Deep Learning-based Enhanced Visual Servoing for automated fruit-sorting robot

This article presents a novel application of deep learning in automated fruit-sorting robotics, improving real-time object recognition and manipulation. The robot, with integrated advanced neural networks, achieves heightened precision in distinguishing diverse fruits. This innovation addresses fruit variability, offering adaptability for improved sorting accuracy. The fusion of deep learning and visual servoing represents a significant advancement in automated fruit-sorting technology with potential benefits for optimizing agricultural processes. The project aims to control a fruit-sorting robot using information obtained from image processing. The methodology integrates sliding mode and neural network (NN)-based control approaches for the automated manipulator. A Convolutional Neural Network (CNN) is trained on a diverse dataset of fruit images to classify fruit types. The results of CNN can be used to design the optimal path of the robot. The sliding mode control handles uncertainties, guiding manipulator movements with a sliding surface. Simultaneously, an NN is trained to control joint angles. The comparison of the two controllers on a simulated robot revealed insights. The NN-based controller demonstrated superior accuracy and speed, adapting to varying fruit configurations through learned mapping of joint angles. The sliding mode controller, while robust and stable during dynamic movements, exhibited sensitivity to uncertainties, impacting sorting precision. The hybrid system, seamlessly integrating both controllers, will enhance adaptability by combining the precision of the NN-based approach with the stability of sliding mode control for optimal fruit-sorting performance in diverse scenarios. Results highlight choosing the right controller, balancing precision and speed, based on specific application needs.

کلیدواژه ها:

Image classification ، NN-based control ، Sliding mode control ، Visual servoing