The automated detection of tea leaf diseases is a very crucial issue to enhance crop production and to facilitate the sustainability of tea production, but it is difficult because of the inter-class similarity, complicated background, and different lighting conditions. The paper develops an automated tea leaf disease classification system on a modified deep convolutional neural network (DCNN) that will ensure the system is highly accurate yet less complex in terms of computational procedures. The suggested model is trained and assessed using multi-class tea leaf data that contains diseased and healthy leaf categories. A detailed comparative study is done against the already developed deep learning architectures, which are AlexNet, VGG16, and ResNet50, in the same experimental conditions to justify its effectiveness. The experimental findings also prove that the proposed modified DCNN is significantly better as compared to the benchmark models, whereas it reaches an accuracy of 99% in terms of classification, and AlexNet, VGG16, and ResNet50 achieve an accuracy of 77, 72, and 74, respectively. The excellent results of the proposed model are due to its optimized convolutional layers, effective representation of features, and generalized fine-grained leaf disease patterns. In addition, the architecture is lightweight and is thus appropriate in real-time and resource constrained agricultural applications. The results prove that the suggested method offers a strong and effective solution to automated tea leaf disease classification and can be used to facilitate early disease diagnosis and smart decision-making on the precision agriculture systems.