Real-time plant disease identification via CNN-based leaf image classification and mobile deployment

A key problem of agricultural productivity is crop diseases as it decreases productivity which affects food security and farmer&s;s earnings greatly. Manual detection is time-consuming, expensive, and has high error margins. This is especially true in bigger fields and when the symptoms of the diseases are hard to notice. Thus, this study leverages the CNN approach and focuses on automated detection of diseases from images of plant leaves. The model uses the Plant-Village dataset that has images of leaves that are labeled as either healthy or infected. The study take advantage of the VGG-19 architecture as it has proven to provide strong results in classifying images. The study uses Keras to do the necessary preprocessing and augmentation to further strengthen the model. Adam is used with categorical cross-entropy to optimize the model and applied early stopping to promote better generalization. The system achieves an accuracy of 95.6% on 50th epoch. The model deployed Tensor-Flow Lite and combined it with an Android application which facilitates real-time detection using smartphone cameras even when there is no internet connection. This system is an affordable and easy to use system which classifies images with high accuracy in addition to offering pointers and confidence levels. This model enables better precision farming through proper use of crop management.