Current Articles
2022, Volume 47, Issue 8
Display Method:
2022,
47(8):
1155-1164.
doi: 10.13203/j.whugis20210573
Abstract:
Objectives The complex and dangerous mountainous environment, uncertain geographical and geological conditions are the key factors affecting the safety, quality and schedule of railway tunnel construction. Methods Orient to intelligent and precise construction management, this paper proposes a top-down and bottom-up combination of railway tunnel drill safety-quality-schedule knowledge graph construction method, clarifies the conceptual connotation and semantic relationship for the five key elements of man-machine-material-method-environment(4M1E) related to safety-quality-schedule during the construction of railway tunnels, and designs a two-way collaborative construction method of mode layer from top to bottom and data layer from bottom to top, then introduces key technologies such as data acquisition and knowledge extraction. Results Taking the construction event of the Kangding No. 2 railway tunnel exit work area as an example, we construct a case knowledge graph. The results show that the knowledge graph constructed by the method in this paper finely depicts the key element attributes that affect safety-quality-schedule, the semantic relationship between the elements, and the mutual feedback relationship, etc. Conclusions Our proposed method provides key support for the overall systemic intelligent management of safety-quality-schedule in the whole process of railway tunnel drilling and blasting construction, and also lays the foundation for the digital twin of railway tunnel engineering.
2022,
47(8):
1165-1175.
doi: 10.13203/j.whugis20210714
Abstract:
Objectives Aiming at the problem of difficult effective management and rapid application between different data products of land and resources, the study uses the graph database to store the public land cover datasets, including GlobaLand30, FROM-GLC10_2017, GLC_FCS30_2020, etc., on the semantic level to establish a knowledge graph of land resources. It provides a new processing framework for the management, rapid application, and data quality assessment of land and resources data. Methods A new application framework for land cover data product management, knowledge extraction, and data acquisition and update based on administrative divisions is proposed. Anomaly data retrieval algorithms based on graphs are used to explore the consistency of different products, and a knowledge-based fast retrieval algorithm for graph nodes of interest (GNOI) in the graph. Results Through the introduction of the knowledge graph, a dynamically updateable nationwide land resource knowledge graph containing 447 817 nodes and 447 816 relationships has been formed, and it is found that the data accuracy of 92 units may have large errors in the 2 875 administrative units covering the whole country. Conclusions The research has greatly improved the utilization rate of multi-source land cover data products, shortened the time of data preprocessing for researchers, and provided new ideas for the knowledge management and application of land resources.
2022,
47(8):
1176-1190.
doi: 10.13203/j.whugis20210652
Abstract:
Objectives In the remote sensing (RS) big data era, intelligent interpretation of remote sensing images (RSI) is the key technology to mine the value of big RS data and promote several important applications. Traditional knowledge-driven RS interpretation methods, represented by expert systems, are highly interpretable, but generally show poor performance due to the interpretation knowledge being difficult to be completely and accurately expressed. With the development of deep learning in computer vision and other fields, it has gradually become the mainstream technology of RSI interpretation. However, the deep learning technique still has some fatal flaws in the RS field, such as poor interpretability and weak generalization ability. In order to overcome these problems, how to effectively combine knowledge inference and data learning has become an important research trend in the field of RS big data intelligent processing. Generally, knowledge inference relies on a strong domain knowledge base, but the research on RS knowledge graph (RS-KG) is very scarce and there is no available large-scale KG database for RSI interpretation now. Methods To overcome the above considerations, this paper focuses on the construction and evolution of the RS-KG for RSI interpretation and establishes the RS-KG takes into account the RS imaging mechanism and geographic knowledge. Supported by KG in the RS field, this paper takes three typical RSI interpretation tasks, namely, zero-shot RSI scene classification, interpretable RSI semantic segmentation, and large-scale RSI scene graph generation, as examples, to discuss the performance of the novel generation RSI interpretation paradigm which couples KG and deep learning. Results and Conclusions A large number of experimental results show that the combination of RS-KG inference and deep data learning can effectively improve the performance of RSI interpretation.The introduction of RS-KG can effectively improve the interpretation accuracy, generalization ability, anti-interference ability, and interpretability of deep learning models. These advantages make RS-KG promising in the novel generation RSI interpretation paradigm.
