Design and predictive modeling of a veterinary drug detection sensor in paddy field water based on artificial neural networks

For rapid real-time detection of veterinary drug residues in paddy field water, we developed a novel sensor system using interdigitated electrodes as detection probes and the STM32F405RGT6 microcontroller as the core processing unit. The hardware architecture integrates multiple functional modules including excitation signal generation, signal detection, signal processing, LoRa coupled with 4G wireless communication, voltage regulation, and lithium battery charging. The system acquires three types of measurement data (amplitude ratio, phase difference, and their combination) from water samples containing sulfamethazine, ofloxacin, doxycycline hydrochloride and tetracycline hydrochloride across a broad frequency spectrum from 200 Hz to 100 MHz. Through Competitive Adaptive Reweighted Sampling (CARS) for feature selection and artificial neural network modeling, we established a multi-input multi-output concentration prediction model. Comparative analysis demonstrated superior performance when using phase difference data as model input, achieving prediction coefficients of determination (R²) between 0.7831 and 0.8713 with root mean square errors of prediction (RMSEP) ranging from 22.0759 to 28.1526 mg/L. Studies showed that this sensor device could effectively detect the contents of four veterinary drugs, namely sulfamethazine, doxycycline hydrochloride, ofloxacin, and tetracycline hydrochloride, in paddy field water, thus realizing the rapid and real-time monitoring of veterinary drugs in paddy field water.