Recognition of tree species based on image analysis

Abstract

The article presents a method for the automatic recognition of tree species based on the analysis of tree trunk images captured using unmanned aerial vehicles (UAVs). The authors employ convolutional neural networks (CNNs) to classify seven tree species: birch, spruce, pine, aspen, maple, oak, and larch. The study aims to overcome the limitations of traditional identification methods, such as subjective expert assessment and high labor intensity. The methodology includes data collection and preprocessing, image augmentation to increase variability, and model training using ResNet, EfficientNet, and MobileNet architectures. The results demonstrate the model’s high efficiency, with an average classification accuracy exceeding 90%. The best performance was achieved for oak (97,6%) and birch (95,4%), while visually similar species (spruce and pine) showed lower accuracy (around 85%), indicating the need for further optimization. An analysis of external factors revealed the significant impact of shooting angle and lighting conditions on classification accuracy. The proposed method can be applied in forestry for automated monitoring and resource management. To enhance performance, the authors recommend expanding the dataset, incorporating additional environmental parameters, and conducting cross-validation on independent datasets.