Improving Thai Herb Image Classification using Convolutional Neural Networks with Boost up Features

The use of herbs is incredibly important for human and animals. Herbs have high similarity in shape, color, and texture so that there are difficulties for normal people to classify herbs. Even though plant classification technology has been developed for years, this technology is still unsolved and considered as challenging problem. This research focuses on improving herbal leaves classification. The proposed methodologies include a simple CNN structure, preprocessing process and a boost up feature technique. The proposed technique was used to improve the performance of simple CNN model. The results show that the model trained by using boost up feature technique outperforms the model conventionally trained by approximately 10% on the average.