Flowers Classification with Low Carbon Footprint Using Deep Learning Pretrained Models

The aim of this paper is to investigate making flower classification using low carbon emissions. This is done by using pretrained models for feature extraction, then, using these features for classification. The pretrained models tested are VGG16, Xception, and MobileNet-V2, support vector machines (SVM) was adapted for the classification. Moreover, the effect of adding Batch Normalization (BN) layer before classification is done and also tested. Performance is assessed using accuracy, precision, recall and F1-Score and the results show that “MobileNet-V2 and SVM”, “MobileNet-V2 with BN and SVM”; and “Xception with BN and SVM” achieves the highest accuracies while MobileNet-V2+SVM being superior. Regarding carbon emissions, MobileNet-V2+BN+SVM had the least carbon emissions during test whereas VGG16 had the least carbon emissions during training.