A rapid household mite detection and classification technology based on artificial intelligence-enhanced scanned images

Abstract

Household mites, recognized as a principal allergen, can induce allergic rhinitis in over 90 % of patients worldwide. It is indispensable to accurately assess mite pollutant exposure within living environments to heighten awareness regarding mite prevention. Current techniques for household mite detection and quantification, however, suffer from limitations such as complex sampling requirements, time-consuming analysis processes, and high costs, which ultimately contribute to a lack of awareness among residents. Therefore, this study develops an innovative artificial intelligence (AI) technique with multi-feature fusion for household mite detection and classification to evaluate indoor mite infestation levels. This system incorporates a symmetric Generative Adversarial Network (GAN) and multiple Image Signal Processing (ISP) models to not only enhance the visual quality of images obtained from scanned Dust Mite Traps but also facilitate data augmentation, thus significantly improving the detection, classification, and quantification accuracy of two prevalent household mite species: dust mite and Cheyletid mite. With the enhanced You Only Look Once (YOLO) model, the integrated AI framework demonstrates rapid and precise mite detection and quantification, achieving an accuracy rate of 85.4 % and a counting error of only 7.1 %. Furthermore, the visualization process improves human visual interpretation, effectively raising awareness about dust mite contamination for indoor environment quality. The proposed AI models offer a cost-effective, efficient tool for assessing mite infestation within homes and increase awareness about mite protection, thereby reducing the risks of exposure to indoor allergens.