DLCD: Deep learning-based change detection approach to monitor deforestation

• Published in: Signal, image and video processing image and video processing, 2024, Vol.18 (Suppl 1), p.167-181
• Main Authors: Srivastava, Saurabh, Ahmed, Tasneem
• Format: Article
• Language: English
• Subjects: Algorithms; Change detection; Climate change; Computer Imaging; Computer Science; Deep learning; Deforestation; Descaling; Drought; Error analysis; Forest management; Forests; Global warming; Hydrologic cycle; Image classification; Image Processing and Computer Vision; Machine learning; Monitoring; Multimedia Information Systems; Original Paper; Pattern Recognition and Graphics; Rainfall; Signal,Image and Speech Processing; Soil erosion; Soil fertility; Transpiration; Vision
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-024-03140-1

The large-scale removal of trees from forests to make way for human activities is known as deforestation, given that it may result in soil erosion, natural habitat deterioration, biodiversity loss, and water cycle disturbance, this is a major environmental problem. As a source of food, clean water, oxygen, and medicines as well as an essential component of the hydrological cycle—they supply water to the atmosphere through transpiration—forests are a contributing factor to climate change and global warming. In addition to decreasing soil fertility and rainfall, deforestation increases the likelihood of floods and droughts and has a major effect on global warming. Deforestation monitoring is an important input for forest management that helps to prepare an action plan, but monitoring is still a challenging task. Hence, there is a need for an accurate deforestation mechanism to monitor those areas that have been converted from forest to non-forest areas. Therefore, in this paper a deep learning-based forest monitoring approach has been proposed, which is implemented in two steps: (i) a machine learning-based classification technique has been applied to the Sentinel-2 images to classify the forest and non-forest areas, and (ii) a deep learning-based change detection technique is proposed to detect the changes occurred during 2017–2022 of the Kukrail forest range situated in India. The performance of the proposed algorithm is assessed by estimating the error measuring parameters like Precision, Recall, and F1 Score, and it is observed that the proposed approach is quite suitable for forest area change monitoring.