# High Accuracy Quantification of Aflatoxin B1 via a Compact Smart Gas Sensing System Assisted by Dual-Branch Convolutional Neural Network

## TL;DR

A compact smart gas sensing system with a dual-branch neural network accurately detects and quantifies aflatoxin B1 in contaminated grains, enabling real-time monitoring.

## Contribution

A novel compact gas sensing system with a dual-branch CNN for real-time, non-destructive aflatoxin B1 quantification in grains.

## Key findings

- The system achieves 100% accuracy in identifying grains infected with Fusarium graminearum and Aspergillus flavus.
- The DB-CNN model shows high quantitative performance (RMSE = 1.0292 μg/kg, R2 = 0.9994) for Aflatoxin B1 detection.
- The detection system supports wireless transmission and completes the process within 4 minutes.

## Abstract

Mycotoxin contamination of grains during storage and transportation represents a significant threat to global food security. Conventional detection methods exhibit limitations in terms of real-time monitoring. This study presents a compact smart gas sensing system for mycotoxins, facilitating non-destructive testing of corn infected with fungi by analyzing the volatile organic compounds (VOCs) emitted during fungal growth. It also facilitates the precise quantitative detection of Aflatoxin B1 (AFB1). Additionally, a dual-branch convolutional neural network (DB-CNN) model has been developed to conduct an in-depth analysis of the temporal and spatial characteristics of VOCs signals. The system achieves 100% accuracy in identifying grains (corn, peanuts, wheat, and rice) infected with Fusarium graminearum and Aspergillus flavus by extracting the characteristic fingerprint spectra of fungal VOCs. In the quantitative analysis, the DB-CNN exhibits good performance (RMSE = 1.0292 μg/kg, R2 = 0.9994). In addition, the designed detection system supports wireless transmission and can be connected to a smartphone for data transfer, thereby facilitating data storage and remote monitoring. The entire detection process is completed within 4 min. This study provides an innovative technical foundation for dynamic real-time monitoring of fungal contamination in the food supply chain, contributing to early warning systems and quality control measures.