detection of drones with yolov4 deep learning algorithm

Drones or unmanned aerial vehicles (uavs) have rapidly spread all over the world and are becoming widely popular in major cities for personal and commercial use. it has also been widely used for military purposes in the last decade. thus, it has become difficult to maintain control over them and the risks they pose to privacy and security. in this paper, we present a solution to detect drones before they can reach a sensitive area or residence using the latest yolov4 deep learning algorithm while using darknet as a backbone. we trained our model on different images at different distances and climatic conditions and trained our model to detect birds and aircraft that are very similar to drones at higher distances that may cause confusion, and also train the system at close distances and at very low and high image quality. for all available cases, our dataset was collected from three global and certified datasets in aircraft detection systems and the result was a dataset containing all cases. however, the collection of drones, birds and aircraft datasets is not easy to obtain. the proposed method achieved an accuracy of 98.3% with the main challenge of detecting similar small objects near and far in all conditions.