The system is built on an Arduino Mega 2560. A 12 V lithium-ion battery supplies two LM2596 step-down modules, producing a 6 V rail for three MG995 servo motors and a separate 5 V rail for the Arduino, load cell amplifier, IR sensor and control buttons, joined at a common ground. Separating the rails prevents servo inrush current from resetting the microcontroller.

Weight is measured by a 1 kg single-point load cell with an HX711 24-bit amplifier connected to pins 34 and 35. An LCD 1602 I2C display is connected through SDA pin 20 and SCL pin 21, showing the real-time reading and the running count of each weight category. Green and red LEDs indicate the Light and Heavy classifications.

The firmware uses a non-blocking state machine based on millis() timing rather than delay(), so the IR sensor and control buttons remain responsive throughout the cycle. A stability detection routine accepts a weight reading only after ten consecutive samples each fall within 0.5 g of the previous one, which removes the bounce that occurs when fruit drops onto the platform. The hopper rotates in 20-degree increments with a 350 ms pause between steps to prevent fruit from being thrown out. The Reset function requires three confirmations to prevent accidental clearing of recorded data.

The classification threshold is implemented as a configurable parameter rather than a fixed value, allowing farmers to adjust the grading boundary to suit their own market conditions.

Testing across ten complete runs produced a standard deviation of 0.09 g to 0.24 g per sample and 100% classification accuracy over 50 fruit cycles. Known limitations include behaviour near the 64–66 g boundary, which was not verified, and the weighing-location dependency of a single-point load cell.