The absence of identifiers such as barcodes or QR codes for fruits in small retail stores often leads to manual product selection, inef f iciencies, and pricing errors. To address this challenge, this study devel ops a low-cost intelligent weighing system that integrates automatic fruit recognition directly on a resource-constrained edge device. The system is built on the ESP32-S3 microcontroller, which incorporates a camera for image capture, local processing, and result display without dependence on external computing infrastructure. A dataset of 11 fruit categories was collected from the domestic market and augmented to 11,000 samples, en suring robustness against variations in size, lighting, and background. A lightweight MobileNet V1 128Ö128 0.25 model, optimized through INT8 quantization, was deployed to balance accuracy and computational effi ciency. The system achieved 95.7% overall accuracy and up to 100% for most classes, with an average latency of 192 ms per cycle. These results demonstrate the feasibility of real-time, low-cost fruit recognition on em bedded platforms, offering a practical and scalable solution for intelligent checkout in small-scale retail environments.