2022,
47(8):
1191-1200.
doi: 10.13203/j.whugis20220120
Abstract:
Objectives The rapid development of information and communication technology has facilitated the online tourism service and massive web text, which provides a new opportunity for tourism sector planning and personalized recommendation. However, owing to the characteristics of semantic vagueness and low signal-to-noise ratio, the web text is difficult to get utilized directly. Therefore, how to integrate the technologies of knowledge engineering, natural language processing and machine learning, so as to form a formalized domain knowledge graph from abundant tourism text, has attracted much attention. Methods This paper proposes a tourism knowledge graph construction method based on tourism domain ontology and transfer learning. Firstly, the ontology of tourist attractions is defined based on the domain specifications and standards, which support a comprehensive and systematic description of the semantic characteristics of attractions. Secondly, a transfer learning method is adopted to transform the pre-training language model into a customized knowledge extractor to acquire knowledge triples accurately from web text, which is integrated with the scattered tourism-related information including tourist check-ins and POI (point of interest) attributes to build a systematic knowledge graph. Results Experimental results show that the proposed knowledge extractor improves the accuracy (average area under the curve) and integrity (the number of sematic characteristics) of acquisition of sematic knowledge by 50.7% and 670%, respectively, compared with the common LDA (latent Dirichlet allocation) model. The constructed knowledge graph of tourist attractions contained 77 039 entities, 16 types of relationship, and total 10 971 810 triples. Conclusions Through the unified organization paradigm of triplet knowledge, the study realizes the fusion and integration of multi-source heterogeneous tourism data, and addresses the potential systemic risk in the decision-making process based on a single data source. It is argued that the constructed knowledge graph can fully capture the real tourism scene, support in-depth analysis of tourist behaviors and demands at different scales and granularities, and provide decision support for sustainable developments of tourist destinations.
2022,
47(8):
1201-1212.
doi: 10.13203/j.whugis20220038
Abstract:
Objectives Accurate, timely and effective monitoring of the growth and yield of winter wheat over a large area can help optimize the wheat planting structure, adjust the regional layout and ensure the country's food security. Therefore, it is very important to further improve the estimation accuracy of winter wheat yield. Methods Vegetation temperature condition index (VTCI) and leaf area index (LAI) at the main growth period of winter wheat, which were simulated by the CERES (crop environment resource synthesis)-Wheat model and retrieved from MODIS (moderate resolution imaging spectroradiometer) data, were assimilated by using ensemble Kalman filtering (EnKF) algorithm and particle filtering (PF) algorithm. In addition, the principal component analysis combined with the Copula function was used to develop univariate (VTCI or LAI) and bi-variate (VTCI and LAI) winter wheat yield estimation models, and the optimal model was selected to estimate winter wheat yields from 2017 to 2020. Results The experimental results show that, at the sampling-sites scale, both VTCI and LAI after assimilated can comprehensively reflect the variation characteristics of MODIS observed and model simulated values, and the application of PF algorithm has a better assimilation effect. At the regional scale, the bivariate yield estimation model developed by using PF algorithm has the highest accuracy. Compared with the accuracy of the models constructed by VTCI and LAI without assimilation, the root mean square error of the optimal assimilation model is reduced by 56.25 kg/hm2, and the average relative error is reduced by 1.51%. Conclusions The above results indicate that the model can effectively improve the accuracy of winter wheat yield estimation and has good applicability for large area yield estimation.
2022,
47(8):
1213-1219.
