A Mobile Deep Convolutional Neural Network Combined with Grad-CAM Visual Explanations for Real Time Tomato Quality Classification System

This study develops a control system to classify tomatoes based on deep learning algorithms. In the main part, the CNN model was designed and trained to classify RGB image of tomato from 2D camera. In the hardware part, a conveyor system was design to implement and test the algorithms. This system contains a conveyor belt, pneumatic pistons, camera, embedded computer and its control circuit. In the third part, the CNN model was deployed into product through the embedded computer to interactive with actuators. To validate in practice, the system was tested to run in real-time and the authors measured the classify capability of this system. As the outcome, the system worked well with high speed and high accuracy, additionally, it is very intuitive with the visualization of model prediction.