Advanced surveillance systems are essential in sensitive environments for identifying military vehicles, including the detection of aerial attacks in the form of drones and military aircrafts. This research focuses on building a comprehensive model for identifying threats in military environments by differentiating between military and civilian vehicles, including trucks, tanks, aircraft, helicopters, and drones, as well as distinguishing drones from birds. Transfer learning has been utilized through Convolutional Neural Network (CNN) models, including MobileNet V2, InceptionV3, XceptionNet, and VGG16, used together with custom layers for enhanced classification accuracy. Techniques such as resizing, blurring, and data augmentation have been used for pre-processing data to enhance adaptability for generalized scenarios. Of all the implemented models, MobileNet V2 achieved maximum accuracy at 95.41%, and also obtaining the highest precision and recall values, while InceptionV3 and XceptionNet models performed well. The system had also incorporated an alert mechanism, so it showed the capability of effectively detecting the threat.