Optimized Convolutional Neural Network Architecture For Image Classification

Abstract

An optimized Convolutional Neural Network (CNN) is employed for the image classification task using the CIFAR-10 dataset as the foundational dataset. The initial layer of the CNN preprocesses the dataset to the training data. To categorize the images in the dataset into ten classes under the labels aircraft, bird, cat, deer, dog, frog, horse, ship, truck and vehicle. The CNN model architecture is made up of convolutional layers, max-pooling layers and fully connected layers. The activation functions used are ReLU and sigmoid. The proposed system has been trained over 10 epochs with the usage of binary cross-entropy loss and the Adaptive Moment Estimation (Adam) as the optimizer. When tested, the model's test accuracy is 94.32 percent. This paper evaluates the binary classification between images classifying between real and AI generated shows how CNN performs image classification tasks efficiently.