Deep Convolutional Neural Network Model for Tea Bud(s) Classification

摘要

Tea production exerts a huge impact on the economy of countries like China, Kenya, and Sri Lanka as they are involved in the world-wide tea production in a substantial manner. They are also amongst the countries, where production of tea is done in a huge scale. However, there is a myriad range of problems associated with tea picking. For instance, there is no proper procedure for selecting tea leaves, inability to guarantee the integrity of tea buds and inability to achieve the picking standards of conventional standards. Further, conventional tea should be plucked at a precise time. The convolutional neural network (CNN) is a deep learning method that performs better in image processing and classification tasks and widely used in the recent literature. Therefore, this study proposes an approach, based on CNN to develop a model that identifies and predicts the suitability of tea buds for the plucking as a solution to the aforementioned problems. First, the suitable and unsuitable tea buds are identified visually before the process of picking. The image samples used here, are created, and preprocessed to identify the hyperparameters. After that, the best combination of hyperparameters was identified for the optimal model. Then, the optimal trained model was evaluated using test data. Finally, an interactive software was developed for tea bud(s) classification. The experimental results show that the accuracy of the CNN model is 70.15% for 10000 image samples, while the accuracy of Support Vector Machine (SVM) and Inception V3 is 65.86% and 68.70% respectively. Hence, the CNN based classification performs better in classification and can improve the classification efficiency of tea buds effectively.