Time Efficiency Improvement for Rice Seed Identification Using Hybrid Machine Learning Models and Principal Component Analysis

This paper proposes a hybrid machine learning model to improve the time efficiency in identifying rice seed varieties. In a conventional hybrid method, the features of the rice seed images extracted by VGG16 deep neural network are fed to support vector machine model to carry out classification task. Different from the conventional method, in the proposed method, the number of feature dimensions of the rice seed images extracted by VGG16 is reduced using principal component analysis (PCA). Then the PCA-based features are fed to logistic regression or support vector machine models to identify the rice seeds. PCA helps to reduce the number of feature dimensions of images while keeping significant information. As a result, the proposed model achieves similar accuracy and better time efficiency in compared with the conventional hybrid model. Experiments are carried out to confirm the effectiveness of the proposed model. The experimental results show that the proposed model achieves similar accuracy to the conventional model while reducing the processing time of training and prediction 1.5 times.