Deep Learning-Based Detection of Juniperus Procera Dieback in UAV Imagery Using an Ensemble Approach

Juniper trees are essential in many ecosystems, and any changes in the vegetation cover of juniper trees must be detected. Deep learning models were used in this study for detecting dieback in Juniperus procera in natural forests in the southern region of Saudi Arabia was examined. The goal in this study was to develop a reliable method for automated tree cover detection for large-scale environmental monitoring. Various deep learning models, including DenseNet121, InceptionResNetV2, custom convolutional neural network, Xception, ResNet50, and EfficientNetB2, were applied to classify images of juniper trees. The performance of each model was evaluated using metrics including accuracy, precision, recall, F1 score, specificity, and the Jaccard index. An ensemble model was created by combining the four highest- ranking models (Xception, ResNet50, InceptionV3, and EfficientNetB2) to further increase the classification accuracy. The EfficientNetB2 model outperformed the other models on all evaluation metrics, achieving the highest accuracy (98%), precision (98%), recall (98%), F1 score (98%), specificity (98.63%), and Jaccard index (92.86%). InceptionV3 also performed well, with an accuracy of 96%. The ensemble model achieved an accuracy of 97%, demonstrating effectiveness in combining the strengths of the top-performing models. Deep learning models, particularly EfficientNetB2, offer a highly effective approach for classifying juniper trees into normal and dieback categories. These models provide a promising solution for automated ecological monitoring for preserving and managing juniper tree populations.