doi: 10.13203/j.whugis20220153
Abstract:
Objectives With the rapid development of earth observation system and space information network, the integration system of space and earth observation has been built in China. High-resolution remote sensing data have changed from GB-level to TB-level. The limited bandwidth capacity and storage space on the satellite seriously limit the development of the intelligent and real-time service based on remote sensing information. Methods Firstly, we introduce the characteristics of remote sensing data and the bottleneck of satellite-ground data transmission. Secondly, the limitations of traditional on-orbit compression algorithm are presented, we further discuss the importance of using high-ratio intelligent compression processing to realize low latency data transmission. Then, we introduce task-oriented intelligent compression method and procedure on Luojia-3(01) satellite. The compression framework obtains the observation region through high-quality imaging and high-precision geometric positioning, and captures the region of interest (ROI) using information extraction model. Finally, the mask of ROI is used to guide the compression model to achieve adaptive bit-rate allocation, and generate bits-stream file for transmission to the ground. Results According to the requirements of different tasks, using adaptive bits allocation can realize the intelligent compression of remote sensing image with high compression ratio, so as to realize the fast data transmission between satellite and ground. Conclusions Luojia-3(01) satellite has an extensible application software module that provides a common data interface, and provides an on-orbit verification environment for high-ratio compression algorithm, which is of great significance to the industrialization and commercialization of remote sensing technology.
2022,
47(8):
1220-1235.
doi: 10.13203/j.whugis20220335
Abstract:
Objectives Intelligent interpretation based on high-resolution remote sensing is an important way to realize the fine generation and rapid update of geographic information. Based on the background of fine-accurate geographical application using remote sensing, this paper analyzes the limitations of remote sensing information products in production and application. We explain the necessity of remote sensing serving geographical research and the key of coupling the fine shape of the map and the accurate content of the spectrum. Methods Based on the basic understanding that geography guides the research of intelligent remote sensing, we first put forward the development direction and technical ideas of high-resolution remote sensing geoscience analysis based on fine geographical scenes. Then, we propose an intelligent computing mode via the space-time/satellite-ground collaboration. Furthermore, we take the evaluation of rocky desertification cultivated land in Guanling County, Guizhou Province, China as an application case. Results Through this case study, three basic models, namely zoning-stratified perception, spatiotemporal synergistically inversion and multi-granular decision-making, are used to show how to carry out fine-accurate application by high-resolution remote sensing collaborative computing in complex mountain areas from a comprehensive perspective of refinement, quantification and topicalization. Conclusions Combined with the previous work experience and practical cognition, several key scientific problems are discussed, such as spatial expression using irregular grids, multimodal reconstruction of temporal features, multi-source uncertainty analysis and iterative optimization guided by uncertainty. We give some potential study directions and research ideas in order to establish a more complete and feasible theoretical system and explore the development paths for the research on intelligent remote sensing under the guidance of geography and the fine-accurate geographical application with remote sensing data.
2022,
47(8):
1236-1244.
doi: 10.13203/j.whugis20210506
Abstract:
Objectives The scale of buildings and their distribution is key indicators to measure the economic and social development of a region. Therefore, it is significant to study the extraction of buildings based on remote sensing images. Existing neural network methods still have shortcomings in the completeness of building extraction and the accuracy of building edges. To solve the above problems, this paper proposes a multi-level feature fusion network (MFFNet) based on high-resolution images. Methods Firstly, we use edge detection operators to improve the ability of the network to recognize the boundaries of buildings. Secondly, we use a multi-path convolution fusion module to extract building features from multiple dimensions, and introduce a large receptive field convolution module to break through feature extraction. The process is limited by the size of the receptive field. After fusing the extracted features, the convolutional attention module is used to compress them, and the global features are further mined by pyramid pooling, so as to achieve high-precision extraction of buildings. Results The current mainstream UNet, pyramid scene parsing network (PSPNet), multi attending path neural network (MAPNet) and multiscale-feature fusion deep neural networks with dilated convolution (MDNNet)are used as the comparison methods, and we use Wuhan University Aerial Image Dataset, Satellite Dataset II (East Asia) and Inria Aerial Image Dataset as experimental data for testing. Compared with the other four methods, MFFNet improves intersection over union, precision, recall, F1-score and mean average precision by 1.53%, 2.65%, 2.41%, 3.32% and 1.19% on average, achieves a better effect. Conclusions MFFNet not only accurately captures the detail features of buildings, but also strengthens the extraction and utilization of global features. It has better extraction effect on large buildings and buildings in complex environment.
