Abstract: In order to supplement the limitations of active positioning technologies such as UWB indoor and outdoor positioning and vehicle-mounted strap-on inertial navigation in signal blind spots and non-cooperative target monitoring, this paper proposes lightweight target detection and multi-object tracking algorithms, and constructs an intelligent UAV inspection system. In the design of the detection model, the deformable convolutional DCNv2 and AFPN are introduced in the backbone network, and the progressive feature pyramid network AFPN is used to strengthen the multi-scale feature extraction ability. The lightweight detection head LSDECD-Head is designed to improve the detection accuracy of small and occluded targets combined with the Focaler-GIoU loss function. The model compression is realized by the LAMP pruning algorithm, and the performance of mAP50 of 0.868 and inference time of 196 ms is still maintained at 30% pruning rate, which adapts to the constraints of UAV computing resources. In terms of multi-object tracking, the ByteTrack algorithm is improved, and the Apparence-Spatial Similarity Matrix, (ASM) that integrates the spatial position, operation state and appearance features of the target is introduced, and the trajectory prediction is optimized by the acceleration correction function, which increases the multi-object tracking accuracy (MOTA) by 2.6% and reduces the number of ID switches by 21 times. In addition, a multi-level inspection system is built, integrating data collection, real-time detection and multi-machine collaborative scheduling functions, and relying on 5G and ad hoc network technology to achieve data transmission and remote monitoring. The experimental results show that the proposed scheme significantly improves the accuracy and stability of equipment detection and tracking in the open-pit mine scene, and provides a practical technical solution for intelligent and unmanned inspection of mines.