The rapid growth of drones, characterized by their increasing complexity and affordability, has unintentionally fostered an increase in drone-related criminal activity. Because of these unauthorized drone usage, advanced counter-drone systems have been developed to defend important infrastructure, public safety, and national security. As drone technology advances, so must anti-drone systems developed to eliminate possible threats. In this paper, deep feature fusion based model is proposed for real-time malicious drone identification. The proposed model begins by resizing the image. Next, features are extracted using ResNet50, VGG16, InceptionV3, and Xception. These features are then combined into a single vector. The combined data is split into 70% for training and 30% for testing. A grid search is used to optimize hyper-parameters for various algorithms. Finally, the model’s performance is evaluated using the test set. The proposed model is evaluated on a malicious drone image dataset with five classes and tested on two other benchmark datasets. The proposed model achieved 97% accuracy on the malicious drone image dataset, and 100% and 98% accuracy on the other two datasets. It demonstrated competitive performance, utilizing a neural network and grid search, compared to other machine learning algorithms and deep learning architectures.