IDENTIFICATION OF SOME TYPES OF LONGAN (THROUGH LEAVES) USING IMAGE AND DEEP LEARNING TECHNOLOGY

Abstract

In the Mekong Delta region, besides rice being the most important tree, longan has been planted in a large area with various families, bringing high income to the farmers. Each of them has different properties in harvesting time, flowering, and supervising. In fact, the farmers must regularly monitor, take care and rely on their experience to choose the right time to intervene in each type of longan for increasing productivity. From longan pictures taken in the gardens, it is not easy to determine the type of longan excluding knowledgeable people and scientists. Therefore, it is necessary to help apply technology to assess the kind of longan-by-leaf photographs taken of longan. The study used data set of 3 types of longan leaves, namely Ido, Thach Kiet, and Dimocarpus Longan Lour in the total of 2182 collected pictures. Using deep learning techniques with the VGG16 model to train the obtained data, the accuracy result was 98.3%. Thus, the research results can help agronomists and researchers have special measures to support and cooperate with the farmers in classifying and orienting to plant suitable longan with the highest efficiency.