[Significance] Data resources have become pivotal in modern production, evolving in close synergy with advancements in artificial intelligence (AI) technologies, which continuously cultivate new, high-quality productive forces. Remote sensing data intelligence has naturally emerged as a result of the rapid expansion of remote sensing big data and AI. This integration significantly enhances the efficiency and accuracy of remote sensing data processing while bolstering the ability to address emergencies and adapt to complex environmental changes. Remote sensing data intelligence represents a transformative approach, leveraging state-of-the-art technological advancements and redefining traditional paradigms of remote sensing information engineering and its applications. [Analysis] This paper delves into the technological background and foundations that have facilitated the emergence of remote sensing data intelligence. The rapid development of technology has provided robust support for remote sensing data intelligence, primarily in three areas: the advent of the big data era in remote sensing, significant advancements in remote sensing data processing capabilities, and the flourishing research on remote sensing large models. Furthermore, a comprehensive technical framework is proposed, outlining the critical elements and methodologies required for implementing remote sensing data intelligence effectively. To demonstrate the practical applications of remote sensing data intelligence, the paper presents a case study on applying these techniques to extract ultra-high-resolution centralized and distributed photovoltaic information in China. [Results] By integrating large models with remote sensing data, the study demonstrates how remote sensing data intelligence enables precise identification and mapping of centralized and distributed photovoltaic installations, offering valuable insights for energy management and planning. The effectiveness of remote sensing data intelligence in addressing challenges associated with large-scale photovoltaic extraction underscores its potential for application in critical fields. [Prospect] Finally, the paper provides an outlook on areas requiring further study in remote sensing data intelligence. It emphasizes that high-quality data serves as the foundation for remote sensing data intelligence and highlights the importance of constructing AI-ready knowledge bases and recognizing the value of small datasets. Developing targeted and efficient algorithms is essential for achieving remote sensing intelligence, making the advancement of practical data intelligence methods an urgent research priority. Furthermore, promoting multi-level services for remote sensing data, information, and knowledge through data intelligence should be prioritized. This research provides a comprehensive technical framework and forward-looking insights for remote sensing data intelligence, offering valuable references for further exploration and implementation in critical fields.

[Significance] The extraction of Cartographic-Level Vector Elements (CLVE) is a critical prerequisite for the direct application of remote sensing image intelligent interpretation in real-world scenarios. [Analysis] In recent years, the continuous rapid advancement of remote sensing observation technology has provided a rich data foundation for fields such as natural resource surveying, monitoring, and public surveying and mapping data production. However, due to the limitations of intelligent interpretation algorithms, obtaining the necessary vector elements data for operational scenarios still heavily relies on manual visual interpretation and human-computer interactive post-processing. Although significant progress has been made in remote sensing image interpretation using deep learning techniques, producing vector data that are directly usable in operational scenarios remains a major challenge. [Progress] This paper, based on the actual data needs of operational scenarios such as public surveying and mapping data production, conducts an in-depth analysis of the rule constraints for different vector elements in remote sensing image interpretation across a wide range of operational contexts. It preliminarily defines "cartographic-level vector elements" as vector element data that complies with certain cartographic standard constraints at a specific scale. Centered on this definition, the content of the rule set for CLVE is summarized and analyzed from nine dimensions, including vector types, object shapes, boundary positioning, area, length, width, angle size, topological constraints, and adjacency constraints. Evaluation methods for CLVE are then outlined in four aspects: class attributes, positional accuracy, topological accuracy, and rationality of generalization and compromise. Subsequently, through literature collection and statistical analysis, it was observed that research on deep learning-based vector extraction, while still in its early stages, has shown a rapid upward trend year by year, indicating increasing attention in the field. The paper then systematically reviews three major methodological frameworks for deep learning-based vector extraction: semantic segmentation & post-processing, iterative methods, and parallel methods. A detailed analysis is provided on their basic principles, characteristics and accuracy of vector extraction, flexibility, and computational efficiency, highlighting their respective strengths, weaknesses, and differences. The paper also summarizes the current limitations of remote sensing intelligent interpretation methods aimed at CLVE in terms of cartographic-level interpretation capabilities, rule coupling, and remote sensing interpretability. [Prospect]Finally, future research directions for intelligent interpretation of CLVE are explored from several perspectives, including the construction of broad and open cartographic-level rule sets, the development and sharing of CLVE datasets, the advancement of multi-element CLVE extraction frameworks, and the exploration of the potential of multimodal coupled semantic rules.

