To address the high rates of missed detections and false alarms, as well as the poor robustness in fatigue driving detection for open-pit mine truck drivers, a fatigue driving detection model for mine truck drivers (EBS-YOLO) based on the improved YOLOv8 is constructed to enhance the overall performance of fatigue detection. Firstly, YOLOv8 was used as the basic model for fatigue detection, and a small target detection layer was added to enhance the model's attention to small targets. Secondly, the bottleneck attention module (BAM) was used to improve the model performance to extract small target features, especially eye features. Finally, all cross-stage aggregation modules (C2f) in the backbone network were replaced with efficient multi-scale attention (EMA) modules, thereby effectively reducing model parameters and computational overhead to meet the requirements of a lightweight model. The results showed that the improved YOLOv8 model had a great detection effect with the accuracy, recall rate, and average detection accuracy reaching 93.6%, 93.9%, and 96.5%, respectively, and the memory size of the model was only 4.9 MB. Compared with the YOLOv8 model, the improved model can quickly and accurately identify the fatigue state of mining truck drivers, meet real-time requirements, and effectively prevent fatigue-driving accidents.

To improve the interaction ability of traffic vehicles in the cut-in scenario，a method for constructing a vehicle hazardous cut-in strategy based on deep reinforcement learning was proposed. Firstly，a simulated environment was built based on scalable multi-agent reinforcement learning training school(SMARTS) simulation platform. Then，twin delayed deep deterministic policy gradients (TD3) algorithm was adopted to train an agent to cut in a randomly chosen target vehicle hazardously. The algorithm was compared with proximal policy optimization (PPO) and deep deterministic policy gradient (DDPG) algorithms. The trained model was tested in seven different scenarios with varying traffic densities. Finally，a multi-agent testing environment was built，and the trained model was applied to validate intelligent driving strategies. The results show that the success rate of hazardous cut-ins reaches 80.35% in model training with TD3 algorithm，outperforming both comparative methods. In model testing，except for the 2 700 vehicle/h test scenario，the model achieves a hazardous cut-in success rate of over 80% in the other three test scenarios that were not used in training，demonstrating good generalization ability. Meanwhile，the time to collision values between the ego vehicle and the target vehicle at the moment of lane changes are concentrated within the range of 0 to 6 seconds，with 95% falling within this bracket. The proportions of time to collision values in the intervals of (0，2]，(2，4]，(4，6]s are 60%，30%，and 5% respectively，covering test conditions with different collision risk. In the validation of intelligent driving strategies，the traffic vehicle controlled by the trained model can actively perform cut-ins in front of the test vehicles，exposing it to the risk of a rear-end collision and helping in identifying safety vulnerabilities in intelligent driving strategies.

To avoid worsening the consequences of oil and gas pipeline accidents due to emergency failures，the causes of emergency failure in 27 accidents at home and abroad were analyzed using the HFACS model. Based on the results of grounded theory (GT) statistical coding analysis，a classification model of failure causes of emergency response in oil and gas pipeline accidents was proposed. SNA was used to develop the relationship network of the causes of emergency failures in oil and gas pipeline accidents. The core-periphery，centrality，and association direction index analyses were used to identify core factors and factors with high association and strong mediating roles in the classification model of the causes of emergency failures in oil and gas pipeline accidents. The results indicated that the classification model of emergency failure causes in oil and gas pipeline accidents was divided into five levels: government and emergency department factors，operator organizational factors，operator unsafe supervision，preconditions for unsafe behavior of on-site personnel，and unsafe behavior of on-site personnel. The emergency failure causes were further divided into 16 bottom-level factors，among which there were 9 core factors: inadequate safety supervision by government and emergency departments，ineffective emergency rescue，regulations defects，insufficient supervision by pipeline operators，technical environment，and skill errors. Skill errors，regulations or procedure defects，technical environment，and insufficient supervision by operators were highly associated factors. Moreover，pipeline operators' regulation defects，procedure defects，technical environment，insufficient supervision，improper resource management，and decision-making errors were strong mediating factors.

To prevent accidents and disasters, such as the collapse of existing roads nearby, caused by deep and large foundation pit excavation and support, it is necessary to ensure the safe operation of adjacent roads during the construction of deep and large foundation pits. Taking a deep and large foundation pit project near a city trunk road in Kunming as an example, on the basis of in-depth research at the site, the three-dimensional finite element numerical simulation software-new eXperience of GeoTechnical analysis system(MIDAS GTS NX) is used for simulation and calculation, and combined with on-site monitoring, to analyse the force and deformation characteristics of the foundation pit support structure and the deformation characteristics of the adjacent existing road under the existing excavation and support scheme. The study results indicated that the forces and deformations of the supporting piles and anchor cables were within design limits after the foundation pit excavation and support were complete. The displacement near the existing road increased with the increase of pit excavation depth, and become stable after the excavation was completed. The maximum deformation of the existing road occurred at a position 2.5 times the excavation depth from the foundation pit boundary. Furthermore, the deformation did not reach the alarm threshold for road displacement caused by foundation pit excavation. Thus, the existing support scheme can ensure the safety of both the foundation pit and the adjacent road.

