Robots and strawberries: how AI increases field yields

The world's population is growing rapidly, and according to UN forecasts, by 2030 it will reach 8.5 billion people. World Bank analysts believe that by 2050 we will need to increase the amount of food by 50 percent to support the growing population of the planet, and climate change will reduce the yield by 25 percent outdoors. But the areas best suited for growing cultivated plants are already being processed. Finding new places is difficult, and achieving a significant increase in yield is even more difficult.



This problem must be solved with the help of new technologies. And here the most promising direction seems to be the use of neural networks and artificial intelligence to create agricultural robots and crop control systems.



Why precisely neural networks? They are best suited for solving applied problems. We will not describe the technical details of their operation, we better describe the advantages. The neural network is not programmed in the classic sense of this process. It “learns” by finding patterns in the loaded data and is able to use them in further work.



Like a person, a neural network can quickly recognize images of photos and videos, can predict and make decisions. At the same time, artificial neural networks operate with large amounts of data faster and more efficiently than humans. What you need to optimize farmland, where areas are measured in hundreds of hectares, personnel - in thousands of employees, and livestock - in millions. Yes, the number of sheep in the country - this is Big Data. Primary information for training is enough for almost any company in the industry. The main thing is to assemble it in a format that is understandable for learning and integrate into workflows.

')

The quality and quantity of the crop, the increase in livestock numbers, depend on many factors. Analyze them all to make the right decision, the person is not able, no matter how experienced he is. So the need for modern technologists is obvious. Especially since there are already a large number of successful developments that help farmers to harvest, monitor livestock and build forecasts. We will tell about the most interesting projects with robots and AI in agriculture.

Rural Robots

Let's start with the robots. They are large and small, there are even duck robots .



Agrobot







The Spanish company Agrobot has proposed a robot for the automatic collection of delicate strawberries. The device is completely autonomous and can navigate in space. Robotic arms (there can be up to 24 of them) work independently, removing one berry from the bush. To assess the maturity of the berries, the robot uses artificial intelligence technology. Sensors analyze the berries, and graphic processors evaluate the color of the fruit and its presentation, and the data on each fruit is recorded in the database.



For three days, Agrobot is able to pick strawberries from 800 acres. After each collected row, it stops and sends information to the operator. The machine quickly copes with the task and is suitable for different farmer sites. The first successful test strawberry robot held on a farm Driscoll in California.



Dogtooth Technologies







Competitor Spanish robot, created in the UK. The device is designed to collect soft fruits. It is able to autonomously move through the rows of crops, to find and collect ripe fruits, sort the collected berries and pack them. After collecting the berries, the video camera examines the fruit from all sides in order to determine the grade, shape, measure the mass, detect defects (dents, mold, etc.). Rejected fruits are placed in trash containers.



Sorting of fruits and their packaging occurs on the spot, so additional costs for the work of sorters are excluded, and the product arrives on the shelves faster. Orientation in space takes place using high-precision GPS coordinates.



The robot has some interesting features. For example, the British are used to buying ripe strawberries with a small portion of the stem in the retail chains. The machine takes this feature into account when picking a berry with a small part of the stem.



Vegebot







Vegebot is a working prototype of the Iceberg robotic lettuce picker, created by engineers from the University of Cambridge. The device can independently recognize the intact cabbages of lettuce, as well as carefully process and assemble them. Read more about the robot told denis-19 in a recent article on Habré.



Wall-ye vin







The brainchild of the Burgundian inventor Christophe Milloth (France) works hard in the vineyards. A device with four wheels, two hands and six cameras weighs 20 kilograms, selects the path automatically and uses artificial intelligence to determine what to do at the moment. A day can prune up to 600 vines.



Wall-Ye VIN is engaged not only in pruning and pasynkovaniem, but also accumulates important data on the condition and vitality of the soil, fruits and vines. He moves from vine to vine, reveals certain features of the plant, photographs and records data from six cameras, marking each vine, after which its manipulators are included in the work.



