Development of a Neural Network-Based Camera for Tomato Harvesting Robots

• Main Author: Mamat, Aziz
• Format: Thesis
• Language: English; English
• Published: 2008
• Online Access: http://psasir.upm.edu.my/id/eprint/7301/1/FK_2008_88a.pdf; http://psasir.upm.edu.my/id/eprint/7301/

Automated tomato harvesting robots were rapidly developed recently. Most of the designs were more focused on positioning of the end of robotic arm by using various methods such as combination of the sensor and vision system. This project concentrated on the artificial intelligent via the Neural Network, in order to provide a better decision making system for tomato harvesting robot. The objective of this study was to develop 3 degree of freedom tomato harvesting robotic system complete with gripper and motion program. The development of software for tomato pattern identification, determination of the X and Y coordinates from web camera captured and the determination of the tomato and tomato ripeness using decision making from Neural Network also become the main objective. The approach is to detect the desired object using vision system attached to the cylindrical automation system and perform image analysis. These features will serve as inputs to a neural net, which will be trained with a set of predetermined ripe tomato. The output is a command for harvester arm to make the movement for harvesting. The position determination was done with a conversion of the distance in pixel into a distance in metric unit (mm) of the tomato image. Whereas the depth of the tomato distance (z direction) was done by moving the actuator system towards the calculated tomato position until the object sensor senses the present of the tomato. AWIsoft07 software was developed to view the harvester vision, display the captured image analysis on the harvester vision, and display the numerical analysis output and neural network output. The harvester system with 3 degree of freedoms (3DOF) equips with specially designed tomato gripper named as AWI2007 Tomato Harvesting Robot was developed in order to realize the data from the AWISoft07 developed software. Several calibrations were made to ensure the accuracy of the AWI2007 Tomato Harvesting Robot. The AWIsoft07 and AWI2007 Tomato Harvesting Robot were subjected to several harvesting tests under the laboratory environment. The harvesting result shows the ability of the software and the harvester. Consequently, AWI2007 Tomato Harvesting Robot with the camera vision was able to recognize the tomato ripeness intelligently via neural network analysis and moved to the harvesting position. These situations provided new improvements for tomato harvesting system compared to the previous findings. Therefore the application of the neural network based on camera vision was successful perform as artificial intelligent for tomato harvesting robotic system.