Deep Learning-Assisted Sensor Array Based on Host–Guest Chemistry for Accurate Fluorescent Visual Identification of Multiple Explosives

Accurate and rapid discrimination of multiple explosives with high precision is of paramount importance for national security, ecological protection, and human health yet remains a significant challenge with conventional analytical techniques. Herein, we present an innovative deep learning-assisted artificial vision platform based on cyclodextrin-protected multicolor fluorescent gold nanoclusters (CD-AuNCs) with four distinct emission wavelengths, enabling the highly accurate discrimination of seven explosives. The sensor array leverages the host-guest interactions between the cyclodextrin ligands on the AuNCs' surface and the target explosives, generating unique fluorescence fingerprint patterns. Mechanistic studies reveal that the fluorescence enhancement of CD-AuNCs is attributed to ligand rigidification, while fluorescence quenching is primarily caused by photoinduced electron transfer between CD-AuNCs and explosives. The multicolor fluorescence responses are captured by using a smartphone, and the corresponding RGB values are simultaneously extracted. To enhance the recognition accuracy, a dense convolutional network (DenseNet) algorithm with advanced image recognition capability is integrated with the fluorescence sensor array. This platform achieves remarkable 100% recognition accuracy at a concentration of 200 μM, enabling the rapid and precise visual classification of explosives. The proposed strategy not only provides a powerful tool for on-site explosive monitoring but also offers a versatile platform for the intelligent detection of diverse analytes, demonstrating significant potential for real-world applications in environmental and security monitoring.