AI and Food Industy - The new Agriculture (Part I)

Artificial Intelligence is proving to be one of the keys to the next human revolution. We can find AI everywhere we go or even do. In these days, almost everything that runs a software uses AI. Thus, we are in contact with AI every day, even if you only use a smartphone. 

Food - Photo by Dan Gold on Unsplash

It uses AI in a bunch of apps, starting on the camera app. Although we are using AI in so many devices, some people are not aware of it. An example is the use of AI in the Food Industry. Thereby, I am starting this series of posts to show you where AI is being used and how it is evolving the Food Industry.   

Food is a human being need. Do you know how many are we? We are very close to being 8 billion people on earth. Moreover, the projections point to more than 9 billion people in 2050, and more than 10 billion by 2100. 

We all need food to survive. Agriculture occupies 50% of the habitable land and requires 70% of the water in the world. Some studies indicate the necessity of increasing food production by 60% to fulfil the human's food needs by 2050 [1,2]. As you can see, using today's methods, we will need one brand new planet very soon. Off course there is no such thing as a new planet. Thus, we must use our intelligence and tools to overcome the food-need challenge. Indeed, as you expected, AI is one of the principal keys, perhaps the main one, in the food production and distribution development. 

Artificial Intelligence can be applied in almost every step of the food chain, starting in the food cultivation, passing by its treatments and validations, and finishing in the food recommendation or delivery.

When it comes to the growth or cultivation of the food products, we need to differentiate them from the final industrialised food products. We are talking about the ones essentially produced in the farms without pos-crop human manipulation. These foods are natural and grown by farmers, such as carrots (natural here is not affirming biologic). The increase in the world population generates some challenges for farmers and the food industry in general. How to produce more with the same amount of land? How to be more productive? How to reduce the water and fertilizers and still get more quantity of product from the same soil?

Farming was revolutionized with machines to increase its outcome, however, AI approaches will come up with new tools to achieve better results in terms of food quantity, quality, and land smart usage. This is not only an improvement for human self-satisfaction, is the demand for food that requires these new techniques.

Agriculture - Photo by Naseem Buras on Unsplash

Big farms require big and fast tools to handle all the production of new foods. A great example is the tractors and its implements of different types. However, tractors are not autonomous. We need people to drive them and perform some decision making, as putting the tractor implements down, up, to turn them on, and so on. Moreover, some of the tractor attachments still need improvements to complete their purpose in a better way. Yes, you got it right! AI can leverage the tractors use by decreasing the time spent to complete some tasks, increasing the foods production either in quantity and quality, and economising in different types of resources. Let's see some examples.

Well, we need tractors to be more autonomous, so... autonomous tractors using Computer Vision (CV) are arriving at the farms. Indeed, the use of Computer Vision and deep learning techniques in so many new agricultural tools is the major factor of AI's growth in the agriculture market.

These self-driving tractors are developed to autonomously detect their ploughing position into the fields or decide the best speed needed to complete the task. They can also avoid obstacles like irrigation objects, humans, rocks, other vehicles, and animals while performing their tasks. The companies behind these autonomous tractors are also deploying tracking systems that allow the farmers to know the tractor's current position and also to interact with them if necessary. Connected to the driverless tractors, we have new implements to automate the farmers' tasks even more. Companies like Blue River Technology and Harvest CROO Robotics are making robotic devices that can control undesired crops or weeds to assist farmers in selecting or collecting crops with bigger volumes. 

To view these machines working, I recommend watching these videos:

• Autonomous Tractors 
• Blue River Technology's

In the previous examples, we have the use of AI applied through CV techniques. But, what is Computer Vision and how it works? Indeed, it is a topic with much to say. For now, I will compile some of the most important points. As its name states, Computer Vision is the computer's ability to understand the surrounding environment through images or video. The main goal is to, somehow, mimic the human's visions capabilities. This process combines different tasks for acquiring, processing, analysing and understanding the digital images captured by the system's camera. Note that in the case of real-time CV, the system reduces the video to images. The video is split by each frame and the algorithm process each frame as an image. After having these images to process, the algorithms convert them into numerical or symbolic information, as we can see in the following two example's images. 

