Abstract: Addressing the issue of excessive sagging of conductors in cross-line environments and the subsequent accidents caused by vehicles snagging and potentially leading to wire breakage or tower collapse, this paper proposes a visible light image target recognition model to reduce the complexity of target recognition under variable conditions in cross-line environments, particularly improving the speed and accuracy of vehicle recognition in visible light images. The proposed algorithm is a single-stage object detector featuring a fast target feature extraction module and a cascaded regression module for locating vehicle targets in visible light images. Additionally, an improved composite loss function is introduced to optimize the learning process and enhance accuracy. This research is highly beneficial for cross-line monitoring as it can automatically and accurately monitor vehicle movements near power transmission or distribution lines without human intervention.