This work presents a real-time counterfeit purse detection system built on the You Only Look Once(YOLOv12) object detection framework, designed for deployment on mobile Android devices. We address the challenge of limited labeled authentic/counterfeit purse data by creating a custom synthetic dataset that enables effective model training. Our approach involves training the YOLOv12n model over 20 epochs with optimized hyperparameters (learning rate 0.001, weight decay 0.0005) to perform simultaneous object detection and binary classification of purses as authentic or counterfeit. To make the system suitable for mobile deployment, we implement a comprehensive optimization pipeline that includes L1 unstructured pruning at (30%) sparsity across convolutional layers, followed by fine-tuning for another 20 epochs at reduced learning rate (0.0001) to maintain accuracy. The model undergoes further compression (75%) through post-training INT8 quantization using PyTorch’s Facebook General Matrix Multiplication(FBGEMM) backend, significantly reducing memory footprint while preserving detection performance. We evaluate our system using precision, recall, and mean average precision (mAP)of (94%) metrics on held-out test data, demonstrating reliable performance at confidence thresholds of 0.4 with an overall accuracy of (86%). The final optimized model is prepared for Android deployment, enabling efficient on-device inference for real-time counterfeit detection applications. Our implementation successfully achieves accurate purse authentication while maintaining real-time frame rates suitable for mobile embedded systems, offering a practical solution for anti-counterfeiting applications in retail and security environments.