Most of current self-checkout systems rely on barcodes, RFID tags, or QR codes attached on items to distinguish products. This paper proposes an Intelligent Self-Checkout System (ISCOS) embedded with a single camera to detect multiple products without any labels in real-time performance. In addition, deep learning skill is applied to implement product detection, and data mining techniques construct the image database employed as training dataset. Product information gathered from a number of markets in Taiwan is utilized to make recommendation to customers. The bounding boxes are annotated by background subtraction with a fixed camera to avoid time-consuming process for each image. The contribution of this work is to combine deep learning and data mining approaches to real-time multi-object detection in image-based checkout system.