Oil Palm Plantation Detection in Satellite Image Using Deep Learning

Abstract

The traditional approach, such as manual surveys, used by farmers and
governments to detect where oil palms are planted has proven to be
ineffective, tedious and time consuming. This problem can be
addressed if images obtained from satellites can be used to scan
through the forest to detect oil palm plantations. In this paper, we
developed a web based deep learning system that is capable of
processing a satellite image of land to detect the presence of oil palm
plantations in Nigeria using high-resolution satellite imagery. This
research designed a detection system which uses a convolutional
neural network to extricate important features, and a classifier trained
using satellite images. Results showed exceptional effectiveness with
a training loss of 0.11 and an accuracy of 99.0%. Utilizing different
images for validation taken from diverse elevations, the model
reached a training loss of 0.245 and an accuracy of 82.9999% on
validation data, while on test data we got 1.59 in loss and an accuracy
of 87.5%. Thus, the proposed approach is seemingly effective within
the field of precision agriculture.