M.S. AAI Capstone Chronicles 2024
9
process. The pooling layer is used to reduce the computational complexity and the number of
parameters by reducing the spatial size of the input. The final fully connected layers flatten the data into
a one-dimension vector and process the data for classification (Goodfellow et al., 2016). An outline of
the simple model architecture can be seen in Figure 5.
Figure 5
CNN architecture
For the model training process, the data is split into training, validation, and testing datasets in a
70:20:10 ratio respectively. The training and validation datasets are used during the training process,
and the testing dataset is reserved for the final evaluation of the model. After the images are pre
processed, they are used as an input into the CNN model built using TensorFlow Keras ( “Convolutional
Neural Network (CNN)”, n.d.). The model is evaluated on accuracy, precision, and recall with additional
emphasis on precision to prevent false alarms which may trigger unnecessary UAV maneuvers. Due to
the multi classification nature of the problem, a categorical cross-entropy loss function defined by
TensorFlow is implemented ( “ tf.keras.losses.CategoricalCrossentropy ”, n.d. ).
123
Made with FlippingBook - professional solution for displaying marketing and sales documents online