Deep Recurrent Residual U-Net with Semi-Supervised Learning for Deforestation Change Detection

“Deforestation” refers to the systematic removal of trees from forests to facilitate significant human activities. Expansion of agriculture, infrastructure and logging are the main causes of deforestation. Deforestation causes greenhouse gas emissions, water pollution and loss of biodiversity. The forest area is gradually decreasing with time. To overcome the above challenges, detection of forest change is crucial. Recently, due to advances in satellite technology, deforestation images have become accessible. In this study, we proposed a deep recurrent residual network to detect how much forest has been cleared over a period of time. Initially, an image of a particular forest area at two different time periods is provided as input to the change detection technique. The input image is iteratively segmented using a novel recurrent residual U-net. It then aggregates these segmented results and provides input to a semi-supervised learning method. From this, the semi supervised MLP method detects how many changes the current image has undergone from the previous image. Our proposed work is implemented on the platform of Python. Our proposed task is evaluated in terms of precision, accuracy, Fscore, recall, specificity, and sensitivity. Our proposed work achieved a high accuracy rate of 94.26%.