The hand with shears is not only designed to cut branches, with the help of it, it can protect against thieves. A gyro is built into the device, and if it is lifted from the ground, it will be defended with a secateur, will erase all data from the hard disk and send a signal to the owner. In addition, the built-in GPS receiver will not allow him to go beyond the working area.



Unnamed apple robot by Abundant Robotics







California robot, which still has no name despite the impressive investment from the GV (formerly Google Ventures), was created to collect apples. The device moves through the rows between the apple trees with the help of a lidar that draws the world with lasers and depicts fruit using machine vision.



Operators can adapt it for a particular apple variety, after consulting with a farmer who knows from experience which color corresponds to the mature one. After recognizing the degree of maturity of apples in real time, the robot sucks the fruit from the tree with a vacuum tube, sending it down the conveyor to the basket. A robot can pick apples 24 hours a day, skipping un-ripe fruits to return to them later, as a collector would have done.



ecoRobotix







Swiss ecoRobotix is a robot designed for automatic thinning of weeds and weeding. The idea has been in the air for a long time. A neural network can be taught to distinguish beneficial crops from weeds. Having “examined” several million photographs of healthy and diseased plants at different stages of growth at the start, the system can determine whether a healthy shoot or a weed is in front of it with a video camera in a few milliseconds. It will also be able to assess the degree of threat to the crop and suggest ways to solve the problem if there are visible signs of crop contamination.



ecoRobotix is ​​equipped with a computer vision system designed to identify weeds. Orientation in space takes place using GPS and touch sensors. Able to process about 3 hectares of crops per day. When detour "possessions" in case of need sprays a weed with a small dose of herbicide. This approach reduces the use of chemicals by 2-3 times.



Weeds are generally a sore subject for farmers, so there are other projects in this area. For example, an Indian smart garden sprayer using an ultrasonic sensor system determines the size of the tree and the distance to it. The obtained information is analyzed and affect the jet power and the amount of the sprayed substance. Testing showed high efficiency of the system, while reducing to 26% consumption.



And Bayer and Bosh are developing smart spraying technologies. It will differ from systems on the market due to its ability to distinguish weeds from crops. It is assumed that the system will “recognize” the weed and determine the type and required amount of pesticide, taking into account the programmed application parameters.



Similar technologies are used by IBM. And successfully used! For one of its clients in Southeast Asia, the company was able to predict crop stress in the region due to pest / disease contamination. Then it took the ground crew a few hours to just get to this place.

Smart Systems

AI systems also benefit farmers. The range of their use is slightly wider than that of robotic devices, however, the tasks are often different. Although there are intersection points.



Sonoma







Yes, this is not a robot, but technology. However, also worthy of attention. Sonoma, owned by Microsoft, won the Autonomous Greenhouse Challenge greenhouse experiment, which took place in the Netherlands from August 27 to December 7. 5 IT giants have figured out how machine learning technologies can cope with growing plants and how realistic it is to use these technologies in “traditional” gardening.



Automated harvesting systems have been around for quite some time. However, in the experiment it was about the complete control of AI over production. Technology team Sonoma allowed to grow 50 kg of cucumbers per square meter. The neural network controlled irrigation, gas composition, feeding, temperature and other aspects that affect the growth of cucumbers.



Tencent's iGrow team and the Chinese Academy of Agricultural Sciences ranked second. Intel's Deep Green team ranked last.



Taranis







Israeli startup Taranis allows you to monitor the condition of the plants, timely identify negative factors and eliminate them. For monitoring, readings from field sensors, meteorological data, and aerial photography are used. The analysis uses images with ultra-high resolution (up to 8 cm per pixel) of Mavrx.



The study of large amounts of data allows you to localize crop areas with oppressed growth, identify plant diseases, pest problems, determine nutrient availability of plants, potential yields, etc. The system not only offers solutions to detected problems, but also determines their optimal timing based on meteorological forecast carrying out.