In order to solve the problems of complex structure, large scale and difficulty in balancing detection accuracy and efficiency of the current forest fire detection algorithm based on deep learning, a lightweight forest fire detection algorithm based on YOLOv5s was proposed. Firstly, an optimized background difference technique was used to eliminate the interference of fire-like objects in the background image, thus reducing the time required for image analysis. Secondly, a group blending strategy was designed to optimize the conventional convolution, and an efficient channel attention (ECA) mechanism and depthwise separable convolution were incorporated into the C3 module of feature extraction, which enhanced the ability of image feature extraction and fusion and at the same time effectively reduces the number of model parameters. Then, a dynamic non-monotonic focusing mechanism was used to optimize the WIOU loss function, reducing the harmful gradients generated by low-quality samples. Finally, sufficient experimental comparisons between the proposed algorithm and other algorithms on the constructed forest fire dataset. The results show that the proposed algorithm shows good generalization in various scenarios, and the detection accuracy of the flame target can reach 86.1%, which is 2.7% higher than that of the standard YOLOv5s, and the detection speed is increased by 11.4%, which effectively reduces the fire false alarm rate and enhances the detection performance of the model.

To ensure that the equipment support system of professional emergency rescue teams could meet the requirements of rescue tasks and gradually adapt to complex and variable disaster risks, a text mining method was applied to analyze the factors influencing the equipment support capacity of emergency rescue. Based on technical personnel support capability, equipment resource support capability, equipment and facility support capability, information resource support capability and management system support capability, an assessment index system for emergency rescue equipment support capacity was developed. To reduce the impact of fuzziness, randomness, and subjective-objective bias on assessment results, a combined weighting method was adopted to determine the weights of each assessment index. A comprehensive assessment method was established using the matter-element extension model and the integrated cloud model. Professional emergency rescue team A was selected as an example for application to verify the scientific validity and effectiveness of the model. The results show that the index system comprehensively and accurately reflects the overall level of emergency rescue equipment support capability of professional teams. The assessment model reasonably and effectively assesses the capability level and identifies weaknesses in the current equipment support system, providing improvement points and theoretical support for the development of the team's equipment support system.

In order to reduce the losses caused by landslide disasters, taking Changde city of Hunan province as an example, the fractal theory and information method were applied to evaluate the regional landslide susceptibility based on field investigation and historical landslide data. The sensitivity of influence factors was quantitatively studied by fractal theory. The information values of each secondary impact factor were obtained by using the information method, and the comprehensive information values were obtained by combining the fractal dimension value and the information value. Based on the values, the susceptibility zoning of the study area was carried out. The results show that the slope, engineering geological rock group, elevation and vegetation are the influencing factors that have a second-order cumulative and fractal distribution with the landslide, while other influencing factors have a first-order cumulative and fractal distribution with the landslide. The areas of very low, low, medium, high and very high susceptibility areas respectively account for 5.24%, 8.84%, 35.06%, 39.21% and 11.65% respectively. Annual rainfall greater than 1 600 mm, slope of 20-30° and elevation of 900-1100m are important factors.

In order to improve the evacuation efficiency of the teaching building, the classroom structure was optimized through control experiments and numerical simulations to enhance evacuation efficiency. Emergency evacuation tests were used to obtain the movement characteristics of students aged 6-7 years old. And Pathfinder simulation software was used to study the impact of desk layout, classroom door position, and exit position on evacuation. The results indicate that for a single classroom, although shortening the pre-action time can reduce the overall evacuation time, it cannot improve the congestion caused by the building structure. Appropriate evacuation routes and desk layouts can significantly reduce evacuation time. For buildings with classrooms on one side of the corridor, increasing the width of the corridor and exit is the most effective way to improve evacuation efficiency. For buildings with classrooms on both sides of the corridor, the structure of the evacuation corridor inside the building, including the number of corridors and the intersections inside the corridors, is the most important factor affecting evacuation time. Therefore, it is recommended to develop optimization plans for classroom evacuation structure from different aspects.

A thermal safety warning system was established based on a thermoregulation model to mitigate the thermal safety risk for elderly people under high-temperature conditions. Firstly, the improved model suitable for the elderly was established by replacing the physical model and adjusting the active and passive systems in the classical model. The improved model was validated using publicly available experimental data. Secondly, the model simulated the temperature changes in elderly people in high-temperature environments. Statistical analysis was used to assess the impact of various parameters on thermal safety. Finally, based on the analysis results and the improved model, a thermal safety risk warning system for elderly people was developed. The warning system was demonstrated through a case study. Results indicate that the improved model accurately simulates the body temperature of elderly people, with a root mean squared error less than 0.12 ℃. Physical activity intensity significantly impacts thermal safety, with a standardized regression coefficient β larger than 0.8. As heat exposure time increases, the impact of activity intensity on thermal safety is decreased (β decreases from 0.945 to 0.806), while the influence of environmental factors is increased (β of temperature and humidity increases from 0.249 and 0.137 to 0.370 and 0.348). In the case study, the safe duration of continuous activities for the resting and working elderly people in Baoding/Hong Kong is 172/175 minutes and 108/122 minutes, respectively. The highest thermal safety risk period for elderly people on that day is between 17:00 and 18:00.

An improved object detection algorithm for high-consequence areas was proposed to solve the problems of the sensitive and complex external environment of the overseas section of the China-Myanmar oil and gas pipeline, difficulty in manual inspection, and high-risk factors. Firstly, a convolutional block attention module was used to adaptively learn channel and spatial attention to enhance the network's perception and generalization capabilities. Then, focal and efficient intersection over union(Focal-EIoU) loss was used to comprehensively consider the target features and their associations to deal with the issues of class imbalance, reduce the interference of easy-to-classify samples, and enhance the robustness of the model. Finally, the improved model was used to intelligently recognize regional attributes of China-Myanmar oil and gas pipeline remote sensing images. Furthermore, the proposed YOLO model was validated against related ablation experiments. The results showed that for the feature recognition of remote sensing images of the China-Myanmar oil and gas pipeline, the proposed model reached a mean average precision (mAP) of 68.2% for the field, green space, settlement, and river. The model performance was improved by 29%, 21.6%, and 10.7% compared with YOLOv5, YOLOx, and YOLOv8, respectively.