This study presents an imaging system that integrates lateral shearing digital holographic microscopy (LS-DHM) with a convolutional neural network (CNN)-based deep learning classifier to detect and classify nanosized virus-like particles. The system was tested on three distinct particle types: RNA-encapsulated nanoliposomes, non-RNA nanoliposomes, and nano-hexagonal boron nitride powder, used as a size-control sample. The virus-mimicking particles were specifically prepared for this research. To facilitate classification, time-dependent concentration amounts (CA) were calculated from binary phase images reconstructed from captured holograms. Two experimental scenarios were conducted: in the first, each sample was introduced and classified individually; in the second, two different particles were simultaneously sprayed into the system to evaluate classification performance in mixed environments. In the individual classification scenario, accuracy ranged from 89.28% to 100%, depending on the particle type. When two particle types coexisted, classification accuracy remained high, achieving 91.42% and 96.00% for the respective mixtures. A key distinction of this study from prior works is its capability to classify airborne particles of the same size but different content in real time, without requiring surface immobilization or labeling. Furthermore, this work marks the first attempt to compare two virus-like particles with different RNA content against a real-time detection using LS-DHM. The system demonstrated an ability to differentiate particles based on physical parameters such as size and stiffness. These results indicate the system's strong potential for earlystage, label-free, and real-time virus-like particle detection applications.