Watson







The IBM Watson Decision Platform for Agriculture advises farmers in the processing of remote sensing data. Using AI to combine data from multiple satellites, the IBM solution is able to detect inefficient areas of culture with almost the same accuracy as the terrestrial sensors of the Internet of Things (IoT). Watson from IBM will determine for the farmer the type, quantity and optimal timing for pesticide treatment of the affected areas.



Will help in carrying out preventive treatment. Using the high resolution plant activity index (HD-NDVI), assess the condition of the plant, determine the necessary preventive measures (fertilization, nutrients, etc.). By combining humidity data (HD-SM) with terrain data and meteorological measurements, the dynamics of changes in soil moisture are simulated. The farmer also receives a forecast of yields, the dynamics of changes in yields on the basis of images and information from past seasons, etc.



Health Change Maps and Notifications







Developed by Farmers Edge, the AI ​​platform, Health Change Maps and Notifications, informs the farmer about the efficiency of the equipment, the state of the plants, the appearance of pests or diseases, nutritional deficiencies, etc. The program processes satellite images and sends the user messages about possible risks and necessary measures.



ET Agricultural Brain







Alibaba's Swinish AI project allows you to detect a swine pregnancy, which allows farmers to determine the date of farrowing and prepare for the subsequent process of pregnancy and litter birth of healthy piglets. The system deploys intelligent surveillance cameras in barns, and machine learning algorithms produce results based on monitoring sows' sleep, standing position and feeding conditions. For example, a sow is likely to be pregnant if she sleeps on her back, stands still and runs a little and consumes a constant amount of food. Alibaba engineers are also planning to add a prediction of litter quantity based on the characteristics of the figure of a pregnant pig.







The system uses computer vision techniques to set up profiles for each pig — documenting their breed, age, weight, nutritional conditions, intensity and frequency of exercise, and movement trajectories. Meanwhile, voice recognition algorithms are used to monitor the health of piglets and protect against asphyxiation, which reduces the death rate by three percent and increases the annual production level by three pigs per sow.



The financial division of another large Chinese holding, JD.com, also aimed at cattle breeding. Last month, the company introduced a set of AI-based agricultural solutions.



Cainthus







Continuing the theme of animal husbandry, we will tell about a very interesting Irish project Cainthus, which journalists dubbed “Facebook for cows”. Identification of cows on the individual features of the muzzle allows you to collect a variety of information about each animal, ranging from the characteristics of their behavior, ending with appetite. The data can be used by farm owners to monitor the health of dairy cows and increase their milk yield.



The company offers farmers to improve the comfort of the cows throughout the entire life cycle, keeping track of their individual needs and immediately signaling the need for intervention if something is wrong with the animal. For observation, a computer vision system is used.



According to the developers, the platform is extremely relevant and in demand. They also claim that she is one of a kind. However, this is not the case.



Cattle care







Another interesting project with Russian roots, Cattle Care , offers similar functionality. The developers have created a video monitoring system for cow health and productivity based on computer vision. Video analytics for dairy farms allows for the most comfortable conditions for each female.



The principle of operation is quite simple. The pattern on the cow's skin is unique, like human fingerprints. Using this feature, the system trained in the photographs of the wards collects information from video cameras installed on farms, detects and identifies each specific cow. By counting the number of steps, chewing movements, the amount of feed consumed, the water consumed and other behavioral patterns, the computer draws up a medical record of each cow. Thanks to Cattle Care, the farmer immediately sees if his ward has something wrong.

Conclusion

As can be seen, artificial intelligence and robots are fully capable of increasing the efficiency of agriculture and simplifying the work of farmers. However, can these technologies solve the potential threat of food shortages? I wonder your opinion.



What else is useful you can read in the blog Cloud4Y



→ VNIITE planet of the whole: how in the USSR the “smart home” system was invented

→ 5 open-source security event management systems

→ How neurointerfaces help humanity

→ Cyber ​​insurance on the Russian market

→ Light, camera ... cloud: how clouds change the movie industry



Subscribe to our Telegram channel to not miss another article! We write no more than twice a week and only in the case.