Conveyor belt systems facilitate the movement of goods and materials, making them an essential part of various fields. Deep learning control methods are therefore indispensable to assure their energy efficiency and optimal motor speed. Here, training is conducted via a convolutional neural network with transfer learning using VGG16 base net. Feature extraction layers are fixed, whilst fully connected layers are replaced. Thirty-three appropriate object classes are considered for classification, and images are acquired using an overhead camera. The obtained weight values are transmitted to Arduino that uses these data to control the stepper motor speed for optimal performance and minimum energy consumption, achieving approximate estimates of weight and efficient control of the motor. This work integrates the following hardware components: an Arduino microcontroller, a NEMA17 stepper motor, a GY-219 digital current sensor (INA219), an L298N motor driver and a high-resolution camera that processes images using Python with TensorFlow and OpenCV. These components enable the conveyor belt system to realise precise classification and instantaneous regulation of objects. Experimental results show the effectiveness of the proposed system in energising and managing the conveyor belt with an accuracy of 99.47% and a loss of 0.04874, highlighting how we can use the developed system to maximise energy economy and classification accuracy. For academics, engineers and business leaders, this study provides important new perspectives on the current scene and expected future developments in this expanding field.