9.7 billion. That will be the population of our planet in 2050. This exponential growth in the world's population will require an increase of nearly 70% in food production in the coming decades.

This implies a redoubling of efforts by the agricultural sector to address not only its productivity and competitiveness issues, but also emerging challenges with regard to environmental protection.

The good news is that in terms of production methods the agricultural sector is on a par with other, reputedly high-tech, industries. Increasingly innovative solutions, developed in the spirit of more efficient and responsible farming, are gradually taking over fields and meadows around the world.

Machine learning, smart farming’s latest weapon

Agriculture is already connected

Following the example of industry, the agricultural sector is on the threshold of its fourth "revolution". Today's farms, already highly robotised, automated and connected, are becoming ever more intelligent through a combination of leading edge technologies and increasingly autonomous production tools.

This is smart farming, or precision agriculture, which incorporates new technologies into agricultural techniques to increase productivity and economic performance while improving product quality and reducing the environmental impact.

Algorithms to safeguard crops

Smart farming is especially reliant on increasingly efficient image treatment technologies. Considerable progress has been made in the quality and resolution of these images, whether provided by satellites, drones, or cameras mounted on farming machinery or on the ground. Today digital imaging, now infra-red, ultraviolet or hyperspectral, allows farmers to obtain extremely detailed data about their crops and monitor them with far greater accuracy.

But the factor making these technologies truly intelligent is their association with machine learning and deep learning algorithms, which enable the enormous analytical and predictive potential of the images collected to be exploited.

Computer vision in action in the agricultural sector

According to Research & Markets, the artificial intelligence market in the agricultural sector will be worth over $2.6 billion by 2025, an annual growth rate of 22.5 %. Combined with computer vision and visual inspection techniques, machine learning is revolutionizing modern farming through its unprecedented potential for detection, classification, self-learning and automation. Demonstration:

Phenotype, for more resistant crops

In their search for productivity and quality, farmers have always practised plant breeding or crossing to obtain plants which are more resistant to disease, less demanding in resources, more nutritive, bigger, smaller, redder or greener.

To accelerate and improve this process, agronomists and researchers like those at  Limagrain use machine learning to select the best seeds and ensure the quality of future harvests. They employ increasingly efficient mathematical models to automatically record, measure and classify plant phenotypical properties (physiological and morphological characteristics, colours, etc.) from millions of photos.

Early detection of parasites, diseases, etc., to reduce the use of aggressive treatments

Early detection of crop contamination is also a key challenge for farmers. The traditional method involved examining the fields manually. This is obviously open to errors or belated identification which can affect the entire crop.

Digital visual inspection and machine learning technologies enable these contaminations to be identified far more rapidly and effectively than can the human eye alone, and enable the implementation of a more responsible and less aggressive treatment strategy.

For example, NatureSweet, an American tomato grower, called upon Prospera to monitor its greenhouses. A dozen cameras photograph the tomato plans continuously, using Prospera's artificial intelligence software to recognise any infestation.

Similarly, the European consortium Flourish Project has developed an extremely accurate crop monitoring and classification system which is able to detect each plant individually and identify areas contaminated by parasites or disease for localised field treatment.

Classification of products to optimise the marketing process

Even if ugly fruits and vegetables are currently popular, the aesthetic characteristics of agricultural produce remain a key issue for farmers supplying the mass retail market. Machine learning software produced by RSIP Vision or Aweta are increasingly used to detect the aesthetic properties of fruit, vegetables and even flowers (size, weight, shape, colour, texture, etc.) to eliminate defective products and select only those meeting the established standards.

These product classification and recognition systems also allow the selection, particularly according to their degree of ripeness, of products for specific markets (local or export).

Monitoring of cattle herds while minimising human intervention

Smart farming also addresses cattle breeding, where practices sometimes accused of ignoring animal welfare are often in the news. RSIP Vision  has addressed the issue and developed a dairy cow health monitoring system which enables each animal’s fat ratio to be measured non-invasively from a photo of its back, and to adjust its feeding regime accordingly, so that it is in the best possible condition to produce the best milk, while Cainthus, has developed a facial recognition package based on machine vision, driven by artificial intelligence. The system allows each animal in a large herd to be counted and simplifies monitoring by detecting sick or injured animals, while minimising direct contacts between the animal and humans, which is now recognised to affect the well-being of cattle.

Today machine learning and deep learning, associated with computer vision technologies, are bringing profound changes in farming practices by improving efficiency, performance and long term viability. One of the key benefits of machine learning in the agricultural sector is its ability to automate tasks which for the human operator, however well-trained, are time-consuming, laborious and subject to errors. This opens up enormous perspectives for progress and innovation which could represent a key turning point in the agricultural sector’s capacity to feed the growing world population.

Originally published on July 23, 2018 Topics: Machine Learning Deep Learning GeoSpatial Computer vision


You may also like: