The supply of emergency rescue materials plays a pivotal role in post-earthquake relief operations. However, varying disaster scenarios generate differentiated demands for emergency resources, where factors such as seasonal impacts, geographic environment of affected areas, population density, and requirements of emergency response protocols significantly influence the categories and quantities of required supplies. Current research predominantly focuses on essential materials, while determinations of scenario-specific material demands rely heavily on expert empirical inference, resulting in discrepancies between model predictions and actual requirements. This study proposes a reasoning-enhanced large language model (LLM) framework integrated with Retrieval-Augmented Generation (RAG) technology for post-earthquake emergency material demand prediction. By simulating expert decision-making processes, we construct an emergency knowledge base amalgamating standardized protocols, historical seismic case data, and disaster scenario characteristics extracted from digital resources. The framework employs RAG to enhance domain-specific knowledge integration within the reasoning model, utilizing Chain-of-Thought generation to produce differentiated prediction schemes that specify material categories and per-capita demand metrics. Through dynamic updating of population statistics and material requirements via post-disaster network information monitoring, the system achieves real-time demand prediction based on evolving victim counts and individualized allocation parameters. Validation through expert evaluations using the 2013 Ya’an Lushan earthquake and simulated disaster scenarios demonstrates effectiveness, with successful practical implementation observed in the January 7, 2025, Dingri M6.8 earthquake case.

To address the problems of the text of earthquake emergency information keeps changing incrementally with the time of an earthquake’s occurrence and there being more and more information categories, thus making it difficult to identify earthquake emergency key information, this paper proposes an intelligent recognition algorithm of earthquake emergency information based on the optimized BERT-BiLSTM-CRF algorithm. Based on the historical seismic emergency information dataset of the past 10 years, first, the BIO sequence labeling method is used to classify the seismic entities, and the BERT pretraining model is constructed to represent the seismic emergency text with sentence-level feature vectors. The BiLSTM algorithm is used to obtain the contextual information of the bidirectional seismic emergency text, and we introduce the attention mechanism to enhance the recognition effect of the seismic emergency key information in the statements. Finally, we use conditional randomization to enhance the recognition of earthquake emergency key information in the utterance. The conditional randomization algorithm is applied to extract the dependency relationship between adjacent vectors and improve the accuracy identification to realize the intelligent recognition of earthquake emergency information. The experimental results show that our model can extract earthquake emergency information from online media efficiently and accurately, with better performance than other baseline models.

To enhance the effective use of teaching video information in digital campuses, provide timely and accurate assessment of students’ classroom learning status, and offer technical support for teaching management, this paper introduces an improved SSD (Single Shot MultiBox Detector) object detection algorithm for intelligent analysis of students’ real classroom behavior. This article examines classroom teaching videos from four months of a semester at a certain university, collecting over 10,446 1080p video frames to create a dataset for annotation and training. To address the poor detection performance of the SSD model when identifying four behavior states—listening, sleeping, playing, and writing—of students in combined classes due to object occlusion and complex classroom environments, this paper introduces ResNet residual modules and a dual attention mechanism into the VGG-16 network. This enhancement improves the model’s ability to detect student states in video frames under conditions of object occlusion. An FPN (Feature Pyramid Network) fusion model is constructed to address the low recognition accuracy of the SSD model for small objects in the back rows, thereby improving image recognition accuracy. The experimental results show that the improved SSD model has improved feature extraction ability and small object recognition accuracy compared to the original SSD model, with mAP increasing from 51.62% to 63.75%. The research results of this project can provide real-time information reference for university teaching management, help teachers understand students’ learning status promptly, and adjust teaching methods on time to improve teaching quality and classroom effectiveness.

To address the challenges of semantic parsing of multi-source heterogeneous information and the delayed emergency response decisions caused by insufficient relational reasoning capabilities in earthquake emergency management, this study proposes a domain knowledge extraction method for earthquakes based on a large language model combined with a three-level prompt engineering system (TPES-LLM) of "instruction fine-tuning - demand awareness - case matching."The method deploys a local large language model using LangChain+QWEN2.5-7B, integrates earthquake domain knowledge through LoRa fine-tuning based on earthquake experts’ classifications and industry standards, and injects seismic knowledge into the model. The multi-head attention mechanism weights are optimized based on the co-occurrence frequency of historical earthquake entities, and demand-aware knowledge identifies key textual features that significantly impact knowledge extraction. Training is performed on 36 known earthquake disaster events to learn the association patterns of entities, relationships, and events hidden within the earthquake case data for case matching. This method significantly enhances the accuracy of entity recognition and the efficiency of relation extraction for complex disaster-related texts.Additionally, a bidirectional graph attention network (Bi-GAT) is designed to enable bidirectional propagation and dynamic aggregation of node features. The path confidence constraint algorithm (PCCA) is used to achieve deep semantic associations of earthquake disaster elements. Based on the Neo4j graph database, an earthquake emergency knowledge graph is constructed. Experimental results from real earthquake events such as the 2022 Luding 6.8-magnitude earthquake, the 2024 Jishishan 6.2-magnitude earthquake, and the 2025 Dingri 6.8-magnitude earthquake show that the accuracy of intelligent Q&A retrieval for the earthquake emergency knowledge graph reaches 89.62%, 87.28%, and 90.23%, respectively.The earthquake emergency knowledge graph based on large language models constructed in this study provides intelligent decision support for earthquake emergencies, with significant application value.

In this paper, according to the status of China's need to improve the ability of earthquake disaster prevention, this paper puts forward the implementation plan of earthquake disaster prediction system of Langfang city based on GIS. Based on the GIS spatial database, coordinate transformation technology, GIS spatial analysis technology and PHP development technology, the seismic damage factor algorithm is used to predict the damage of the city under different intensity earthquake disaster conditions. The earthquake disaster prediction system of Langfang city is based on the B / S system architecture. Degree and spatial distribution and two-dimensional visualization display, comprehensive query analysis and efficient auxiliary decision-making function to determine the weak earthquake in the city and rapid warning. The system has realized the transformation of the city's earthquake disaster reduction work from static planning to dynamic management, and improved the city's earthquake and disaster prevention capability.

After a natural disaster, it is crucial to locate the affected areas and assess building damage for emergency rescue operations. Automatic analysis technology based on remote sensing imagery improves the effectiveness of building damage assessment. However, existing methods face challenges in determining building damage levels across different types of disasters owing to complex appearance features in remote sensing images. These methods for extracting building damage features are insufficient and lack attention to subtle differences between some visually similar categories. Therefore, this study proposes a multi-level damage assessment universal framework (MLDA) for mixed-hazard buildings. MLDA designs a global spatial feature guidance (GSFG) module to extract high-level semantic features. GSFG pays more attention to the spatial deformation of buildings by long-distance context modeling. It extracts common features of multiple damaged buildings by calculating the deformation similarity under a global perspective, accurately determining their damage levels without considering the impact of the type of disaster. Then MLDA constructs a damage change feature attention (DCFA) module to enhance heterogeneous features between pre- and post-disaster images. DCFA identifies subtle differences between different damage levels in the process of multi-scale feature fusion by channel and spatial dual attention mechanisms. Compared with state-of-the-art methods, MLDA assesses the damage level of mixed disaster-stricken buildings effectively. Especially for minor damaged buildings, the accuracy rate improved by about 2.3%. The source code is available at https://github.com/DongzheHan-cv/MLDA.