The mango holds significant economic and ecological value in India, as the country exports a substantial amount of mangoes. Each year, diseases and pests cost the mango sector a significant amount of money. A multitude of illnesses severely impair the appearance, flavor, and economic viability of mangoes. It is quite challenging to identify the disease with the naked eye. The lightweight VGGNet model is proposed for the detection of mango leaf disease. The process of training and testing the proposed lightweight model involves a combination of sophisticated algorithms, datasets, and validation techniques. We first apply data preprocessing and then split the data into training and testing datasets. The training of the CNN model will take place in its three layers, which are the convolutional layer, the maxpool layer, and the fully connected layer. We then apply the ReLU activation function to start the testing of the model. For this effort, a dataset with 1196 images has been used. Mango leaf images were collected from the PlantVillage website (which is an open database of images). The results show that the efficacy of the proposed lightweight VGGNet model has achieved an accuracy of 99.61%.