<![CDATA[Sign language recognition systems play a crucial role in bridging communication between individuals with hearing impairments and the general public; however, their development often faces challenges in balancing accuracy and computational efficiency on mobile devices. This study aims to design, implement, and evaluate a lightweight Convolutional Neural Network (CNN) model based on SSD MobileNet V2 to optimize the accuracy and efficiency of real-time Indonesian Sign Language (BISINDO) alphabet recognition on Android devices. This research adopts an applied experimental approach with a deep learning–based system development design. The research subjects consist of 3,878 hand gesture images collected from various contributors, representing 26 BISINDO alphabet letters with variations in hand size, skin tone, and gender. Data were collected through image acquisition and labeling using LabelImg, followed by analysis using transfer learning and performance evaluation via the TensorFlow Object Detection API. Data analysis involved measuring precision, accuracy, and model learning rate to assess system effectiveness. The results demonstrate that the proposed model achieved an average precision of 89% and a real-time recognition accuracy of 93%, outperforming the baseline MobileNetV3 model. These findings confirm that lightweight CNN architectures can provide an efficient and reliable solution for sign language recognition on low-power devices. Overall, this study concludes that the integration of deep learning and mobile technology can deliver inclusive innovations that expand communication accessibility for individuals with hearing disabilities. The implications of this research highlight the importance of developing user-friendly artificial intelligence–based systems that support digital equity in educational and social contexts.]]>
