Lemon leaf diseases pose a significant threat to agricultural productivity, highlighting the need for early and accurate diagnostic solutions. This study proposes SwinEffNet, a hybrid deep learning architecture that integrates the local feature extraction strength of EfficientNetV2-S with the global contextual modeling capability of the Swin-Tiny Transformer. To address data scarcity and class imbalance, a custom dataset of 3,266 high-resolution lemon leaf images collected from Bangladesh was curated, covering six classes: Algal Leaf Spot, Black Spot, Citrus Canker, Citrus Pest, Greening, and Healthy. The Synthetic Minority Over-Sampling Technique (SMOTE) was employed to balance the dataset, while a hierarchical multistage feature fusion strategy was utilized to enhance feature representation. Experimental evaluations demonstrate that SwinEffNet achieves a high classification accuracy of 99.19%, with consistently strong precision and recall across all disease categories, outperforming standalone architectures such as DenseNet201 and ConvNeXt-Tiny. The proposed framework exhibits strong robustness and generalization capability, making it suitable for large-scale citrus disease monitoring. Overall, this research presents an effective and scalable automated solution that can support sustainable citrus farming practices and contribute to economic stability within the citrus industry.