To prevent miners from mistakenly entering dangerous areas，a lightweight underground miner object detection model based on YOLOv5s-MPD was proposed，which combined with depth camera to locate miner targets and detect whether miners had entered dangerous areas in real time. Specifically，the MobileNetv3 lightweight neural network was used as the backbone feature extraction network to significantly reduce the model size. Secondly，Polarized Self-Attention (PSA) module was introduced to enhance the perception of targets. Finally，Deformable Convolution Network v2 (DCNv2) was used to replace the standard convolution in the C3 module of the feature fusion layer，solving the problem of partial feature information loss in conventional convolution. The improved model was used in combination with the color images obtained by the depth camera to detect miner targets and obtain the spatial three-dimensional coordinates of the target center points. The results show that compared with YOLOv5s，the improved model reduces the number of parameters and computation by 83.54% and 77.03%，respectively. The model size is only 3.4 MB，and a detection speed of 70.2 f/s，which is increased by 54.97%. The mean average precision is 0.825. Compared with mainstream object detection models，the improved model has a more balanced number of parameters，computation，model size，detection speed，and mean average precision. In the actual positioning accuracy test，within a range of 1-8 meters，the average absolute error and average relative error of the distance between the camera and the miner target were 0.11 meters and 1.74%，respectively. The maximum absolute error and maximum relative error were 0.25 meters and 2.96%，respectively. In the dynamic detection，the miner target could be detected and its location information output，with a detection success rate of 97.5%.

In order to improve urban governance and promote sustainable development，the resilience measurement of core cities in Chinese mainland was analyzed based on panel data from 25 core cities (municipalities directly under the central government，provincial capitals，and regional capitals) between 2011 and 2020. Technique for order preference by similarity to an ideal solution(TOPSIS)-Entropy Weight Method was applied. The resilience situation for 2026 and 2029 was predicted using a BP neural network model. This research aimed to explore the dynamic spatial differentiation of regional resilience. The results show that the standard deviation of the resilience index across cities fluctuates around 0.180，with the resilience disparity between cities remaining relatively stable. However，some cities show a downward trend in their resilience index year by year. The standard deviation of the predicted resilience index for 2026 decreases to 0.173，indicating a reduction in the resilience disparity between cities and a narrowing of the resilience gap. In the four time points of 2014，2020，2026，and 2029，the spatial heterogeneity of urban resilience evolves relatively stably. The urban resilience rankings are as follows: Eastern region > Central region > Western region > Northeastern region. Among them，the economic and infrastructure resilience in the Eastern region is the highest，while the social and ecological resilience in the Central region is the highest.

In order to further study the detonating effect of UAV，and improve the fire extinguishing efficiency of fire bombs，ANSYS Workbench software was used to establish a finite element model of fire bombs，and the explosion and dispersion process of fire extinguishing agent was simulated to clarify the impact of different detonation heights on the dispersion characteristics of fire extinguishing agent. Meanwhile，full-scale experiments on the vertical dropping of fire extinguishing bombs by unmanned aerial vehicles at different detonation heights were conducted. The results show that the simulation data such as the explosion process of fire extinguishing bomb，the throwing process of fire extinguishing agent and the spreading radius of fire extinguishing agent are in good agreement with the full-scale test results. After the fire extinguishing bomb explodes，the extinguishing agent disperses in the air in a cone shape and spread evenly in the horizontal direction，and the spreading shape is roughly circular. With the increase of detonation height，the landing time of fire extinguishing agent increases，and the horizontal velocity of fire extinguishing agent decreases. The increase of landing time plays a dominant role relative to the decrease of horizontal velocity，resulting in the increase of the spreading radius of fire extinguishing agent with the increase of detonation height. When the detonation height of the fire extinguishing bomb increases from 5 to 12 m，the numerical simulation data of the fire extinguishing agent spread radius increases from 2.04 to 3.56 m，and the error between the numerical simulation data and the experimental data is within 5%.

During the attack of public health emergencies，the surge in material demand in isolated communities leads to a series of problems，i.e.，serious shortfalls in supply，fairness of distribution and special distributional needs. To address this issue，a multi-period，multi-type，cross-regional emergency material dynamic allocation problem was studied based on a hierarchical distribution network of "material supply point-distribution center-isolated community". Especially，the demand time windows were applied to characterize the realistic requirements of isolated communities for the delivery time，and the special needs of some vulnerable groups，e.g.，aging populations，were considered. Thus，a multi-objective non-linear dynamic program was developed by integrating the three dimensions of considerations：economy，material distribution satisfaction rate and distribution equity. In view of the multi-objective structure，the Epsilon constraint algorithm was used to solve the proposed model，and the material allocation during the attack of public health emergencies in Shanghai was taken as a case study to analyze the Pareto front of the optimal solutions and conduct the sensitivity analysis on some key parameters in the model. The results show that，under the constraint of limited emergency material supply，the material allocation strategy considering the aging populations helps to prioritize the material supply of vulnerable groups. However，it also contributes to an increase in the unfairness rate of the isolated communities as a whole. For the material allocation strategy considering the delivery time windows of isolated communities，the total operation cost of the cross-regional distribution network and the unmet rate of the material demand in the isolated communities are high，but the overall distribution of supplies in the system is relatively fair，and it could better meet the needs of isolated communities in terms of delivery of supplies，which provides a useful reference for the decision makers to balance the individual demand and the overall interests.

