Counterfeit notes remain a persistent concern for financial systems, particularly in countries with high currency circulation. To address this, we explore a deep learning approach tailored for Indian banknote recognition. The study evaluates three models: MobileNet alone, MobileNet combined with a Support Vector Machine (SVM), and MobileNet paired with Random Forest. MobileNet is chosen for its lightweight structure, which allows efficient processing of image data without requiring heavy computational power. The role of the MobileNet in the hybrids is the extraction of the intrinsic visual features, and the SVM and Random Forest are the decision-making classifiers making the decision more precise. This layered design makes it possible to strengthen classification accuracy and reduce errors. The overall contribution lies in showing how compact neural networks, along with classical machine learning methods, can form practical solutions for reliable counterfeit detection.