Flower Recognition Model based on Deep Neural Network with VGG19

Abstract

Computer vision is one way to streamline processes like robotic process automation and digital asset management and has come a long way in terms of its capabilities and what it can provide and do for different industries. Applications provided by computer vision include object detection and image detection. This field of technology is still relatively young and faces many challenges however. Challenges faced in this field include the lack of comprehensively annotated images to use for training the optimal algorithms, and lack of accuracy for application to real-life images which differ from the training dataset. To tackle these issues, this paper is aiming to adjust pre-trained machine learning models, which are ResNet50 and VGG19 respectively, while also training and tuning a new SqueezeNet inspired model to create a flower recognition model that is able to process and remember large amounts of flower species data. From the research carried out, VGG19 was discovered to have the best performance on both the 5 Categories and Flower-102 dataset, with an accuracy of 88 percent and 84 percent respectively.