ScholarWorks@성균관대학교

초록

Image recognition is a novel task that helps computers understand human vision and make decisions from that perception. There are lots of deep neural network algorithms and computer vision techniques. Convolutional Neural Network (CNN) is one of the best deep learning approaches for image categorization which has been proven to achieve the maximum precision and accuracy in visual recognition. CNN is a widely used deep learning method for mining high-level attributes from raw data. Unfortunately, to develop a Convolutional Neural Network, the model requires a large amount of data. Which is time-consuming while training, and provides a higher miss rate in detection if the model is fed with a lower amount of image data. In this study, we developed a combined approach based on Shallow Convolutional Neural Networks (SCNN) and traditional machine learning classifiers to tackle this massive problem. The SCNN was employed as a feature extractor in this study, and the obtained features were then supplied into the classic machine learning classifiers network, Random Forest (RF), which has an ensemble learning mechanism that helps to enhance picture classification performance. Our SCNN-RF model has achieved an accuracy of roughly 93.5% for the bigger dataset and 85% for the smaller one.

키워드