Drones and Birds Detection Based on InceptionV3-CNN Model: Deep Learning Methodology

In Unnamed Ariel Systems (UASs) technology, recognizing flying robots known as ‘drones’ has become a central issue and has attracted the attention of many researchers to be used in protecting natural resources and integrating green initiatives. The bird strike with small flying Drones is one of the most common five reasons for drone crash accidents so, discriminating between those sky objects is a challenging research problem, which is known as Drone versus Bird” detection problem. Discriminating between drones and birds at long heights is not an easy task as many believe. In addition, classifying birds and drones under different aerial constraints is a challenging issue. In response to this important problem, a novel deep learning model called Inception-V3-CNN is designed and implemented to accurately solve the “Drone versus Bird” detection problem. The proposed model has been trained and tested using a benchmark dataset consisting of 1243 images for drones, birds, and nests. The experimental results confirmed the efficiency and superiority of the proposed Inception V3-CNN compared to other standard deep learning models such as ResNet 50, VGG 19, Xception, and Inception V3 in solving Drone versus bird detection problem, where the proposed model achieved testing accuracy reached to 99% while distinguishing flying birds than small drones.