Accurate and timely diagnosis of strawberry diseases is crucial for minimizing crop loss and reducing the cost of plant treatment. The use of unmanned aerial vehicles (UAVs) equipped with multispectral cameras opens up new possibilities for noninvasive plant health monitoring. The objective of this research is to explore the feasibility of detecting diseases in strawberries through multispectral imagery obtained from UAVs combined with deep learning techniques. Leaf spot disease, caused by Ramularia tulasnei Sacc., was chosen as a case study. A DJI Phantom 4 Multispectral quadcopter was used to survey an experimental plot (37 × 22 m) from an altitude of 3 m, achieving a resolution of 2.2 mm per pixel. Preprocessing of the acquired data involved orthomosaic building using Agisoft Metashape Pro and manual region-of-interest (ROI) extraction in ENVI 5.2. A dataset comprising 360 multispectral images, each of size 64 × 64 × 5 (with spectral bands at 450, 560, 650, 730, and 840 nm), was created and separated into three classes: healthy plants, diseased plants, and soil. Using this dataset, a U-Net-based model – supplemented with the attention gate, squeeze-and-excitation, and residual connections modules – was trained. Additionally, to assess the contribution of multispectral channels, another U-Net model was trained on RGB images, as well as a classic Random Forest model based on 40 GLCM texture features. The U-Net model trained on multispectral data achieved an overall classification accuracy of 96.04 %, which is significantly higher than the performance of Random Forest (76.64 %) and the U-Net on RGB images (94.14 %). The resulting disease distribution maps enable site-specific fungicide application. The proposed approach confirmed the effectiveness of deep learning for diagnosing leaf spot disease in strawberries. Future research will focus on expanding the dataset (different cultivars and disease stages), testing more advanced segmentation architectures, and scaling the method to commercial plantations.