Fine-Tuning Convolutional Neural Network Models for improvement of Object Detection Accuracy
Chamarty Anusha1, P. S. Avadhani2
1Chamarty Anusha, Computer Science & Systems Engineering, Andhra University College of Engineering, Andhra University, Visakhapatnam (AP), India.
2P S Avadhani, Computer Science & Systems Engineering, Andhra University College of Engineering, Andhra University, Visakhapatnam (AP), India.
Manuscript received on 18 June 2019 | Revised Manuscript received on 25 June 2019 | Manuscript published on 30 June 2019 | PP: 2000-2004 | Volume-8 Issue-5, June 2019 | Retrieval Number: E7663068519/19©BEIESP
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Deep Learning is getting a lot of attention these days as these algorithms are out-performing humans in Object Detection. It has entered into several diverse areas like Automobile Industry for making self – Driving cars, Automation industry in Robot development, Medical field for disease identification and for security enhancement in defense sector etc. Deep Learning has become the solution for a variety of problems like Natural language processing, Speech Recognition and Visual Recognition [1,2] etc. Advancement in Deep Learning is especially due to the improvement of Convolution Neural Network Architectures and implementation of new algorithms. Deep Learning is a class of Artificial Neural Networks that learn features from the weights obtained from neurons. In this paper a dataset of animal is taken and it is passed through different Deep Learning Neural Network models like Visual Geometry Group-16 (VGG-16), Visual Geometry Group-19 (VGG-19), Inception, Xception and their accuracy of prediction is compared. To increase the accuracy of prediction further, a method named Fine-Tuning [3,4,5] is applied. Fine-Tuning Neural Network model means varying various hyper-parameters like ‘Number of Top Layers to be Trained’, ‘Learning rate’, ‘Number of Epochs’, Drop-out rate, Optimization algorithm, Activation Function, Training Batch Size, Validation Batch Size, Testing Batch Size, Training dataset Size, Validation dataset Size, Testing Dataset Size etc. and their accuracies of prediction before and after FineTuning are analyzed. There is a significant improvement observed in accuracy of prediction of considered animal dataset after Fine-Tuning of the various Neural Network Models. Based on the experimental results obtained, it can be concluded that the Object Detection accuracy can be enhanced and the detection error rate can be reduced to a greater extent by fine-tuning method.
Keywords: VGG16, VGG19, Inception, Xception.
Scope of the Article: Deep Learning