Surface electromyography (sEMG) based recognition of hand gestures controls robotic arms or prosthetics and assists in rehabilitation. Controlling these devices requires data acquisition and a deep learning model to classify gestures for precise control. However, deep learning models are highly sensitive to the raw sEMG signal data and struggle with variations in the magnitude of sEMG signal features. Therefore, this study proposes an approach to address this issue. This approach combines preprocessing of the raw sEMG signal data from two channels, normalization, and a sliding window technique to provide meaningful data to the deep learning model (1D-CNN). Specifically, the proposed framework preprocesses the data (DWP), normalizes it using Min-Max normalization (MMN), and applies a sliding window (SW) approach known as “DWMS” to classify the gestures. The DWMS achieves the highest accuracy compared to other scenarios, including those without preprocessing and with preprocessing using other normalization techniques such as Z-score, Robust scaling, and Root Mean Square. The results of the DWMS, in terms of accuracy, precision, recall, and F-score, are 81.72%, 81.47%, 81.72%, and 81.38%, respectively.