Machine Learning Assessment of Wildlife Recovery Using Camera Trap Data

Abstract

Introduction and Objective: Wildlife assessment is of great importance in the evaluation of ecosystem recovery, biodiversity conservation and habitat management. Conventional field-based observations are laborious and not efficient in terms of time as well as continuity in gathering information at the population level. The objective of this research is to establish a machine learning-based framework for evaluating wildlife recovery through camera trap data by combining automatic image processing and ecological assessments. Methodology: This framework involves camera trap image acquisition, image processing, machine learning-based species detection and classification, and population trend analysis. The obtained images have been preprocessed using resizing, noise removal, normalization, and image augmentation processes for improving image quality and the generalization of the model. The implementation of a deep learning-based detection and classification model has been done for the automatic identification of wildlife species. Results: The developed framework was able to attain an accuracy rate of 92%, a precision score of 90%, a recall rate of 91%, an F1 score of 90%, and mAP value of 93%. The results show the ability of the model to correctly classify the wildlife species and population estimation through camera trap pictures. Conclusion: The study shows that using camera traps in wildlife recovery assessment with the help of machine learning is effective and scalable.