[Objectives] Remote Sensing Intelligent Interpretation (RSII) often encounters challenges when applied for practical resource and environmental management, especially for complex scenes. To address this, we start from the explanation of why remote sensing interpretation is needed, and clarify that the mission of RSII is to achieve more rapid interpretation to build the digital twin earth with lower cost compared to manual interpretation. However, most RSII systems operate as a unidirectional process from remote sensing data to geoscience knowledge, lacking the feedback from knowledge to data. As a result, remote sensing information extracted from data often mismatch the knowledge of existing geoscience, creating a trust crisis between RSII researchers and geoscience researchers. And the crisis becomes more severe with the uncertainty of remote sensing information. [Analysis] We believe that an agreed upon representation model of geoscience knowledge between RSII researchers and geoscience researchers is necessary to alleviate the crisis. Based on this analysis, we propose a framework using geo-science zoning as the bridge to connect RSII researchers and geoscience researchers. In this framework, knowledge from geoscience could be transferred into the RSII system through geo-science zoning so that the interpretation results could be more coincided with geoscience knowledge. The framework mainly relies on (a) the scene complexity measurement, (b) the knowledge coupling of geographic regions to form the geological zoning method for remote sensing intelligent interpretation, and (c) the sampling specification of regional samples. The scene complexity measurement provides quantitative features for geoscience zoning and sampling weights assignment. Existing zoning data, such as ecological zoning data, geographic elements, and multisource remote sensing images are the main data inputs for geoscience zoning. The main principles for constructing zoning methods include (a) the geoscience elements type, (b) the scale of geoscience zoning, and (c) the process of information flow from data to knowledge. [Prospects] With these models, we can realize regional RSII guided by the knowledge. Preliminary experiments on complexity and optimization sampling, image segmentation scale optimization, cultivated land type fine classification, etc., reveal that this framework has great potential in improving the geoscience knowledge acquisition by RSII, enhancing the accuracy of the state-of-the-art RSII by 6%~10%, especially for the high-complexity nature scenes. However, the superiority of the framework may disappear if the scene for interpretation is simple, like the first level land use/cover classification, which is mainly caused by the inefficient samples after geoscience zoning. Therefore, more attention is needed in sampling when developing geoscience zoning framework.

[Objectives] As remote sensing classification and interpretation technologies continue to advance, the intelligent interpretation of natural resources in complex environments has become a critical research focus. The accuracy and reliability of remote sensing data interpretation depend fundamentally on the quality and representativeness of the samples used in the analysis. In China, diverse terrain, complex meteorological conditions, and fragmented land surface structures introduce significant spatiotemporal variability, making the selection and quality of remote sensing samples particularly challenging. Traditional sampling methods often fail to adequately represent the full spectrum of characteristics inherent in these diverse landscapes, leading to substantial biases and inaccuracies in interpretation outcomes. [Methods] To address these challenges, this study offers a comprehensive review of key elements in remote sensing classification, encompassing methods for sampling labeled data, techniques for multi-scale morphological transformations to augment samples, and strategies for evaluating the quality of labeled samples. The research emphasizes the critical importance of optimizing sample selection to reduce bias and improve interpretation accuracy. It explores the theoretical foundations for sample optimization, highlighting the necessity of obtaining representative samples that accurately capture the complexity and variability of the land surface. [Results] One of the primary contributions of this study is the development of a novel sampling optimization method that integrates terrain complexity into the sampling process. By considering the diverse and intricate nature of the landscape, our approach enhances the representativeness of the samples, thereby reducing errors introduced by sampling bias and significantly improving the accuracy of remote sensing interpretation. In particular, we emphasize the role of multi-scale morphological transformations, which allow for the expansion of sample diversity and the generation of more robust and generalizable remote sensing models. This process is crucial for creating high-quality labeled samples that can better support complex interpretation tasks. The effectiveness of this complexity-based sample optimization approach is demonstrated through a series of experiments. These experiments reveal significant improvements in interpretation accuracy when compared to traditional sampling methods. This substantial enhancement underscores the value of incorporating terrain complexity and multi-scale transformations in the sampling and interpretation process. [Conclusions] By following the principles and methodologies outlined in this research, practitioners and researchers can obtain high-quality, representative labeled samples that significantly improve the precision and efficiency of remote sensing classification models. The findings of this study provide a solid theoretical and technical foundation for advancing remote sensing intelligent interpretation technology. Furthermore, the research offers practical insights and guidelines for applying these optimized sampling strategies to the classification of natural resources in complex scenarios, ultimately contributing to more accurate and reliable interpretation outcomes in the field of remote sensing.

