Tea is one of the world's most consumed beverages, and tea plantations face significant economic losses due to various plant diseases. Early and accurate detection of these diseases is crucial for implementing effective treatment strategies and minimizing crop damage. This study leveraged deep learning techniques to automatically identify diseases in tea leaves from images, providing a tool that could assist farmers and agronomists in disease management. This research presents a deep learning technique for identifying tea leaf sicknesses, a crucial issue affecting tea manufacturing and yield. The proposed hybrid model integrated EfficientNetV2S and ResNet152V2 architectures through a novel feature fusion strategy, enhancing feature representation and generalization beyond conventional ensemble methods. It addressed challenges such as limited training samples and class imbalance by leveraging pre-trained models with regularization techniques, attention mechanisms, and an optimized focal loss function. The dataset used in this study was comprised of 885 labeled images covering eight categories of tea leaf conditions: white spot, red leaf spot, gray light, anthracnose, brown blight, bird eye spot, algal leaf, and healthy leaves. The presented study achieved 97.61% accuracy, 97.69% precision, 96.27% recall, and 96.97% F1-score, demonstrating strong and balanced classification performance. The results indicate that the proposed hybrid deep learning model surpassed conventional and recent benchmark techniques, providing an effective framework for early disease detection in tea plantations. This research contributes to the development of intelligent agriculture solutions by offering a reliable system for automated tea leaf disease identification.