Abstract

Early leaf diseases are hard to identify based on crop type, environment, or symptoms. While there have been a number of deep learning algorithms applied to this topic, the vast majority of them are region-only and unable to differentiate between different kinds of crop types or diseases. To address this, this work employs an improved Convolutional Neural Network (CNN) model to recognize and classify plant leaf diseases. The improved CNN model uses leaf pictures and hyperparameters like learning rate and optimizer to identify illnesses. The technique was tested using the New Plant Disease Dataset. The study also showed how learning rate and optimizer affect model training. On the dataset for leaf disease, new optimized CNN model with enhanced reduced Learning Rate (LR) feature got the accuracy of 99.58% when the learning rate is 0.0001 and got 95.59% accuracy where the learning rate is 0.0005 with a multi-class average 99.02%. The proposed model was also evaluated on New Plant Disease Dataset, other sequential CNN models, and transfer learning models and established higher accuracy and reduced the computational complexity. Precision from the model can be predicted as 97.53% and recall also as 97.53%, whereas, the F-measure can be predicted as 97.33%. According to the intrinsic analysis, it can be concluded that the impact of the optimizer option lies in the fact that the model reaches approximately the same performance level when it is set on the corresponding batch size and learning rate. Furthermore, it was found that lower learning rates provided a better performance than the higher learning rates indicating that using small batch size in conjunction with a small learning rate, produced the most effective model.