Recent advancements in deep learning have revolutionized image processingtasks such as segmentation and classification. This study investigates theperformance of U- Net-CNN models in multi-class aircraft segmentation andclassification using polygon and bounding box annotations. Military aircraftclassification is crucial for defense applications, as it aids in rapid and accuratedecision-making during critical missions. This study investigates howthese annotation methods affect training time, segmentation accuracy, andclassification performance in multi-class segmentation and classification tasksinvolving military aircraft. The research compares polygon and bounding boxmethods to evaluate their effectiveness in capturing object details and computationalefficiency. While polygon annotations achieved superior precision witha mean test accuracy of 0.987 and lower loss of 0.041, bounding boxes excelledin computational efficiency. Future research should expand datasets and exploreadditional annotation techniques to further generalize these findings.