Rooftop Photovoltaic Panel Segmentation using Improved Mask Region-based Convolutional Neural Network

Solar energy production has significantly increased in recent years in world wiled, accounting for 20% of the total in 2023. In recent year, developing the models for identify the PV panels using aerial images. However, the existing models are still suffering from segment the tiny PV panel areas due to the black appearance presented in the PV panel dataset. In this paper proposed improved Mask Region-Based Convolutional Neural Network (R-CNN) for effectively segment the tiny and large area of PV panels. Intend of using the ResNet model to assign the models backbone. Its extract the features and divided into 32 groups. Subsequently Feature Pyramid Network (FPN) is utilized for sums the higher-level features with lower-level features, its cause to reduce the network topology. Finally, Loss function method of binary cross entropy is used to solving the classes imbalance issue. The proposed method is evaluated on such metrics Intersection over Union (IoU) of 91.83%, Precision of 95.97%, Recall of 96.15% and F1-Score of 90.58%. suggested model is compare with various existing models such as DeepSolar-CNN, Advanced MANet and PKGPVN models.