Unmanned aerial vehicles (UAVs) have demonstrated substantial improvements in the efficiency of power line inspections. The application of sophisticated deep learning object detection algorithms facilitates automated anomaly identification through drone imagery, significantly expediting inspection processes. This not only enhances operational efficiency but also leads to cost reductions and diminished associated risks. Nonetheless, challenges arise, particularly in the face of adverse weather conditions. Images captured by drones showing snow, rain, or fog introduce noise, compromising image clarity and quality, which is expected to be a significant obstacle for object detection models. Our research is centered on two main objectives: (i) assessing the impact of weather disturbances on the model’s efficacy, and (ii) determining the necessity of training the model in a noisy context for detecting broken porcelain insulators—a critical component of power line infrastructure. The study evaluates model performance metrics trained on clear images, tested on noise-free datasets, and those with simulated noise (snow, fog, and rain). Additionally, recent object detection models (YOLOV8 and RTDETR) trained on three noisy datasets (snow, fog, and rain) are tested on their respective noisy test datasets, with benchmarking against a model exclusively trained on clear images. The findings reveal a slight superiority of the model trained in a clear context over its noisy counterpart in the corresponding test environment. However, marginal efficiency gains are observed when training a model in a noisy environment and testing it in the same context. Ultimately, the results affirm the model’s generalization ability, indicating that training in a noisy context is not imperative, considering the limited added value compared to potential risks associated with data collection operations in extreme conditions. The code and dataset can be accessed at: https://github.com/phd-benel/weather_powerline_insulator .