[Objectives] With the development of deep learning technology, the ability to monitor changes in natural resource elements using remote sensing images has significantly improved. While deep learning change detection models excel at extracting low-level semantic information from remote sensing images, they face challenges in distinguishing land-use type changes from non-land-use type changes, such as crop rotation, natural fluctuations in water levels, and forest degradation. To ensure a high recall rate in change detection, these models often generate a large number of false positive change polygons, requiring substantial manual effort to eliminate these false alarms. [Methods] To address this issue, this paper proposes a natural resource element change polygon purification algorithm driven by remote sensing spatiotemporal knowledge graph. The algorithm aims to minimize the false positive rate while maintaining a high recall rate, thereby improving the efficiency of natural resource element change monitoring. To support the intelligent construction and effective reasoning of the spatiotemporal knowledge graph, this study designed a remote sensing spatiotemporal knowledge graph ontology model taking into account spatiotemporal characteristics and developed a GraphGIS toolkit that integrates graph database storage and computation. This paper also introduces a vector knowledge extraction method based on the native spatial analysis of the GraphGIS graph database, a remote sensing image knowledge extraction method based on efficient fine-tuning of the SkySense visual large model, and a polygon purification knowledge extraction method based on the SeqGPT large language model. Under the constraints of the spatiotemporal ontology model, vector, image, and text knowledge converge to form a remote sensing spatiotemporal knowledge graph. Inspired by the manual operation methods for change polygon purification, this paper developed an automatic purification method of change polygons based on first-order logical reasoning within the knowledge graph. To improve the concurrent processing and human-computer interaction, this paper developed a remote sensing spatiotemporal knowledge graph management and service system. [Results] For the task of purifying natural resource element change polygons in Guangdong Province from March to June 2024, the proposed method achieved a true-preserved rate of 95.37% and a false-removed rate of 21.82%. [Conclusions] The intelligent purification algorithm and system for natural resource element change polygons proposed in this study effectively reduce false positives while preserving real change polygons. This approach significantly enhances the efficiency of natural resource element change monitoring.

