Smart agriculture mainly stresses the improvements in early plant disease diagnostics, crop classification and management, and effective pest control. Maize being an important staple crop necessitates early and accurate disease detection on its leaves. Hence, this paper proposes a novel Convolutional neural network model based on EfficientNet-B4 model and multi-head attention for the effective classification of maize leaf diseases. The proposed model focuses on early-stage disease patterns with the help of a large amount of data. Compared to other architectures, it attains an outstanding F1-score of 96.75%. The proposed system not only helps farmers to get timely diagnosis but also offers them useful information to mitigate risks and improve production. This study benefits agricultural resilience, food security and farmer’s welfare.