M.S. AAI Capstone Chronicles 2024

14

The results and conclusions drawn from this analysis are limited by the computational resources

available. The use of a larger dataset for training the CNN model for the SAA task presented is expected

to yield more promising results as the issue of overfitting may be mitigated. A larger training dataset

would also benefit the ViT model by allowing more objects to be successfully detected in the images.

Additionally, due to the environment in which the model is expected to be deployed, further exploration

is required to reduce the latency of the model outputs. Once the model metrics are optimized, the

integration of the computer vision model with the UAV can be established. A secondary interface needs

to be developed to translate the results produced by the model into an action which the UAV can take.

128