Monitoring the viscosity of liquids is crucial for quality control across various industrial sectors. Traditional methods, like capillary viscometers, are commonly used but pose challenges due to their cost and complexity, especially when real-time viscosity information is essential. This paper introduces a deep learning model for viscosity estimation from droplets images, incorporating transfer learning with the MobileNet architecture. We prepared fourteen solutions with varying water and Polyvinylpyrrolidone (PVP) ratios, measured their viscosities, and recorded videos of their droplets. Extracting fully developed droplet images from the videos, we trained a MobileNet Enhanced model to estimate viscosity values for the water-PVP solutions. The model demonstrated strong performance in capturing droplet features and was able to accurately estimate the viscosity values of samples of unseen chemical formulations with the same composition.