My uncle works in organic market gardening so I have already witnessed the hard work involved in this activity. Artificial intelligence and more precisely here the learning machine is a real step forward, which can play a decisive role in the market gardening activity. Thus, I basically wanted to combine machine learning and robotics technologies with permaculture in order to optimize water and nutrient management and reduce labor-intensive tasks. That is why I would like to create and develop an organic market gardening robot helped by machine learning.
Here, we will not go as far as I would have wished, that is to say to the step where the robot comes out because that is not the purpose of the assignment. Hence, in a first step I would like to create an algorithm which can recognize the product, know if it is ready to be picked up. If he is ready, the algorithm will then delegate the task to the robot and if it is not, he will be able to know what it needs to be picked up, such as :
It will be for professional market gardeners who want to optimize resource management, without having toresort to too many employees, to bring organic management closer by avoiding intensive agriculture. To allow the development of the sale of market garden products of rare species (not possible in intensive farming).
We are going to use the supervised machine learning thanks to Ml5.js which is a JS library thatprovides access to machine learning algorithms. We are then going to use the classification and our knowledge of beans. That is why we need to know very perfectly the cycle of each bean so as to have the best image database we can.
We will have a camera on a gridded area of the cultivated land, then thanks to ml5.js and Mobilenet we will classify what the camera shows. We need to teach to the AI all the parameters we want it to recognize and to analyse.
Let's then focus on the tomato cultivation and let's try to find this crucial information so as to know if our tomato is ready to be picked up or not.
First, we need to know if the tomato is ready to be picked up. To do so, we will focus on its color. According to careplant the following image let us know about the link between maturity of the tomato and its color. One should teach the machine to learn that when the tomato is green (at the left of the picture), it is not ready to be picked up. But when it is red (at the right of the picture), it is actually ready to be picked up
Now that we know if the tomato is ready to be picked up or not, we could go further and try to know what does the tomato need so as to be picked up such as :
It is a little bit complicated to find such a classification so we are going to leave our project here.
a) Farmbot
One similar project would be the Farmbot robot which is basically designed to make a vegetable garden self-sufficient. The difference will be one addition of a parameter ultimately that is the permaculture resource management, allowed by the robot. Here we have been asked to focus on a little part of the robot project so Farmbot is not exactly a direct competitor, but I wanted to underline how interesting the project could be if it goes as far as the robot step.
Introducing Farmbot genesis, source : Farmbot
b) Pl@ntnet
A competitor at our level would rather be Pl@ntnet
Indeed, it is an application that allows you to recognize all types of plants very easily and quickly. You just have to upload an image and select whether it is a flower, a tree, a leaf or a fruit. As it has a database of more than 4,000 species, it tells you which one corresponds to the name, family and geographical area to which it belongs But actually, our project would go much further with the ability to know if the product is ready to be picked up. It will then be very useful to the farmers.
c) FarmIA
as shown by this article : FarmIA
To feed 9.2 billion people in 2050, agriculture will have to become more productive and efficient, and artificial intelligence can play an important role in this trend. technology is already being used to increase yields: drones, thermal cameras and other moisture sensors are already part of the daily life of farmers around the world. With the explosion in the amount of data from these tools, artificial intelligence is becoming essential to analyze them and help farmers make the right decisions. The robot is programmed to analyze the plants and detect their needs, in terms of care and feeding.