An Augmented Approach to Vehicle - Type Classification using Deep Transfer Learning Approaches with adam Optimizer

Intelligent Transportation Systems (ITS) for an Internet of Things (IoT) Smart City are primarily concerned with enhancing the safety, efficiency via information and communication technology. It is vital to have systems in place that can gather road data and monitor traffic. In this research, we used data from the Stanford AI automobile dataset, where we were tasked with classifying 196 vehicles. We employed deep transfer learning algorithms, including ResNet50, MobileNetV2, and DenseNet169. This study offers a technique to enhance the detection of occluded vehicles. Due to the increased traffic & number of cars to classify that resulted in blockage or occlusion, we applied the ADAM (adaptive moment estimation) optimizer approach to obtain the best results. The results are assessed for both the training & validation phases based on accuracy, precision, root mean square error, loss, recall & Fl score. During the training phase, DenseNet169 scored the best accuracy (99.83%), highest recall (95.46%), lowest loss (0.01), and rmse value (0.10) & 97.89
^%
precision, 95.46% recall, and 90.59% Fl score during validation phase.