Low-cost RF sensor and machine learning system for early detection of nematode infestation in walnut trees

Overview

This solution combines a low-cost, pocket-sized radio frequency (RF) tridimensional sensor with machine learning algorithms to enable early, non-destructive detection of nematode infestation in walnut trees. Nematodes are a major threat to the walnut industry, and early identification is critical for effective management decisions and crop protection. By measuring RF reflectance from walnut leaves and applying classification models such as Neural Networks, Support Vector Machines, and Random Forest, the system classifies trees into four infestation levels based on root-lesion nematode counts.

The approach offers a scalable, field-deployable alternative to traditional laboratory-based nematode assessment, which is time-consuming, destructive, and expensive. Growers and agricultural managers can use this tool to monitor orchard health more frequently, identify hotspots earlier, and apply targeted interventions before infestations spread.

Technical specifications

• Sensor hardware: Walabot, a programmable pocket-sized RF tridimensional sensor that captures reflectance data from walnut leaves across multiple radio frequencies
• Machine learning pipeline: Classification models built with scikit-learn, including Neural Networks, Support Vector Machines, and Random Forest
• Sampling protocol: Multiple measurements per tree (five replicates) across multiple blocks to reduce error and account for variability
• Classification output: Four nematode infestation levels based on root-lesion nematode counts, enabling tiered management responses
• Current best performance: Neural Networks model achieving 82% classification accuracy on field-collected data
• Data collection site: UC Kearney Agricultural Research and Extension Center, with additional field experiments planned

Technology readiness level

The technology is at an early-to-mid stage of development (TRL 3-4). Proof-of-concept validation has been completed using 2019 field data collected from 15 sampling trees across six blocks, demonstrating measurable correlation between RF reflectance and nematode infestation levels. The team is planning additional year-long field experiments to expand the training dataset, which is expected to improve model accuracy and robustness. Further validation in diverse orchard conditions and across multiple growing seasons is needed before commercial deployment. The research group is open to collaboration with industry partners, agricultural technology companies, and walnut growers to advance field validation, expand data collection, and co-develop a deployable product.