High Resolution Satellite Image Based Seagrass Detection Using Generalized Regression Neural Network

Abstract

Seagrass plays an important role in maintaining the sea ecosystem. Seagrass provide habitats for fish and invertebrates, provides food, purifies water, stabilize sediment so its sustainability needs to be maintained. Due to human activities like sewage input, dumping of solid waste on the shoreline and anchoring of boats, the population of seagrass has been decreasing continuously which has led to instability in the marine ecosystem. Maping and classifying the seagrass from the remote place and from satellite image is very complex task. Because seagrasses are live in the in depths of 3 to 9 ft (1 to 3 m) to at depths of 190 ft (58 m) in the oceans. Remote Sensing is a technique of mapping any place, whithout being making physical contact with that place. So this research uses Andaman & Nicobar’s remotely sensed high resolution satellite image, which has taken from Google Earth. The proposed research paper uses machine learning as a tool for mapping and classifying the Seagrass from Satellite image. Generalized Regression Neural Network (GRNN) algorithm is a part of Artificial Neural Network which is used for classification of Satellite image. The research uses RGB image features for the classification of the seagrass. The training model is created by the extracting the RGB values of the Satellite image. By applying this training model on Generalized Regression Neural Network the system has classified the imaged. The system has classified the image into two group, one is Seagrass and another is Non-Seagrass. This classification shows 100% accuracy and 1.0 Kappa coefficient. So this research shows the very good accuracy for the classifiaction of the seagrass from satellite image.