“…Alternatively, deep learning algorithms, such as AlexNet, googLeNet [ 20 ] can directly process images coming from sensors and classify them according to internal representation learning processes, which are orchestrated through multiple convolutional and fully connected layers. Throughout the years, many studies tackled TSR [ 21 , 22 , 23 ] using different feature descriptors and ML-based classifiers. Different combinations of such classifiers and features have been proven to generate heterogeneous classification scores [ 15 , 19 , 24 , 25 , 26 , 27 ] motivating the need for comparisons to discover the optimal classifier for a given TSR problem [ 3 , 28 , 29 ].…”