2022,
47(8):
1245-1256.
doi: 10.13203/j.whugis20210172
Abstract:
Objectives The middle and lower reaches of the Yangtze River, including Hubei, Hunan, Jiangxi, Anhui, Jiangsu, Zhejiang and Shanghai, are important planting bases of commercial grain in China. However, at present, there are relatively few studies on agricultural drought in this region, and there is a lack of attention to the response of land cover types to drought. Moreover, in the context of climate change, the evolution and trend of agricultural drought in the middle and lower reaches of the Yangtze River need further discussion. Methods This study used MODIS (moderate resolution imaging spectroradiometer) V6 products to construct vegetation condition index (VCI), temperature condition index (TCI) and vegetation health index (VHI) to monitor the temporal and spatial evolution of agricultural drought in the middle and lower reaches of the Yangtze River from 2001 to 2019, and further explored the drought sensitivity of different vegetative types.Based on the concept of climate change, this study analyzed the drought trends in six provinces and one city in the middle and lower reaches of the Yangtze River. The results of the above three indices were further evaluated by standardized precipitation index (SPI) on different time scales, obtained and calculated from the CHIRPS V2.0 dataset. Results The results show that the VCI and TCI could monitor the long-term abnormal vegetation growth and heat anomalies, respectively, but neither index could provide comprehensive overview of drought conditions. Combining the advantages of both indices with the weights of 0.7 and 0.3 for VCI and TCI, respectively, the VHI, was more effective in agricultural drought monitoring in the middle and lower reaches of the Yangtze River. Different vegetation showed different drought sensitivity in study. Crops have the highest sensitivity to drought, forests are the lowest, and grasslands are somewhere in between. In the context of climate change, Jiangxi, Hunan, Hubei, Zhejiang, and Anhui show an intense wet trend in the past 20 years, while the Jiangsu and Shanghai show a weak wet trend. Conclusions Drought indices should be integrated to provide comprehensive evaluation of agricultural drought in the middle and lower reaches of the Yangtze River. Jiangsu province and Shanghai city are still at drought risk due to the weak wet trend and the local agricultural department should take drought mitigation measures to prevent economic losses. In the middle and lower reaches of the Yangtze River, croplands have the most obvious response to drought, indicating that crops are most sensitive to drought than grasses and forests, more attention should be paid to agriculture management. The relevant results can provide reference for the early warning of drought in various provinces and cities in the middle and lower reaches of the Yangtze River and help the management of regional agricultural production.
2022,
47(8):
1257-1270.
doi: 10.13203/j.whugis20220243
Abstract:
Objectives On 8th January 2022, a large earthquake (Mw 6.7) struck Menyuan County, Qinghai, China, causing serious damage to Lanzhou-Xinjiang high speed railway and forcing the closure of the railway for repairs, which has attracted highly domestic and international attention. Methods We presented a technical framework to determine earthquake surface ruptures by integrating optical, synthetic aperture radar (SAR) and unmanned aerial vehicle (UAV) images as well as light detection and ranging (LiDAR) data, and evaluated its damage to traffic networks. Firstly, we acquired a range of datasets including GF-1, GF-7, Sentinel-2 optical images and Sentinel-1A SAR images. GF-1 and GF-7 images were used to determine the spatial distribution characteristic of the surface ruptures. Secondly, we employed to estimate 2D surface displacement fields using optical pixel offset technique. One in the east-west (EW) direction and the other in the south-north (SN) direction. SAR pixel offset technique was utilized to acquire surface displacements in the range and azimuth directions whilst interferometric SAR(InSAR) was mainly for surface displacements in the radar line of sight (i.e. the range direction). Structure from motion (SfM) was used to process UAV images to obtain high precision digital surface models (DSMs). Finally, all the abovementioned information was used to precisely determine the spatial distribution and surface displacement characteristics of the earthquake surface ruptures. Results Our results show that the maximum surface displacement in the EW direction was about 2.0 m, the maximum in the range direction was approximately 1.5 m, and the total length of the surface ruptures was around 36.22 km. Furthermore, we performed an assessment of traffic inefficiency in Menyuan and its surrounding areas based on the distribution of historical geohazards as well as the earthquake surface ruptures using machine learning methods support vector machine models. Conclusions The Menyuan earthquake had the greatest impacts on highways, and the least impacts on rural roads. The southeast sections of the major highways G0611 and G338 had high risks. The technical framework demonstrated in this paper appears to be promising to precisely map surface ruptures, which in turn will directly benefit to earthquake disaster reduction.
