Classification of flower species using CNN models, Subspace Discriminant, and NCA

Abstract

Flowers have an important place in human life. Because flowers can appear at every stage of human life. People want to know these types of flowers that they come across even in daily life. However, due to a large number of flower types, there are difficulties in recognizing these types. We used deep learning methods in this study to overcome these difficulties. Deep learning methods have been widely used in different fields recently. In this study, we used 3 different deep learning methods. In the first stage, we performed the classification process using the pre-trained Efficientnetb0, MobilenetV2 and Alexnet architectures. In the second step, we extracted the feature maps of the images in the dataset using these three pre-trained deep learning models. Then, we optimized these features using the NCA size reduction method to save time and cost. Next, we classified these optimized features in the features Subspace Discriminant classifier. In the final stage, we combined the features we obtained with three pre-trained deep learning architectures. After optimizing these combined features with the NCA method, we classified the features in the Subspace Discriminant classifier. In the first step, the highest accuracy we achieved in the three pre-trained deep learning architectures was 83.67%, while our accuracy rate was 94% in this hybrid method we recommend. This shows that our proposed model is successful. © 2021 IEEE.