Customized Deep CNN for Foliar Disease Prediction Based on Features Extracted from Apple Tree Leaves Images

India is an agricultural country, almost 80% of rural people are depending on agriculture only. From the years a lot of technologies evolved that are used in designing various machineries, identifying the suitable crops based on the soil by testing the soil etc. Even though a lot of information technology-based evolvements done in the agriculture field, still it needs to be improved to have a better crop yield, to identify various diseases due to which the crop effects, fertilizers and pesticides to be used etc. Apple is one of the natural crop products, which is effected with various diseases causing the reduction in the crop yielding. If the diseases are identified at early stages and treated properly, the crop yielding can be increased by that price can be reduced and it can make available to even poor people. A customized deep CNN was developed to predict the foliar diseases in apple trees by studying various features in the leaf images. The dataset consists a total of 3642 apple leaf images belongs to 4 classes. Healthy, Multiple disease, Rust and Scab are the class labels. The efficiency of developed customized deep convolutional neural network was studied by implementing the network with various epoch values. The proposed customized CNN model performs with accuracies 0.9376 and 0.9290 on training and validation sets with the number of epochs 30 respectively. From the results obtained, it is noticeable that the trained model predicts the foliar disease in apple tree leaves correctly. In future, the process of predicting the foliar disease in apple tree leaves can be automated by the development of mobile apps and by incorporating the model in IOT.