A UAV Image Matching Algorithm Considering log-Polar Description and Position Scale Distance Feature
2022,
47(8):
1271-1278.
doi: 10.13203/j.whugis20200362
Abstract:
Objectives Few corresponding points can easily affect the calculation of image pose information, increase the difficulty of constructing a regional network in an aerial triangulation solution, so that lead to problems such as image stitching misalignment, incorrect bundle adjustment results or even failure. In order to better complete the matching of unmanned aerial vehicle (UAV) images, this paper proposes a robust UAV image matching algorithm considering log-polar description and position scale distance. Methods Firstly, a Gaussian multi-scale image collection is established and feature points are extracted. Secondly, the descriptors are constructed using log-polar coordinates, and a descriptor suitable for UAV image characteristics is established. Then, the feature matching is performed by the distance function of position and scale constraints. Finally, the mode seeking and fast sample consensus method are used to eliminate the outliner and complete the extraction of correspondence. Results The image obtained by four-rotor UAV is used as the data source, and a comparison experiment of image matching with scale invariant feature transform (SIFT) algorithm and synthetic aperture radar-scale invariant feature transform (SAR-SIFT) algorithm is carried out. The experimental results show that a 210-dimensional log-polar coordinate descriptor is constructed through the gradient location and orientation histogram. The descriptor can better describe the feature points in 10 directions through the circular neighborhood, making the matching results more robust. The position scale Euclidean distance matching function established by integrating factors such as position and scale can better calculate the UAV image matching relationship, and match more correct corresponding points. In terms of the number of correct corresponding points extracted under the same parameter settings, the proposed algorithm is significantly more than the other two algorithms, and in terms of the root mean square error of the matching results, the algorithm in this article is also significantly better than the two compared algorithms. Conclusions The proposed algorithm can better extract the corresponding points of UAV images.
2022,
47(8):
1279-1286.
doi: 10.13203/j.whugis20200292
Abstract:
Objectives The application and development of remote sensing image requires higher and higher image quality. Different processing methods and parameters are often needed for different quality remote sensing images, which is not suitable for the intelligent demand. Through the classification of remote sensing image quality level, it can provide prior information for remote sensing image processing, evaluate the objective quality of remote sensing image, assessment the effect of sensor imaging. With the development and popularization of deep learning theory, it is possible to evaluate the quality of digital images by using deep convolution neural network. Methods We propose a classification model of quality classification for remote sensing images based on deep convolution neural network. It is established by improving the feature extraction network and classification design.After the quality classification pretreatment, the classical deep learning method is used to detect the target, and the detection accuracy is obviously improved, which can effectively solve the problem of unbalanced quality of the training set data. Results The experimental results show that this proposed method is better than the traditional method. The highest score value of accuracy, recall, precision and F1_score can reach 0.976, 0.972, 0.974 and 0.973 on the remote sensing image data set of Northwestern Polytechnic University. Conclusions The classification of remote sensing image quality by convolution neural network extends the application field of deep learning. It provides a new method for the quality evaluation of remote sensing image. The classical deep learning method is used to detect the target, and the detection accuracy is obviously improved though quality classification. It provides a way to solve the problem of imbalance in remote sensing image quality.
Automatic Hiding Method of Sensitive Targets in Remote Sensing Images Based on Transformer Structure
2022,
47(8):
1287-1297.
