Transfer Learning Based Object Detection and Effect of Majority Voting on Classification Performance

Abstract

The use of traditional machine learning techniques in the classification tasks of image-based automatic object species requires primarily extracting the feature set. This requires deciding which set of features to use, and is a toilsome process. In this paper, we present a transfer learning based deep learning approach to overcome object classification problems. Various well-known CNN models are used during the experimental study. We also presented the majority voting scheme to improve the performance of the proposed method. According to the obtained results, the highest performance was achieved with the VGG-19 architecture with 98.85% accuracy among the fine-tuned models. Moreover, the majority voting approach improved performance by about 0.2% achieving 99.03% accuracy.