Eyes of Demeter: Low-Cost Multi-Modal Early-stage Crop Elemental Stress Sensing with Spectrometer and Camera Fusion

Crop Element Stress (CES), caused by the excessive accumulation of essential nutrients such as nitrogen, phosphorus, or potassium in crops, plays a critical role in crop growth, development, and final yield. Existing CES detection methods lack standardized measurement techniques, and large variations in crop characteristics limit their effectiveness. Prior research has proposed diverse approaches; however, these methods are limited because they either detect CES only at a late stage, after irreversible yield loss has occurred, or require costly equipment exceeding $20,000. These limitations create a substantial gap between existing CES detection methods and their practical deployment in real-world agricultural settings. This paper introduces EDemeter , a low-cost and user-friendly system that uses spectral and RGB imaging to detect early-stage CES, enabling smallholder farmers to conduct high-precision crop monitoring. First, we design a dedicated signal processing pipeline to align the dimensions of spectral and RGB images and perform forward and backward fusion to generate feature images. We then develop a transformer-based encoder to capture local sequential dependencies among adjacent spectral bands and extract group-level spectral features, which are subsequently fed into a classifier for CES detection. Moreover, we augment the training dataset to improve the model's generalization capability.; AB@The EDemeter system employs a low-cost ($36), lightweight spectrometer operating across the 400-760 nm wavelength range, enabling field-proximal diagnosis of elemental stress under controlled illumination. Experimental results demonstrate that ten days after the onset of elemental stress, EDemeter achieves a detection accuracy of 89.06% on a single sample. When five samples collected from the same field are tested together, the accuracy increases to 98.9%, only 1% lower than that achieved by expensive equipment. Considering that a large number of samples can be collected from each field in real agricultural scenarios, this system can serve as a cost-effective solution for early-stage elemental stress detection. Our code is available at: https://anonymous.4open.science/r/EDemeter.