Abstract: The installation and maintenance of split-type spacers on overhead transmission lines is a critical task to ensure the long-term operational safety and stability of high-voltage power grids. In this respect, the use of unmanned aerial vehicles (UAVs) to transport and deploy installation robots to work on the conductor has emerged as a promising solution to replace manual operations at high elevation and has been shown to significantly improve the efficiency and reduce safety risks. However, the reliability of UAV-assisted deployment depends critically on accurate visual perception of the target conductor. During heavy fog, the visibility drops sharply, conductor edges become blurred, and the contrast between the line and the background is severely degraded, which make it extremely difficult for UAVs to accurately guide split-type spacer installation robots onto the conductor. To address this problem, this paper proposes a method for conductor detection and line-landing control based on the cooperative optimization of a lightweight dehazing network and YOLOv8. First, a lightweight dehazing module was constructed to mitigate the adverse effects of fog on aerial imagery. Using an end-to-end feature mapping mechanism, the module directly learns the transformation from hazy inputs to clear outputs without relying on the explicit estimation of atmospheric scattering parameters. This approach effectively improves the overall image clarity and enhances the contrast between the conductor and the background, while its lightweight architecture ensures that the robot remains suitable for deployment on resource-constrained airborne platforms. Subsequently, the dehazed images are fed into a YOLOv8 detector, to which a cooperative optimization mechanism is introduced to jointly train the dehazing module and the detection backbone. This coupled strategy allows the dehazing process to be guided not only by image-quality objectives but also by detection-oriented feedback, thereby strengthening the feature representation of conductors and improving the detection robustness under fog of varying densities. Finally, based on the geometric relationship between the center-of-detection bounding box and the camera-imaging model, a guidance control strategy is designed to convert pixel-level deviations into corresponding velocity and attitude commands for the UAV. This strategy, which enables closed-loop visual servoing, allows the upper module of the installation robot to descend smoothly and land precisely on the target conductor. Extensive experiments were conducted in different fog densities to verify the effectiveness of the proposed method. The results demonstrate that the method achieves detection accuracies of 89.65%, 86.07%, and 81.76% under light, medium, and heavy fog conditions, respectively, which represent an average improvement of approximately 5percentage points over the original YOLOv8 baseline. In terms of computational efficiency, the introduction of depthwise separable convolutions shortens the inference time from 11.8 ms to 8.3 ms, which corresponds to an improvement of approximately 30% in the computation speed and substantially enhances the real-time applicability of the algorithm to airborne computing platforms. Furthermore, in line-landing guidance experiments, the proposed method consistently succeeds in stably and precisely guiding the upper module of the robot onto the target conductor, thus fully satisfying the requirements of practical engineering operations. These results confirm that the proposed framework provides a feasible and effective solution for autonomous spacer installation under adverse weather conditions and offers valuable technical guidelines for the intelligent operation and maintenance of overhead transmission lines in complex environments.