Plant Sorting System



Plantation Zwethlande has 30 years of experience in the growing of Ficus pot plants. Because of these years of experience in this field, the company is famous for their high quality plants. The company has the ambition to deliver a high diverse assortment of green plants and blossoming plants, where quealtity is of the utmost importance.


Because Zwethlande is always seeking for better and more innovative solutions for their plants to improve quality and reduce costs, they handed in a Assignment in at SMR to automate sorting Yucca plants. In the current situation the plants are sorted semiautomatic. The plants are supplied on a conveyor belt where they are sorted on height. Plants that are to small will be transported back to the greenhouse. The other plants will are sorted manually in the categories one, two and three branches. For us the task to classify these plants and put them in the baskets.


The system can be divided in to two mayor categories the vision and the robot system. The vision system classifies the plants in the three categories. For the classification machine learning is used. At the customer a dataset was created of about 600 plants. From each of these plant-photo’s 26 feature are calculated with visionsoftware (plantcv and opencv was used). The machine learning algorithm will train on this dataset. The machine learning model that was created from the dataset will predict the category of a new taken photo with the calculated feature input.

The robot system is build in ROS (robot operating system) to make the system as universal as possible. Using the MoveIt package with the trac_ik resolver we use path planning the move the robot to the correct position. The robot is aware of the environment, so it will never collide with the objects in the environment. Using ROS we can also easily swap any component (robot control, vision, plc) for another component. For example, we could easily use another robot without changing the code too much.

You can also see the results in the webapplication. The webapplication is real-time (SocketIO), so you see the changes immediately.


Screen Shot 2017-10-24 at 13.59.15