The increasing popularity of photovoltaic (PV) setups stems from their capacity to generate clean and cost-effective electricity. However, various factors can either totally or partially disrupt the production of PV panels. To address this challenge, this study proposes a ResNet-50 model with dynamically adjusted hyperparameters to classify real-time captured images of PV panels into efficient and non-efficient categories. The hyperparameter tuning within the ResNet-50 model is conducted across three distinct cases, revealing that the most optimal classification results are achieved with the following settings: 50 epochs, a learning rate of 0.001, and a batch size of 32. The highest weighted average metrics, including accuracy (96%), recall (96%), precision (97%), and F1-score (96%), were obtained under these settings.