The identification of banknotes is one of the main problems for the blind person. Money is a tool that is used in everyday life to carryout buying and selling transactions by all humans in every part of the world, so this makes money as a primary item for everyone, even for people with disabilities such as the visually impaired ones. The weakness of the visually impaired person in seeing and identifying money can cause into money confusion, mis-taken, or even deceived at the time of the transaction. Therefore, tools are needed to ease the visually impaired to identify the value of money. The purpose of this study is to propose one of the deep learning-based methods of convolutional neural networks (CNNs) that can be used to detect the value of Indonesian currency banknotes. Data sets used in the form of paper money images with an amount of 1000, 2000, 5000, 10000, 20000, 50000, 100000 thousand rupiah, which amounted to 100 images each, bringing the total images used as a dataset reaching 700 images. At the training stage, the CNN model that was built received 560 paper money image inputs (80% of the total dataset), while, 140 images used at the testing stage (20% of the total dataset). Once the trial completely conducted, it is obtained 94. 29% as the best accuracy in the 60th epoch with the kernel size in the first convolutional layer is 3x3 and the kernel size in the second convolutional layer is 2x2.