Deforestation rate estimation using crossbreed multilayer convolutional neural networks

Deforestation is an important environmental issue that involves the removal of forests on a large scale, resulting in ecological imbalance and biodiversity loss. Synthetic Aperture Radar (SAR) images are widely used as a valuable tool to detect deforestation effectively. The SAR technology allows capturing high-resolution images irrespective of weather conditions or daylight, making it helpful to monitor remote and densely vegetated areas. Recently, deep learning techniques used on SAR images have showcased promising results in the automation of deforestation detection and mapping processes. By leveraging neural networks (NNs) and machine learning (ML) systems, these approaches examine SAR data to recognize deforestation patterns and estimate deforestation rates over time. Therefore, this study develops a cross-breed multilayer convolutional neural network (CNN) for deforestation rate estimation in the Amazon. The proposed model initially preprocesses the input SAR data to remove the speckle noise using a box car mean squared sparse coding filter (BCMSSCF). Besides, crossbreed multilayer CNN (CM_CNN) is used for mapping and segmentation of the deforested area. To determine the pace of deforestation in the Amazon region, a widespread experimental analysis was performed on the LBA-ECO LC-14 dataset. A detailed comparative result analysis of the proposed model is made with recent approaches. The experimental results stated that the proposed model shows promising results in terms of different performance measures. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.