Flower and an example of an image converted to numerical information - Photo by Honey Yanibel Minaya Cruz on Unsplash

It is important to note we always have three matrixes of information, for each image, like the one in the example of the flower. Each matrix corresponds to each RGB channel. Thus, we have a matrix for the red, blue, and green channels. This means we work with three-dimensional matrixes. These images are then the inputs to the AI model that returns, for instance, if some item is inside the image or not. Before we have a model ready to give us the computer predictions, we need to train the computer so it can learn the morphology of the things we want the computer recognises. In the training phase, we need to collect big amounts of labelled images to feed and train the AI models used to categorise the images, assuming we are using a Supervised Learning approach. Just to keep in mind, some CV systems are now being tested using Unsupervised Learning techniques to make them more robust and generic when recognising images. Normally, the AI models we use in CV systems are built with convolutional layers. The convolutional neural networks are specialised in finding patterns, as we need in the CV field. If you do not know how CNNs work check my post about it at Neural Networks, how do they work?. Thereby, based on the given training data, the algorithm finds patterns in the images and correlates them with the given label/description. Then, when the algorithm receives an image, he looks for patterns he knows. If he finds something that matches both full or partially some pattern, he recognizes the item with an associated value of certainty. Very briefly, and roughly explained, this is how CV works. Later in this blog, I will post more technical information about CV.

Cabbage after a pest attack - Photo by Markus Winkler on Unsplash

One of the biggest threats to the plants' growth is pests. It can be fungi, bugs, or even bigger wild animals destroying and chewing everything. AI companies are now starting to develop systems to help farmers with this problem. They use satellite or drone images to find pests or alert if some pest is in the neighbourhood. Using these AI systems, the farmers are now able to act timely, preventing crop losses and, off course, getting more profit. 

Pests can also make the plants sick, however, they are not the only cause. The continuous degradation of ground quality, and the deforestation itself, are a big challenge in many countries. The malnourished soil is not suitable for seed or plant development. Therefore, we are now seeing the first companies coming out with AI systems to analyse the ground and report for nutrient deficiencies. PEAT is one of these companies that has an app based on image recognition, where the user takes photos of plants with a smartphone. The app, called Plantix, tells the user if the plant has nutrient deficits or even if it is being attacked by a pest. Moreover, the app also suggests how to treat the soil or the plant, based on the cause it had identified. In the same line of thinking, another company applied a similar approach to viniculture. SkySiquirrel, then acquired by VineView, uses the image recognition approach to drones. The drone flies above the fields giving reports to the user. 

While using these algorithms and getting all of this data, the AI algorithms can also predict the weather conditions, and analyse the crop sustainability for certain regions. Moreover, they can evaluate the presence of diseases, nutrients deficit, or even pests, based on data such as temperature, precipitation, wind, and solar radiation. As we can see so far, AI reduces the time spent to perform the farmer's tasks and increases the amount of crop using the same land. Moreover, the food quality is also better because the plants or seeds are treated according to their needs and not according to an overall need. Now, farmers can have specialised treatments for each sector of the field. Indeed, this means fewer pesticides applied, less wasted water to apply the pesticides, and less money spent to grow the farm.

After all this information, what do you think it is the next most needed AI solution to agriculture? Please let me know in the comments below.

References:

1. How much of the world’s land would we need in order to feed the global population with the average diet of a given country? - Our World in Data. (n.d.). Retrieved June 11, 2020, from https://ourworldindata.org/agricultural-land-by-global-diets
2. Water for Sustainable Food and Agriculture A report produced for the G20 Presidency of Germany. (n.d.). Retrieved from www.fao.org/publications