Recognition and Classification of Broiler Droppings Based on Deep Convolutional Neural Network

Table 2

The applied mage augmentation techniques.

Method	Description
Reversal	Flip the image horizontally or vertically at random
Blur	Gaussian blur (sigma between 0 and 3.0), average and or uniform blur (kernel size between 2 by 2 and 7 by 7), and median blur (kernel size between 3 by 3 and 11 by11)
Sharpen	Sharpen each image and overlay the result with the original
Noise	Add Gaussian noise to some images
Gray-scale transformation	Add -10 to 10 to each pixel
Brightness	Change the brightness of images (50-150% of the original value)
Contrast normalization	Improve or worsen the contrast of images
Image rotation	Random rotation of angles around the center of the image
Superpixel	Sample between 20 and 200 superpixels for each image and replace with the average value