[Objectives] The active discovery and scientific assessment of the ecological damage caused by mining disturbances in coal mining areas is a key focus in the research of "intelligent mining and green mines". However, traditional analysis methods, characterized by a reliance on monitoring, limited discovery capabilities, dependence on experts, and retrospective assessments, struggle to meet the regulatory authorities’ actual needs for intelligent identification and rapid early warning. This article aims to verify the adaptability of knowledge graph-based spatial reasoning methods for actively detecting and intelligently identifying ecological damage in coal mining areas, while exploring innovative approaches and technologies for ecological environment governance in the modern era. [Methods] By integrating multi-source monitoring data from "Space-Air-Ground-Human" systems and summarizing knowledge on the location, form, group distribution, distribution patterns, and spatiotemporal evolution of ecological units in coal mining areas, an indicator system for describing these units is designed. Additionally, intelligent identification and reasoning rules for ecological damage are constructed using knowledge graph technology. [Results] Coal mining subsidence is a typical land use/cover change phenomenon caused by coal resource extraction. Remote sensing technology is crucial for extracting subsidence and its variations, yet traditional methods often overestimate the affected area due to misclassification of natural water surfaces as mining-induced subsidence. To address this, new knowledge and spatial reasoning rules were introduced to accurately differentiate between mining subsidence areas and natural water surfaces. Using a coal mining area in Shanxi Province as a case study, spatial reasoning rules were applied to identify subsidence units accurately. Experimental results demonstrated that the proposed method improved the precision and intelligent recognition accuracy of mining disturbance units. Compared with traditional recognition approaches, the proposed method reduced false positives by 21.43%. [Conclusions] Knowledge graph technology proves highly adaptable for analyzing and evaluating ecological environments in coal mining areas. It offers technical support for the proactive discovery and accurate identification of ecological damage caused by mining disturbances. Furthermore, it provides new technological tools and ideas for building advanced ecological governance models.

[Objectives] Simultaneous Localization and Mapping (SLAM) using LiDAR technology is a pivotal technology in the fields of positioning and navigation and spatial data acquisition. It offers distinct advantages, such as reliable performance and the ability to generate intuitive maps. However, in environments lacking geometric features or with high feature similarity, traditional LiDAR SLAM techniques often yield inaccurate results, because the current LiDAR degenerate environment detection methods require heuristic thresholding, calculation of indirect evaluation indexes, and have low detection efficiency. In order to predict such anomalies and address this issue, it is crucial to effectively detect LiDAR degenerate environments. [Methods] This study proposed a LiDAR degenerate environment detection method based on improved XGBoost algorithms. It achieved direct degradation detection of individual-frame point clouds by analyzing the geometric structure of the environment. Specifically, a classification feature system was established utilizing LiDAR point cloud data, followed by the construction of XGBoost decision trees. Subsequently, the Fuzzy Comprehensive Evaluation (FCE) algorithm was employed to compute a comprehensive importance metric for each feature, thereby facilitating the formation of an effective subset of features which leads to enhanced detection accuracy. Meanwhile, a bidirectional feature screening strategy based on Spearman's rank Correlation Coefficient (SCC) was implemented to construct the feature subset and increase model training efficiency. Moreover, by integrating Sliding Window (SW) and Boyer-Moore Voting (BMV) strategies, we performed secondary correction on the initial XGBoost detection outcomes, further enhancing the precision of LiDAR degenerate environment detection. The validity of the proposed method and its effectiveness in detecting LiDAR degenerate environments were verified through the collection of real scene data. [Results] Our results demonstrated the efficacy of components of the method in this paper. The success rate of LiDAR degenerate environment detection reached 94.41%, and the false detection rate of well-conditioned environment reached 1.24%. Compared to the LOAM degradation detection module, the detection success rate was improved by 10.91%, the false detection rate was reduced by 95.26%, and the detection efficiency was improved by 56.97%. [Conclusions] These findings highlight the achievement of efficient and precise LiDAR degenerate environment detection. The improved XGBoost algorithm in this paper is also applicable to other fields for improving data processing efficiency and accuracy. The LiDAR degenerate environment detection method proposed in this paper can not only be effectively applied to the LiDAR SLAM process but also play an important role in the fields of Geographic Information System (GIS) and Remote Sensing (RS).

