Smarter algorithms could cut power use in gas-detection devices by half
Researchers have shown that optimizing neural network algorithms—rather than building better hardware—can dramatically improve how well electronic nose sensors detect dangerous gases and spoiled food. For manufacturers, the finding suggests a faster, cheaper path to wider adoption of these safety-monitoring devices across industries from food production to chemical plants.
Originaltitel: Neural network for gas recognition based on MOS electronic nose: Algorithm design and hardware deployment
<p>Electronic noses typically comprise gas sensor arrays, signal-acquisition electronic circuits, and patternrecognition algorithms. Compared with other similar ones, metal-oxide-semiconductor-based electronic noses offer long lifespans, high accuracy, and competitive cost; they have been widely employed in safety monitoring, food quality monitoring, and so on. However, high power consumption, complex software, and inconsistent performance in classification and regression currently limit the wider application of this technology. Algorithm optimization is found to be a critical route for enhancing gas analysis performance in electronic noses, as it can reduce extra hardware costs while sustaining high gas recognition performance. This review compares two key algorithm architectures: artificial neural networks and spiking neural networks. A wide range of popular artificial neural networks is covered in detail, including multilayer perceptron, convolutional neural networks, and recurrent neural networks. The article discusses the pros and cons of these networks, along with the most current findings from their electronic nose applications. We also includ reviews and comparisons of gas classification/ regression models that use spiking neural networks, highlighting their architectures and accuracies. Ultimately, we review the current state of gas classification and regression using artificial neural networks and spiking neural networks, and provide the future trends of neural network for electronic nose applications.</p>