“…We performed 19 augmentation operations on the web-scrapped images using Python Imaging Library (PIL) version 9.2.0, and scikit-image library version 0.19.3. to increase the number of images and introduce variability in the data. Our augmentation operations include (1) brightness modification with a randomly generated factor (range [0.5, 2]), (2) color modification with a randomly generated factor (range [0.5, 1.5]), (3) sharpness modification with a randomly generated factor (range [0.5, 2]), (4) image translation along height and width with a randomly generated distance between -25 and 25 pixels, (5) image shearing along height and width with randomly generated parameters, (6) adding Gaussian noise of zero mean and randomly generated variance (range [0.005, 0.02]) to images, (7) adding zero-mean speckle noise and randomly generated variance (range [0.005, 0.02]) to images, (8) adding salt noise to randomly generated number of image pixels (range [2%, 8%]), (9) adding pepper noise to the randomly generated number of image pixels (range [2%, 8%]), (10) adding salt & pepper noise to randomly generated number of image pixels (range [2%, 8%]), (11) modifying the values of the image pixels based on the local variance, (12) blurring an image with a Gaussian kernel with a randomly generated radius (range [1, 3] pixels), (13) contrast modification with a randomly generated factor (range [1, 1.5]), (14) rotating all images by 90 ◦ , (15) rotating images at random angle (range [-45 ◦ , 45 ◦ ]), (16) zooming in an image by about 9%, (17) zooming in an image by about 18%, (18) flipping images along the height, and (19) flipping images along the width. In Table I, we show the number of original and augmented images per class in our database.…”