CIFAR-10 Image Recognizer

Read the full report.

The CIFAR-10-Based Image Classifier is a computer vision, convolutional neural network model designed to classify images from the CIFAR-10 dataset. This dataset contained ten different types of images to classify: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck. We also utilized transfer learning to build a secondary neural network based on the VGG16 model, and performed an examination of their performance.

Development

Feature extraction is performed using a convolutional neural network (CNN) architecture, with 10 possible outputs. The CNN is comprised of multiple hidden layers with a 3x3 convolution kernel with strides 1 and 2, using relu activation, several 2x2 max pooling layers, dropout layers, and compiled with categorical crossentropy and the adam optimizer. The CIFAR-10 based model achieves a 76% accuracy rate, outperforming the transfer learning VGG16-based model, which achieved a 71% accuracy rate.