Implementation of Deep Convolutional Neural Network in Multi-class\n Categorical Image Classification

Abstract

Convolutional Neural Networks has been implemented in many complex machine\nlearning takes such as image classification, object identification, autonomous\nvehicle and robotic vision tasks. However, ConvNet architecture efficiency and\naccuracy depend on a large number of fac- tors. Also, the complex architecture\nrequires a significant amount of data to train and involves with a large number\nof hyperparameters that increases the computational expenses and difficul-\nties. Hence, it is necessary to address the limitations and techniques to\novercome the barriers to ensure that the architecture performs well in complex\nvisual tasks. This article is intended to develop an efficient ConvNet\narchitecture for multi-class image categorical classification applica- tion. In\nthe development of the architecture, large pool of grey scale images are taken\nas input information images and split into training and test datasets. The\nnumerously available technique is implemented to reduce the overfitting and\npoor generalization of the network. The hyperpa- rameters of determined by\nBayesian Optimization with Gaussian Process prior algorithm. ReLu non-linear\nactivation function is implemented after the convolutional layers. Max pooling\nop- eration is carried out to downsampling the data points in pooling layers.\nCross-entropy loss function is used to measure the performance of the\narchitecture where the softmax is used in the classification layer. Mini-batch\ngradient descent with Adam optimizer algorithm is used for backpropagation.\nDeveloped architecture is validated with confusion matrix and classification\nreport.\n