Wet grasslands across Europe are undergoing rapid structural changes driven by disrupted hydrological balance and altered trophic status, often associated with the spread of expansive plant species such as Molinia caerulea. Efficient, high-resolution monitoring of this species is relevant for conservation and habitat management. We developed a lightweight deep learning approach to detect M. caerulea using high-resolution UAV RGB imagery acquired over a heterogeneous wet grassland complex in eastern Poland. A MobileNetV2-based convolutional neural network was developed and evaluated using 3564 manually labelled image tiles representing the presence and absence of the species, including 2856 training tiles and 708 validation tiles. The model achieved high classification performance (accuracy 0.944, F1-score 0.940) and showed strong robustness confirmed by bootstrap validation. Both continuous patches and isolated tussocks of M. caerulea were reliably detected, with few false positives. Our results demonstrate that computationally efficient and lightweight deep learning models combined with UAV imagery provide a practical and scalable tool for species-level monitoring in complex grassland ecosystems, supporting conservation planning and management of wet grasslands.