Image Detection by Deep Learning with Convolutional Neural Network (CNN)

Abstract

Image detection is the process that uses deep learning to identify an object and to classify it. It is widely applied in digit recognition, facial recognition, and many other areas. The convolutional neural network (CNN) is one of the deep learning techniques that is frequently used in image detection because it can effectively extract characteristics and develop pattern recognitions of an image. In this paper, a CNN model is developed using Python to classify the images of cats and dogs from a Kaggle machine learning competition held in 2013. The impact of the training epochs and batch sizes to the performance of the CNN model had been studied in this research. Overall, an increase in the number of training epochs will improve accuracy, but when a certain limit is reached, the CNN model may overfit the training data. Batch size can affect training speed and model optimisation. The smaller the batch size, the longer the training time. For a fixed learning rate, there is an optimal batch size that maximises the performance. Therefore, considering the training time, accuracy, and loss value, it is suggested to use a CNN model with a fixed learning rate of 0.001, batch size of 32, and training epoch number of no more than 10.