AI-Assisted Image-Based Lichen Health Assessment for Environmental Sensing Using Naturally Acquired Field Images

Lichens are widely recognized as biological indicators of environmental quality because their physiological condition reflects long-term atmospheric and ecological changes. However, conventional lichen assessment relies primarily on manual visual inspection by trained specialists, limiting the scalability of environmental monitoring. This paper presents an AI-assisted image analysis framework for automated lichen health assessment using naturally acquired field images as a step toward intelligent environmental sensing. A dataset consisting of 140 images of Lobaria pulmonaria was collected from the public iNaturalist database and manually annotated according to health and moisture conditions. An end-to-end framework integrating dataset construction, image augmentation, deep feature learning, health classification, and quantitative evaluation was developed and evaluated using stratified five-fold cross-validation. Binary health assessment achieved an average classification accuracy of 82.1%, while binary moisture and four-class lichen-status classification achieved accuracies of 75.7% and 64.3%, respectively. To further evaluate the robustness of the proposed framework, representative convolutional neural network, transformer-based, and classicalmachine learning classifiers were compared under an identical experimental protocol, with EfficientNet-B0 achieving the highest overall performance. The results demonstrate that AI can reliably assess lichen health from naturally acquired field images and further reveal that environmental moisture introduces substantial visual ambiguity for automated health assessment. This work establishes a practical and reproducible baseline for future AI-assisted environmental sensing, ecological monitoring, and image-based environmental bioindication.