Anomaly detection in hydroponic maize fodder through image processing

Abstract

This work presents a specialized automated monitoring system for hydroponic maize green fodder that addresses key limitations of existing plant disease detection approaches. Unlike prior works focusing on field crops with high-end computing infrastructure, our system operates on low-cost ESP32-CAM hardware while achieving competitive accuracy. The system integrates classical HSV-based segmentation with MobileNetV2 classification, exploiting multiple camera viewpoints (superior and inferior) for enhanced diagnostic reliability. Fixed-size patches (250×250 pixels) extracted from segmented regions serve as input to the binary classifier (healthy/diseased). The network achieved 95.01\% accuracy for superior view and 97.28\% for inferior view, comparable to state-of-the-art approaches using computationally expensive architectures. Key contributions include: (1) specialized preprocessing for hydroponic imaging conditions, (2) edge-compatible deployment maintaining high accuracy, (3) integrated height measurement for comprehensive crop monitoring, and (4) multi-view assessment enhancing diagnostic confidence. The system demonstrates the feasibility of deploying efficient deep learning solutions in resource-constrained agricultural environments.