An Optimal Retinanet Model For Automatic Satellite Image Based Missile Site Detection

Satellite image processing is a manually tedious job and offers scope for automation as part of the information extraction process from satellite images. The involves object detection one challenges ascertaining minimum number images required to train deep learning model achieve certain accuracy. To best authors’ knowledge, work in missile site relatively limited, with an existing exploration latest one-shot methods, such RetinaNet, being absent. This proposes optimal based on RetinaNet framework training minimal dataset. A comparative analysis previous paves road future research methods optimally trained models. As study, key findings are that scheme dataset possible. step enables reduction time development concerned. One many techniques determine plotting versus mean average precision. same validated our work. Further, hybrid two-model concept tested wherein prioritizes Recall while other Precision. Thus combination both models detect set targets provides detection. Lastly, study finds single-stage algorithm advantage balancing speed accuracy over erstwhile two-stage methods.