To improve the accuracy of human thermal injury assessment and protect rescuers' safety in thermal radiation environments，a coupled heat and mass transfer model of skin-microenvironment-firefighting clothing system was proposed to predict skin burn injuries under dynamic conditions. Based on mechanism of heat and moisture transfer in porous media，the periodic movement of fabric caused by human activities and its impact on heat and mass transfer in the skin-microenvironment-firefighting clothing system were considered. Furthermore，the proposed model was used to simulate skin temperature，time of skin burn，and the distribution of temperature and humidity in the fabric layers for both dry and wet cases in real time. The results show that the relative error between simulated values predicted by the model and the experimental measurements presented in the literature is only 3.79%. When exposed to 8.5 kW/m² thermal radiation environments，the time to second-degree burn for the dry case is 33.7 s earlier than that for the wet case. When firefighters approach a 20 kW/m² radiant heat source at a speed of 1 m/s，the heat transferred is impeded by the increase in thermal layer thickness. This extends the time for second-degree burns to occur by 10.9 s and reduces the heat absorbed by the skin surface by 20%. When the air gap thickness in the microenvironment is the same as the amplitude of the periodic motion of fabric，the skin temperature increases rapidly and fluctuates significantly，and the time to second-degree burn occurs 43.7 s earlier. Human body movement and moisture in fabric layers affect heat transfer process between the human body and thermal environment，thereby their impact on the accuracy of rescue assessments cannot be ignored.

In order to enhance emergency management in the field of gas pipeline networks，Gas-kBERT model was proposed. The model incorporated data from the gas pipeline network field expanded by Chat Generative Pre-Trained Transformer，(ChatGPT)and Chinese Gas Language Understanding Subject-Predicate-Object(CGLU-Spo) and related corpora were constructed in this field. By altering the model's masking (MASK) mechanism，domain knowledge was successfully injected into the model. Considering the professionalism and specificity of the gas pipeline network field，Gas-kBERT was pre-trained on various scales and contents of corpora and fine-tuned on named entity recognition and classification tasks within this field. Experimental results demonstrated that，compared to the general BERT model，Gas-kBERT exhibited significant performance improvements in F₁-score in text mining tasks in the gas pipeline network field. Specifically，in the named entity recognition task，the F₁-score was increased by 29.55%，and in the text classification task，the F₁-score improvement reached up to 83.33%. This study proves that the Gas-kBERT model performs exceptionally well in text mining tasks in the gas pipeline network field.

In order to improve the safety cognition level of underground caverns construction workers，safety requirement and safety capability were introduced as mediating variables，and occupational burnout was used as a moderating variable to construct a moderated chain mediation model. A structured questionnaire was designed using 5 scales: risk perception，safety requirement，safety capability，occupational burnout and safety cognition. A questionnaire survey and data analysis were conducted on 312 underground caverns construction workers，and SPSS 26.0 and AMOS 26.0 software were used to test the mediating effect and moderating effect.The results show that risk perception directly and positively affects the safety cognition of underground cavern construction workers，while occupational burnout plays a negative moderating role between the two. Risk perception also affects the safety cognition of construction workers through the independent mediating effect of safety requirement and safety capability，as well as through the chain mediating effect of safety requirement and safety capability. Therefore，motivating and improving the risk perception，safety requirement and safety capability of construction workers，while reducing their occupational burnout，can effectively improve their safety cognition level.

In order to examine how construction workers' safety motivation affects their behavior，SDT was used to divide the safety motivation into two types: autonomous and controlled. Subsequently，a hypothetical model linking workers' safety motivation，opportunity，ability，and behavior was constructed by integrating the analytical framework of the AMO theory. Finally，469 frontline construction workers were taken as the research subjects for data collection，and structural equation modeling (SEM) was used for hypothesis testing and empirical analysis. The results show that the autonomous safety motivation of construction workers is significantly positively correlated with safety behavior，and the controlled safety motivation of construction workers is significantly negatively correlated with safety behavior. Safety motivation and safety competence play a positive moderating role between workers' autonomous safety motivation and safety behavior，while they show a significant negative moderating relationship between workers' controlled safety motivation and safety behavior.

To reduce the safety risk of fireworks production enterprises and eliminate the subjectivity and static limitations of the current safety evaluation mechanism，an evaluation indicator system of the safety risk for firework production enterprise was constructed based on the grounded theory (GT). The evaluation values and weights of the evaluation indicators were calculated by applying the frequency-based Analytic Network Process-Back Propagation Neuron Network (ANP-BPNN) model. The dynamic evaluation of the safety risk level of firework production enterprise was achieved based on SD，and verified with an enterprise involved in a major fireworks explosion accident of fireworks. The results show that the dynamic evaluation model of safety risk for fireworks production enterprises can accurately capture the development trend of the safety risk. The interval between safety risk evaluations of fireworks production enterprises is flexibly adjusted based on predicted time for safety risk level to reach the higher risk. The continuous dynamic evaluation of safety risk can ensure that the enterprise always maintains a safe production state.

Taking improving the safety and efficiency of tower crane operation as an example，a method of integrated identification of fatigue state and unsafe behavior was proposed in order to detect the potential safety hazards of drivers in real time. A live video stream was captured via a camera，and the video was analyzed and pre-processed to extract critical information for identifying subsequent fatigue and unsafe behavior. In terms of fatigue state recognition，the analysis method based on the state of eyes and mouth was used to monitor the physiological indicators such as the state of eyes opening and closing，the blink frequency and yawn frequency. In terms of unsafe behavior identification，computer vision and deep learning technology were combined to detect the potential dangerous operations of drivers in real time，thus ensuring timely detection of safety risks. The results show that the performance of the optimized YOLOv5-ECA(Efficient Channel Attention) model is significantly improved in fatigue state and unsafe behavior recognition. The accuracy rate and recall rate of the model on the test set are more than 90%，showing good recognition ability.