doi: 10.13203/j.whugis20220219
Abstract:
Objective Decryption is the key to ensure the safe sharing of remote sensing resources. To solve the problems of incomplete target detection, unreliable complementary results, high resource consumption and difficulty of training in the traditional methods of sensitive target hiding in remote sensing images, an automatic hiding method of sensitive targets in remote sensing images is proposed based on the ability of Transformer structure to deal with global information. Methods Firstly, the optimized Cascade Mask R-CNN instance segmentation model with Swin Transformer as the backbone network is used to detect sensitive targets and generate mask regions. After improving the generalization capability of the model, RSMosaic (remote sense Mosaic), a data synthesis method to reduce the dependence on manually labeled data is designed. Secondly, the mask region is expanded by using the shadow detection model based on HSV(hue-saturation-value) space, and the MAE(masked autoencoders) model is introduced to achieve target background generation. Finally, the generated images are spliced with the original images to obtain the decrypted images. Results The sub-meter remote sensing images collected by Google Earth are used as test data, and the results show that this proposed method generates reliable hiding results while reducing dataset dependence and training resource consumption. Compared with the traditional method, the AP (average precision) values of bounding box and pixel mask are improved by 13.2% and 11.2% respectively in sensitive target instance segmentation, and the AP values can be improved by another 9.39% and 14.16% respectively after using RSMosaic, which is better than other repair models in terms of objective index and index variance in the field of image repair, especially in mean absolute error and maximum mean discrepancy indexes which are improved by more than 80%. It achieves the effect of automatic hiding of sensitive targets with reasonable structure and clear texture. Conclusions The proposed method reduces manpower, data and computing resources, and achieves better results in both subjective visual effects and objective indexes, which can provide technical support for real remote sensing image sharing.
2022,
47(8):
1298-1308.
doi: 10.13203/j.whugis20210001
Abstract:
Objectives Forest canopy density is an important factor in forest resource surveys. It plays an important role in forest quality evaluation and forest resource management. In recent years, artificial intelligence technology and remote sensing technology have developed rapidly and they have been successfully applied to forestry remote sensing quantitative estimation. It is significant to study how to use deep learning methods to effectively integrate the remote sensing data with different spatial coverage capabilities in regional forest canopy closure estimation. Methods This paper proposes a deep learning model (UnetR) which aims to estimate forest canopy closure based on deep learning model with the high-density light detection and ranging and high spatial resolution satellite imagery. We optimized the loss function of Unet for image classification, and added a batch normalization layer after the convolution layer, the model had the ability to quantitatively estimate continuous variables. Results The comparative evaluation results with fully convolutional networks, random forest and support vector regression models show that the root mean square error of the UnetR model was lower, the estimation accuracy was higher determination coefficient is 0.777, root mean square error is 0.137, estimation accuracy is 75.60%. Conclusions This paper provided a low labor cost and high degree of automation estimation model for remote sensing monitoring of regional forest canopy closure.
2022,
47(8):
1309-1317.
doi: 10.13203/j.whugis20210524
Abstract:
Objectives Multi-source image matching is primarily disturbed by nonlinear intensity difference, contrast difference and inconspicuous regional structure features, while the significant differences of texture features result in lack of part structure seriously between light detection and ranging(LiDAR)depth map and aerial image, and this problem causes a mutation in the phase extremum, which further increases the difficulty of matching. Methods In this paper, a method of efficient matching of LiDAR depth map and aerial image based on phase mean convolution is proposed. In the image feature matching stage, a histogram of phase mean energy convolution(HPMEC) is established, which extended the phase consistency model in order to solve a mean convolution sequence and phase maximum label map by constructing phase mean energy convolution equation. Then the nearest neighbor matching algorithm was completed the initial match and marginalizing sample consensus plus was used to remove outliers. Based on the thread pool parallel strategy, the images were matched by dividing the overlapping grid. Multiple sets of LiDAR depth map and aerial image with different types of ground coverage are used to as dataset to experiment with position scale orientation-scale invariant feature transform (PSO-SIFT), Log-Gabor histogram descriptor (LGHD), radiation-variation insensitive feature transform (RIFT) and histogram of absolute phase consistency gradients (HAPCG) methods respectively. Results The results show that the performance of HPMEC method is superior to the other four methods in the matching of LiDAR depth map and aerial image, the average running time is 13.3 times of PSO-SIFT, 10.9 times of LGHD, 10.4 times of HAPCG and 7.0 times of RIFT, at the same time the average correct matching points are significantly higher than the other four methods, the root mean square error is lightly better than the other four methods within 1.9 pixels. Conclusions The proposed HPMEC method could achieve efficient and robust matching between LiDAR depth map and aerial image.
