Mango, a widely cultivated tropical fruit, is susceptible to various foliar diseases that adversely affect yield, quality, and market value. Early and exact disease identification is crucial for effective crop management and sustainable production. Conventional diagnostic methods, primarily dependent on manual visual inspection, are often inefficient and liable to error. To overcome these challenges, the current study proposes a lightweight convolutional neural network (CNN) model for automated detection and classification of mango leaf diseases using image data. A dataset of 4,000 images comprising 3,500 diseased and 500 healthy samples across 8 categories, including anthracnose, die back, bacterial canker, and powdery mildew. Comparative analyses with pretrained models (DenseNet169, DenseNet121, and InceptionV3) showed high accuracies. Among them, DenseNet121 and InceptionV3 reach approximately 99.92%. A custom 13-layer CNN with 55,184 trainable parameters was developed, achieving 100% accuracy and outperforming all benchmark models in precision, recall, and F1-score. The proposed model demonstrates strong diagnostic effectiveness and computational efficiency, offering a practical solution for real-time, field-level disease monitoring in mango cultivation. The proposed approach combines high diagnostic accuracy with computational efficiency, making it practical for real-time, field-level disease monitoring. This progress supports precision agriculture by providing accessible and user-friendly plant health assessment tools that promote sustainable mango production.