Convolutional neural network in image analysis for determination of mangrove species

Abstract

Information on mangrove species plays crucial role for sustainable management of coastal ecosystems. However, the current in-depth data acquisition for sustainable management of coastal ecosystems is still collected manually. The increased demand to obtain mangrove environmental data in a short time and at affordable cost has encouraged our research to develop an automatization method for determining the species based on the images of mangrove leaf. In this paper the authors used a deep learning method that uses the Convolutional Neural Network (CNN) to overcome manual leaf sample identification during image recognition process. CNN is used to process the machine learning on a personal computer. Stages on CNN were data input, preprocessing and training. CNN was implemented by using tensorflow libraries through the transfer learning process to recognize three mangrove species of northern coast of Probolinggo, East Java, Indonesia. The recognition process is based on images of the mangrove leaf shape. This method was simple and can be reproduced by anyone without the need for in-depth computer programming knowledge. In a relatively short time, the method has been proven to give high accuracy of predicted results. Field test showed that this method can determine and distinguish the leaves of the three species of mangrove well. In the future this method will be developed to identify mangrove plants using Unmanned Aerial Vehicle (UAV).