Abstract: To improve the positioning and mapping accuracy of inspection robots in the complex environment of screening buildings in open-pit coal mines, which is characterized by dense equipment and frequent dynamic occlusions, and to solve the problems of positioning drift and map distortion caused by insufficient sensor data processing accuracy, poor front-end pose fusion performance, and low computational efficiency of the traditional Cartographer algorithm in such scenarios, an improved Simultaneous Localization and Mapping (SLAM) algorithm based on multi-level and multi-sensor fusion is proposed. Firstly, the Extended Kalman Filter (EKF) technique is used to preprocess the raw angular velocity and acceleration data of the Inertial Measurement Unit (IMU), optimize its attitude solution process, effectively suppress sensor noise and bias, and thus obtain more accurate and reliable robot pitch, roll and yaw attitude information. Secondly, an EKF-based pose fusion module is designed at the SLAM front end to perform tightly coupled fusion of the optimized IMU attitude data with the displacement information provided by wheel odometry, which provides higher-precision initial pose estimation for LiDAR scan matching and improves the success rate and efficiency of matching. Finally, aiming at the variability and complexity of the environment, a system-level fusion framework is constructed to fuse LiDAR scan data, optimized IMU pose, odometry information and loop closure detection information. Global graph optimization is carried out using the Ceres solver, forming a redundant and complementary perception system. To comprehensively verify the algorithm performance, mapping experiments were carried out in a long corridor environment simulating the structure of an open-pit coal mine screening building. Ten feature points were selected for measurement, and the average relative error of the map constructed by the improved algorithm was 0.432%, much lower than 2.042% of the traditional algorithm. Finally, deployment verification was conducted in a real open-pit coal mine screening building scenario. The real distances between 40 groups of manually marked points were measured by a high-precision total station and compared with the corresponding distances in the map constructed by the algorithm. The average absolute error was no more than 2 cm, and ghosting and distortion of the map were effectively eliminated. The results show that the proposed improved algorithm enhances positioning accuracy, map consistency, mapping accuracy and overall system robustness in environments with strong occlusion, dynamic interference and repeated features through EKF-optimized IMU attitude, fusion of IMU and odometry data, and multi-level sensor information fusion, which meets the engineering requirements of high-precision autonomous navigation for inspection robots in open-pit coal mine screening buildings.