From the past few years, object detection in satellite imagery focuses on accurately identifying and classifying objects. This enhanced applications in environmental monitoring by improving the reliability and precision of automated analysis in remote sensing. Traditional approaches for object detection in satellite imagery had faced several issues which include occlusions, class imbalance. Therefore, this research proposes Efficient Object Detection (EfficientDet7) for detecting objects. Initially, the data is collected from Dataset for Object Detection in Aerial Images (DOTA) dataset, which includes aerial images of various pixel range. Further, the collected data is preprocessed using various techniques such as data augmentation, min–max normalization, and noise reduction. Then, the filtered data is embedded into the self-distilled vision transformer (DinoV2) model for extraction of certain features of the input data, such as spatial connections, which are essential for differentiating objects intricate aerial sceneries. Then, the extracted features are fed into the EfficientDet7 which incorporates a Feature Pyramid Network (FPN) to enhance multi-scale feature representation and increase the accuracy of object detection across a range of sizes satellite imagery. The proposed EfficientDet D7 attained better results than existing COCO-based Neural Network (COCO-Net) with 99.94% of accuracy, 99.93% of precision, and 99.89% of recall, respectively.