Intrusion Detection Systems (IDS) play a crucial role in network security defense.However, a significant challenge for IDS in training detection models is the shortage of adequately labeled malicious samples.To address these issues, this paper introduces a novel semi-supervised framework GANGRL-LLM, which integrates Generative Adversarial Networks (GANs) with Large Language Models (LLMs) to enhance malicious code generation and SQL Injection (SQLi) detection capabilities in few-sample learning scenarios.Specifically, our framework adopts a collaborative training paradigm where: (1) the GAN-based discriminator improves malicious pattern recognition through adversarial learning with generated samples and limited real samples; and (2) the LLM-based generator refines the quality of malicious code synthesis using reward signals from the discriminator.The experimental results demonstrate that even with a limited number of labeled samples, our training framework is highly effective in enhancing both malicious code generation and detection capabilities.This dual enhancement capability offers a promising solution for developing adaptive defense systems capable of countering evolving cyber threats.

Although Multimodal Large Language Models (MLLMs) have demonstrated remarkable achievements in recent years, they remain vulnerable to adversarial examples that result in harmful responses. Existing attacks typically focus on optimizing adversarial perturbations for a certain multimodal image-prompt pair or fixed training dataset, which often leads to overfitting. Consequently, these perturbations fail to remain malicious once transferred to attack unseen image-prompt pairs, suffering from significant resource costs to cover the diverse multimodal inputs in complicated real-world scenarios. To alleviate this issue, this paper proposes a novel adversarial attack on MLLMs based on distribution approximation theory, which models the potential image-prompt input distribution and adds the same distribution-fitting adversarial perturbation on multimodal input pairs to achieve effective cross-image/prompt transfer attacks. Specifically, we exploit the Laplace approximation to model the Gaussian distribution of the image and prompt inputs for the MLLM, deriving an estimate of the mean and covariance parameters. By sampling from this approximated distribution with Monte Carlo mechanism, we efficiently optimize and fit a single input‑agnostic perturbation over diverse image‑prompt pairs, yielding strong universality and transferability. Extensive experiments are conducted to verify the strong adversarial capabilities of our proposed attack against prevalent MLLMs spanning a spectrum of images/prompts.

In order to implement the State Council and the MEE's requirements on strengthening the integration and sharing of government information resources, this paper is to explore a unified management method of multi-source environmental information based on GeoSOT geographic grid. Since more than 90% of the environmental information has spatial attributes, based on the spatial position identification, it can realize the unified data management, relevant retrieval and spatial correlation analysis inside environmental protection department and around relevant ministries. Multi-scale spatial grid identification can also meet the requirements of different levels of management from macro to fine. The unified identity and management scheme of multi-source environmental data is expected to provide basic technical support for the integration of environmental protection information.

With the rapid development of Earth observation, remote sensing and geographical information system, people can obtain massive spatial data. The data people use is not yet limited to a local area, which gradually expands to the whole earth. Growth in the amount of data makes how to manage the global spatial data and how to supply user efficient data retrieval service becoming a key issue for academic community. This thesis proposes a spatial data index and retrieval method based on GeoSOT global discrete grid system. We use GeoSOT grid code in creating spatial data identification. Then the data can be managed in ordinary one-dimension database based on its GeoSOT identifications. This method can realize a unified management of global spatial data. In another aspect, it also proposes a solution for cross-database associative spatial data retrieval.

Roads are traffic lifelines for emergency rescue and disaster relief. After major earthquakes, it is very significant to extract road damage rapidly and accurately in disaster areas by remote sensing for emergency rescue. Because road damage caused by earthquake is ever-changing,there is no common spectral characteristic of it in remote sensing images. Meanwhile, there are many phenomena of “synonyms spectrums” and “different spectrum characteristics with the same object” in remote sensing images. Thus, traditional methods by spectrum characteristics are usually with low accuracy and not universal. This paper proposes an automatic approach to extract road damage rapidly based on sidelines using high resolution satellites images and road maps. Road sideline is one of stable geometric features in both pre-earthquake and post-earthquake images, and the change of road sideline is a remarkable evidence of road damage exists. The approach firstly extracts sidelines of undamaged road from images acquired after earthquakes, and then these road sidelines are compared with the road lines before earthquakes supplied by road maps. The damaged segments can be extracted through comparison. The performance of the method is evaluated by an experiment with QuickBird images in the WenChuan earthquake disaster area.

With the development of Earth observation technology, the traditional local area research methods are facing great challenges. How to effectively manage and use the massive spatial data has become a key issue for academic and applied science community. Global discrete grid system divides the earth's surface into a series of different scale regions (grids) according to some rules. Based on global discrete system, a new global multi-scale data model can be established. This thesis analyzes the current scientific research and commercial global grid systems, such as the platonic solid discrete grid systems, World Wind, Google Earth and Terraserver. According to its advantages, global discrete system may become a feasible option of global data management.

After the occurrence of natural disasters, objective and accurate assessment of road damage is the key to emergency rescue. However, in the traditional road damage assessment, the assessment factor and the form of assessment result are single. It can only give an overview of damaged roads, but can't further provide their traffic capacities, which are essential for the disaster relief department. A new approach of road damage assessment is presented in this paper. The damage levels are graded based on the magnitude of actual traffic capacities of post-disaster roads, and the detailed damage information is output at three levels - damage block level, single-road level and region level. The proposed method is applied in the damage assessment of roads in Wenchuan County, China, which was hit by an 8.0-magnitude earthquake. The experimental result shows the validity of this method.

In high spatial resolution satellite images, shadows are usually cast by elevated objects such as buildings, bridges, and towers, especially in urban region. Shadows may cause loss of feature information, false color tone and shape distortion of objects, which seriously affect the quality of images. Hence, it is important to segment shadow regions and restore their information for image interpretation. This paper presents an effective and robust approach for shadow segmentation and compensation in color satellite images with high spatial resolution. The approach uses normalized saturation-value difference index (NSVDI) in hue-saturation-value (HSV) color space to detect shadows and exploits histogram matching to recover the information under shadows. Experimental results by applying the proposed approach in the IKONOS color images of urban area demonstrate the effectiveness and feasibility of the proposed approach.

Common methods for DEM generation are based on collinearity equation, and rational function. Either of them require registered stereo pair as input. In order to register each point precisely, this paper proposes a corresponding image object constrained registration method based on object oriented image segmentation. Compared with current methods, the proposed method provides one geometric image transform (GIT) function for each pair of building roof because one GIT could fit well for just one roof. Object oriented segmentation not only helps us set the effective area for each GIT, but also provide us enough feature points from the boundary of a segmented building points to fit the coefficients of its GIT. The result by applying the proposed method on IKONOS stereo pair demonstrates the feasibility of generating building DSM with acceptable accuracy.

Research on road extraction from digital imagery is motivated by the need for data acquisition and update for geographic information systems (GIS). Roads usually have parallelism of road sides, and on the images, especially the edge map, there are dual edges for each road. In this paper, we propose an approach for automatically extracting road from ETM panchromatic image with a resolution of 15 meters based on Dual-Edge Following. Our approach uses the edge detector with embedded confidence (EDEC, Peter Meer, 2001) to detect road edge, then traces road to generate road candidates by Dual-Edge Following, next exploits the perceptual organization based on probability to link the road segments. Dual-Ddge Following use edge information of both road sides to search for road segments which can improve the precision of road segments. The experiment with ETM panchromatic image in Xinjiang, China shows the validity of the approach.