[Objectives] Scene understanding based on 3D laser point clouds plays a core role in many applications such as object detection, 3D reconstruction, cultural relic protection, and autonomous driving. The semantic classification of 3D point clouds is an important task in scene understanding, but due to the large amount of data, diverse targets, and large-scale differences, as well as the occlusion of buildings and trees, this task still poses challenges. The existing deep learning models for point cloud classification face several challenges due to the unstructured and disordered nature of point clouds. These challenges include inadequate extraction of local and global features and the absence of an efficient mechanism for context feature integration, making it challenging to achieve fine-grained classification of ground objects. Therefore, this study introduces a novel point cloud feature classification approach that incorporates a multi-scale convolutional attention network for both local and global features. [Methods] To address the lack of structure in point clouds, we construct a local weighted graph to model the positional relationships between central points and their neighboring points. This graph facilitates dynamic adjustments of kernel weights, enabling the extraction of more representative local features. Simultaneously, we introduce a global graph attention module to account for the overall spatial distribution of points, address the disorder of point clouds, and effectively capture global contextual features, thereby integrating information at different scales. Furthermore, we design an adaptive weighted pooling module to facilitate the seamless fusion of local and global features, thus maximizing the network's classification performance. [Results] The proposed method is evaluated using the publicly available Toronto-3D point cloud dataset and a campus point cloud dataset obtained from real measurements. We compare its performance against various network models, including Pointnet++, DGCNN, RandLA-Net, BAAF-Net, and BAF-LAC, The experimental results show that the OA and MIoU of our method in the Toronto-3D dataset are 97.21% and 85.46%, respectively. Compared with network models such as Pointnet++, DGCNN, RandLA Net, BAAF Net, and BAF-LAC, OA has improved by 1.99% to 8.21%, and MIoU has improved by 3.23% to 35.86%. In the campus dataset, the OA and MIoU of our method in this paper are 97.38% and 85.70%, respectively. OA has improved by 0.58% to 10.53%, and MIoU has improved by 2.01% to 32.01%. [Conclusions] These results surpass those achieved by the comparison networks and effectively overcome problems such as large changes in target scale and building occlusion, establishing our method's capability to achieve high-precision and efficient fine classification of ground objects in complex road scenes.

[Objectives] Training a high-performance remote sensing scene classification model requires a massive amount of remote sensing data samples. However, it is usually challenging to ensure the security of data samples collected from different sources, due the risk of dataset poisoning. Furthermore, remote sensing datasets are often enormous, and there is a growing demand for large remote sensing models that require massive parameters and significant computational resources. As a result, distributed learning is developed for efficient model training. [Methods] On these grounds, this study proposes a backdoor defense method for remote sensing scene classification models through distributed contrastive learning. The remote sensing dataset is partitioned across different worker nodes for distributed training. Each node uploads its trained model parameters, including the weights of the feature extractor and predictor, to a server for parameter aggregation. This iterative process of node training and server aggregation continues until the model converges. During the training phase on worker nodes, a two-stage training process is executed. In the first stage, contrastive learning is performed on two augmented version of local dataset to train two feature extractors. One feature extractor participates in the global aggregation and updates, while the other only updates locally, ensuring that it remains unaffected by any backdoors that may be introduced during global model aggregation. In the second stage, the feature extractor is frozen, and the predictor is trained using local dataset and labels. Considering that, when model training stabilizes, the gradient update directions of aggregator weights show significant differences between nodes using a poisoned dataset and those using a clean dataset. Based on this observation, a server-side secure aggregation method is designed. During the server parameter aggregation phase, the server calculates the similarity between the predictor weights uploaded by each worker node and the aggregated weights from the previous round. If the similarity repeatedly falls below the threshold, the node is considered to contain poisoned data. The poisoned data is then flagged, and the node's aggregator weights are excluded from the aggregation process, ensuring secure aggregation. [Results] This study conducted experiments on three remote sensing scene classification datasets: EuroSAT, NaSC-TG, and PatternNet, and three widely used models: ResNet-50, VGG16, and GoogleNet. The experimental results demonstrated that the proposed method significantly reduced the average backdoor attack success rate from 99.77% to 1.36%, effectively defending against five typical backdoor attacks: BadNets, Blended, SIG, Trojan, and WaNet. [Conclusions] This study provides effective theoretical and methodological support for backdoor defense in the training process of remote sensing scene classification models.