A Vision-Driven Approach for Plant Health Monitoring via Convolutional Neural Networks

Abstract

Abstract: The early and accurate identification of plant diseases plays a critical role in enhancing agricultural productivity and minimizing crop losses. In this study, a deep learning-based solution has been developed using Convolutional Neural Networks (CNNs) to automate the classification of plant leaf diseases. A total of 17,088training images and 4,273test images were used, covering multiple disease categories. The system employs an image preprocessing pipeline that includes data normalization and augmentation techniques such as rotation, flipping, and shifting to improve the robustness of the model against real-world variations. A custom CNN architecture was constructed using sequential layers including convolutional, max pooling, dropout, and global average pooling layers, followed by dense layers for classification. The model was trained using the Adam optimizer and categorical cross-entropy loss function over 50 epochs. It achieved a classification accuracy of approximately 93.4% on the test dataset. The results demonstrate the model's ability to generalize across different disease types with high reliability. Visual validation through test samples further confirmed its effectiveness. The trained model is exportable for integration into real-time agricultural advisory systems. This study showcases a scalable and lightweight architecture suitable for deployment in resourceconstrained environments, including mobile and IoT platforms. The proposed system has the potential to support farmers in making timely and informed decisions for crop disease management.

Code and data availability

The paper describes a CNN for plant leaf disease classification using a dataset (17,088 training / 4,273 test images) and an HDF5-trained model, but provides no dataset name, repository link, code deposit, or availability statement. No paper-specific public asset is actionable.

No evidence-backed public reproduction asset is currently recorded.