A (zh) * 2022-04-06 2022-05-03 四川省冶勘设计集团有限公司 应用于矿山环境恢复治理的数据处理方法及系统
CN114492210B (zh) * 2022-04-13 2022-07-19 潍坊绘圆地理信息有限公司 一种高光谱卫星星载数据智能解译系统及其实现方法
CN114492210A (zh) * 2022-04-13 2022-05-13 潍坊绘圆地理信息有限公司 一种高光谱卫星星载数据智能解译系统及其实现方法
CN114943303A (zh) * 2022-06-16 2022-08-26 福州大学 一种基于多传感器遥感的时序aod重构方法
CN115588140B (zh) * 2022-10-24 2023-04-18 北京市遥感信息研究所 一种多光谱遥感图像多方向目标检测方法
CN115588140A (zh) * 2022-10-24 2023-01-10 北京市遥感信息研究所 一种多光谱遥感图像多方向目标检测方法
CN115830446A (zh) * 2022-11-25 2023-03-21 中国水利水电科学研究院 一种动态水体产品融合方法、装置、设备及可读存储介质
CN115830446B (zh) * 2022-11-25 2023-06-13 中国水利水电科学研究院 一种动态水体产品融合方法、装置、设备及可读存储介质
CN115641442A (zh) * 2022-12-02 2023-01-24 耕宇牧星(北京)空间科技有限公司 一种基于融合网络的山体图像山脊线分割方法
CN115641442B (zh) * 2022-12-02 2023-04-18 耕宇牧星(北京)空间科技有限公司 一种基于融合网络的山体图像山脊线分割方法
CN115984655A (zh) * 2022-12-19 2023-04-18 中国科学院空天信息创新研究院 一种基于跨域共享的多模态遥感数据融合方法
CN115984655B (zh) * 2022-12-19 2023-06-13 中国科学院空天信息创新研究院 一种基于跨域共享的多模态遥感数据融合方法
CN115688053A (zh) * 2022-12-30 2023-02-03 速度时空信息科技股份有限公司 一种基于数据融合矿山环境动态监测管理方法及系统
CN115688053B (zh) * 2022-12-30 2023-08-11 速度科技股份有限公司 一种基于数据融合矿山环境动态监测管理方法及系统
CN116363502B (zh) * 2023-01-31 2023-10-20 中国科学院地理科学与资源研究所 融合多源地理大数据的采矿地多维信息获取方法及装置
CN116363502A (zh) * 2023-01-31 2023-06-30 中国科学院地理科学与资源研究所 融合多源地理大数据的采矿地多维信息获取方法及装置
CN115860975A (zh) * 2023-02-15 2023-03-28 南京航天宏图信息技术有限公司 一种基于卫星遥感的盐湖锂矿项目产能监测方法与装置
CN116341986A (zh) * 2023-04-07 2023-06-27 北京一点五度科技有限公司 一种多源多时相传感器数据融合的碳汇项目审计与评级方法
CN116341986B (zh) * 2023-04-07 2024-01-12 北京一点五度科技有限公司 一种多源多时相传感器数据融合的碳汇项目审计与评级方法
CN116091953A (zh) * 2023-04-11 2023-05-09 耕宇牧星(北京)空间科技有限公司 一种基于分组小波校准网络的建筑垃圾识别方法
CN116977868B (zh) * 2023-06-07 2024-03-01 珠江水利委员会珠江水利科学研究院 一种基于特征匹配的影像乘积融合方法、系统及存储介质
CN116977868A (zh) * 2023-06-07 2023-10-31 珠江水利委员会珠江水利科学研究院 一种基于特征匹配的影像乘积融合方法、系统及存储介质
CN116580309B (zh) * 2023-07-13 2023-09-15 湖北省地质环境总站 结合深度学习与面向对象分析的露天矿山采场提取方法
CN116580309A (zh) * 2023-07-13 2023-08-11 湖北省地质环境总站 结合深度学习与面向对象分析的露天矿山采场提取方法
CN116958814A (zh) * 2023-07-24 2023-10-27 河北工程大学 一种湿地典型植物生长情况实时监测方法及系统
CN117216641A (zh) * 2023-09-20 2023-12-12 四川省公路规划勘察设计研究院有限公司 一种高位崩滑地质灾害智能识别方法
CN117237256A (zh) * 2023-09-27 2023-12-15 中国自然资源航空物探遥感中心 一种浅海珊瑚礁监测数据采集方法、装置及设备
CN117422991B (zh) * 2023-10-07 2024-06-04 中色地科蓝天矿产(北京)有限公司 基于大数据的智能矿山检测系统、方法和可读存储介质
CN117422991A (zh) * 2023-10-07 2024-01-19 中色地科蓝天矿产(北京)有限公司 基于大数据的智能矿山检测系统、方法和可读存储介质
CN117218558A (zh) * 2023-11-07 2023-12-12 北京市生态环境监测中心 一种多模态数据融合的工地扬尘精细化核算监管方法
CN117218558B (zh) * 2023-11-07 2024-02-23 北京市生态环境监测中心 一种多模态数据融合的工地扬尘精细化核算监管方法
CN117315433B (zh) * 2023-11-30 2024-02-13 中国科学院空天信息创新研究院 基于分布一致性约束的遥感多模态多空间泛函映射方法
CN117315433A (zh) * 2023-11-30 2023-12-29 中国科学院空天信息创新研究院 基于分布一致性约束的遥感多模态多空间泛函映射方法
CN117422936A (zh) * 2023-12-15 2024-01-19 广州蓝图地理信息技术有限公司 一种遥感图像分类方法及系统
CN117422936B (zh) * 2023-12-15 2024-04-02 广州蓝图地理信息技术有限公司 一种遥感图像分类方法及系统
CN117576581A (zh) * 2024-01-17 2024-02-20 山东元鸿勘测规划设计有限公司 基于图像处理的地质勘探遥感监测方法
CN117576581B (zh) * 2024-01-17 2024-04-05 山东元鸿勘测规划设计有限公司 基于图像处理的地质勘探遥感监测方法
CN117592004A (zh) * 2024-01-19 2024-02-23 中国科学院空天信息创新研究院 Pm2.5浓度卫星监测方法、装置、设备及介质
CN117592004B (zh) * 2024-01-19 2024-04-12 中国科学院空天信息创新研究院 Pm2.5浓度卫星监测方法、装置、设备及介质

