Impact of Hyper-parameter Optimization in Flower Classification through HOG features using Support Vector Machines

Abstract

The flower classification is crucial. Pharmaceutics, cosmetics, Ayurvedic medicine, horticulture, botany, obtaining patents, and gardening are some of the areas which benefit greatly from the study of flower classification systems. Therefore, this paper aims to improve accuracy by Hyperparameter optimization for SVM Kernels using extracted HOG features. This model employs feature extraction procedures based on the histogram of gradient (HOG). The proposed framework is evaluated on a benchmark dataset (from Oxford University), which has 1360 images of 17 flower species. Hence, 1568 features were abstracted from each of the images with size 224X 224. The SVM hyperparameters are tuned using the GridSearchCV optimization technique. To finish the classification processes, SVM is used across three distinct kernels by varying their hyperparameters. The experimental data shows that the polynomial kernel of SVM model with the hyperparameter values for C, gamma and degree as 0.1,1 and 2 respectively is highly accurate, to the tune of 98.53%. To this end, SVM model is tested with the above value to check with the selected hypermeter value, it shows a 99.26% of accuracy. Further, Deep Learning and Transfer Learning classification models may be incorporated into the proposed model to further advance its performance in the future.