Dog and Cat Classification with Deep Residual Network
Yao Yu chen · 2020
I With the development of artificial intelligence, the deep neural network(DNN) has achieved excellent results in image processing domain such as image classification[1] and objct detection[2]. The convolution neural networks(CNN) [3] is a Representative algorithm of DNN which have the representation learning ability. According to its convolutional structure, input information is extracted with translation invariance. Based on the widely used CNN, there are many efficient models. For image classification there are Lenet-5[4], VGG[5], Resnet[6] and so on. For object detection, the yolo series[7] is well-known. Also few well known datasets are proposed to measure their performance such as ImageNet and Cifar-10[8]. These data sets are dedicated to the classification of multiple objects in natural scenes. Nowadays, pets play an increasingly important role in our life, so we built a cat and dog dataset, each of which categories with 12500 samples which is larger then 1260 in Imagenet. For our dataset, we trained an image classification model. We focus on the performance of distinguish dog and cat In different scenes, lighting and noise. Our method achieved an accuracy of 92.7 percent and remained robust under adversarial attack.