A Faster R-CNN Model for Multi-Class Classification and Detection of Land, Air, and Sea Vehicles

Automatic vehicle detection and classification systems have recently been utilized in various fields. Although the applications in this field generally focus on civilian land vehicles (mostly automobiles), relatively few studies address civil and military land, air, and sea vehicles together. This study developed a multi-class vehicle detection model using faster region-based convolutional neural networks (Faster R-CNN) on a tailored dataset. The dataset includes 13,577 images, with a total of 21,070 objects identified. Among these, 8,194 objects are classified as land, 7,203 as sea, and 5,674 as air. The faster R-CNN model yielded an F1-score value of 0.869 for air, 0.838 for land, and 0.894 for sea vehicles. The results showed that the developed model is a promising approach for identifying civil and military vehicles on land, at sea, or in the air.