Abstract
Satellite images present object detection as a significant challenge because of the sheer magnitude and high quality of the remote sensing data as well as the confusing backgrounds. This research introduces the Faster Region-based Convolutional Neural Network's (Faster R-CNN) architecture to enhance the precision and speed of object detection on satellite images. The suggested approach is based on the combination of the state-of-the-art feature extraction strategies and a Region Proposal Network (RPN), where the given method can save substantial analytical cost at the expense of the detection performance. To improve the model's capacity to recognize objects of different sizes and resolutions, the multi-scale feature fusion model is suggested. In addition, the targeting design is more efficient when targeting different objects such as automobiles, buildings, and plants. It has been proved experimentally that the suggested Faster R-CNN model not only achieves the state-of-the-art detection findings, but it is also more resourceful in terms of the processing speed compared to the standard R-CNN models. The model has been demonstrated to function well in real-time applications like urban planning, environmental monitoring, and catastrophe management when tested on publically accessible satellite image datasets. Overall, the proposed framework is a scalable and robust solution, to precisely and efficiently detect objects in large-scale satellite data.