Recently, analyzing multiple types of fake banknote recognition and detection is a key concern in finance and business. Fake detection is an increasing methodological approach with the significance and technologies in an enormous amount of banknote image data with high dimensionality and unprecedented speed, which leaves a massive data gold ore waiting to be mined. Therefore, in this paper, we proposed a deep CNN technique to differentiate between real and fake banknotes using the fake detection method by examining the computer vision features of the digital content for detecting fake banknotes using smartphone cameras in a cross-dataset environment. The proposed CNN model is used to classify and detect real and fake banknotes datasets for Ethiopian banknotes confirming that the proposed algorithm demonstrates a higher detection accuracy. The detection model sequence includes image acquisition, Image size normalization, grayscale conversation, and histogram equalization, which support to reducing the number of parameter counts in the convolutional layer in the DL framework with high performance. The proposed model architecture results in less computational complexity during hardware deployment and model training. The impact of parameter reduction on model accuracy is analyzed by evaluating the proposed Customized model. We used the percentage method to split the banknote dataset into training (80%), validation (10%), and testing (10%). After a different experimental iteration of the proposed model, we get 99.9% training accuracy, 99.4% Validation accuracy, and 97.6% testing accuracy.