The power transmission network is vital to a country's infrastructure, and real-time detection of transmission towers is essential for ensuring transmission safety. Currently, deep learning has dominated in object detection, and most deep learning-based transmission tower detection methods rely on supervised learning. However, supervised detection of transmission towers requires a large amount of tower data to construct the dataset, and requires professionals to perform annotation, which can be a time-consuming and labor-intensive process. In most cases, detecting the exact class of towers is not necessary, but only their location. At this point, weakly supervised tower detection demonstrates its significant advantage. This paper proposes a weakly supervised binary classification method for detecting transmission towers and collects and constructs a binary classification tower dataset. The paper tests the weakly supervised detection performance on the constructed binary classification dataset using two weakly supervised networks. The experimental results show that weakly supervised transmission tower detection based on binary classification can complete the localization task of transmission towers and is feasible in cases where tower class is not required.