Almost two-thirds (65.2%) of Japanese farmers are over 65 years old. Japan’s rapidly aging society and the steady migration of young citizens to the cities pose a significant threat to the country’s agricultural traditions and the industry’s survival. To address the problem farmers are turning to “smart agriculture” labour-saving techniques and devices and large-scale production. Artificial intelligence is playing an important role in accelerating and improving the technologies driving Japan’s smart agriculture.
The Internet of Things and AI can provide farmers with real time analysis of weather, temperature, humidity, and price data. They can suggest ways for example to maximize the value of crops by fully understanding local resources. Data collected by field sensors and unmanned aerial vehicles can provide massive and valuable information about soil, seeds, livestock, crops, costs, farm equipment, water and fertilizers. In this article we look at the development and application of AI in Japanese agriculture, especially in automatic picking, crop classification and agricultural monitoring.
Automatic picking is an important technology in agricultural production. In the last century, automatic picking techniques have been mainly used in equipment such as rice harvesters. Huge harvesting equipment however requires manual manipulation and is brutish, unable to harvest fragile and vulnerable agricultural products such as strawberries, tomatoes, or various fruits. One of the challenges in harvesting robot development is to have the system judge when and which tomatoes are ready to be harvested. Existing general picking robots cannot make the subtle colour differentiations required to distinguish between unripened and ripened. By adding image recognition technology and deep learning, a robot’s ripeness assessment performance can approach human level.
Japanese electronics giant Panasonic showcased its “Tomato Harvesting Robot” at the 2017 International Robot Exhibition in Tokyo. The company says that adding AI to the robot has increased its harvest efficiency rate from 80 percent to 96 percent. As tomatoes gradually mature from green to red, the robot’s cameras record images, and the system checks the changing color against samples created by the plantation to make its harvest judgments. It can also deftly pick a tomato without bruising the surface. The Tomato Harvesting Robot has improved quality and uniformity, and also optimized its sample settings and improved traditional methods.
Although the current picking rate of one tomato per six seconds is only about half the expert human picker rate, the Tomato Harvesting Robot can run nonstop for 10 hours and work at night (using a dedicated light source in the colour analysis). This gives it a clear edge over the farmer who can only manage daytime shifts of three to four hours before taking a break.
Crop classification is an essential process, as different grades of agricultural products are sold at different prices. Also, some products don’t make the grade for direct sale in supermarkets and are instead shipped to factories for processing into canned foods, preserved products, etc. A problem in the crop classification process is setting clear criteria. Classification results can thus largely depend on the agricultural product producers.
Makoto Koike is a cucumber farmer in the western city of Shizuoka. He says it takes more than eight hours to sort 4,000 kilograms of cucumbers during the harvest season. A former IT engineer, Koike thought there must be a better way, and so boldly built his own automated cucumber sorting system using TensorFlow. His cucumbers are classified into nine grades according to thickness, length, curvature, and so on.
Koike’s cucumbers are first photographed by a trio of cameras. This data is then transmitted to an AI system trained on thousands of cucumber images, which determines the grade using deep learning. Finally, a conveyor belt transports the cucumbers to a robot arm which places each in its appropriate packing crate.
Farm monitoring technology includes production forecasting, monitoring of pests and diseases, weather forecasting, etc. In smart agriculture, real-time monitoring and control of farm production is done through drones, surveillance cameras and other sensors installed on the farm. AI can leverage the data to detect temperature, humidity, crop disease and pest status, while also intelligently analyzing crop growth in real time. This is useful for example for rapid response pest control to reduce damage caused by pests and diseases. These automation technologies can greatly reduce time and toil, an important consideration for aging agricultural workers.
This crop cultivation solution developed by Tokyo-based Optim Cloud IoT company is divided into air monitoring with management (Agri Manager) and a crop record assistant (Agri Assistant). Through aerial photography and sensors, Agri Manager can remotely confirm temperature, sunshine volume and other data. Using image processing technology, it can discover and predict 27 different pests and diseases. The Agri Assistant uses voice recognition technology to upload comprehensive crop information that will reduce farmers’ workloads and increase planting efficiency.
Japan’s aging population and steady rural depopulation are producing a number of measures designed to shore up the country’s agricultural sector, such as improving the efficiency and reducing physical hardship for the existing labor force. The large-scale application of artificial intelligence in agriculture can accelerate this process and could also motivate more young people to work in the industry as it sheds its “difficult and dirty” image.
Agriculture will change dramatically in the coming years, and AI may entirely reinvent it in Japan.
Analyst: Yuu Rirou | Editor: Michael Sarazen