An electronic nose based on adaptive fusion of transformer-ELM with active temperature modulation algorithm for accurate odor detection in refrigerators

Accurate detection of food spoilage in refrigerators is crucial for ensuring food freshness and safety. However, due to the wide variety of gases emitted by decaying food and their uneven distribution of gases within the refrigerator, current mixed odor detection methods are not satisfactory. This study proposes a dedicated algorithm for a refrigerator electronic nose that enables precise classification of mixed food odors and prediction of their intensity. To achieve this objective, a dataset of food odor samples was collected from refrigerators, and sensory identification as well as gas chromatography-mass spectrometry analysis were performed to obtain freshness and intensity labels. The developed electronic nose algorithm incorporates key techniques, including active temperature modulation and an adaptive fusion model of lightweight Transformer-ELM, to enhance sensitivity, selectivity, and global modeling capabilities for identifying abnormal odors in volatile compounds of mixed gases. Experimental evaluations on a large-scale dataset demonstrate the effectiveness of the proposed method in classifying refrigerated food freshness and predicting odor intensity. This research contributes to the field of electronic nose technology and has potential for applications beyond refrigerator odor detection.