Advanced forest image classification using generative adversarial network-based data augmentation

The objective of this study was to develop a data augmentation framework that improves forest image classification by addressing limited labeled data, class imbalance, and the high variability of forest scenes caused by differences in vegetation density, illumination, seasonal conditions, and image acquisition perspectives. The generated dataset was intended to enhance model generalization and support the development of reliable forest monitoring applications. A dataset containing multiple forest and vegetation cover categories was expanded using an adaptive conditional generative adversarial network optimized with the Adaptive Jaya Algorithm, together with conventional image augmentation methods such as rotation, flipping, scaling, translation, and brightness adjustment. The images were preprocessed through resizing and Z-score normalization before training an EfficientNet-B0 model for automated feature extraction and classification. The augmented dataset improved class representation, particularly for minority categories, resulting in a classification accuracy of 98.48% . The enhanced dataset reduced overfitting, strengthened feature learning, and improved recognition across diverse forest conditions. All experiments were implemented using Python. The resulting dataset provides a valuable resource for researchers developing robust forest image classification models and supports applications in environmental monitoring, intelligent forest management, and sustainable resource assessment.