Sparrow Mapping using Hyperspectral Remote Sensing and Deep Learning

Biodiversity loss, driven by habitat destruction, poses a significant threat to ecological balance and the survival of endangered species. Accurate identification and monitoring of critical habitats are essential for effective conservation efforts. However, traditional field-based habitat mapping methods are often time-intensive, expensive, and spatially constrained. This work aim to address these limitations by integrating hyperspectral remote sensing with advanced deep-learning methodologies for precise habitat identification and mapping. This work integrates hyperspectral remote sensing with deep learning methodologies, specifically Convolutional Neural Networks (CNNs) and Artificial Neural Networks (ANNs), to analyse high-dimensional hyperspectral data. CNNs are utilized for extracting spatial features, effectively capturing both spectral and spatial variations in the imagery. ANNs are applied for spectral pattern recognition, enabling accurate classification of habitats based on their unique spectral signatures. The pre-processing pipeline includes resizing and grayscale conversion of hyperspectral images to optimize them for deep learning analysis. Validation using field data ensures the reliability of the model outputs. The methodology demonstrated promising results, with ANN achieving a superior accuracy of 99% and CNN achieving an accuracy of 84%. These results are intended to support conservation planning, resource allocation, and evidence-based decision-making. This work contributes to advancing biodiversity conservation and underscores the potential of deep learning in ecological monitoring and management.