The growing demand for multi-nozzle inkjet printing has created a need for automated status monitoring algorithms. This study proposes an efficient and accelerated deep learning model for inkjet printing monitoring, which utilizes a residual neural network with a translation-insensitive, linear transform for input size reduction. Three representative ejected droplet statuses – none, non-spherical, and spherical – are classified by this model. The condition of the jetting is inspected by analyzing each frame of the process video in sequence. Thanks to the reduced input size, both training and inference speeds are improved by 4.4-fold and 2.4-fold, while it preserves more features than simple interpolation methods. The model also achieves an inference accuracy of approximately 96% within 5 epochs, whereas the uncompressed model can only obtain ∼92% of accuracy. In addition, a total of 125 consecutive frames acquired from realistic inkjet printing are examined by this model, and the comparison with the ground truth results in 97% of accuracy. The resulting inference speed was 14-fold faster than the recording speed (25 fps) of CCD cameras, highlighting its real-time application capability. To extend the applications, a multi-class droplet categorization was also provided, which can more precisely sort geometries based on tail length. The present study will help bridge the gap between deep learning algorithms and industry by reducing computational costs.