2022,
47(8):
1318-1327.
doi: 10.13203/j.whugis20220235
Abstract:
Objectives Most of the remotely sensed night light data is acquired by satellites, and the spatial resolution is always coarse and the overpass time is fixed. These limitations hinder understanding night light patterns with insight. Compared to satellite images, camera image data has higher temporal and spatial resolution, and web camera images provide more information on night light and details of urban economic activities. However, such data has rarely been reported for academic researches. Methods For real-time video data, we design a specific Crawler program for downloading images from the Earth Camera website. Consequently, the acquired time series camera images are analyzed to find different trend components and their spatial distribution by using principal component analysis (PCA). Results Generally, the Crawler program can be run stably to obtain public camera image data from the Earth Camera website. Through analyzing city light in a region inside Tokyo based on the proposed PCA method, we find that the city light dynamic is complex and random in some extent, and the revealed pattern shows a general and significant decreasing trend, while some regions have more complicated temporal patterns such as increasing at first and then decreasing. In addition, the different building façades have different city light dynamic as well. Conclusions This study proposes a technical framework for acquiring and analyzing urban public camera images at night, and it suggests that urban camera can effectively provide city light change information from a micro perspective, which will provide new data source for supporting quantitative remote sensing of night light.
2022,
47(8):
1328-1335.
doi: 10.13203/j.whugis20220119
Abstract:
Objectives Mars is the main target object for deep space exploration. Mars rovers, or roving probes, are important tools for surface exploration and scientific research on Mars. For the growing amount of remote sensing data collected by Mars rovers, there is an urgent need for a method that can intelligently detect novel targets of scientific value from the massive amount of images, reduce the time cost of detection planning, and provide information for subsequent scientific analysis. The traditional novel detection methods mostly include distance-based metrics and image-based reconstruction methods, distance-based metrics calculate novel scores pixel by pixel without considering spatial contextual information, and image-based reconstruction methods focus on reconstruction of typical landscape backgrounds, and novelty is manifested by image reconstruction errors, which is not effective in extracting small novel targets such as boreholes and dust removal points. Methods To address the above problems of traditional novel detection methods in Mars rover novel target detection, this paper proposes an improved Mars rover multispectral image depth novel target detection method, uses full convolutional self-coding neural network to extract deep features for typical landscape reconstruction, and joints Mahalanobis distance for novel target and typical landscape background separation, fully exploits the spatial and spectral dimensional features to improve the accuracy of Mars rover novel target detection results. Results The experiments use the multispectral image dataset of Curiosity rover released by NASA (National Aeronautics and Space Administration), and the proposed convolution auto-encoder combined Mahalanobis distance method(CAE-M) is compared with Reed-Xiaoli detector, principal component analysis, convolution auto-encoder convolution, and generative adversarial networks on the surface of Gale crater. The results show that CAE-M outperforms previous detection methods in terms of detection accuracy and visual interpretation, and has a balanced and stable performance in different classes of novel target detection. Conclusion The proposed CAE-M method comprehensively utilizes spatial and spectral information of multispectral images, which can help Mars rover exploration missions to quickly and intelligently screen and sort novel data with scientific value in massive data, save the time and cost of route planning, improve scientific returns.
2022,
47(8):
1336-1348.
doi: 10.13203/j.whugis20210157
Abstract:
Objectives In recent years, hyperspectral images classification based on deep learning has made important progress. In view of the scarcity of training samples for hyperspectral image classification, this paper proposes a lightweight attention depth-wise relation network (LWAD-RN) to solve the problem of small sample hyperspectral image classification. Methods The network consists of an embedding layer and a relation layer. In the embedding layer, a lightweight convolutional neural network combining attention mechanism is used to extract pixel features, and a dense network structure is introduced. The relation value is calculated in the relation layer for classification, and the task-based mode is used to train the network. Three groups of public hyperspectral image datasets are used to implement experiments. Results and Conclusions The results show that LWAD-RN can effectively improve the classification accuracy under the condition of small samples (5 training samples per category), and the efficiency of model training and classification is improved.The proposed LWAD-RN can obtain ideal classification accuracy under the condition of small samples, and the lightweight network structure can improve the model training and classification efficiency. However, under the condition of small samples, the quality of training samples will have an important impact on the performance of the model. Therefore, follow-up studies should be conducted on how to select high-quality training samples more accurately and efficiently to ensure the stability of the model and better meet the needs of practical application.