Also Published As

Publication number Publication date
CN111742329B (zh) 2023-09-12
ZA202208476B (en) 2022-11-30
CN111742329A (zh) 2020-10-02

Similar Documents

Publication Publication Date Title
WO2021226977A1 (zh) 一种基于多源遥感数据融合和深度神经网络的矿山开采典型地物动态监测方法及平台
Huang et al. Multi-level monitoring of subtle urban changes for the megacities of China using high-resolution multi-view satellite imagery
Li et al. A 30-year (1984–2013) record of annual urban dynamics of Beijing City derived from Landsat data
Wieland et al. Estimating building inventory for rapid seismic vulnerability assessment: Towards an integrated approach based on multi-source imaging
Jianya et al. A review of multi-temporal remote sensing data change detection algorithms
Tian et al. A global analysis of multifaceted urbanization patterns using Earth Observation data from 1975 to 2015
Liu et al. Large-scale mapping of gully-affected areas: An approach integrating Google Earth images and terrain skeleton information
Stilla et al. Change detection of urban objects using 3D point clouds: A review
Hecht et al. Automatic derivation of urban structure types from topographic maps by means of image analysis and machine learning
CN116504032B (zh) 一种基于实景三维的滑坡危险性监测预警方法及系统
Brook et al. Modelling and monitoring urban built environment via multi-source integrated and fused remote sensing data
Darem et al. Development of a map for land use and land cover classification of the Northern Border Region using remote sensing and GIS
CN103035006A (zh) 一种LiDAR辅助下基于LEGION的高分辨率航空影像分割方法
Li et al. Fusing taxi trajectories and RS images to build road map via DCNN
CN116415843A (zh) 用于弱网络环境的多模态遥感辅助矿山生态环境评价方法
Liu et al. Improved road centerlines extraction in high-resolution remote sensing images using shear transform, directional morphological filtering and enhanced broken lines connection
Li et al. Identification of undocumented buildings in cadastral data using remote sensing: Construction period, morphology, and landscape
Chen et al. Leveraging Chinese GaoFen-7 imagery for high-resolution building height estimation in multiple cities
Sui et al. Processing of multitemporal data and change detection
Sharifi et al. Efficiency evaluating of automatic lineament extraction by means of remote sensing (Case study: Venarch, Iran)
CN114842356B (zh) 一种高分辨率地表类型样本自动生成方法、系统及设备
Forghani et al. Extracting terrain categories from multi-source satellite imagery
CN113780175B (zh) 高植被覆盖区台风暴雨型滑坡的遥感识别方法
Zhang et al. A novel knowledge-driven automated solution for high-resolution cropland extraction by cross-scale sample transfer
Forghani et al. Object-based classification of multi-sensor optical imagery to generate terrain surface roughness information for input to wind risk simulation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20935455

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20935455

Country of ref document: EP

Kind code of ref document: A1