The rapid urbanization and expansion of power grid infrastructure have heightened the need for accurate and efficient power line monitoring. Traditional inspection methods are labor-intensive, time-consuming, and prone to errors. While aerial imagery offers a promising alternative by providing comprehensive views of extensive infrastructure, detecting power lines in such images remains challenging due to variable weather conditions, complex landscapes, and potential visual obstructions. Current methodologies, often based on edge or line segment detection, struggle with accuracy and reliability in real-world scenarios. This paper proposes a deep learning-based approach to power line detection, leveraging the combined strengths of pre-trained VGG19 and ResNet152 models. By employing rigorous data preprocessing and augmentation techniques, our method enhances model robustness and generalization. An initial model using a standard convolutional neural network (CNN) achieved an accuracy of 92.20%. Building on this, the customized CNN model that integrates features from VGG19 and ResNet152 achieved a high accuracy of 95.40%, outperforming existing models, such as DenseNet121 and AlexNet. This approach not only facilitates near-real-time power line detection but also contributes to improved reliability and efficiency in power grid maintenance, ensuring a resilient and sustainable electrical infrastructure.