A Hybrid CNN–Autoencoder–Random Forest Framework for Multimodal Plant Health Classification in Precision Agriculture

Abstract

This research presents a deep learning combination of Convolutional Neural Networks (CNN) with Autoencoders as well as Random Forest (RF) for plant health classification with multiple data sources. Both visual indicators from RGB leaf images and compressed non-visual data from physiological and environmental attributes are effectively incorporated by this system through its deep autoencoder. Random Forest classifies plant conditions as Healthy, Moderate Stress or High Stress after merging all features together. The hybrid model demonstrates a simulated accuracy rate of 98.5% according to experimental results while maintaining balanced precision and recall and F1-scores throughout all classes. Various tests including confusion matrices and PCA visualizations and epoch-wise performance trends support the stable performance of the proposed approach. The combination between image and sensor-based data delivers superior classification outcomes which establishes a practical solution for plant health monitoring systems in precision agriculture applications.