Abstract: With the advancement of smart construction site development, construction sites have placed greater demands on the automation of safety inspections and personnel management. Drones possess advantages such as high manoeuvrability and extensive field of view coverage, but their limited endurance restricts their ability to perform long-term, stable operations. Ground mobile robots, while offering strong endurance, lack sufficient high-altitude inspection capabilities. Addressing these issues, this paper designs a land-air collaborative inspection robot system tailored for smart construction sites. The system comprises a UAV, a robot dog, and an autonomous charging platform, establishing a land-air collaborative inspection architecture. It employs a vision-based landing method for the UAV, validated through experiments on a physical UAV platform. In 30 autonomous landing experiments, the unmanned aerial vehicle (UAV) successfully landed 22 times, with a success rate of 73.3% and an average deviation of 9.05 cm. This indicates that the method has good stability in construction site scenarios and can meet the requirements of autonomous recovery and charging of UAVs. Meanwhile, a large language model was used to achieve automatic annotation of construction site images, and a target detection model for inspection was constructed and verified from the perspective of the UAV. The results show that the model's mAP@0.5 on the validation set is 76.1%, and the detection success rates for personnel, reflective vests, and safety helmets are 92.3%, 88.5%, and 73.1%, respectively. Overall, the proposed land-air collaborative inspection robot system has good feasibility and engineering application value in the application of smart construction sites.